Methods

Study design

This study is a descriptive systematic review investigating risk criteria in HSS. The 27-item Preferred Reporting Items for Systematic Review and Meta-analysis statement (PRISMA) 2009 checklist is used as a reference¹². The study protocol was approved by the Higher Education Council of School of Public Health at Tehran University of Medical Sciences.

Search methods for identification of studies

In the present study, “hospital “refers to the legal institution that provides 24-h medical services, including accepting, visiting, admission, and treatment of injured and/or sick individuals^13,14. “Site selection” refers to an operational problem solving research in which the researchers will find out the best location that meets the assigned preferences¹⁵. “Risk” refers to both the probability of an event (hazard) and its impacts on the exposed community (vulnerability)¹⁶.

Initially, four electronic databases (MEDLINE through PubMed, Scopus, Science Direct and Google Scholar) were searched up to December 31; 2015.The search strategy was based on the PubMed database model. The key terms were adopted from Medical Subject Heading (MeSH) when possible; otherwise appropriate key words were selected according to the expert idea. The expert team is including the authors’ team and 3 more volunteer PhD students at Health in Emergencies and Disasters from Tehran University of Medical Sciences with previous field experiences in disaster medicine. Titles and abstracts were searched with the following syntax:

Study eligibility

The inclusion criteria were articles that were published in peer-reviewed journals and had site selection criteria for a hospital. Gray literature including conference proceeding papers and thesis were included as well. All non-English articles were excluded.

Data collection and analysis

The key terms were searched in the databases separately and all articles were imported to a bibliographic management program (EndNote X3). Duplications were then omitted. The titles and abstracts were evaluated by two authors (MJM and BR) independently; in the cases that the exclusion criteria could not be applied, the full article was reviewed. If there was disagreement about the eligibility of a particular article between MJM and BR, the third author (AA) was asked to adjudicate. The electronic search was conducted from November 2015 to December 2015.

In the next step, the data were extracted out of the full-text of the included articles. This data included the first author’s name, the year of publication, and the first author’s country, the geographical scope, site selection criteria and the method of study. The references of selected articles was hand searched. Finally, the extracted site selection criteria were categorized through a focus group discussion by the expert team which has been described above already. The final results were sent to the expert team for confirmation through e-mails.

Quality assessment

A 9-question checklist was produced to assess the quality of the retrieved publications by authors (Table 3). The Quality-related questions investigated the following components: number of applied criteria for HSS, categorization of the applied criteria (Yes or No), source of applied criteria (i.e. the authors themselves, experts idea or review of literature) , GIS based method (Yes or No), explanation of why the analysis method was used (Yes or No), description of the candidate regions for HSS (Yes or No), number of candidate regions for HSS (i.e. the number of regions or a countless sites), discussion about the limitation (Yes or No), and discussion about the generalizability of the study (Yes or No).

The checklist was filled out by two assessors independently (MJM and BR) and “AA” adjudicated when there was disagreement.

Results

In the first step 41 studies were retrieved through the bibliographic search. After removing seven irrelevant and nine duplicates, 25 were remained. Sifting process left 15 eligible studies that were published up to the end of December 2015, 2 of them were identified from hand searching of references of included articles. Figure 1 illustrates the related PRISMA flow diagram. The result of quality assessment is summarized in Supporting Information file, S1 Table.

Table 1 summarizes the main results of the present study. The first article was published in 1995¹⁷. The maximum number of articles were published in 2013, i.e., 6 articles. The total number of authors of the included articles was 38 (2.7 authors per article, in average). The first/corresponding authors were affiliated with different universities; among these, only K.N.Toosi University of Technology (in Iran) was associated with two publications^2,18.

HSS studies performed in China were 2, Iran (5), USA (2), Taiwan (2), Tunisia, South Africa, Israel and Bangladesh (1). Regarding the geographical scope, three studies were performed at the provincial or state level, two at the country level, five at the city level, and five at a district of a city.

With regard to the methodology, 10 articles used Geographical Information System (GIS), 5 studies applied AHP, and one used fuzzy AHP for site selection¹⁹. One study combined the AHP with the Ordinary Least Squares (OLS) method¹⁸ while another one combined AHP with the Rank Order Method (ROM)²⁰. Other models that were used included mathematical modeling, Belief-Desire-Intention (BDI)², Technique for Order of Preference by Similarity to Ideal Solution TOPSIS¹, goal programming¹¹, fuzzy Analytical Network Process (ANP)⁷, and travel-time methodology²¹.

Through a focus group discussion, with the above described expert team, all 39 HSS criteria were classified into six main groups namely cost, demand, environmental, administrative, disaster risk, and other concerns. The application percentage of each of these groups was 100, 93.3, 53.3, 33.3, 20.0, and 13.3 respectively (Figure 2).

The six mentioned groups were subsequently divided into 16 subgroups (Table 2).

Discussion

This systematic review was conducted to find out the disaster risk related criteria for HSS. According to the results, despite the increasing trend of worldwide disasters, risk related criteria are not taken into consideration to the extent that they should.

It is not mandatory and also not realistic to apply all these factors for HSS. Depending on the strategy for building medical centers, the planners may consider only some of these. For example, Kim et al (2013) conducted HSS in the construction of a hospital for the aging population and considered factors that looked at the real health requirements of the target group; thus, some criteria such as environmental issues (air and sound pollution and sewerage system) were not considered²². Soltani et al (2011) used several criteria such as urban planning, traffic volume, and travel time for site selection of a hospital in district five of Shiraz, Iran; however they did not consider environmental and land specification issues in their study⁷. As the main goal of Wu et al (2007) was ensuring competitive advantage for HSS, they considered administrative criteria such as regulations, policymaker’s attitude, and even demand for the hospital personnel. They did not consider environmental concerns and accessibility to infrastructures such as main roads, as well²³.

Alavi et al. (2013) considered two groups of factors that affect accessibility (roads and social services) and also distance from potential hazards (faults and industrial centers) for one region of the capital city of Tehran, Iran. However they did not use future planning development, real demand for hospital in the region, existing health services, and other land specifications¹. Also Jing-Er Chiu and Hang-Hao Tsai (2013) used MCDM to determine the optimal location for expansion of a regional teaching hospital in Yunlin County, Taiwan. They used highly detailed criteria including the demand for medical service, cost, transportation, sector support, and future development. The main theme considered in their study was increasing the competitive advantage for the hospital²⁴.

In 2014, Xiao-Hua Hu et al developed a model to identify the proper location for medical and health services in a large group of islands in Hainan Province in China. This model was based on minimizing the travel distance for clients of these services. In this study, the real demand and environmental issues were not considered²⁵.

To achieve sustainable development, a community should consider important issues in building and utilizing a new hospital. In HSS procedures, environmental considerations are important issues^26,27,28. While important, the majority of publications devoted more attention to cost and demand rather than environmental issues such as air and noise pollution as these have sever negative effects on hospital functions. In addition, following the completion of the hospital, these types of pollution would be aggravated.

Regarding the cost concerns, most of the studies devoted attention to accessibility by main roads and arteries. Other cost subgroups were proximity to infrastructures, and land specifications (availability, use, texture of the ground, being vacant, and ownership). However, none of the articles discussed the beneficial aspect of the proximity of the hospital location to airports or seaports.

Concerning the demand category, health service utilization of the community was assessed according to the total number of required beds, the patient transfer rate, and the number of patients rejected by hospitals^11,18,22. In this category, epidemiological indices such as “burden of disease” and forecasting demand for health care based on demographic factors, economic growth in the area, and even new technologies in health system could be considered. For example, oil and energy activities are a great source of economic development but require a certain type of health care infrastructure when it comes to emergency medical services including trauma and cardiac patients.

As construction of a hospital is a kind of investment, the investors wait for future profit. Hence, threats to this investment should be taken into account. For more than 25 years, WHO has promoted and supported the efforts to the purpose of safe hospitals to improve the function of hospitals in emergencies and disasters. Besides, in Sendai Framework (2015-2030)¹⁰ the application of the principles of universal design and standardization of building materials in critical facilities such as hospitals is considered with the aim of disaster risk prevention and reduction. Unfortunately, few articles discussed hazards such as faults and industrial areas ^1,11,27. Other potential hazards, such as floods and man-made disasters, should be considered for at-risk areas. Considering the “risk”, rather than merely hazards, is highly recommended in future permanent and field hospital site selection studies.

The two unclassified criteria, namely unforeseen circumstances and other competitive hospitals were considered in the “Other” category. It is recommended that the safety and security of the candidate locations, rural versus urban areas, local investors, accessibility of communication systems, and availability of competent and qualified staff be considered in this category in further studies.

Conclusions

Despite the critical role of hospitals in health service delivery in disasters and emergencies and the effect of hospital location on the quality of these services, few articles have considered hazards as the criteria for hospital site selection (HSS). Cost and demand are two groups of criteria that have been addressed more frequently in HSS studies. The decision makers should prospectively match the main objectives of hospital building with the HSS criteria according to the strategy of site selection and the availability of data and resources. Undoubtedly, being safe and remaining functional in emergencies and disasters should be one of the main objectives in HSS in line with Sendai Framework. More comprehensive criteria like “risk of hazards” and “burden of diseases” are suggested to be considered in future studies.

Limitations

Non-English articles were not included in this study.

Supporting Information

Corresponding Author

Ali Ardalan, MD, PhD

E-mail: [email protected]

Tehran University of Medical Sciences, Tehran, Iran

Data Availability

All relevant data are in the article.

Competing Interests

The authors have declared that no competing interests exist.

Behavioural (likely public behaviour)

Traditionally, planning and policy for mass decontamination has focused on the technical aspects of mass decontamination, such as developing and testing decontamination equipment, with little consideration of likely public behaviour. Where likely public behaviour has been considered in planning for incidents involving mass decontamination, there has been a reliance on common myths about crowd behaviour, such as inherent public disorder and mass panic ². However, over 50 years of research has shown that panic occurs very rarely during mass emergencies and disasters, and that people are much more likely to behave in a helpful and cooperative way ^3,4,5,6,7. This reliance on assumptions about panic has resulted in little attempt to develop strategies for communicating with members of the public during decontamination; if people are going to behave in an irrational way then there won’t be any point in trying to communicate with them. Indeed, research suggests that assumptions about panic may lead to information being withheld from members of the public, which may actually create the very disorder ^8,9 and non-compliance ¹⁰ which authorities are hoping to prevent.

Evidence from small-scale incidents involving decontamination shows that those people who are affected may refuse to comply with decontamination procedures if they are not provided with sufficient information about why decontamination is necessary, and what the process involves ². Findings from large-scale field exercises and field trials involving mass decontamination suggest that, if sufficient information is provided and the people involved believe that responders are being open with them, members of the public are likely to be willing to comply with decontamination ^11,12, and also willing to help others to undergo decontamination ^12,13.

Overall, these research studies show that public behaviour is likely to be contingent on the way in which emergency responders manage the incident. Emergency responders who communicate effectively with members of the public and show respect for their needs change the relationships and that, consequently, results in more positive outcomes from the incident, in terms of reduced public anxiety and increased public compliance and cooperation. Thus, if managed well, members of the public can actually be an asset to emergency responders in their attempts to successfully manage the incident.

Communication and privacy

Effective communication is essential during mass decontamination – failure to communicate effectively can result in reduced public compliance and cooperation, increased confusion, and even attempts to challenge responders’ authority. Several research studies have examined how different perceptions of responder communication can affect public compliance and cooperation, as well as levels of public anxiety, during mass decontamination ^11,14,15.

A recent mass decontamination field experiment specifically tested three different responder communication strategies, in order to try to identify what makes some communication strategies more effective than others ¹². Results from this study revealed that, for a communication strategy to be effective during mass decontamination, it should include open and honest information about the nature of the incident, health-focused explanations about the importance of decontamination, and sufficient practical information to enable those people who are affected to successfully undergo decontamination. Crucially, the inclusion of an effective communication strategy not only resulted in more positive psychological outcomes for those affected, it also resulted in improved speed and efficiency of decontamination on objective measures ¹². Therefore, this could save lives during a real incident.

As well as the provision of effective responder communication, another key issue during mass decontamination will be whether members of the public feel they have sufficient privacy ^16,17,18. It is crucial that emergency responders show that they are doing what they can to respect public needs for privacy and modesty; failure to do so can result in reduced public compliance and cooperation ^{2,12,13,14,19}.

Several research studies have examined the mechanisms underlying the relationships between effective responder communication, sufficient privacy, and positive outcomes during mass decontamination ^12,14,20. Findings show that effective communication from emergency responders and the provision of sufficient privacy are crucial because they enhance public perceptions that responders are behaving in a legitimate way. Enhanced perceptions of responder legitimacy facilitate increased identification between members of the public and emergency responders, as well as amongst members of the public, and it is this identification which leads members of the public to actively engage with the decontamination process.

The findings from these research studies have been used to generate recommendations for emergency responders when managing incidents involving mass decontamination ¹⁵, and so far have been included in US decontamination guidance documents for emergency responders ^21,22,23.

Conclusion

This research programme can be used to inform best practice for managing casualties during mass decontamination. However, as noted in Cibulsky et al. (2016)¹, there is a need for further research in this area to optimise casualty management strategies for mass decontamination. Ongoing work related to the development of optimal management strategies for mass decontamination is being carried out as part of the NIHR Emergency Preparedness and Response Health Protection Research Unit (EPR HPRU) (https://epr.hpru.nihr.ac.uk/). This research includes a qualitative study of factors which affect perceived public acceptability of different decontamination methods, part of which involves testing the effect of different responder management strategies on perceived public acceptability of mass decontamination. The research also examines cultural factors which may affect public willingness to comply with different decontamination methods. Further research could include analysis of data from real incidents involving decontamination, as well as new field studies and exercises testing different responder management strategies. In particular, future research should consider practical aspects of communication, such as how best to deliver information to members of the public during the decontamination process, and also how best to protect casualties’ privacy and so promote compliance with the decontamination process.

Competing Interests

The authors have declared that no competing interests exist.

Correspondence

Holly Carter: [email protected]

Literature Review

Social Media and Crisis

The rise and proliferation of Web 2.0 applications revolutionized the way people interact and collaborate with each other. At the core of the Web 2.0 framework, are the social media, “a group of Internet-based applications that build on the ideological and technological foundations of Web 2.0, and that allow the creation and exchange of User Generated Content (UGC)”⁶. Social media nowadays includes collaborative projects, blogs and microblogs, content communities, social networking sites, virtual game worlds and virtual social worlds. Through these applications, users create online communities to share information, ideas, personal messages, and other content. This user-generated content in a changing and dynamic virtual community creates a massive and unstructured data⁶.

In modern society, the significance of the social media in everyday life has increased dramatically, turning the world into a “global village”⁷. Since the late-20th century, new methods of mass media including the internet, cellular technologies and personal portable devices, have been considered as fulfilling a significant role in agenda setting and framing⁸, creating social capital⁹, and affecting politics, society and culture¹⁰. According to a recent report that summarizes social, digital & mobile statistics among global users¹¹, 42% of world’s population are active internet users, and above one-quarter have active social media accounts. With regard to social networking platforms, the report shows that there are 1.36 billion Facebook users, 629 Million Qzone users (Chinese social network), 343 million Google+ users, 300 million Linkedin users and 271 million Twitter users. Twitter, a microblogging communication platform, for instance, reaches an estimated total of over 50 million tweets per day¹², and during crisis twitter activity increases exponentially¹³.

Social media has become an essential mean of communication during disasters and it has been suggested that best practices need to be established to assist organizations, government and decision makers in optimizing risk and crisis communication in this era ¹⁴. Houston et al.¹⁵ developed a framework for the practice and study of disaster social media. They note that disaster social media users in the framework include communities, government, individuals, organisations, and media outlets. Those players seek to gain insights from the massive user-generated content published in the communication channels, provided by the social media platforms. This published timely information is quickly spread across multiple social networks, and may affect, shape or direct public behavior without official involvement or even in contradiction to formal instructions ¹⁶. Although many challenges exist during social media message processing and analysis, (e.g. management of information overload, credibility issues, and prioritizing different classes of messages), state of the art computational methods exist to carry out complex information processing operations. ¹⁷

Crisis Communication

Crisis communication can be defined broadly as the collection, processing, and dissemination of information required addressing a crisis situation. Crisis communication is a research and practice field in various setting and crises; including national¹⁸, organizational crises^19,20 , natural²¹ , man-made²² and technological disasters²³. Effective communication with the public remains a crucial role in public safety through crisis preparation and response²⁴. The media are critical for facilitating pre-disaster preparedness, disseminating warning messages of pending disasters, providing information to citizens about the disaster, and facilitating recovery efforts²⁵. Moreover, according to Firestone & Everly³ , crisis communications can play an important role in mitigating, or exacerbating, the psychological and behavioral reactions to critical incidents and disasters. One important adjacent field of crisis communication is Risk Communication, defined as any purposeful exchange of information about health or environmental risks between interested parties (e.g. governments, agencies, media, citizens and more)²⁶. Reynolds and Seeger²⁷ prposed a merged and comprehensive approach called “crisis and emergency risk communication”.

Social Media as a crisis communication platform

Social media has been radically changing the communication landscape over the past several years and as a result, crisis communication is undergoing substantial change²⁸. It becomes clear that social media serves as ultimate space for crisis communication processes. Implementation of traditional crisis communication activities, was identified as one of the functions of disaster social media¹⁵. A recent survey²⁹ of about 288 government officials in the US revealed that 71% of them were using social media during crisis, with Facebook as the popular social medium. Officials used social media during various types of crises, including public health³⁰, natural disasters, transportational, political, social and criminal crises. Researchers found that the degree of social media use, rather than the number of tools used, was positively correlated with local city officials’ evaluations of their ability to control a crisis situation and the strength of their responses²⁹.

Another research team³¹ systematically investigated crisis messages collected from Twitter. Relevant tweets for 26 different crisis situations that took place in 2012 were sampled from the available public twitter stream, and for each situation types of information and sources of that information were examined. During crisis situation, eyewitnesses, government, NGOs, business, media & news organizations and outsiders participate and publish different information types in social media ³¹. Types of information include data about affected individuals (e.g. people trapped, casualties, people missing, found or seen), infrastructure & utilities (e.g. damages, reports about environment, and availability of services), donations & volunteer (e.g. donations of money, goods of services; requesting help; shelter needed, food shortage/distribution; volunteer information); caution & advice (e.g. warnings, preparation; caution & advice; tips; safety), sympathy & emotional support (e.g. concerns and condolences; gratitude, prayers; support; emotion-related info) and other useful information (e.g. flood level; weather, wind, visibility; information verification). Based on the available data of that study (https://crisislex.org/tweet-collections.html), it seems that the Government, being the official source of information during crisis, was responsible for less than 5% of all published information. Of the applicable information types, the majority (32%) of the communication efforts by the government were about useful information (32%) and caution and advice (27%). Only 6% of the published tweets concerned about sympathy and support.

Effective Crisis Communication.

Seeger³² demonstrated ten best practices of crisis communication, drawn from a literature review and verified by expert crisis communication panel that reached high consensus. Seeger³² divides the recommendations into three broad categories: strategic planning, proactive strategies and strategic response. These are used as principles or processes that underlie an effective crisis communication plan and an effective crisis response.

Covello, McCallum, & Pavlova³³ note that detailed, in-depth knowledge and understanding of both the characteristics of target audience and the community in which the target audience resides, are needed for effective message development. Target audience characteristics include knowledge, attitudes, perception, behavior, beliefs, values, needs and concerns, while community characteristics include information about social networks, opinion leaders and community dynamics. Covello³⁴ suggested several risk communication templates to utilize in the process of risk communication: 1) CCO (Compassion, Conviction, Optimism) template, which is particularly useful when responding to a question indicating a high level of emotion or outrage⁴. 2) Primacy/Recency template³⁵ emphasizes the first and last messages in the communication due to the restriction of recall of information. 3) 27/9/3 template states that the combination of the three key messages should equal a total of 27 words, 9 seconds spoken aloud, and 3 key messages³⁴. 4) AGL-4 (Average Grade Level minus 4)⁴ template recommends phrasing the message at four reading grade levels below that of the stakeholder – taking into account the national reading grade level. 5) 1N=3P (Negative equals 3 positives) template counterattacks the weight of negative messages by introducing three positive messages for every negative message. 5) TBC (Trust, Benefits, Control) template suggests to phrase three messages that are phrased with specific order and content (trust first, benefits second and control last). These templates, however, were not empirically tested and thus limit our certainty regarding effective messaging.

The media plays a crucial role in shaping public response to terrorism and other disasters²⁵. Effective crisis communication suggests the need of an unhindered but purposeful exchange of information within and between authorities, organizations, media, involved individuals, and groups before, during, and after a crisis³⁶. Covello et al.³³ also argue that risk communication is no longer a neglected topic within government. With the changing media environment and the developing online atmosphere, traditional media are shifting to practices that are more adapt to a social media environment³⁷. Furthermore, to improve services and communication with the population, government officials seek to leverage these new media channels. Nevertheless, Graham, Avery & Park²⁹ emphasize that government’s engagement through social media should be more active and reflect a clear response priority in crisis communication plans. Ambiguous or unreliable communications can cause damage and serve to exacerbate publics’ mental health reactions and a delay in operational response and recovery³⁸To conclude, it is important to incorporate the use of citizen generated content into any crisis plan and learn to respond to the media and public even quicker than before³⁹

Models and theories of Crisis Communication. Over the years several models of crisis communication were developed⁴⁰, such as the Situational Crisis Communication Theory (SCCT⁴¹). SCCT “predicts the reputational threat presented by a crisis prescribes crisis response strategies designed to protect reputational assets”⁴¹, especially in organizational context.

Chaos Theory⁴² was described as a general framework for understanding crisis communication⁴³. CT argues that chaos or disorder may be the necessary precursor of a higher level of order. CT functions best at the broad level of a paradigm for understanding the behavior of complex systems. Seeger⁴³ adds that in the context of CT, small variance in communication processes, message phrasing, distribution, timing or other factors may produce extensive fluctuations in systems, leading to bifurcation.

Reynolds & Seeger²⁷ presented the Crisis and Emergency Risk Communication (CERC) five-stage integrative model. The model blends crisis and risk communication together into a processes model (see²⁷), assuming that crises will develop and progress in a predictable and systematic ways. According to the working model of CERC, communication processes occur during five stages: 1) Pre-crisis, 2) Initial Event, 3) Maintenance, 4) Resolution and 5) Evaluation. In each stage, communication can have different aims, strategies and target audiences.

With the advancing new media, the social media aspect was integrated in these models. For instance, the Social Media Audience Sharing Model (SMA)²²⁴ aims to increase the reach for messages disseminated via a given social media platform in the context of an emergency. Social mediated crisis communication model (SCCM)⁴⁴ is another model that serves as a framework for crisis communication management in the changing media landscape and explains how the source and form of crisis information affect organizations response options and provides recommended social-mediated crisis response strategies. The Networked crisis communication model⁴⁵ examines the influence of communication strategy and media type on damage to reputation, as well as secondary crisis communication and secondary crisis reactions. Although these models recognize new players in the crisis communication processes (e.g. influential social media creators, follower, and the growing empirical evidence that emphasizes the psychological functions of social media during emergency management – during most of the time, less or no consideration is given to psychosocial aspects during crisis communication. The need to focus on other participants (such as the public) rather than the organization or the responding authority, is echoed in Liu & Fraustino⁴⁶ who suggests that scholars should move beyond predominantly focusing on image management, emphasized by dominant crisis communication theories

Accommodated Communication

Considerable research, using text derived from social interactions, such as natural conversations and social media conversations, suggests that individuals tend to converge in various dimensions such as posture, pause length, utterance length, self-disclosure, head nodding, backchannels and linguistic style⁴⁷. Niederhoffer & Pennebaker⁴⁸ studied psychometric properties of language in dyadic interactions and assessed the degree to which people coordinate their word use in natural conversations, derived from internet chat and laboratory conversations. Based on a text-analysis software, they found that individuals in dyadic interactions exhibited linguistic style matching (LSM) on both the conversation level as well as on a turn-by-turn level. LSM found in research to be a predictor of social dynamics in small online and face-to-face groups⁴⁹. LSM, also found to be a robust marker for romantic relationship stability, stressing the importance of similarity in the way people converse with each other in the context of interpersonal processes⁵⁰. Lord, Sheng, Imel, Baer & Atkins⁵¹ saw that’s language style synchrony between client and therapist was predictive of empathy ratings during evidence-based behavioral treatments like motivational interviewing (MI). Therefore, it is apparent that synchronized verbal behavior holds an important role in various interpersonal interactions.

LSM, in fact is a derivative of Communication (or speech) Accommodation Theory (CAT), an intergroup theory of interpersonal communication, provides a framework for understanding how and why people adapt their communication toward and away from others and the social consequences of doing so. The theory holds that people tend to preform accommodative and non-accommodative moves, to reduce or increase significant social distances between the speakers, depending on the different circumstances. Convergence and divergence, called in the CAT approximation strategies, include the adaptation (or the alteration) of communicative behaviors in terms of wide range of linguistic-prosodic-nonverbal features. The theory also proposes different patterns of accommodation (upward vs. downward, and symmetrical or asymmetrical), based on the reciprocity of the speakers⁵². During the last years, examination of CAT progressed to the domain of electronic communication, such as e-mail, text messages, voice mail and recently also was adapted to electronic communications and the social media sphere. Danescu-Niculescu-Mizil et al.⁴⁷ examined and verified the hypothesis of accommodated communication in the context of twitter conversations. The researchers developed a probabilistic framework that enabled to model accommodation and measure its effects. They focused on linguistic style feature, derived from the Linguistic Inquiry and Word Count⁵³.

To the best of our knowledge, CAT was never integrated in any crisis communication models published in the academic literature. The closest domain LSM, was mainly researched in the context of crisis negotiations of suicidal and surrender outcomes, and hostage taking negotiation outcomes^54,55,56. In these studies, the researchers analyzed the correlations between the linguistic behavior of each communicator across 18 linguistic dimensions, like word count, prepositions, negations, emotionally toned words and others. LSM was assessed by the strength of the correlation and averaged across all conversations taken into account. The studies showed that negotiators (police officers and hostage takers) tended to show greater levels of linguistic style matching in successful negotiations compared to unsuccessful negotiations⁵⁷, and an overall consistency in the linguistic behavior of subjects and police negotiators in surrender and suicide incidents⁵⁴.

Interestingly, linguistic matching or accommodated communication is not emphasized in the processes models of crisis and risk communication, and possibly is taken for granted. Even when a certain aspect that is related to crisis communication with the public is presented, such as crisis communication strategies from the Situational Crisis Communication Theory⁴¹, linguistic features and linguistic accommodation are not explicitly mentioned as being part of the processes. Some research considered other aspects like number of words, number of sentences and words per sentence⁵⁸. In other words, when a crisis communication model, like the networked crisis communication model⁴⁵ and social-mediated crisis communication model⁵⁹, mentions elements like message form or message strategy, it ignores the linguistic characteristics of the message. The best practices in crisis communication³² introduce general principles of crisis communication, such as strategic planning, proactive strategies & strategic response, and each include sub categories that can be directly related to public communication, e.g “be open and honest”, “communicate compassion” and “provide self-efficacy”. Since speakers’ verbal style also influences how messages are perceived⁶⁰, we suggest that even if crisis managers will act according to these principles, they should employ certain linguistic styles and consider delivering messages in public accommodated language

Social Computing

“Computational social science is an emerging research area at the intersection of computer science, statistics, and the social sciences, in which novel computational methods are used to answer questions about society”⁶¹ (p. 257). Ericson⁶² argued that social computing refers to “systems that support the gathering, processing and dissemination of information that is distributed across social collectives. Furthermore, the information in question is not independent of people, but rather is significant precisely because it linked to people, who are in turn associated with other people.” Scholars⁶³ stated that social computing represents a new research frontier for information systems. Social computing environments present settings for data collection on a wide variety of aspects for researchers interested in online behavior of individuals, both in natural observations and for controlled experiments.

Behavioral Targeting.

Behavioral targeting (BT), also called online profiling⁶⁴ or hypertargeting⁶⁵ uses historical user behavior to predict user behavior and affinities in web applications such as targeting of online advertising, content personalization and social recommendations⁶⁶. BT is used by online advertisers to increase the effectiveness of their campaigns, and is playing an increasingly important role in the online advertising market⁶⁷. Through series of experiments, compared to standard run of network advertising, BT advertising was found to be more successful, creating greater utility for consumers from more relevant advertisements and clear appeal for advertisers from increased ad conversion⁶⁸. User profiling is performed using novel computation techniques^66,69, commonly fall inside the social computing paradigm, an approach to analyze and model social behaviors on different media and platforms⁷⁰. Van Dam & van de Velden⁷¹ proposed that social networks, like Facebook, can be “operationalized to gain insight into the individuals connected to a company’s Facebook site” (p. 60). In their study they describe a user profile data collection framework that uses “Facebook insights” (accessed by the admin) and other personal public information on Facebook to cluster users. They propose that their methodology can be implemented into an analytical customer relationship management (CRM) framework aimed at the analysis of customer characteristics that may help improve a firm’s customer management strategies.

Recommender systems.

Ricci, Rokach, & Shapira⁷² defined Recommender Systems (RS) as “software tools and techniques providing suggestions for items to be of use to a user” (p. 1). They add that since recommendations in most cases are personalized, different users or user groups receive different suggestions. RSs try to predict what the most suitable products or services are, based on the user’s preferences and constrains (learned and collected from users. Many product purchasing sites like Amazon and E-Bay use advanced recommendation engines⁷³, but social networking sites (e.g. Facebook) use RS as well, to “push” relevant social content based on use preferences, actions and texts⁷⁴.

Recommender systems in Emergency. In the context of emergency, the existing published literature suggests that recommendation systems can be integrated for improved disaster management⁷⁵, Construction and real estate crisis management⁷⁶ and supply distribution in emergency⁷⁷. However, we could not find existing literature of how crisis communication can be aided by RS. More specifically, how crisis messages are selected for individuals or groups of individuals based on some characteristics. Research shows that integrating data obtained from popular social media networking websites significantly can improve results of existing recommendation systems⁷⁸.

Computational social science in the face of disasters.

Over the past years technologies that rely on citizen sensing have been playing a major role in real life applications, such as public and environmental health surveillance⁷⁹, and other participatory social activities^80,81,82. Purohit, Castillo, Meier & Sheth⁸³ noted that with the explosion in social media and the universal mobile access– researchers have unique opportunities to “extract social signals, create spatial-temporal mappings, perform analytics on social data, and support applications that vary from situational awareness during crisis response, preparedness and rebuilding phases to advanced analytics on social data, and gaining valuable insights to support improved decision making”. (p. 1).

Social computing, analytics, visualization for crisis communication.

In the era of big data analytics and social media, incorporation of business intelligence systems in the organization are very common. These systems are responsible for data processing, analysis and visualization for better decision making and business function. Based on a recently published taxonomy in the domain of text analytics and visualization⁸⁴, one can find various analytical tasks and visualization tasks with specific aim. The tasks are common text analytical procedures^85,86, and include: 1) Text Summarization / Topic Analysis / Entity Extraction, 2) Discourse Analysis; 3) Sentiment (Opinion) Analysis, 4) Event Analysis; 5) Trend / Pattern Analysis; 6) Lexical / Syntactical Analysis; 7) Relation / Connection (Association) and 8) Translation / Text Alignment Analysis. These tasks later serve for visualization procedures for: 1) Region of Interest, 2) Clustering / Classification / Categorization, 3) Comparison, 4) Overview; 5) Monitoring; 6) Navigation / Exploration; 7) Uncertainty Tackling.

Tables 1 & 2 summarize current state of the art of how social computing and text analytical tools can be integrated in the best practices of crisis communication^27,32. We focus on the objectives in the processes of crisis and emergency communication²⁷ that occur in different times of the crisis life cycle and have direct association to communication processes with the public. Finally, we describe how computational and analytical tools can be utilized to accomplish these aims.

Monitoring and recognition of emerging risk.

Information extraction tools using pre-defined lexicons of crisis related terms^87,88, as well trend analysis (e.g. hashtag or term anomaly detection) could be a useful way to spot potential risks in the environment, based on social media feeds and streams, that previously have shown themselves to be valuable sources of real-time information about what is happening in the world⁸⁹. Social Media feeds can represent a hybrid form of a sensor system that allows for the identification and localization of the impact area of the event^90,91. Next, using text visualization techniques, geo-located information can be mapped and depicted in a way a potential risk can be categorized according to its geographical occurrence. As people will explicitly mention different terms related to potential risks, dynamic map would show categorizations of the terms into risk topics, allowing crisis managers to monitor risks.

General public’s understanding of risk.

Public’s knowledge of risk is conducted most of the times through surveys³³. However, it can also be inferred from social media discussions. Emergency authorities can, for instance, publish question regarding different issues related to specific risk in their feed (e.g. lifesaving behavior in wild-fire) and later investigate the discussion followed by post publishing. Research shows that social media discussions tend to be topic dependent^92,93, especially when the page posts direct questions and asks the general public to discuss an issue. Crisis managers could train text responses over documents that an expert panel would consider as relevant to the original question (e.g. “what is the correct lifesaving behavior during wildfire”), and also classify documents according their answer correctness using Natural Language Processing (NLP)⁹⁴ and classification algorithms⁸⁶. Overview level of the public understanding of risk, based on social media discussion, then could be visualized using simple pie charts.

Message development in crisis communication.

Based on the variety of approaches presented above, this aspect can be aided by discourse analysis⁹⁵, linguistic analysis and natural language processing. Using lexical approach⁹⁶ and psycholinguistic tools⁵³, crisis managers can analyze messages from different perspectives– from syntactic features of the text and various word meaning categories (e.g. LIWC categories). We propose that integration of such tools would enable to observe whether all pre-designed crisis messages are adapted to the target audience’ communication capacities.

Affected public opining mining.

To gain feedback from the affected public, one should monitor social media and extract signs from citizen who speak and discuss the emergency. Temnikova et al.⁸⁸ presented a terminological resource, EMTerms, which include over 7,000 terms used in Twitter to describe various crises, classified into 23 information-specific categories (e.g. caution and advice, infrastructure damage, supplies needed or offered, personal updates, safety and security and more). Such resource can assist decision makers monitor different categories of information, and later communicate to the public on the most prominent information. Once information is retrieved, automated text analytical approaches, like opinion mining, can assist in capturing public’s perceptions. Opinion mining refers to the extraction of emotion, appraisal and opinion words that are associated with certain social issue, people or entity (e.g. product, public figure, and event) and the classification of the words into different opinion categories (e.g. positive vs. negative, support vs. against)⁹⁷. Seeger³² argued that capturing public’s perceptions are important due to the tight association between beliefs and actions. He adds that monitoring public’s risk perceptions and opinion prior and during crisis is essential for crisis response and message adaptation to public’s needs and concerns. This text analysis procedure has various application in a review summarization & classification, market and brand analysis⁹⁸, political opinion analysis and decision making. Opinion and sentiment classification is performed on a document and sentence levels by extracting relevant features (e.g. single and/or multiple words, emoticons) from the text and applying computational techniques for estimating the overall polarity or direction of the text. The most common methodology involves pre-defined sentiment or opinion lexicons that carry thousands of domain-specific words, or machine learning techniques that use supervised learning for sentiment classification.. In the context of natural/human made disasters, the assessment of sentiment is somehow scarce. Nagy, Valley, & Stamberger⁹⁹ compared methods for evaluating sentiment in disaster microblogs and explored patterns of change in emotion of the crowd during a technological disaster. Unfortunately, assessment of sentiment towards the general crisis event provides only partial information for crisis communication practitioners regarding crowd opinion, since it is not matched to specific risk or issue, and more

In the case of health crisis and pandemics, however, considerable research is found. For instance, in the case of swine flu pandemic, Salathé & Khandelwal¹⁰⁰ used publicly available data from users of online social media and measured spatial-temporal sentiment towards a new vaccine over a period of six months. They further found strong correlation between sentiments expressed online and CDC-estimated vaccination rates by geographical region. This provides good example of how policy makers can use publicly available data to learn about public’s perception regarding risk and its’ related behaviors.

Misunderstanding/Rumors Corrections.

Possibly one of the most researched topics in the domain of disaster computing and social media research during crisis, is information credibility^101,102. Castillo et al.¹⁰² analyzed information credibility of news propagated through Twitter and suggested that by using automated methods a tweet can be classified with up to 80% precision as being credible or not. Their automatic credibility estimation was done by extracting message-based features (e.g. text length, sentiment ratio in the text, inclusion of hashtag), user-based features, topic-based (e.g. aggregates computed based on message and user based features), and propagation-based features (e.g. depth of a retweet in the network of message spread, initial tweets per topic).

Evaluation and assessment of responses, including the effectiveness of communication.

Similarly to the previous section, evaluation of public responses can be achieved through continuous social media monitoring using opinion mining, to infer on concerns, panics, and the emotional impacts of interactions among social media users¹⁰³. Recent studies demonstrated capabilities to track changes in sentiments of affected public during natural disasters^99,103,104.

Proposed Process of Hyper-targeted Crisis Communication in Social Media (HCCS)

In the above paragraphs, we described how advanced analytical tools can be integrated in the processes and aims of crisis communication. We suggest that the availability of the computational techniques can improve communication with the public by a process of hyper-targeted crisis communication. Figure 1 demonstrates how the data posted and shared through social media is collected for monitoring, analyzed and prepared for a tailor-made crisis communication.

In the process, massive amounts of textual user-generated content in social media is continually generated and monitored for crisis or emergency related communication. Once detected, content is collected and transferred to the analytical phase, where it is preprocessed, analyzed according to various text mining analytical methods. The information then is either transferred for reporting where it can be visualized and interpreted or moves to the communication management module. Authorities, crisis managers and designated spokespersons personnel can communicate with the public, using the communication management module, which include a recommender system that proposes matched messages. This sub-system is also responsible for target audience profiling (i.e. hypertargeting), that enables accommodated communication to specific groups in the population. Bi-directional communication is then monitored for matching accuracy and secondary messages in the communication. During the conversation, a crisis messaging recommender system formulates messages based on conversation history and the other available parameters.

Benefits of Targeted Crisis Communication

Automatic categorization of unstructured vital information is of high importance for speeding up disaster management. More precisely, psychosocial information extraction is valuable for decision makers to understand the endurance of the public, public’ psychological needs and psychological risk, using validated theoretical models in the field of disaster psychology¹⁰⁵. Being able to capture these aspects, emergency authorities can be more synchronized with population needs and thus more effective in their response to the public during disaster. By integrating disaster related psychosocial aspects in a command and control disaster management systems, they will be in a better position to manage the situation and speed up recovery.

Successful detection of language choice patterns from social media content may assist in guiding the authorities and service providers to respond on the same communication “channels” of the public, and enhance responsiveness and interactivity. This can further lead to a more cooperative public and enhanced public morale. It is our assumption that such focused messaging may not only improve the reception of the message, but also contribute to the ensuing behavior or compliance which is critical in disasters.

Limitations of the present review and currents challenges

One should remember that beyond text, additional forms of data are published on social media, sometimes far more influencing, since pictures (regular photo or meme) may generate more engagement than other forms of data (including text and video)¹⁰⁶. Research documents applicable tools for multimodal analysis of social media content, that include both textual and visual data, e.g. Flicker photos, YouTube videos and more¹⁰⁷. Thus, more precise public monitoring could be accomplished when integrating data from different modalities and content qualities, as adjacent task in the crisis communication processes. The above literature focused mainly on text that appears in the social media, that can be extracted and analyzed. As for now, some social media platforms offer easy means of extracting data, like the Twitter Streaming API¹⁰⁸. One limitation in social media analysis is the inaccessibility to Facebook public feed API, which is restricted to a limited set of media publishers. This limitation is restricting research institutes and companies to obtain public posts, in a similar way that is available in Twitter. It is possible, though, to extract data from Fan Pages and Groups using publically available applications, like NetVizz¹⁰⁹.

Decision makers should not solely rely on text analytics in social media when doing crisis communication. Though social media is very popular and has very high penetration rate¹¹, some social and age groups in the population are not connected to the internet¹¹⁰. As this population tends to receive the information and guidelines during real time emergency via TV, radio, newspaper and word-of-mouth -the instructions will still need to use traditional means of mass communication..

Despite recent anecdotal reports indicating that authorities (e.g. Home-front command) increase their presence in the social media and continuously interact with the public in times of calm and emergency, still authorities prefer to communicate with the public in non-interactive means. This was previously mentioned as an obstacle for crisis communication¹¹¹ in light of the raising of social media .Nonetheless, decision makers should not solely rely on methods of text analytics and data mining in social media when doing crisis communication and follow a well-organized risk and crisis communication plan³⁴.

It is almost impossible not to consider privacy issues when discussing targeted communication¹¹². Collecting information about the behavior of social media users, for research or commercial purposes, is considered invasive by the public as well as inappropriate¹¹³. Recently, the use of Facebook as a research platform for massive social experiment¹¹⁴, was criticized for not passing the ethical board and not providing informed consent to the participants. Conversely, Facebook advertising system is using profile information and interests (extracted from activity patterns of the users), to deliver targeted messages. Hence, authorities have to deal with this issue using sound legal advice and should address the privacy issues in their communications with the public to build trust and credibility.

Recommendations

Both user profiling (i.e. behavioral targeting) and recommender systems are closely related. In both processes, user preferences are used for content recommendation. However, the application of behavioral-targeted recommendation system in the field of crisis communication is novel and deserves future studies. We propose that behavioral-targeted recommendation systems in the domain of crisis communication would be applied on two levels; public level and decision maker level. On the public level, personalized and adapted crisis messages would pop-up or appear in popular websites and social media based on two possible parameters: 1) the written content people provide on these sites (e.g. users’ public posts, reviews, and participation in talkbacks; 2) user behavior on site – such as “like”/”share’ or “re-tweet, or other forms of ratings. This approach would be effective if the user is active and engaged, but also may be relevant for passive users that mainly consume content rather than commenting on it. This proposal will be feasible in social media platforms which allow the tracking of users or has an agreement for data sharing with authoritative bodies. On the decision maker level, designated spokespersons would be assisted by the messaging recommender system to adapt their linguistic style to the user who approaches them (or vice versa) thus “tailoring” the response and assisting people in real-time to get the appropriate help, directions or support.

While certain online platforms for communication with the public in emergency were proposed^111,115 they all significantly lack the user-profiling element, that might be in practice, an important factor in communication efforts¹¹⁶. Future web-based or mobile application-based platforms for disaster risk communication should include user-profiling module, based on different parameters extracted from users’ published text and data. We suggest social media management tools²⁹ as suitable platform for crisis communication, as some of the tools enable the integration of extensions and apps that assist in social media data extraction, analytics and interpretation.

Conclusions

In very fast pace, social media have acquired a prominent role in media and our daily life⁷¹. During crisis and emergency, people tend to approach social media not only because of the need for quick information, but also due to the human tendency for storytelling, which allows people to experience their lives as coherent, orderly and meaningful. It is what makes people’s life more than a blooming, buzzing confusion¹¹⁷. Communication with public during emergency is critical for successful emergency management¹¹⁸ life saving, rescue and recovery. On the other hand , one should take into account that 2.08 billion people world-wide are social media users¹¹ and significant number of people are expected to participate in the online convergence, posting massive amount of user-generated content making it an almost impossible web of information. Using state of the art computational tools makes this mission possible. Being one of the fundamental tools of emergency management¹¹⁸, crisis and risk communication play an important role in mitigating, or exacerbating, the psychological and behavioral reactions to critical incidents and disasters. Thus the task of continues monitoring of the social web for crisis related information can significantly improve the efforts of emergency management. This is the reason for officials to be eager to exploit social media content to gain insight from the public for the optimization of risk and crisis communication¹¹⁹. Risk and crisis communication’ best practices are well documented and include various tasks during different crisis stages (e.g. understanding public’s characteristics, monitoring emerging risk, feedback from affected publics etc.). These aims and practices can be improved and become efficient when utilizing linguistic computational tools that can quickly assist decision makers in designing emergency campaigns, or/and managing communication with the public during emergency. Officials should also remember that the most effective communication are those designed for a specific audience³³. Research shows that interpersonal communication that is linguistically matched between the communicators is not only more successful, but also reflects our natural tendency to adjust ourselves to the listener. Finally, decision makers and crisis managers need to acknowledge the potential of the massive generated textual information published during times of emergency in the social media. This knowledge is not only beneficial for situational awareness but also for user profiling, which in our case, is not intended for marketing purposes. Behavioral analyzed information from text posted on social media can leverage targeted communication during crisis, by automatic target audience characterization and adapted, psycholinguisticly matched risk message preparation. Thus, as we keep facing major disasters and emergencies that have became more frequent during the last decade, governments can be more effective in transmission and dissemination of warning messages or/and survival information, better manage different public reactions triggered by the crisis, gain credibility from the public, and improve public cooperation during emergencies.

Competing Interests Statement

The authors have declared that no competing interests exist.

METHODS

An environmental scan was conducted through a peer-reviewed literature search of PubMed, Scopus, EMBASE, Web of Science, Cochrane databases, and an open Internet search for training program websites. The landscape literature search was conducted by combining the search terms “international,” “disaster,” “complex humanitarian emergencies,” “training,” and “humanitarian response.” All searches were conducted from January2, 2013 to September 12, 2013. Initially, 681 citations were retrieved from the above named search engines. Screening by titles narrowed the focus to 165 articles and a further screening by abstract led to 40 articles, of which a full text review revealed 14 articles that contained information regarding training programs for responders to international disasters and CHE.^{8,11,12,13,14,15,16,17,18,19,20,21,22}

RESULTS

A total of 14 peer-reviewed articles mentioned or described eight training programs, while open Internet search revealed 13 additional programs. In total, twenty-one training programs were identified as currently available for responders to international disasters and CHE.

Each of the programs identified has different goals and objectives, duration, expenses, targeted trainees and modules. Seven programs (33%) are free of charge and four programs (19%) focus on the mental aspects of disasters. The mean duration for each training program is 5 to 7 days. Fourteen of the trainings are conducted in multiple locations (66%), two in Cuba (9%) and two in Australia (9%). The cost-reported in US dollars- ranges from $100 to $2,400 with a mean cost of $480 and a median cost of $135. Most of the programs are open to the public, but some are only available by invitation, such as the International Mobilization Preparation for Action (IMPACT) and the United Nations Humanitarian Civil-Military Coordination (UN-CMCoord) Field Course. The target audience for IMPACT should satisfy three conditions: 1) have met the established criteria (technical competency in an international emergency response context in one of the “core functions”, “field experience”, and available for four-week voluntary deployment), 2) successfully completed screening interviews, and 3) completed World of Red Cross (WORC) training program.²³ UN-CMCoord Field course participants should satisfy the following five conditions: 1) completed the UN-CMCoord course, 2) have a minimum of five year experience with civil-military coordination in the humanitarian assistance, 3) fluency in oral and written English; 4) fluency in other UN language and 5) currently performing a task that require interaction with the Military.²⁴

Table 1 displays the twenty-one structured training programs (in alphabetical order), along with the organizer, location(s), cost, duration, and year established.

Table 1 abbreviations:

ADPC= Asian Disaster Preparedness Center; CDAC= Capacity Development across Cultures; CDHRTP= Canadian Disaster and Humanitarian Response Training Program; CHART= Combined Humanitarian Assistance Response Training; CHR= Contingency Henry Reeve Program; CIHC =Center for International Humanitarian Cooperation DMTC= Disaster Management Training course; DMHP= Disaster Mental Health Program by the American Red Cross. ELAM Escuela Latino Americana de Ciencias Médicas (in Spanish; in English Latin American Medical School); FEMA= Federal Emergency Management Agency; HELP= Humanitarian Emergencies for Large Populations; IHAT= International Humanitarian Action Training; IHPT= International Humanitarian; Protection Training; IDM= Integrated Disaster Management Program; IMPACT= International Mobilization and Preparation for ACTion; ICISF= International Critical Incident Stress Foundation; ICRC= International Committee of the Red Cross; IDHA= International Diploma in Humanitarian Assistance; IFRC= International Federation of Red Cross and Red Crescent Societies; ICTC=In country training course; IHE= International Health Exchange programs; IRC= International Rescue Committee; MIHA=Masters in International Humanitarian Action; MHCE=Mental Health in Complex Emergencies; NI=No information; OCHA= United Nations Office for the Coordination of humanitarian Affairs; PHCE= Public Health Complex Emergencies ROC= Recovery Orientation Course, SIU= Symbiosis International University TAP= Training assistance program; UN-CMCoord= Humanitarian Civil-Military Coordination WE Inc.= World Education, Inc. WORC= World of Red Cross; Yr. Est.= year established.

Below are descriptions of the objectives, target audience, course size, and training content of the twenty structured training programs listed in the table above. Information is listed alphabetically and depends on its availability from the various sources queried.

Asian Disaster Preparedness Center (ADPC)²¹ : The objective is to develop participants’ skills with the aim that they should gain a solid grasp of disaster management processes and be able to address key implementation issues in disaster. Typical participants include government officials responsible for disaster management plan and policy; and personnel involved in defense forces and emergency services. There is a maximum of 30 participants per course. The main topics include disaster preparedness planning and disaster risk reduction.²⁵

Capacity Development Across Cultures (CDAC): The objectives include developing skills for building relationships, understanding capacity, and working collaboratively with partners and colleagues to achieve capacity development outcomes. The participants are Red Cross staff, volunteers, and anyone interested in building capacity internationally. There is no course size restriction.²⁶ The main topics are not delineated in details.

Canadian Disaster and Humanitarian Response Training Program (CDHRTP): The objectives include developing capacity for essentials during a humanitarian response, managing acute issues in complex disasters, and learning research skills. This course also places participants in a three-day simulation complex humanitarian emergency to train their capacity building and problem solving skills. The participants are anyone interested in humanitarian work. There is no course size restriction listed. Topics include gender based violence, leadership, human rights, monitoring and evaluation, and, refugee protection.²⁷

Combined Humanitarian Assistance Response Training (CHART)^11,28: The objective is to develop civilian and military response to humanitarian crises in a more efficient manner. It is designed to provide responders with the tools and knowledge to work effectively in rapidly evolving situations. Participants are civilian and military disaster planners, and the maximum number of participants admitted per course is sixty. The main subjects include the complex environment, security, logistics, nutrition, public health, international standards, responder readiness, planning, peacekeeping issues, and international humanitarian law.²⁹

Contingency Henry Reeve Program (CHR): The objective is to develop responders who can aid in saving lives, especially in remote areas. The training program focuses on delivering primary care in disasters and epidemics. The type of participants is not documented. There are no restrictions to number of participants. The main topics includes disaster medicine courses for all types of emergencies.³⁰

Disaster Mental Health Program (DMHP): The objective is to develop resilience among responders by ameliorating acute post disaster stress responses among Red Cross workers and the disaster victims they serve. Participants are comprised of professionals from diverse mental health backgrounds. The number of participants and course titles are not delineated.³¹

Disaster Management Training Course (DMTC): The specific objectives are to develop participant understanding as to why children are among the most vulnerable populations in disasters and to identify the most important problems and priorities. The program targets health care professionals and physicians as participants. There is no restriction to the number of participants. Lecture topics included epidemiological assessment, triage during disasters, and malnutrition.

Escuela Latino Americana de Ciencias Médicas (ELAM)¹⁴: The objective is for participants to develop skills necessary for disaster response. The program has a strong clinical component. Participants worldwide are welcome if they have an interest in disaster preparedness. There are no restrictions to the number of participants. The main topics include disaster medicine courses for all types of emergencies.³⁰

Humanitarian Emergencies for Large Populations (HELP)^11,12,16,32: This course is a multicultural and multidisciplinary learning experience created to enhance professionalism within humanitarian assistance programs in the setting of disasters and CHE. Participants are medical and public health professionals, environmental health engineers, and epidemiologists. Twenty-five to thirty participants are admitted per course. The main topics include public health activities, communicable diseases, epidemiology, international humanitarian law (IHL), security, humanitarian principles, and ethics.³³

Integrated Disaster Management (IDM): The objectives are to ensure and develop awareness of the nature, type, and management of disasters, design disaster management plans as well as hands-on training in handling medical and non-medical emergencies. The participants are undergraduate and post-graduate students in Symbiosis International University (SIU), India. The number of participants and main topics were not outlined.¹⁵

International Humanitarian Action Training (IHAT)/ International Humanitarian Protection Training (IHPT). IHAT: The objective is to develop and equip individuals and organizations with the knowledge needed to respond effectively and implement recovery after disasters. It is open to current and aspiring aid workers, volunteers, and those interested in international emergency response. There are no restrictions to the number of participants per course.²⁶ The main topics were not listed. IHPT: The objective is to develop understanding and competencies for protection into humanitarian action (Sphere 2011) in four key areas: child protection, gender-based violence (GBV), sexual exploitation and abuse (SEA), and other human rights abuses. Participants are humanitarian aid workers, program officers/coordinators, caseworkers, policy officers or others working with people affected by disaster. There are no restrictions to the number of participants per course. The IHAT course is a pre-requisite to this course.³⁴ The main topics include gender-based violence (GBV) and child protection among others.

International Health Exchange programs (IHE): The objective is for aid workers around the world to develop the latest skills to ensure the quality and effectiveness of humanitarian programs. The participants are health workers and any professional groups who wish to work in emergency relief. There are no restrictions to number of participants per course. The main topics include humanitarian essentials, security, and technical training.³⁵

International Mobilization and Preparation for ACTion (IMPACT)¹⁹: The objective is to provide additional training to participants by taking an in-depth look at the roles of the ICRC, the Federation, and National Societies in times of disaster and conflict. Another objective is to provide contextual and capacity building skills for relief and development activities. Participation is by invitation only and limited to 20 applicants. Interested parties who meet the established criteria, successfully complete screening interviews, reference checks, and WORC? may be invited to participate in the course. The course covers practical subjects such as dealing with stress, security, and cross-cultural awareness.²³

Masters in International Humanitarian Action (MIHA & OTHER PROGRAMS): Center for International Humanitarian Cooperation (CHIC) offers several humanitarian training programs. The objective is for humanitarian workers and members of international relief agencies to perform more effectively in conflict and post-conflict areas. The CHIC designs training courses tailored to organizations or geographical location. Participants are members of international relief organizations and humanitarian workers. There are no restrictions to the number of participants per course. Main topics include mental health issues, disaster management, and negotiation techniques.³⁶

Public Health Complex Emergencies (PHCE)¹²: The objectives are to develop and sharpen practical problem-based skills in an interactive group setting and to help participants become well-informed decision-makers and managers of public health policy in complex emergencies. Participants are non-governmental organization (NGO) staff who are responsible for making decisions that affect the health of refugees, internally displaced persons, and those affected by CHE. Other participants include District Medical Officers and other Ministry of Health staff working in regions affected by complex emergencies. There are no restrictions to the number of participants per course. The main topics include context of emergencies, reproductive health, epidemiology psychosocial health, communicable disease weapons, violence and trauma, and environmental health protection and security and nutrition coordination.³⁷

Recovery Orientation Course (ROC): The ROC uses an intensive experiential teaching approach to develop a hands-on experience for working in the humanitarian sector. It is mandated for prospective Medair employees but open to the public. There are no restrictions to the number of participants. The main topics include an overview of Medair and understanding humanitarian aid, others and self. The cost of the course is dependent on the participant’s experience with relief work.³⁸

Training Assistance Program (TAP)¹⁹: This training program is similar to the CHART program. The difference is that the host organization helps with the designing of the curriculum and develops a draft agenda based on the training objectives and target audience.³² The main topics are similar to CHART, and the class size is not outlined.

United Nations Humanitarian Civil-Military Coordination (UN-CMCoord) Course and Field Course³²: The objective of these courses is to equip humanitarian and military personnel with the skills and knowledge necessary to communicate and effectively interact with each other. Participants include governmental and non-governmental organizations, aid agencies, civil protection units, military and civil defense organizations, UN agencies and other intergovernmental bodies, the Red Cross, and Red Crescent Movement. The maximum numbers of participants for the course and field course are twenty and twenty-eight, respectively. The main topics include the roles of military and humanitarian actors in emergencies and complex emergencies. Participation in the field course is by invitation only. The prerequisites are the UN-CMCoord course and a minimum of five years of experience in civil-military coordination.²⁴

Various programs by The Salvation Army: The goal of The Salvation Army’s disaster training program is for individuals to develop the skills needed to serve during times of crises particularly in the field of mental health. Participants include spiritual care officers, disaster workers trained in critical incident stress debriefing, chaplains, and mental health professionals. There are no restrictions to the number of participants per course. The main topics include foundations of emotional and spiritual care, disaster food services, disaster social services, and incident command systems.³⁹

Various programs organized by International Critical Incident Stress Foundation (ICISF): These programs currently offer over 40 different courses selections at conferences and trainings held around the United States. The objective is to develop the skills for handling critical incident stress and mental health issues. Participants include people working in the following fields: crisis intervention, disaster response, education, emergency services, employee assistance, healthcare, homeland security, human resources, mental health, military, spiritual care, transportation, and traumatic stress.The main topics are group crisis intervention, compassion fatigue, emotional, spiritual and psychological first aid.⁴⁰

World of Red Cross (WORC): This online general orientation program introduces the main elements of the Red Cross Red Crescent Movement. The objective is to educate participants on the history and principles of the Red Cross Red Crescent Movement, in order to improve understanding and commitment to the mandate and culture. The course is open to all Red Cross Red Crescent staff, volunteers, government authorities, donors, the media, schools, and the interested public. There are no restrictions to the number of participants per course. The main topics include the origin and history of the Red Cross. This course is one of the three prerequisite for the IMPACT program.²³

DISCUSSION

A successful and efficient response to international disasters and complex humanitarian emergencies depends on the capacity, skills, and prior training those responders have already acquired. As the number of complex humanitarian emergencies and disasters increase, one of the challenges facing responding organizations is to maintain well-trained volunteers and employees. There is an increasing demand for skilled health providers, public health professionals, and field program administrators; this requires a cadre of trainees with both general and specific capacities.¹¹

The SPHERE standards were created in 1997 to define a set of universal minimum standards in core areas of humanitarian assistance.⁴¹ This project was a collaborative effort by many agencies to produce standards that relate to disaster assistance in: 1) water supply and sanitation; 2) nutrition and food aid; 3) shelter and site planning; and 4) health services.⁴² Despite the definition of the SPHERE guidelines, many organizations do not train their personnel in the applications of these guidelines.¹¹

Due to the various objectives of the programs along with the various expectations and interests of the trainees, there is no single training program that is universally accepted as the ideal. The main language used for these trainings is English, but the HELP courses are taught in multiple languages. Regionalized trainings conducted in other languages such as French, Arabic, or Spanish may reduce the burden to find interpreters or obtain visas. This would reduce costs and improve logistics for participants.

The list provided in this paper offers a general description of the various training programs offered worldwide. This will help potential responders –volunteers or non-volunteers- to identify training opportunities that will build their capacities in the specific areas they are responsible for. This will allow them to consider the financial, linguistic, and logistical characteristics. With evolving technology, the responders of the future may not be limited to attending in-person training programs, but have access to virtual environment experiences. However, this modality of training is still infantile for training programs geared towards international disasters and CHE and requires a large amount of financial and technical resources. For the foreseeable future, in-person training programs remain indispensable, and they continue to grow in number, scope, breadth, location and targeted audience

The vast majority of these programs target volunteers and are known worldwide for being offered by large, reputable, and established international organizations; however many more specialized trainings exist as well. NGOs typically do staff training targeted for specific roles in the field, however these training courses are seldom published in peer-reviewed journals and may not even appear on the websites of the NGOs who offer them. A centralized database of all training programs would be beneficial.

Limitations

There is a paucity of peer-reviewed literature with substantial information on structured training programs for responders to international disasters and CHE. Twelve articles on specific training programs were published in peer-reviewed journals. However, it is important to note that the literature search was not conducted in a systematic fashion due to resource constraints. There is no single web-based resource that is comprehensive, robust, and well known enough that potential responders can consider. In addition to the programs reported and analyzed in this report, there are many other smaller programs with different foci.

Additionally, there is no published literature on the recommended consensus-based recommendations for competencies or curricula. Future work could include surveying participants on their experiences in each program.

CONCLUSIONS

Training programs available for responders to international disasters and CHE vary in their objectives, target audiences, modules, geographical locations, and financial cost. Currently, there is no resource that centralizes the descriptions of humanitarian response training programs. This paper presents an overview of available programs and serves as a resource for potential responders and organizations interested in capacity-building training prior to an international disaster response or complex humanitarian emergency.

Methods

Data on the impact of tsunamis were compiled using two methods, a historical review of tsunami events and a systematic literature review for publications relating to the human impacts of tsunamis with a focus on mortality, injury, and displacement.

Historical Event Review

Data for the historical event review were obtained from two sources. The National Oceanic and Atmospheric Administration’s National Geophysical Data Center (NOAA-NGDC) tsunami database ⁸ consists of two sets of related files on tsunami events and a tsunami run-ups. The event file lists the cause (almost exclusively undersea earthquakes) that triggered tsunamis together with the total impact of a single event (i.e. the aggregate impact from multiple wave run-up locations together with the coordinates of the originating event). The run-up file includes wave characteristics and impacts in each affected location (multiple run-up reports per event). Records from the tsunami run-up file that met all of the following criteria were retained: 1) occurred between 1900 and 2009; 2) reported as definitely or probably occurring; 3) had a wave height ≥2.0m; and 4) resulted in ≥1 deaths. This yielded a total of 116 run-up reports from 82 different tsunami events.

The Centre for Research on the Epidemiology of Disasters’ Emergency Events (CRED EM-DAT) was the second data source used in the review. All wave/surge events that were reported between 1900 and 2009 in EM-DAT were included (n=58); data were initially exported in 2008 when CRED reported wave/surge events; this category was subsequently discontinued. For tsunami impacts reported by EM-DAT, zeroes were treated as missing values because they were used as placeholders and their inclusion in the analysis could contribute to the under estimation of tsunami impacts. The NOAA-NGDC run-up database was subsequently searched for events reported by EM-DAT that were previously excluded due to wave height <2.0m or uncertain reporting criteria, and NOAA-NGDC event data were added to the EM-DAT records. This process yielded a total of 151 records, including 58 events reported by EM-DAT and 134 run-ups reported by NOAA-NGDC. The run-up file was used to assess wave characteristics and outcomes. A separate event file comprised of 81 events was created by combining multiple reports of tsunami impact within a country into a single event. To create a summary record for each of event with multiple reports, human impacts at each location were summed, and the maximum wave height and inundation depth were applied. The event file had 94 events, including 58 reported by EM-DAT and 71 by NOAA, and was used to assess frequency and distribution of tsunamis and their impact by country. Findings presented in this review are based on the 151 run up file. Both run-up and source data can be accessed online at https://www.jhsph.edu/refugee/natural_disasters/_Event_Tsunamis.html.

In order to examine country- and event-specific characteristics associated with low and high levels of tsunami mortality, deaths were categorized as follows: low (<10 deaths), medium (11-75 deaths) and high (>75 deaths). Bivariate tests for associations between tsunami mortality and the following characteristics were performed using χ² (categorical measures) and ANOVA (continuous measures): time period (dichotomized, 1900-1955 and 1956-2009), region as defined by the World Health Organization (WHO), income level (World Bank), gross domestic product (GDP), GINI (measure of income inequality), distance from source (quartiles), wave height (dichotomized, <6.65 and >6.65) and earthquake magnitude. For the region variable, only three events were reported in Africa and because all were related to the 2004 Indian Ocean tsunami, they were grouped with Southeast Asian; only one event was recorded in the Eastern Mediterranean region which was grouped Europe .All analyses were performed using Stata Statistical Software, Version 11.0 ⁹

Systematic Literature Review

Key word searches in MEDLINE (Ovid Technologies, humans), EMBASE (Elsevier, B.V., humans), SCOPUS (Elsevier B.V., humans), and Web of Knowledge, Web of Science (Thomson Reuters) were performed to identify articles published in July 2007 or earlier that described natural hazards and their impact on human populations. Key words used to search for natural hazards included natural hazard(s), natural disaster(s), volcano(es), volcanic, volcanic eruption, seismic event, earthquake(s), cyclone(s), typhoon(s), hurricane(s), tropical storm(s), flood(s), flooding, mudslide(s), tsunami(s), and tidal wave(s). Key words included for impact on human populations were affected, damage(d), injury, injuries, injured, displaced, displacement, refugees, homeless, wounded, wound(s), death(s), mortality, casualty, casualties, killed, died, fatality, fatalities and had to be used in either the title, abstract or as a subject heading/key word. The search resulted in 2,747 articles from MEDLINE, 3,763 articles from EMBASE, 5,219 articles from SCOPUS, and 2,285 articles from ISI Web of Knowledge. Results from the four databases were combined and duplicates were excluded to yield a total of 9,958 articles. . One search was done for all the five natural hazards described in this set of papers. This paper describes the results for tsunamis. The systematic review is reported according to the PRISMA guidelines.

Title screening was performed to identify articles that were unrelated to natural disasters or human populations. Each title was screened by two independent reviewers and was retained if either or both reviewers established that inclusion criteria were met. To ensure consistent interpretation of inclusion criteria, percent agreement was assessed across reviewers for a small sample of articles, and title screening began after 80% agreement on inclusion was achieved. A total of 4,873 articles were retained for abstract review. Articles that met one or more of the following criteria were excluded in the abstract screening: language other than English; editorial or opinion letter without research-based findings; related to environmental vulnerability or hazard impact but not human populations; individual case report/study; focus on impact/perceptions of responders; and not related to human or environmental vulnerabilities or impacts of hazards. As with the title screening, overall percent agreement between reviewers was assessed, and abstract screening began after achieving 80% agreement. Each abstract was screened by two independent reviewers and was retained if either or both reviewers established that inclusion criteria were met. During the abstract review, included abstracts were coded for event type, timeframe, region, subject of focus, and vulnerable population focus.

A total of 126 articles were retained for full article review. Articles discussing the impacts of natural disasters on human populations in terms of mortality, injury, and displacement were prioritized for review. A total of 64 articles on tsunamis meeting the aforementioned subject focus criteria were retained for full review. Upon full review, 27 articles were retained including 23 that underwent dual review and standardized data abstraction, two identified as review articles ^10,11, one policy article ¹² and one article on mitigation ¹³ (Figure 1). Following the systematic review, a hand search was conducted using the databases and key words listed above to identify relevant articles published between July 2007 when the initial search was conducted and October 2012; seven additional articles were identified that met criteria for full review. In total, 34 articles relating to risk factors for mortality, injury or displacement were identified; summaries of articles with primary data (n=30) are presented in Table 1.

Results

Historical Event Review

Between 1900 and 2009, 94 tsunamis that affected human populations were recorded. The frequency of events was relatively constant through the 1980s, after which a dramatic increase was reported (Figure 2). This increase is likely the result of improvements to monitoring and reporting systems. Tsunami frequency and mortality were concentrated in the Western Pacific, Southeast Asia, and Americas regions, each of which accounted for almost one third of tsunami events and deaths, but Southeast Asia accounted for 52% of the tsunami-affected population from 1900 – 2009 and 95% of the tsunami affected population from 1980 – 2009 (Figure 4). An estimated 2.5 million people were affected by tsunamis between 1900 and 2009. A sharp increase in tsunami mortality and affected populations was observed from 2000 to 2009 as a result of the 2004 Indian Ocean tsunami (Figure 3). The overall impact of tsunamis on human populations is summarized in Table 2.

Tsunami Characteristics. The physical characteristics of tsunamis were reported by NOAA-NGDC and included distance of the tsunami run-up from source, wave height, and earthquake magnitude. The median distance from source was 119 km (mean 810 km; range 7-10,621 km), and the median wave height was 6.7m (mean 13.0m; range 1.8 -67.1m). The majority of the tsunamis reported were due to earthquakes (95.5%), with small minorities resulting from landslides (3.0%), volcanoes (0.8%), and meteorological events (0.8%). Median magnitude for earthquake generated tsunamis was 8.1 (mean 8.1; range 6.3-9.5).

Mortality. Deaths were reported in 92.6% (n=94) of tsunamis occurring between 1900 and 2009. There were an estimated 255,195 deaths (range 252,619-275,784) resulting from tsunamis recorded in the historic event review, with the 2004 Indian Ocean tsunami accounting for an estimated 227,497 deaths (89%) of all mortality. Deaths were concentrated in Indonesia (62% or 170,689 deaths), Sri Lanka (13% or 35,399 deaths), Thailand (3.2%, 8,876 deaths) and the Philippines (3% or 8,137 deaths). Apart from rare high-impact events such as the 2004 Indian Ocean tsunami, mortality levels in tsunamis tend to be relatively low with a median of 50 deaths per event (mean=2,963,range 0-165,708) when using the highest reported death toll.

Table 3 presents results of the bivariate analyses between tsunami characteristics and mortality. Time period, and GINI coefficient were not statistically associated with tsunami mortality. There were considerable differences in mortality levels by WHO region, with the majority of tsunami events that occurred in the Americas resulting in low (<10) deaths and 50% of events in the South East Asian region resulting in high (>75) deaths (p<.001). Both World Bank-defined income level and per capita GDP were significantly associated with tsunami mortality. Whereas the majority of events in high income countries resulted in ten or fewer deaths, a considerably greater number of events in low and lower-middle income countries resulted in greater than ten deaths (p=.009). Similarly, events that resulted in greater than 75 deaths occurred in countries or territories with per capita GDPs that were more than two times lower than countries that experienced fewer tsunami deaths (p<.001). While the data show an inverse associations between tsunami mortality and distance from source (p=.001) and a positive association between mortality and wave height (p=.016), it is important to note that information pertaining to both measures were missing in a considerable proportion of events, particularly those occurring earlier in the study period.

Injury. Injury reports were only available in 22 (23.4%) events for a total of 48,462 injuries (range 45,466-51,457). A typical tsunami caused 218 injuries (median value) but the distribution was skewed by rare large-scale events (mean injured=2,339, range=0-23,176). To more accurately estimate the total number of injuries due to tsunamis, it was presumed that injuries would occur in events where deaths were reported. There were 87 tsunami events with fatalities; when the median and mean for injuries were applied to the remaining 65 events with fatalities, it was estimated that between 14,170 and 152,035 unreported tsunami-related injuries occurred worldwide between 1900 and 2009.

Systematic Literature Review

Mortality. Of the 33 articles reviewed, 12 reported on mortality, including mortality counts or rates and/or risk factors for death (Table 4). Age and sex were described as risk factors for death in three articles, all from the 2004 Indian Ocean tsunami ^28,32,37. In both Indonesia and Sri Lanka, significantly higher death rates were reported among females who were 1.4 to 2.1 times more likely to perish than males. Significantly elevated risk of death was also observed in children (1.8 to 4.3 times) and older adults (2.1 to 3.1 times) who were more likely to die compared with younger/middle age adults. Other risk factors included education which was inversely associated with mortality risk ²⁸; fisheries-based livelihoods ^28,32indoor location at the time of the tsunami and home destruction ²⁸ and the physical environment ¹⁵. The majority of tsunami deaths were due to drowning ^11,14.

Injury. Tsunami-related injuries were reported in 18 articles though only ten provided detailed information on injury types or risk factors (Table 5). Only one study estimated a population-based injury rate of 8.5% in Indonesia following the 2004 tsunami ⁴². Other articles described the proportion of care seekers with tsunami-related injuries or included only patients with tsunami-related injuries, preventing comparisons in injury patterns across events and locations. Only two studies described risk factors for tsunami-related injury, both with similar findings, in the 2004 Asian tsunami. The first, a study of patients seen by a medical team in Sri Lanka, observed that males and adults (15-64 years) seeking care for injuries were over-represented, suggesting these population sub-groups had higher injury rates ²⁶. The second used household survey data from Indonesia and reported lower injury risk among females (OR=0.81) and the highest injury rates among those 20-49 years of age. The risk factors for injury were opposite those of mortality suggesting that those more likely to survive the tsunami were also more likely to be injured ⁴². The most common types of injuries were wounds due to physical impact with debris, fractures, and near drowning and/or aspiration ^{21,22,23,25,26,30,42} . As compared to other types of natural disasters, tsunamis often result in relatively high mortality but have lower rates of injury ¹¹.

Discussion

Main Findings

From 1900 to 2009, approximately 2.5 million people were affected by tsunamis including over 255,000 deaths and an estimated 50,000 injuries. The mortality and population affected estimates presented in this study are consistent with other reviews of global tsunami events ¹¹ It is likely that the number injured is underestimated, given the low frequency with which this figure was reported, particularly in the first half of the 20^th century. When the mean and median numbers injured were applied to events that resulted in deaths it was estimated that between 10,900 and 116,950 unreported tsunami-related injuries occurred worldwide during this time period. It is important to note, that studies have shown that that ratio of deaths to injuries following a tsunami is typically significantly higher compared with other natural disaster types ¹¹.

This study is the first to examine the influence of place and event characteristics on tsunami mortality. Analyses of tsunami run-up data from 1900 to 2009 reveal that events occurring in the South East Asian region were significantly more likely to result in greater numbers of deaths compared to other regions, and this finding persisted even after excluding the 2004 Indian Ocean tsunami. Examination of the relationship between mortality levels and the two poverty measures (World Bank income level and per capita GDP) demonstrate that risk of mortality event is significantly higher in low-income countries. Lastly, certain event characteristics are more likely to be predictive of mortality than others. Increased wave height and closer proximity to the source were associated with higher mortality levels whereas earthquake magnitude was not associated mortality. A number of other factors that were not examined in this study have been shown to influence tsunami damage and impacts including wave velocity, water depth and submarine topography ¹⁰. Ecological assessments such as this review may mask significant in-country differences in mortality risk (e.g. population density, rural or urban location, geographic variations and individual-level socio-demographic factors).

Findings from the systematic literature review of studies examining tsunami-related mortality and injury contribute to understanding the primary causes of death and types of injury, as well as factors that place certain populations at increased risk. The most common cause of tsunami-related death was drowning, and the most frequently reported injuries included wounds and lacerations, fractures, and near drowning and/or aspiration. When reported the mortality risk was higher among females and the very young and old and injuries were more common among middle-aged men. Additional mortality risk factors included location during the event and fisheries-based livelihood. Few studies assessed or found relationships between socioeconomic status and mortality risk, although that one study found an inverse association between education level and mortality ²⁸ which suggests that low socioeconomic status may place individuals at increase mortality risk. This suggests that preparedness efforts target specific population groups for the prevention of deaths and injuries. Future studies on both tsunami-related injury and mortality risk would benefit from incorporating additional socio-demographic measures to gain a more comprehensive understanding of risk factors.

A number of strategies could be adopted by the international community and vulnerable countries to mitigate the short and long term impacts of future tsunami events. In the 1998 Papua New Guinea tsunami factors that contributed to higher mortality during this event included concentration of populations in vulnerable areas and failure of residents in affected areas to timely evacuate. In contrast, deployment of medical assistance and international support immediately following the tsunami played a large role in preventing further loss of life ¹³.With respect to public health in particular, a greater focus on ongoing disease surveillance, the appropriate targeting of aid to those in most need, and strengthening of health care systems ¹¹ can help to mitigate the medium to long term health impacts of tsunamis. Advanced tsunami warning systems may vastly improve early detection, and education campaigns can play a crucial role in improving awareness about tsunami risk and mitigation ¹⁰. Policies enacted by the Sri Lankan government following the 2004 tsunami highlight some of the challenges to longer term disaster mitigation where enactment of a buffer zone policy that forced the relocation of coastal communities had deleterious social and economic impacts ¹². In adopting disaster mitigation policy, governments should consider the contribution not only of physical vulnerability (i.e. distance of communities from coastal areas) but also social vulnerability, such as livelihoods, to increased disaster risk and the potential short and long term impacts such policies may have on affected communities.

Limitations

Systematic reviews face numerous limitations. The effects of tsunamis are the subject of gross approximations and aggregations which result in a great deal of imprecision. The availability and quality of data has likely increased and improved over time, however, in many events deaths are unknown or unrecorded. For a significant number of events no data are reported for injured, displaced, and affected populations; this likely contributes to a substantial underestimation of the impacts of tsunamis on human populations. Inconsistencies and errors were common in the data files from the two different sources, and in some cases inclusion criteria were not ideal for the purposes of this review which created a challenge in reconciling event lists. Additionally, mainly due to the small number of tsunami run-up events reported over the study period, we were unable to perform more complicated statistical analyses that would have provided estimates of the independent effects of place and event characteristics on tsunami related death. When combined with the relatively small number of tsunami events, uncertainty in the historical record limits the conclusions that can be drawn about the impact of tsunamis on human populations. A principal limitation of the literature review is the fact that only English language publications were included; this likely contributed to incomplete coverage of studies published in other languages originating from low and middle income countries.

Conclusions

From 1900 to 2009, a total of 250,000 tsunami-related deaths and close to 50,000 injuries, respectively, were reported, the majority of which were concentrated in the 2004 Indian Ocean tsunami. An estimated 2.5 million people were affected by tsunamis during this time period. While mortality estimates presented in this study are consistent with those reported in other studies, particularly for the 2004 Indian Ocean Tsunami, the injury figure may be an underestimate of the true value given low reporting levels. The distribution of tsunami related deaths varied greatly by region and economic development level. Findings from the historical event review indicate that the South East Asian region and poorer countries were more likely to experience higher mortality was associated with larger wave height and closer proximity to the source.

The primary cause of tsunami-related mortality was drowning and, although a number of injury types were reported following tsunamis, the ratio of dead to injured is much greater in tsunamis as compared to other natural disaster types. Risk factors for tsunami-related death included female sex and very young and old age. Tsunami losses are likely to increase in future years due to population growth in high risk seismic areas. Increased attention to tsunami prevention and mitigation strategies, with a focus on areas most prone to tsunamis and populations at greater risk is necessary. While strategies that are specific to the development level and country context are important, global initiatives such as early warning systems are essential for further tsunami risk mitigation.

Competing Interests

The authors have declared that no competing interests exist.

Correspondence

Shannon Doocy, Johns Hopkins Bloomberg School of Public Health, 615 N. Wolfe St, Suite E8132, Baltimore, MD 21230. Tel. 410-502-2628. Fax: 410-614-1419. Email: [email protected].

Methods

Data on the impact of volcanic events were compiled using two methods, a historical review of volcanic events and a systematic literature review for publications relating to the human impacts of volcanic eruptions with a focus on mortality, injury, and displacement.

Historical Event Review

A historical database of significant volcanic eruptions between 1900 and 2009 was created from publicly available data. Multiple data sources were sought to ensure a complete listing of events and inclusion of both human and geophysical factors. The two primary data sources were EM-DAT: The Emergency Events Database ³ and the National Oceanic and Atmosphere Administration – National Geophysical Data Center (NOAA-NGDC) Significant Volcanic Eruption Database ¹¹. For an event to be included in the EM-DAT database, one or more of the following criteria must be fulfilled: 10 or more people killed or injured; 100 people reported affected; declaration of a state of emergency; or a call for international assistance. In the NOAA-NGDC database, a significant eruption must meet one or more of the following criteria: caused fatalities; caused moderate damage (approximately $1 million or more); caused a tsunami; or was associated with a major earthquake.

Event lists from both databases were downloaded in July 2007 and August 2009. The reporting format for EM-DAT changed during this time period and revisions were made to some records in the NOAA-NGDC database. Event lists were reconciled to create a combined list of events for each data source; the EM-DAT list included 209 events, and the NOAA-NGDC list included 229 events. Event lists were then merged to create a complete listing of significant volcanic events between 1900 and 2009. Volcano and eruptive characteristics were abstracted from the Smithsonian Institution’s Global Volcanism Program (GVP) and added to each event ⁷; and data on human impacts were added from the Volcanic Disasters and Incidents Database ². To prevent including events with no direct human impact, records where human impact (mortality, injury, or homelessness/displacement) was not quantified by any source were removed. A limitation of using the NOAA database is that events are reported if they are associated with an earthquake or tsunami, regardless of human impact. Similarly, an emergency declaration is sufficient for inclusion in the EM-DAT database, irrespective of if human populations are actually affected. The final list included 192 events reported by EM-DAT and 192 reported by NOAA; 71 events were reported by both sources yielding a total of 313 volcanic events affecting populations between 1900 and 2009. See https://www.jhsph.edu/refugee/publications_tools/index.html for the database of volcano events.

In order to examine country- and event-specific characteristics associated with low and high levels of volcano mortality, deaths were categorized as follows: low (<10 deaths), medium (11-75 deaths) and high (>75 deaths). Bivariate tests for associations between volcano characteristics and human impacts were performed using χ² (categorical measures) and ANOVA (continuous measures), and a multinomial logistic regression model was used to examine the probability of evacuation in volcanic events given volcano type, time period, and certain eruptive characteristics. All analyses were performed using Stata Statistical Software, Version 11.0 ¹².

Systematic Literature Review

Key word searches in MEDLINE (Ovid Technologies, humans), EMBASE (Elsevier, B.V., humans), SCOPUS (Elsevier B.V., humans), and Web of Knowledge, Web of Science (Thomson Reuters) were performed to identify articles published in July 2007 or earlier that described natural hazards and their impact on human populations. One search was done for all the five natural hazards described in this set of papers. This paper describes the results for cyclones. The systematic review is reported according to the PRISMA guidelines. Key words used to search for natural hazards included natural hazard(s), natural disaster(s), volcano(es), volcanic, volcanic eruption, seismic event, earthquake(s), cyclone(s), typhoon(s), hurricane(s), tropical storm(s), flood(s), flooding, mudslide(s), tsunami(s), and tidal wave(s). Key words included for impact on human populations wereaffected, damage(d), injury, injuries, injured, displaced, displacement, refugees, homeless, wounded, wound(s), death(s), mortality, casualty, casualties, killed, died, fatality, fatalities and had to be used in either the title, abstract or as a subject heading/key word. The search resulted in 2,747 articles from MEDLINE, 3,763 articles from EMBASE, 5,219 articles from SCOPUS, and 2,285 articles from ISI Web of Knowledge. Results from the four databases were combined and duplicates were excluded to yield a total of 9,958 articles.

Title screening was performed to identify articles that were unrelated to natural disasters or human populations. Each title was screened by two independent reviewers and was retained if either or both reviewers established that inclusion criteria were met. To ensure consistent interpretation of inclusion criteria, percent agreement was assessed across reviewers for a small sample of articles, and title screening began after 80% agreement on inclusion was achieved. A total of 4,873 articles were retained for abstract review. Articles that met one or more of the following criteria were excluded in the abstract screening: language other than English; editorial or opinion letter without research-based findings; related to environmental vulnerability or hazard impact but not human populations; individual case report/study; focus on impact/perceptions of responders; and not related to human or environmental vulnerabilities or impacts of hazards. As with the title screening, overall percent agreement between reviewers was assessed and abstract screening began only after achieving 80% agreement on inclusion. Each abstract was screened by two independent reviewers and was retained if either or both reviewers established that inclusion criteria were met. During the abstract review, included abstracts were coded for event type, timeframe, region, subject of focus, and vulnerable population focus. A total of 133 articles were retained for full article review. Articles discussing the impacts of natural disasters on human populations in terms of mortality, injury, and displacement were prioritized for review. A total of 59 articles on volcanic events meeting the aforementioned subject focus criteria were retained for full review. Upon full review, 19 articles were retained including 10 that underwent dual review, standard data abstraction and 6 that were identified as review articles (Figure 1). Following the systematic review, a hand search was conducted using the databases and key words listed above to identify relevant articles published between July 2007 when the initial search was conducted and October 2012; one additional article was identified that met criteria for full review. In total, 10 articles with primary data relating to risk factors for mortality, injury or displacement were identified (Table 1) and summaries of (n=6) review articles are presented in Tables 2.

Results

Historical Event Review

Overall, an average of 3 (range 1-10) volcanic eruptions affecting human populations occurred annually. When trends in reporting were assessed by source, the number of events reported annually by NOAA (range 8-27) was more consistent than EM-DAT (range 2-53), where the frequency of reported events increased over time and in particular after 1970 (Figures 2 & 3). The impact of volcanic events across regions from 1979 – 2009 is summarized in Figure 4. The World Health Organization defined regions of the Western Pacific (WPRO), Americas (AMRO) and Southeast Asia (SEARO) each accounted for more than 20% of volcanic eruptions, while both the Africa (AFRO) and the European (EURO) regions each accounted for less than 10% of eruptions; no volcanic eruptions were reported in the Eastern Mediterranean (EMRO) region. When deaths were assessed, the vast majority occurred in the AMRO region, which had 73% of reported deaths for 1900-2008; significant minorities of deaths were reported in the SEARO (16%) and WPRO regions (8%). WPRO was the region with the largest affected, which comprised approximately half of the global affected population in each time period. The AMRO region, which reported the greatest number of volcanic events and deaths for both time periods, had less than 20% of the total affected population. The overall impact of volcanic events on human populations is summarized in Table 3.

Mortality and Injury. When mortality data from the three sources were combined, deaths were reported in 76% (n=237) of eruptions since 1900. Overall, 91,834 deaths (range 85,169-102,372) resulting from volcanic eruptions were reported in the historical event review. For eruptions where mortality was reported, there was a median of 6 (mean = 536; 5% trimmed mean= 117; range 1-30,000) deaths per eruption when using the highest reported death toll. The ten most deadly eruptions accounted for 81.1% of reported volcano mortality (Table 4). Deaths were concentrated in Martinique (30.3% or 31,023 deaths, 3 events), Colombia (23.5% or 24,099 deaths, 7 events), Indonesia (18.4% or 18,840, 70 events) and Guatemala (12.2% or 12,522 deaths, 10 events). Injuries were reported in 59 (18.8%) events where a total of 14,726 injuries (range 11,549-17,917) were documented. In eruptions where injuries were reported, there was a median of 11 (mean = 304; 5% trimmed mean=93, range 1-10,000) injuries per eruption when the highest reported figure was used. To estimate the total number of injuries due to volcanic eruptions, it was presumed that injuries would occur in events where deaths were reported. A total of 237 eruptions with fatalities occurred were reported; when the median and 5% trimmed mean for injuries were applied to the remaining 178 events with fatalities, it was estimated that between 1,958 and 16,643 unreported volcano related injuries may have occurred between 1900 and 2008.

Data to compute ratios of deaths to injured and affected populations was available for 58 and 130 historical events, respectively. Wide variation in ratios was observed, presumably because of variation in reporting quality. For 1900-2008, the median ratio of dead: injured was 0.63 (mean 2.53, range 0-36). When compared to the affected population, there was a median of 0 deaths per 1000 affected (mean = 96, range 0-6709). Point estimates for both ratios decreased by approximately half in the period of 1980 to 2008, which may suggest improvements in disaster response and preparedness; however, these changes were not statistically significant. Multinomial logistic regression for predictors of evacuation indicates that time period of the eruption (among other factors) is significantly associated with evacuation, where eruptions occurring in 1980 and later were 20.25 (CI: 8.01-51.21) more likely to have an evacuation reported as compared to those between 1900-1939, which also suggests improvements in preparedness and response (Table 5).

Affected Population and Other Impacts. An estimated 4.7 million people were reported to be affected by volcanoes between 1900 and 2009, including 4.2 million that were rendered homeless. However, these figures are likely to substantially underestimate the true impact of volcanoes on human populations because estimates of the total affected population and the homeless population were reported in 49.8% (n=156) and 43.3% (n=135) of events, respectively. The distribution of the affected population was highly skewed: on average 22,698 people were affected in an eruption, though the median affected population was significantly lower at 3,000. Other measures of effect on human populations reported by GVP included evacuation and damages in excess of one million dollars; these data were reported for 296 (94.6%) of events. Damage in excess of one million dollars were reported in 75.0% of events (n=222/296) and evacuation was reported in 56.5% (n=177/296) of events.

Eruptive Characteristics. Eruptive characteristics were reported by GVP for 296 (94.6%) eruptions in the event file. When assessed by volcano type, proportions were as follows: stratovolcano 73% (n=228), complex volcano 13% (n=39); shield volcano 7% (n=22); caldera 4% (n=11); cinder cone 2% (n=4); and others 2% (n=4). VEI is a logarithmic scale (0-8) indicating the amount of mass ejected during an eruption and the size of the eruption column. ¹³ In general, VEI 1-3 events generate localized hazardous phenomena and VEI 4-5 events have the potential for disruption on a regional scale; VEI 6+ events can affect the entire planet via their impact on global climate ¹⁰. The average VEI of eruptions was 2.60 (Table 6). In a multinomial regression model, an average of 362 additional deaths (CI: 57-667, p=.020) were associated with a one unit increase in VEI after controlling for evacuation.

Systematic Literature Review

Mortality. A review by Tanguy (1998) estimates 221,907 deaths from major volcanic eruptions over the past two centuries (1783-1997), which includes 90% of recorded deaths throughout history. Volcano related mortality was attributed to primary volcanic phenomena (see table 7 for definitions) including ash and pyrolcastic flows (36%), mudflows or lahars (17.1%), volcanogenic tsunamis (16.9%), and post-eruption epidemics or famines (30.3%). Of the 79,286 deaths due to primary volcanic phenomena, 75.0% were attributed to pyroclastic flows or surges and magma-generated sector collapse, followed by debris avalanches (12.6%), ash fall including ballistic projectiles (11.6%), and lava flows (0.8%). Four volcanic eruptions (Tambora, 1815; Krakatau, 1883; Mt. Pelee, 1902; and Ruiz, 1985) accounted for more than 66% of deaths, and the principal causes of death, including famine, tsunamis, pyroclastic flows, and lahars varied in each event. While famine and epidemics were the principal causes of mortality in earlier volcanic events, they currently present much less of a risk due in large part to international relief and humanitarian assistance efforts ¹⁴.

Another database developed by Witham (2005) considers all volcano-related incidents (n=491) with human impacts between 1900 and 1999, including 296 (60%), which were classified as disasters. Witham found the number of people impacted per event was substantially greater in middle income than in high income countries, with middle income countries reporting the highest numbers of volcano related deaths, injuries, and affected populations ². While fatalities were the most frequent outcome, observed in 53% of events, the largest consequence was displacement, which accounted for 94% of the affected population. Similar to historic findings, the study found fatalities to be concentrated in relatively few large events. The eruptions of Mt. Pelee (1902) and Nevado del Ruiz (1985), which together accounted for more than 50% of deaths, and the top ten events combine to account for more than 90% of deaths. During the 20^th century, pyroclastic flows were the primary cause of death followed by lahars (which were also the principal cause of injury), and tephra was a primary cause for evacuation and displacement ². Tephra, which includes fragmental material from volcanoes and volcanic ash, is also a common cause of trauma-related deaths either by projectile impact or collapse of ash-covered roofs ¹. The three main causes of direct mortality in recent volcanic eruptions include asphyxiation from ash, thermal injuries from pyroclastic flow, and trauma (Table 2) ^15,16. A review of health hazards from volcanic gases indicates that volcanic gases account for less than 1% to 4% of mortality, though this is a likely underestimate because studies neglect volcanic degassing when unassociated with eruptive activity, and because the extent of dispersion of volcanic gases is not always appreciated ¹⁷. Historical estimates in the 20^th Century suggest that 2,000 deaths have resulted from volcanic gases, with the most hazardous volcanogenic gases being carbon dioxide, sulfur dioxide, hydrogen sulfide, mercury, and radon ¹⁸.

Morbidity. Transient increases in trauma-related injuries resulting from traffic accidents and falls, and morbidity, primarily ocular irritations and respiratory symptoms, are observed following volcanic eruptions. Increases in communicable diseases and long-term health effects are not attributed to volcanic eruptions, however, volcanism associated morbidity is likely underestimated ⁹. Ash fall can have health implications for populations as far as hundreds of kilometers away ¹⁹. A recent review of the respiratory effects of volcanic ash suggests that they are short-lived and dependent on the mode of ash generation and particle size, among other factors ²⁰. De novo appearance of ash-related asthma has not been documented, though increases in respiratory symptoms are frequently reported. Higher levels of ash exposure have been associated with increased reports of respiratory symptoms among children ²¹ and in high-exposure occupation groups ^22,23. Increases in emergency room visits for respiratory problems have been observed following recent volcanic eruptions in the United States and Ecuador, where children under five ²⁴ and individuals with pre-existing lung conditions such as asthma or chronic bronchitis were at increased risk for the development of respiratory symptoms ^20,22,23. Recently, adverse cardiorespiratory health effects have been associated with chronic exposure to volcanogenic gases in both Hawaii and Montserrat ^25,26 and historically, where one study suggested that adverse health symptoms in Europe during the summer of 1783 was associated with volcanic gases and fogs ²⁷. Review findings indicate that volcanic eruptions are associated with short-term increases in morbidity, primarily ocular irritations and respiratory symptoms.

Discussion

Main Findings

An increasing trend in the total population affected by volcanoes each decade is observed after 1950; however, this is not unexpected considering this information is reported by EM-DAT where the number of events reported increased over time. Other factors likely contributing to this observation are improved data reporting quality and population growth, which has resulted in a larger population and greater levels of development in at risk areas. This is poses concerns, where lower magnitude eruptions in areas that have experienced significant land use change and high population growth may have greater impacts than anticipated when projections are based on the historical record alone. The 1985 eruption of the Colombian volcano Nevado Del Ruiz is an example of increasing human vulnerability to volcanoes due of population growth in high-risk areas. The town of Armero, which was completely buried by lahars in 1985, experienced similar lahars in 1595 and 1845 and in both instances the community was rebuilt and population expanded ²⁸. In addition to population growth and land use change, globalization is an emerging factor, which may contribute also to increased vulnerability to natural disasters and result in consequences that span larger geographic regions ²⁹.

The increasing frequency of volcanic incidents and impacts on human populations has been documented elsewhere and may reflect numerous factors, including increased reporting, increased use of evacuation in risk mitigation, growth in the population at risk, and actual changes in global volcanic activity ^1,2. A total of 102,373 volcano related deaths and an average of 430 deaths per eruption were reported between 1900 and 2008 when the highest mortality estimate from any source in the historical event review was used (102,140 deaths were reported during the 20^th century). Other estimates of 20^th century volcano mortality range from 91,724 to 98,376, with an average of 845 to 917 deaths per eruption ^2,14. Mortality was concentrated in several significant eruptions, most notably the 1902 eruption of Mt. Pelee on the Island of Martinique which resulted in 28,000 to 30,000 deaths and the 1985 eruption of Colombia’s Nevado Del Ruiz where a lahar caused an estimated 21,800 to 23,080 fatalities. Overall, 45% of volcanic deaths between 1900 and 2008 were reported in the 1900s, 29% in the 1980s, and the remaining 26% in other decades.

Historically, pyroclastic flows have accounted for the majority of mortality in volcanic events, which suggests they will continue to be the most lethal volcanic agent in the near future. It has been widely assumed that pyroclastic flows are unsurvivable, but evidence from eruptions in the 20^th century challenges this assumption, especially at the periphery of flows where impacts are attenuated and protection can be provided by resistant buildings ³⁰. Modeling and reviews of volcanic deaths and injuries in pyroclastic flows from recent decades suggests that survival is possible under certain conditions. Survival limits on heat exposure, inhalation of hot air, and air containing hot particles have been established, and simulations show that in distal areas survival is possible. Close to the volcanic crater, conditions are unsurvivable due to heat, high particle concentrations, and elevated dynamic pressures. In distal areas of pyroclastic flows, death and injury are mainly related to asphyxiating levels of particles; reduced temperature and lower dynamic pressures increase the probability of survival, especially in masonry buildings which should be the most resistant to the impact of pyroclastic flows ³¹.

Since 1980, volcanic disasters have resulted in nearly 30,000 deaths, two-thirds of which were preventable and associated with a single event (Nevado del Ruiz, 1985) which suggests that improved mitigation measures have the potential to reduce loss of life in future eruptions. Effective communication with civil authorities and vulnerable populations and education about the threats posed by volcanoes is equally important as volcano monitoring and forecasting. Initiation of volcano awareness programs in volcanic hazards prone areas and contingency planning are also of central importance, particularly as the size of the population living close to active volcanoes increases ¹⁰. Accurate forecasting of the eruption force and prediction of its occurrence by volcanologists and timely evacuation of populations at risk are vital to effective emergency management, and risk assessment is an integral part of disaster preparedness ³¹. Monitoring has a critical role to play in reducing the impact of volcanic hazards by providing early warning and possibly identifying timeframes of forthcoming eruptions ¹⁰. In 1994, just more than 20% of the approximately 550 active volcanoes were monitored, and the extent to which monitoring has increased remains unclear ⁷. The crucial role of monitoring volcanic activity was exemplified during eruptive activities of Mt. St. Helens and Pinatubo when early warning signs were evident enough to start evacuation of population, involve emergency services, which ultimately minimized the number of victims.

While a wide range of geophysical, geochemical and geodetic techniques are available ^32,33 the ‘core methods’ of seismic and ground deformation monitoring are the best means of tracking magma movement and accumulation. Increased monitoring, either via satellite platforms capable of detecting pre-eruptive ground deformation ³⁴ and thermal anomalies ³⁵ or ground monitoring is of vital importance. Geodetic and oceanographic surveys of active submerged volcanic cones have recently been employed to assess the state of their hydrothermal activity, sample gases and volcanic rocks ^36,37. Periodic assessment of changes in topography of submerged volcanic cones and their geochemical activity will provide early warning signs of possible phreatic explosions. Comprehensive monitoring of volcanoes in conjunction with hazard zonation maps offer the best means of reducing casualties, primarily through ensuring evacuation of people from the threatened areas ¹⁰. Hazard zonation maps show areas of potential volcanic impacts, such as possible propagation of pyroclastic or lava flows, locations of eruptive centers, areas of probable landslides, tsunami effects and ash fallout ³⁸. These maps are invaluable tools in planning of mitigation measures (Table 7). Evidence from recent eruptions suggests that emergency planning for explosive eruptions in urban areas should concentrate on distal areas of predicted pyroclastic flows and areas where the primary risk is death due to asphyxiation from ash inhalation rather than death due to injury. This includes the potential need to rescue survivors suffering from ash inhalation as well as trauma; a smaller number of victims may require treatment for dermal and airway burns, though their numbers may easily exceed regional capacity for treatment ³¹.

Limitations

Systematic reviews face numerous limitations. The effects of volcanic eruptions are the subject of gross approximations and aggregations that have a great deal of imprecision. The availability and quality of data has likely increased and improved over time, however, in many events deaths are unknown or unrecorded. For a significant number of events no data are reported for injured, displaced, and affected populations; this likely contributes to a substantial underestimation of the impacts of volcanoes on human populations. Inconsistencies and errors were common in data files from different sources, and in some cases inclusion criteria were not ideal for the purposes of this review, which created a challenge in reconciling event lists. When combined with the relatively small number of recent volcanic events, uncertainty in the historical record and the relative paucity of primary research focusing on the health-related topics significantly limits the conclusions that can be drawn about volcanic impacts on human populations. A principal limitation of the literature review is the fact that only English language publications were included; this likely contributed to incomplete coverage of studies published in other languages originating from low and middle income countries

Conclusions

The impact of volcanoes on humans in terms of mortality, injury, and affected populations, presented here is a minimum estimate because information for many volcanic events is either unknown or unreported. Data from 1900 to the present suggest that volcanoes have exacted a relatively small toll on the human population when compared to other natural disasters. However, human vulnerability to volcanic hazards is increasing, in large part due to land use change and particularly to the development of densely populated urban areas in close proximity to volcanoes.

Medical treatment clearly has a limited role in volcanic eruptions because severe injuries occur only in a relatively concentrated area, and morbidity experienced in more outlying areas is limited in both severity and duration. A strong emphasis on preparedness strategies is required, and because many major volcanic eruptions are preceded by warning signs, it is possible to plan for these events, which is evident given the increasing trend in evacuations. Because volcano fatalities are concentrated geographically in relatively few eruptions, targeted preparedness efforts in areas that are historically at risk as well as those newly identified via monitoring could be successful. Hazard evaluations for all volcanoes in populated areas, regardless of their active or dormant state, and expanded monitoring could improve preparedness levels. Hazard-specific mitigation strategies such as engineering projects or urban planning could be implemented to reduce potential impacts; however, their costs may be prohibitive when compared to the likelihood of an eruption in the near future. Broader-based awareness and education strategies targeted at the population at risk would likely result in more successful evacuations and may also increase willingness of authorities to implement more costly preparedness measures.

Competing Interests

The authors have declared that no competing interests exist.

Correspondence

Shannon Doocy, Johns Hopkins Bloomberg School of Public Health, 615 N. Wolfe St, Suite E8132, Baltimore, MD 21230. Tel. 410-502-2628. Fax: 410-614-1419. Email: [email protected].

Methods

Data on the impact of flood events were compiled using two methods, a historical review of flood events and a systematic literature review for publications relating to the human impacts of flooding with a focus on mortality, injury, and displacement.

Historical Event Review

A historical database of significant floods occurring from 1980 to 2009 was created from publicly available data. Multiple data sources were sought to ensure a complete listing of events, to allow for both human and geophysical factors to be included, and to facilitate cross checking of information between sources. The two primary data sources were CRED International Disaster Database (EM-DAT) ⁴ and the Dartmouth Flood Observatory (DFO) Global Archive of Large Flood Events database ¹³. For inclusion in the EM-DAT database, one or more of the following criteria must be fulfilled: 10 or more people killed or injured; 100 people affected; declaration of a state of emergency; or a call for international assistance. The DFO database provides a comprehensive list of flood events recorded by news, governmental, instrumental, and remote sensing sources from 1985 to 2009. Inclusion criteria are: significant damage to structures or agriculture, long intervals since the last similar event, or fatalities. Flooding specifically related to hurricane storm surge and tsunamis were excluded.

Event lists from both databases were downloaded in July 2007 and merged to create a single database; the database was updated in August 2009. The EM-DAT and DFO databases included 2,678 and 2,910 events, reported, respectively, between 1980 and 2009. Both EM-DAT and DFO reported the date and location of the event, the affected region and the number dead. In addition, the number affected, homeless, and total affected (sum of injured, homeless, and affected) were reported by EM-DAT. DFO also reported the number displaced, duration of the event (days), and ‘flood magnitude.’ Flood magnitude is a composite score of flood severity developed by DFO that encompasses damage level, recurrence interval, duration of the flood in days and the area affected ¹³. For flood impacts reported by EM-DAT, zeroes were treated as missing values because they were used as placeholders and their inclusion in the analysis could contribute to the under estimation of tsunami impacts. The final list included 2,678 events reported by EM-DAT and 2,910 reported by DFO; 1,496 events were reported by both sources yielding a total of 4,093 flood events affecting human populations. See https://www.jhsph.edu/refugee/natural_disasters/_Historical_Event_Review_Overview.html for the database of flood events.

To assess risk factors for flood-related mortality the following categories were used: no deaths (0 deaths), low (1-9 deaths), medium (10- 49 deaths) and high (≥50 deaths). Bivariate tests for associations between flood mortality and the following characteristics were performed using χ² (categorical measures) and ANOVA (continuous measures): decade, region (defined by the World Health Organization (WHO)), income level (World Bank), gross domestic product (GDP), GINI (measure of income inequality), and flood magnitude. All covariates, with the exception of GINI, which was not strongly associated with flood mortality in adjusted analyses, and GDP, which was highly correlated with per capita World Bank income level, were included in the final multinomial logistic regression model to assess the relative risk of mortality at a given level as compared to events with no deaths. All analyses were performed using Stata Statistical Software, Version 11.0 ¹⁴.

Systematic Literature Review

Key word searches in MEDLINE (Ovid Technologies, humans), EMBASE (Elsevier, B.V., humans), SCOPUS (Elsevier B.V., humans), and Web of Knowledge, Web of Science (Thomson Reuters) were performed to identify articles published in July 2007 or earlier that described natural hazards and their impact on human populations. One search was done for all the five natural hazards described in this set of papers. This paper describes the results for cyclones. The systematic review is reported according to the PRISMA guidelines. Key words used to search for natural hazards includednatural hazard(s), natural disaster(s), volcano(s), volcanic, volcanic eruption, seismic event, earthquake(s), cyclone(s), typhoon(s), hurricane(s), tropical storm(s), flood(s), flooding, mudslide(s), tsunami(s), and tidal wave(s). Key words included for impact on human populations were affected, damage(d), injury, injuries, injured, displaced, displacement, refugees, homeless, wounded, wound(s), death(s), mortality, casualty, casualties, killed, died, fatality, fatalities and had to be used in either the title, abstract or as a subject heading/key word. The search resulted in 2,747 articles from MEDLINE, 3,763 articles from EMBASE, 5,219 articles from SCOPUS, and 2,285 articles from ISI Web of Knowledge. Results from the four databases were combined and duplicates were excluded to yield a total of 9,958 articles.

A multi-stage screening process was used. First, title screening was performed to identify articles that were unrelated to natural disasters or human populations. Each title was screened by two independent reviewers and was retained if either or both reviewers established that inclusion criteria were met. To ensure consistent interpretation of inclusion criteria, percent agreement was assessed across reviewers for a small sample of articles, and title screening began after 80% agreement on inclusion was achieved. A total of 4,873 articles were retained for abstract review. Articles that met one or more of the following criteria were excluded in the abstract screening: language other than English; editorial or opinion letter without research-based findings; related to environmental vulnerability or hazard impact but not human populations; individual case report/study; focus on impact/perceptions of responders; and not related to human or environmental vulnerabilities or impacts of hazards. As with the title screening, 80% overall agreement between reviewers was needed before abstract screening started. Each abstract was screened by two independent reviewers and was retained if either or both established that inclusion criteria were met. Included abstracts were coded for event type, timeframe, region, subject of focus, and vulnerable population focus. A total of 3,687 articles were retained for full article review. Articles discussing the impacts of natural disasters on human populations in terms of mortality, injury, and displacement were prioritized for review. A total of 119 articles on flood events meeting the criteria were retained for full review. Upon full review, 27 articles were retained including 17 that underwent standard data abstraction and 11 that were identified as review articles (Figure 1).

Following the systematic review, a search was conducted to identify relevant articles published after the initial search up to October 2012. This search identified seven additional articles, including three articles with primary data that underwent full review and four review articles. Summaries of abstracted (n=21) and review articles (n=15) are presented in Tables 1 and 2, respectively.

Results

Historical Event Review

Overall, an average of 131 (range 35-287) floods affected human populations annually with the majority (81%) occurred during or after the 1990s. Part of this increase can be explained by improved reporting and by the DFO reporting beginning in 1985. There was great variation in the number of events reported annually between EM-DAT (range 35-213) and DFO (42-235) (Figure 2). While the frequency of flood events increased gradually over time, their impacts on human populations in terms of mortality and affected populations varied greatly between years and were often concentrated around large-scale events (Figure 3). Using the WHO regions the Americas (AMRO) and Western Pacific (WPRO) regions experienced the most flooding events while the fewest were reported in Europe (EURO) (Figure 4). Deaths were overwhelmingly concentrated in South East Asia (SEARO), which accounted for 69% of global flood mortality, though both the Americas (AMRO) and Western Pacific (WPRO) had significant minorities of flood fatalities. The great majority of the flood affected population was in WPRO (59%) and SEARO (35%) of the global total. Overall, the human impacts of floods in Europe, Africa, and the Eastern Mediterranean regions were limited; together the regions accounted for no more than 8% of flood deaths and 4% flood affected populations, respectively. The overall impact of flooding on human populations is summarized in Table 3.

Affected Population. An estimated 2.8 billion people were reported to be affected by flood events between 1980 and 2009, including nearly 4.6 million rendered homeless. However, these figures likely substantially underestimate the true impact of floods on human populations because estimates of the total affected population and the homeless population were reported in only 64.3% (n=2,632) and 14.9% (n=611) of events, respectively. The distribution of the number affected was highly skewed with mean and median affected populations of 1,071,829 and 6,000 per event, respectively, which indicates that the median affected population may better reflect the impact of a typical flood event.

Mortality and Injury. When mortality data from the two sources were combined, deaths were reported in 96.8% (n=3,960) of floods since 1980. This figure excludes 13.9% of floods where no information on mortality was reported; if no deaths are presumed and these events are included, deaths occurred in 65.3% (n=2,673) of floods. 539,811 deaths (range: 510,941-568,680) resulting from flood events were reported. For floods where mortality was reported, there was a median of 9 (mean=135; range 0-138,000) deaths per event when using the highest reported death toll. Mortality exceeded 10,000 in only 4 events and 100,000 in two. The two deadliest events occurred in Bangladesh (138,000 deaths in 1991) and Myanmar (100,000 deaths in 2008). Injuries were reported in 401 (9.8%) events, where a total of 361,974 injuries were documented. In events where injuries were reported, there was a median of 12.5 (mean=904: range 1-249,378) per flood event. To estimate the total number of injuries due to flood events, it was presumed that injuries would occur in events where deaths were reported. There were 2,673 floods with fatalities but only 401 (9.8%) with injuries reported. When the median and mean for injuries were applied to the remaining 3,077 events, it was estimated that between 38,463 and 2,717,681 additional unreported flood related injuries may have occurred between 1980 and 2009.

Bivariate associations between country-level characteristics and flood-related mortality from 1980 through 2009 are presented in Table 4. Findings suggests that the proportion of events with high mortality (>50 deaths) have decreased over time. Income level was also significantly associated with flood mortality, where for both low and lower-middle income countries, a greater proportion of events fell in the medium and high death categories as compared to higher income countries. Higher mortality events were concentrated in the South East Asian and Western Pacific regions.

Findings from the adjusted analyses (Table 5) modeling the relative risk of flood related mortality show that all predictors were significantly associated with flood mortality. The relative risk of medium- and high-level mortality events compared to events with no deaths significantly decreased over time. There was also a significant decreased relative risk of mortality in excess of 50 deaths for events in higher income countries compared with lower income country events. Additionally, as magnitude of a flood increased, so did the risk of having high mortality when adjusting for all other predictors. A flood rated as high magnitude as compared to one with low magnitude was associated with an increased relative risk of having high mortality as compared to no mortality (RR=13.20, 95% CI 8.25, 22.11). Caution should be taken when interpreting such findings, however, as magnitude estimates were missing for a large proportion of events, and missing magnitude was associated with the outcome in this study. Regional differences in reported mortality were also supported by the analysis. Higher mortality events were concentrated in the South East Asian and Western Pacific regions, compared to events occurring in the Americas (Southeast Asia RR=3.35, 95 CI: 2.21, 5.72; Western Pacific RR=2.38, 95 CI: 1.62, 3.34).

Systematic Literature Review

Mortality. Fourteen of the reviewed articles reported mortality data including ten that provided information on direct or indirect causes of mortality and/or risk factors for flood-related deaths (Table 6) ^{15,16,17,18,19,20,21,22,23,24,25,26,27,28} . Most articles provided some information about the distribution of deaths across population subgroups (i.e. gender, age) and/or an individual’s location at the time of the event; seven of these ten articles reported on floods in the United States. Nearly all articles reporting cause of death cited drowning as the most frequent cause of death ^{1,15,18,19,20,22,29} . Cumulatively, drowning accounted for 75% of deaths; other causes of death included falls, electrocution, heart attack, hypothermia, trauma, snake bites, and carbon monoxide poisoning.

All studies in the United States examined mortality related to motor vehicles and found an increased risk of mortality among individuals in motor vehicles during the event, of all deaths 74% were motor vehicle related ^17,18,19,20 . This compares to a motor vehicle related death rate of 63% in a recent review of US flood fatalities between 1959 and 2005 ⁷. Higher proportions of deaths among males (64%) were consistently observed in the United States, except for Puerto Rico where 57% (13/23) of flood related fatalities were female and hurricane Katrina where deaths evenly divided between the sexes (51% male, 49% female) ^{16,18,19,20,28} . In contrast, the one article describing flood mortality in the less developed country of Nepal found that females of all age groups faced increased mortality risk and 58% of all deaths were women ²³ Other factors found to be associated with flood-related mortality included storm course/time storm hit landfall ^19,22 summer months ^17,30, low socioeconomic status ²³, poor housing construction ^16,23,24,31 and timing of warning messages ^19,22.

Injury and Displacement. Injury or morbidity data were reported in ten of the 18 included articles, of which nine provided information on injury type and/or risk factors ^{15,16,24,32,33,34,35,36,54} . The majority of flood-related injuries are minor. The two studies that captured a large number of injuries, both in the United States, found that musculoskeletal injuries were most common (46% and 34%), followed by lacerations (21% and 24%). Other flood-related injuries included abrasions and contusions, motor vehicle related injuries, and falls ^33,34,54. In less developed settings, increased incidence of snake bites and fires were also cited as causes of injury or death ^2,36. Among care seekers in flood-affected areas of Bangladesh 5.1% of wounds were infected. Another review suggested that the proportion of survivors requiring medical attention is less than 2% ². A distribution of injuries across population subgroups was reported by only one study in India which found that injuries were more common in males (67% vs. 33%), that the 11-40 year age group comprised 68% of the injured, and that those age 50 and above accounted for 18% of flood deaths ³⁴. Seven articles reported displacement or evacuation figures however none described risk factors associated with flood-related displacement ^{15,17,21,24,25,35,37} .

Discussion

Main findings

In the past 30 years approximately 2.8 billion people have been affected by floods with 4.5 million left homeless, at approximately 540,000 deaths and 360,000 injuries, excluding an estimated 38,000 to 2.7 million injuries that went unrecorded. While the mortality estimate presented in this study is consistent with the range of estimates presented in other studies ^1,38, approximations of numbers injured and displaced are likely gross underestimates of the true values given the infrequency with which figures are reported. Floods events with high levels of mortality are relatively rare: despite their increasing frequency, there were only four events with >10,000 deaths and 58 events with >1000 deaths between 1977 and 2009. A slight decrease in the average number of fatalities per event was observed which is in keeping with broader natural disaster trends that show an increase in the size of the affected population and a decrease in the average number of deaths per event ⁴. Higher numbers of fatalities were reported in flash floods than river floods, however, river floods affected larger populations and land areas ^3,7. Lower mortality rates in river floods can mostly be attributed to their slower onset allowing for longer time for warning and evacuation ^3,39. The widespread use of effective early warning methods for hydrological events has likely contributed declining flood mortality.

Findings from the historical event review are consistent with previous observations that flood mortality varies by region, economic development level, and the severity of the event ^12,40. The majority of flood-related deaths are concentrated in less developed and heavily populated countries, with Southeast Asia and the Western Pacific region experiencing the highest risk of flood-related deaths. Flood mortality rates are relatively similar across continents, but Asian floods kill and affect more people because they affect substantially larger areas with larger populations ³. At the country level, lower GDP per capita was linked to higher mortality, which is in keeping with the established relationship between poverty and increased disaster risk ⁴¹. Human and social vulnerabilities and inequalities, urbanization, population density, terrain and geo-physical characteristics and variation in the frequency and precipitating causes of floods by region are also factors that contribute flood risk levels ^3,6,12,42 . Temporal changes and development trends have also contributed to changing influences of some of these factors over time ⁴². Economic development increases the risk of disaster-related economic losses however improved emergency preparedness, response, and coping capacity may reduce disaster vulnerability ³. That countries with greater resources are able to better predict and respond to impending flood events suggests that building systems and capacity to detect and respond to floods in less developed countries should be a priority ⁴⁰.

Causes of and risks for flood-related mortality and injury identified in the systematic literature review are consistent with previous reviews on the human impact of flooding ^1,29,43,44 . In comparison, a recent review of 13 flood events in Europe and the United States found that 68% of deaths were due to drowning, 12% trauma, 6% heart attack, 4% fire, 3% electrocution, 1% carbon monoxide poisoning, and 7% other/unknown ¹. Studies reporting the gender breakdown for flood-related deaths, most of which are accounts of flood events in the United States, consistently show a greater proportion of males as compared with female deaths. These observations are aligned with previous studies, including a review of flood events in Europe and the US which estimated that males account for 70% of flood related deaths ^1,44,45,46 . While limited to only a few countries, these findings suggest there may be increased mortality risk for males in more developed settings and for females in less developed countries ^23,47 . An increased risk of death in younger and older populations was also observed which is consistent with broader natural disaster mortality trends ^{7,45,46,48,49} . In Nepal, children had the highest crude mortality rates of all age groups and were nearly twice as likely to die in the flood as their same-sex parent ²³. However, recent reviews of age-specific risk for flood mortality have been inconclusive because attempts to aggregate data were hampered by high proportions of deaths where age is unreported ¹. While the prevailing notion is that women and children are more vulnerable in disasters ⁵⁰, there is a paucity of research in less developed countries where the majority of flood deaths occur. Future research on the human impacts of floods should focus on these less developed settings, most notably Asia where flood deaths are concentrated, with the aim of identifying the most at-risk and vulnerable population sub-groups to better target early warning and preparedness efforts.

The ecological nature of the study of event characteristics did not allow for an examination of specific factors within a country or region that may be associated with increased mortality following a flood event. Population density in coastal regions, which are particularly vulnerable to flooding, is twice of the world’s average population density and many of the world’s coasts are becoming increasingly urbanized ⁵¹. Currently, 50.6% of the world’s population lives in urban settings; by 2050 this figure is projected to increase to 70% with the majority of urbanization occurring in less developed regions of Asia and Africa ⁵². Unabated urbanization and land use changes, high concentrations of poor and marginalized populations, and a lack of regulations and preparedness efforts are factors that will likely contribute to an increasing impact of floods in the future ³⁸. From the natural hazard perspective, climate change is also likely to contribute to future increases in flooding. Increased frequency of intense rainfall, as a result of higher temperatures and intensified convection will likely lead to a rise in extreme rainfall events, more flash floods and urban flooding due to excessive storm water. Additionally, sea level rise and increasing storm frequency will lead to additional storm surges in coastal areas while seasonal changes, notably warmer winters, will contribute more broadly to increased precipitation and flood risk ³⁸. Together, changes in socioeconomic, demographic, physical terrain features and climatologic factors suggests that floods will become more frequent and have greater effects on human populations in the coming decades.

Given that flood losses are likely to increase in future years, increased attention to flood prevention and mitigation strategies is necessary. To date, early warning systems have been an effective mechanism for reducing the impact of floods ³⁸, however, they are not ubiquitous and should be prioritized in less developed countries with large at-risk populations and high frequencies of flooding. It is important that messaging and targeted communication strategies accompany early warnings so that the population understands the impending risk and can respond appropriately. Many flood fatalities are associated with risk-taking behaviors, thus messages to avoid entering flood waters and to curtail risky activities in all stages of the event may be successful in reducing flood fatalities ¹. Additional, improved land use planning and regulation of development can mitigate flood impacts. Studies on the relationships between flood losses, natural hazard characteristics, and societal and demographic vulnerability factors can aid in informing and prioritizing flood prevention and mitigation strategies. Finally, comparisons of the effectiveness of different policies and mitigation strategies can inform future strategy and policy actions and ensure they are appropriate in specific contexts.

Limitations

The effects of flood events are the subject of gross approximations and aggregations that have a great deal of imprecision. The availability and quality of data has likely increased and improved over time and the use multiple data sources increased reporting. However, in many events deaths are unknown or unrecorded; for other outcomes such as injured and affected, reporting frequency is even lower which likely contributes to a substantial underestimation of the impacts of flood events on human populations. While available data is sufficient for a cursory analysis of global flood impacts and trends, improved reporting of flood outcomes, including the development of national systems capable of more accurately reporting mortality and injury would be beneficial. Regarding the measures used in this study, our multivariable model included a broad classification of income level according to the World Bank, as opposed to GDP. While we believe GDP to be a more precise measure of wealth, it was nonetheless excluded in the analysis because we did not obtain GDP estimates that were time specific to each event. Inconsistencies and errors were common in data files from different sources, and in some cases inclusion criteria were not ideal for the purposes of this review, which created a challenge in reconciling event lists. For example, the 2004 Asian tsunami was classified as a flood by Dartmouth but not by EM-DAT; this event was ultimately removed from the data set, however, it represented the highest mortality event in the study period, which has potentially important implications for analysis. Consistent definitions and categorization of events across sources such as that initiated by EM-DAT in 2007 would be useful for streamlining future analysis and comparing the impacts of different types of flood events. Other principal limitations of the literature review are 1) that an in-depth quality analysis of all reviewed articles was not undertaken, and 2) the fact that only English language publications were included which likely contributed to incomplete coverage of studies published in other languages originating from low and middle income countries.

Conclusions

Interpretation of flood fatality data is challenging given the occurrence of occasional extreme events, temporal trends and the completeness and accuracy of available data. The continuing evolution of socio-demographic factors such as population growth, urbanization, land use change, and disaster warning systems and response capacities also influences trends. Between 1980 and 2009 there were an estimated 539,811 deaths (range 510,941 -568,584) and 361,974 injuries attributed to floods; a total of nearly 2.8 billion people were affected by floods during this timeframe. The primary cause of flood-related mortality was drowning. In developed countries being in a motor-vehicle at the time of a flood event and male gender were associated with increased mortality risk. Female gender may be linked to higher mortality risk in low-income countries. Both older and younger population sub-groups also face an increased mortality risk. The impact of floods on humans in terms of mortality, injury, and affected populations, presented here is a minimum estimate because information for many flood events is either unknown or unreported.

Data from the past quarter of a century suggest that floods have exacted a significant toll on the human population when compared to other natural disasters, particularly in terms of the size of affected populations. However, human vulnerability to floods is increasing, in large part due to population growth, urbanization, land use change, and climatological factors associated with an increase in extreme rainfall events. In the future, the frequency and impact of floods on human populations can be expected to increase. Additional attention to preparedness and mitigation strategies, particularly in less developed countries, where the majority of floods occur, and in Asia, a region disproportionately affected by floods, can lessen the impact of future flood events.

Competing Interest

The authors have declared that no competing interests exist.

Correspondence

Shannon Doocy, Johns Hopkins Bloomberg School of Public Health, 615 N. Wolfe St, Suite E8132, Baltimore, MD 21230. Tel: 410-502-2628. Fax: 410-614-1419. Email: [email protected].

Methods

The impact of earthquakes events was summarized using two methods, a historical review of earthquake events, and a systematic literature review for publications relating to the human impacts of earthquakes with a focus on mortality, injury, and displacement.

Historical Event Review

A historical database of significant earthquakes between 1980 and 2009 was created. Four publically available data sources were used to create the most complete possible listing of events, allow for inclusion of both human and geophysical factors, and enable cross checking. The two primary sources were the Centre for Research on the Epidemiology of Disasters International Disaster Database (CRED EM-DAT)⁷ and the National Oceanic and Atmospheric Administration’s National Geophysical Data Center (NOAA-NGDC) Significant Earthquakes Database ⁸. Earthquakes included in EM-DAT met one or more of the following criteria: ≥10 deaths; ≥ 100 affected; declaration of a state of emergency; or a call for international assistance. Earthquakes included in the NOAA-NGDC database met one of the following criteria: ≥10 deaths; moderate damage (US$ 1 million or more); magnitude ≥7.5; Modified Mercalli Intensity X or greater; or the earthquake generated a tsunami. All events reported by EM-DAT were retained (n=706), and zeroes were converted to missing values for injury, homeless, and affected measures; for deaths and total affected, zeroes were converted to missing values only when no other information was reported. Earthquakes from the NOAA-NGDC database were retained if one of the following criteria were satisfied: magnitude ≥5.5; ≥10 deaths; or ≥100 injured (n=579).

Two additional sources, the United States Geological Survey (USGS) Earthquake Hazards Program Global Database ⁹ and the Northern California Earthquake Data Center (NCEDC)¹⁰ were used to collect information on specific earthquake characteristics (coordinates, magnitude, focal depth, additional information when available). When available data from these sources were added for events reported by EM-DAT and/or NOAA-NGDC; new events were added only when mortality was reported by USGS. Earthquakes occurring in uninhabited areas that did not cause injury or death were removed. The final list comprised 953 earthquakes occurring between 1980 and 2009; information on mortality, injury or displacement was reported by one or more sources in 738 events. See https://www.jhsph.edu/refugee/publications_tools/index.html for the database of earthquake events.

The following outcome categories were used to assess risk factors for earthquake-related mortality: no deaths (0 deaths); low (1-9 deaths); medium (10-99 deaths); and high (≥100 deaths). Bivariate tests for associations between mortality and the following characteristics were performed using χ² (categorical measures) and ANOVA (continuous measures): decade, World Health Organization (WHO) region, World Bank income level, gross domestic product (GDP), GINI (measure of income inequality), focal depth, and magnitude. All covariates were significantly associated with earthquake mortality in the univariate analysis and were subsequently included in a multinomial logistic regression model to assess the adjusted odds of mortality at a given level as compared to events with no deaths. Analyses were performed using Stata Statistical Software, Version 11.0 ¹¹

Systematic Literature Review

Key word searches in MEDLINE (Ovid Technologies, humans), EMBASE (Elsevier, B.V., humans), SCOPUS (Elsevier B.V., humans), and Web of Knowledge, Web of Science (Thomson Reuters) were performed to identify articles published in July 2007 or earlier that described natural hazards and their impact on human populations. Following the systematic review, a hand search was conducted to identify relevant articles published after the initial search thru October 2012. One search was done for all the five natural hazards described in this set of papers. This paper describes the results for earthquakes. The systematic review is reported according to the PRISMA guidelines. The key word search included natural hazard(s), natural disaster(s), volcano(s), volcanic, volcanic eruption, seismic event, earthquake(s), cyclone(s), typhoon(s), hurricane(s), tropical storm(s), flood(s), flooding, mudslide(s), tsunami(s), and tidal wave(s). Key words included for impact on human populations were affected, damage(d), injury, injuries, injured, displaced, displacement, refugees, homeless, wounded, wound(s), death(s), mortality, casualty, casualties, killed, died, fatality, fatalities in either the title, abstract or as a subject heading/key word. The search resulted in 2,747 articles from MEDLINE, 3,763 articles from EMBASE, 5,219 articles from SCOPUS, and 2,285 articles from ISI Web of Knowledge. Results from the four databases were combined and duplicates excluded to yield a total of 9,958 articles.

Title screening was performed to identify articles that were unrelated to natural disasters or human populations. Each title was screened by two reviewers and was retained if either or both reviewers established that inclusion criteria were met. Percent agreement was assessed across reviewers, and title screening began after 80% agreement was achieved. A total of 4,873 articles were retained for abstract review. During abstract screening articles that met one or more of the following criteria were excluded: language other than English; editorial or opinion letter without research-based findings; related to environmental vulnerability or hazard impact but not human populations; individual case report/study; focus on impact/perceptions of responders; and not related to human or environmental vulnerabilities or impacts of hazards. Each abstract was screened by two reviewers and was retained if either or both reviewers established that inclusion criteria were met. Again, 80% agreement between reviewers was achieved prior to screening. During the abstract review, included abstracts were coded for event type, timeframe, region, subject of focus, and vulnerable population focus. A total of 3,687 articles were retained for full article review. Articles discussing the impacts of natural disasters on human populations in terms of mortality, injury, and displacement were prioritized for review. From this general review of 395 articles specific to earthquake events meeting the aforementioned subject focus criteria were identified for full review. Upon full review, 150 articles were retained 143 that underwent standard data abstraction; seven that were identified as review articles (Figure 1). Articles that focused on risk factors for specific types of injuries (primarily crush injuries and renal failure) or deaths were excluded because they did not provide insight on overall risk factors for mortality or injury. In total, 70 articles relating to risk factors for mortality, injury or displacement were identified; summaries of articles with primary data (n=60) and review articles (n=10) are presented in Tables 1 and 2, respectively.

Results

Historical Event Review

Overall, 74.1% of events in the database were reported by EM-DAT, 60.8% by NOAA and 25.8% by USGS; reporting by USGS improved dramatically from 2000 onwards. An average of 24.6 (range 13-43) earthquakes affected human populations annually between 1980 and 2009 (figure 2). The frequency of events increased over time, which is attributable to improvements in reporting. The average magnitude was 6.2 (range 4-9; n=493, 66.8% reporting) and focal depth 27.1km (median 19.0, range 0-235.8; n=493; 66.8% reporting). Earthquake mortality increased in parallel with the frequency of events from the 1980s onwards (Figure 3). A rapid increase in earthquake-affected populations was observed after 2000, which is likely a result of both improved reporting and population growth (Figure 3). Earthquakes were most frequent in the Americas, South-East Asia and the Eastern Mediterranean with each region accounting for 20-25% of events; however earthquake impact was greatest in the Western Pacific, which accounted for 44% of deaths and Americas, which accounted for 60% of the affected population (Figure 4). The overall impact of earthquakes on human populations is summarized in Table 3. Of the 738 identified events, the databases recorded 687 (96.9%) causing deaths, 420 (56.9%) causing injuries and 359 (51.4%) leading to homelessness.

Affected Population and Homelessness. An estimated 61.5 million people were affected by earthquakes between 1980 and 2009, including 16 million rendered homeless. These figures are likely to underestimate the true impact of earthquakes because the total affected and homeless populations were reported in only 95.5% and 51.4% of events, respectively. There were an average of 200,783 affected (median 3,7215, range 1-46,000,000) and 46,594 homeless (median 0, range 0-4,000,000) per event [where data were reported] and distributions were highly skewed.

Mortality and Injury.When mortality data from the three sources were combined, information on deaths (inclusive of events with no reported deaths) was recorded in 93.1% (n=687) of earthquakes, with fatalities occurring in 71.0% (n=524). Overall, an estimated 372,634 total deaths (range: 314,531-412,599) were reported; for events with mortality, there were an average of 604 deaths (median=3, range=0-87,652) per event when using the highest reported death toll. Deaths were concentrated in the Western Pacific (190,955 deaths, 44.6%) and Americas (103,679, 24.2%). When assessed by country, the greatest earthquake mortality was observed in China (n=90,106, 21.0%), Pakistan (n=88,314, 20.6%), and Iran (n=86,254, 20.1%). The two most deadly earthquakes (Sichuan China, 2008 and Kashmir Pakistan, 2005) accounted for 40% of all earthquake mortality.

Injuries were reported in only 56.9% (n=420) of events with estimated total of 995,219 reported injuries (range: 845,345-1,145,093). When injuries were reported, there were an average of 3,499 (median=100, range 1-374,171) injuries per event using the highest reported number for the event. To better estimate the total number of injuries, it was presumed that injuries would occur in events where deaths were reported. Among the 687 earthquakes with fatalities reported, injuries were reported in only 420 (61%) events. When the median and mean for injuries were applied to the remaining 267 events, between 29,392 and 1,267,864 additional unreported earthquake related injuries may have occurred between 1980 and 2009.

Bivariate associations between country-level characteristics and earthquake-related mortality are presented in Table 4. All predictors except for earthquake focal depth were significantly associated with mortality. In the adjusted multinomial logistic regression model (Table 5), only magnitude was significantly associated with earthquake mortality. World Bank development level, the Gini Index coefficient, and focal depth were not statistically associated with earthquake-related mortality. The odds of a high mortality event as compared to an event with no deaths increased by 11.93 (95 CI: 5.35-26.57) per additional point on the magnitude scale.

Systematic Literature Review

Mortality. Of the articles reviewed, 27 reported on causes of death (n=17) and/or risk factors including sex (n=8), age (n=15), building or location (n=17) and other risk factors (n=11) (Table 6). The primary cause of death in the majority of studies was building collapse ^{16–20,22,23,25,29,31,33,36,40,46}. Females faced a significantly increased risk of death in three studies ^19,36,41 while several others found no significant difference in mortality by sex ^18,33,38,66 none reported that males faced a significantly increased mortality. Of the 14,709 earthquake deaths where sex was reported in the articles reviewed, 45% (n=6649) and 55% (n=8,060) of deaths were among males, yielding a ratio of 1.0:1.2 for male to female mortality. Age was a risk factor for mortality in many studies. Older populations consistently had higher rates of death ^{13,18,20,26,29,33,36,38,41,66} and in several studies children also faced increased mortality ^19,29,33,44.

A majority of deaths occurred indoors ^{13,16,19,22,25,29,42} often at home, and the rate of complete building collapse was a good predictor of crude mortality rates ^36,38 Construction materials were associated with increased mortality risk, including unreinforced masonry ²², mud and stone walls¹⁹, concrete ¹⁵, panel construction ¹³ and wood construction ^16,17,42; however no clear trend was observed across studies. Location on the ground floor ^20,22 and upper floors ^15,23 of multistory buildings were associated with increased risk of death. Other mortality risk factors included earthquake intensity and distance to epicenter ^31,33,38,44 physical disability ⁴¹ prior injury or illness ²⁹, low socioeconomic status ⁴¹, and being in a car ³¹. Response and health systems characteristics associated with mortality risk included time to rescue¹⁵, per capita availability of physicians and hospital beds ³³, and prior first-aid or rescue training of lay, uninjured survivors ²²; better availability of rescue and early emergency care could prevent a substantial portion of deaths ¹⁷.

Injury. Detailed information on injury was reported in 51 articles among which 42 included data on injury type, 30 on gender risk, and 31 on age risk (Table 7). Soft tissue injuries (including lacerations and contusions) and fractures were the most common types of injury reported ^{15,16,23,25,27,29,43,47,51,52,53,56,58,60,62,67,69,70,71}, and the extremities were the most likely areas of the body to be affected ^{26,46,50,51,59,61}. Crush injuries/syndrome were reported as the most common injury in several in-patient studies, and there was a substantial body of literature on this topic (articles focusing on a specific injury type were not abstracted for this review) ^16,32,45,55 . The proportion of injured by sex and, when combined, suggests a similar injury risk among males in females- Males accounted for the majority in eleven studies ^{13,32,43,46,48,50,51,52,53,71,73} and females in sixteen ^{23,26,28,32,34,35,39,45,47,54,56,58,59,62,63,65,70,72} . However statistically significant differences were observed in a few cases, all of which suggested increased female risk ^28,39,45,55 . Many articles reported descriptive age data, however information on age-related injury risk was reported in fewer articles and included decreased risk among children ^30,46, and increased risk among young and/or working age adults ^50,52,56,62 , the elderly ³⁰, and with increasing age ^{26,29,34,39,54}, Building characteristics and other related risk factors included being indoors ^{13,15,16,23,25}, or on a middle or upper floor ^13,15,23,25, construction type and/or quality ^{15,16,23,26,34,37,39} and low socioeconomic status ^54,56.

Discussion

Main findings

In the 30 year period between 1980 and 2009, approximately 372,634 people died and nearly one million were injured as a result of earthquakes, with potentially an additional 29,392 to 1,267,864 unreported injuries. In this same period, 61.5 million people were affected by earthquakes, including at least 16 million left homeless. While mortality estimates in this study are consistent with those reported by other sources ² the numbers injured and homeless populations are likely gross underestimates given the low frequency with which these figures are reported. Prior review articles either focused on specific regions ^{69,71,73,74,78}, a limited time period ^70,74 mitigation strategies ⁷⁵ mortality only ⁷⁵, or individual injury and treatment ⁷⁷. Only one prior review used data from multiple sources ⁸¹.

Findings from this review, including descriptive statistics of and factors associated with earthquake mortality, are consistent with previous observations that earthquake mortality varies as a function of severity ^74,79, place ^12,77,79,80, time ¹², and development level of the affected area ^79,80. With respect to severity, greater focal depth was inversely associated with mortality, whereas greater magnitude (moment scale) was positively associated with mortality. In terms of place, earthquakes were relatively evenly distributed across the Western Pacific, American and European regions, whereas the plurality of deaths occurred in the Western Pacific, followed by the Eastern Mediterranean region. The largest numbers affected by earthquakes were in the Western Pacific followed by the South East Asia. As observed in previous studies, these findings are skewed by large events, such as the 2005 Pakistan earthquake that resulted in approximately 75,000 deaths ^79,80,81. The Haiti earthquake in 2010, one of the deadliest on record, which falls just outside the scope of the review period, is illustrative of how an occasional high-impact event can drastically change regional impact distributions and study conclusions.

In terms of time trends, the number of earthquakes has increased steadily since the 1980s and a greater number of people have been affected over time. While improved reporting may partly explain an increase in the number of earthquake events, the increases in mortality and the size of affected populations may also be attributable to population growth, urbanization and migration ⁸⁰ and changes in land use patterns ⁷⁶. Similar to other reviews lower economic development level, measured by per capita GDP, was associated with increased mortality which suggests that poorer countries face increased risk due to a variety of characteristics of the built environment ^79,80.

Findings from the systematic literature review of studies examining earthquake-related mortality and injury contribute to an improved understanding of the primary causes of death and types of injury as well as factors that may place certain populations at increased risk. Consistent with prior review articles, this review identified the most common cause of earthquake-related death as building collapse ^{13,74,75,78,80}. In addition, multiple studies highlighted that building type, the rate of collapsed buildings and construction materials were significantly associated with injury and mortality risk. This highlights that building improvements, especially in the design and construction and the enforcement of zoning and building codes, should be central to earthquake prevention and mitigation strategies.

Recurrent characteristics associated with increased risk for both mortality and injury were extremes of age, socioeconomic status and location of individuals at the time of the event. Consistent with the ecological study using the historical event database, individuals and households of lower socioeconomic status were at increased mortality and injury risk. Location, including distance from epicenter, being inside or outside a building, and type of building and location within the building were also strong predictor of earthquake mortality and injury risk. Timing of the event was also associated with mortality and injury risk where earthquakes occurring at night had higher mortality levels than those occurring during the day ^74,77 .The relationship between sex and mortality and injury was less straightforward. While it is tempting to draw conclusions from these findings, it is important to highlight that 70% and 44% of the 27 mortality studies did not report deaths by sex and age, respectively. In addition, few studies performed significance tests, and an even smaller number controlled for other risk factors in the analyses. When considering the extent to which age and sex may interact with other important risk factors, such as location during the event or characteristics of the built environment, accurately characterizing factors that contribute to mortality and injury risk becomes especially challenging.

The important role of the emergency response and health care systems in reducing mortality and injury in the immediate aftermath of an earthquake was highlighted in a number of studies ^18,22,33,81, it is clear that such systems remain inadequate in many earthquake prone countries that are less developed. Health facilities are especially vulnerable from earthquakes due to direct and indirect damage (losses in utilities and infrastructure) that affect that affect their emergency response capacity. The extensive body of literature on earthquake related mortality and injuries could inform response planning for future earthquakes in high risk areas.

An historical event review such as this can elucidate patterns over place and time as well as factors associated with increased mortality risk, but cannot identify more specific associations. For instance, a number of country-specific studies have highlighted significant differences in mortality risk by population density, rural/urban area and across diverse geographic regions ^77,78,79. Particularly in earthquake prone regions or countries, additional research is needed to identify specific characteristics that may place populations at increased risk for mortality or injury during or in the immediate aftermath earthquakes. Nonetheless, statistical models to predict earthquake mortality, can be useful tools for estimating the relative contribution of geographic characteristics and population sociodemographics to earthquake mortality ⁷⁹. Compared to other natural disasters a wealth of data and peer review articles on earthquakes exists and there is a comparatively strong evidence base for drawing conclusions on earthquake impact at global, regional, and in some cases national levels.

Limitations

The effects of earthquakes are the subject of gross approximations and aggregations with a great deal of imprecision. The availability and quality of data has likely improved over time and the use multiple data sources increased reporting. However, underestimation of the impacts of earthquakes is substantial because in many events outcomes such as injured and affected are unreported. In addition, inconsistencies and errors were common in data files from different sources. Several challenges were encountered when attempting to model earthquake mortality including a non-normal distribution, which necessitated analysis with a categorical outcome. Information on 2007-2009 GDP, 2009 World Bank development classification and 2009 GINI index were used for the analysis regardless of event year, and it is possible that many of these values were substantially different in prior decades and some countries are new or have merged with other nations. Many of the island-countries in the Caribbean are territories of European countries, which necessitated the use of GDP, GINI, and development levels representative of the actual earthquake affected areas. Systematic literature reviews are not without their limitations. The articles identified and included in this review is not an exhaustive list, as articles that were not written in English were excluded, and a number of studies meeting inclusion criteria during the abstract review could not be found. Additionally, findings from the included studies are difficult to aggregate because of differences in design, reporting, and study population. Another important series of articles not included in this review is those which report on specific types of injuries and their outcomes; future reviews with an in-depth focus on injury, and to the extent possible, relationships between built environment, injury and outcomes could make an important contribution to the literature.

Conclusions

In the last 30 years, almost 400,000 deaths and 1 million earthquake-related injuries were reported, with an estimated 61.5 million people affected. Approximations of the numbers injured and those made homeless are likely gross underestimates of the true values given low reporting levels. The distribution of earthquake related deaths and injuries vary greatly by region and economic development level with greater magnitude and lower economic development of affected areas associated with increased mortality. Globally, earthquake impact was concentrated in Asia, which had the greatest number of deaths and the largest affected population.

The primary cause of earthquake-related mortality was building collapse most frequently leading to soft tissue injuries, fractures and crush injuries/syndrome. Risk factors for earthquake-related death and injury included very young and very old age, poor socioeconomic status, being indoors and being in a poorly constructed building during the time of the event. Earthquake losses are likely to increase in future years due to population growth of in high-risk seismic areas and in the case low and medium development areas, inadequate construction quality. Increased attention to earthquake prevention and mitigation strategies, with a focus on the built environment in particular, is necessary. Strategies that are specific to the development level and country context are essential. For instance, improved building construction is not a reasonable short term objective for a country like Haiti. Other interim short term strategies need to be adopted in settings where changes in building codes, their enforcement, construction methods, and other characteristics of the built environment may take decades to achieve.

Competing Interests

The authors have declared that no competing interest exist.

Correspondence

Shannon Doocy, Johns Hopkins Bloomberg School of Public Health, 615 N. Wolfe St, Suite E8132, Baltimore, MD 21230. Tel: 410-502-2628. Fax: 410-614-1419. Email: [email protected].

Methods

Data on the impact of cyclones were compiled using two methods, a historical review of cyclone events and a systematic literature review of publications relating to the human impacts of cyclones.

Historical Event Review

A historical database of significant cyclones from 1980 to mid-2009 was created from publicly available data. Multiple data sources were sought to ensure a complete listing of events and to allow for cross checking. The two primary data sources were the Centre for Research on the Epidemiology of Disasters (CRED) International Disaster Database (EM-DAT) ⁷ and the National Hurricane Center (NHC)⁸ because they included information on human impacts. The events in the EM-DAT database include one or more of the following criteria: 10 or more people killed or injured; 100 people affected; declaration of a state of emergency; or a call for international assistance. The NHC database included information on all cyclones in the Atlantic, Caribbean and Gulf of Mexico.

The EM-DAT event list was downloaded in August 2009 and NHC data downloaded in February 2010. Event lists were reconciled to create a combined list of events from both data sources which were then tabulated and summarized for 1980 through 2009. See https://www.jhsph.edu/refugee/natural_disasters/_Historical_Event_Review_Overview.html for the database of tropical cyclone events. A total of 948 events were retained from EM-DAT and 331 from the NHC. For cyclone impacts reported by EM-DAT, zeroes were treated as missing values because they were used as placeholders and their inclusion in the analysis could contribute to the under estimation of tsunami impacts.

To assess risk factors for cyclone-related mortality the following categories were used: none (0 deaths), low (1-9 deaths), medium (10-99 deaths) and high (≥100 deaths). Bivariate tests for association were performed using chi-square for categorical and ANOVA for continuous measures. All covariates, with the exception of World Bank developmental level which was highly correlated with per capita GDP were subsequently included in a multinomial logistic regression model to assess relative risk of mortality at a given level as compared to events with no deaths. Analyses were performed using Stata Statistical Software, Version 11.0 ⁹.

Systematic Literature Review

Key word searches in MEDLINE (Ovid Technologies, humans), EMBASE (Elsevier, B.V., humans), SCOPUS (Elsevier B.V., humans), and Web of Knowledge/Web of Science (Thomson Reuters) were performed to identify articles published in July 2007 or earlier that described natural hazards and their impact on human populations. Following the systematic review, a further search was conducted to identify relevant articles published through October 2012. One search was done for all the five natural hazards described in this set of papers. This paper describes the results for cyclones. The systematic review is reported according to the PRISMA guidelines. Key words used included natural hazard(s), natural disaster(s), volcano(es), volcanic, volcanic eruption, seismic event, earthquake(s), cyclone(s), typhoon(s), hurricane(s), tropical storm(s), flood(s), flooding, mudslide(s), tsunami(s), and tidal wave(s). Key words included for impact on humans were affected, damage(d), injury, injuries, injured, displaced, displacement, refugees, homeless, wounded, wound(s), death(s), mortality, casualty, casualties, killed, died, fatality, fatalities and had to be used in either the title, abstract or as a subject heading/key word. The search resulted in 2,747 articles from MEDLINE, 3,763 articles from EMBASE, 5,219 articles from SCOPUS, and 2,285 articles from ISI Web of Knowledge. Results from the four databases were combined and duplicates were excluded to yield a total of 9,958 articles.

Title screening was performed to identify articles that were unrelated to natural disasters or human populations. Each title was screened by two independent reviewers and was retained if either or both reviewers established that inclusion criteria were met. To ensure consistent interpretation of inclusion criteria, percent agreement was assessed across reviewers for a small sample of articles, and title screening began after 80% agreement on inclusion was achieved. A total of 4,873 articles were retained for abstract review. Articles were excluded if they met one or more of the following criteria: language other than English; editorial or opinion letter without research; not related to human populations; individual case report/study; focus on responders; and not related to human or environmental vulnerabilities or impacts of hazards. Each abstract was then screened by two reviewers and retained if either or both established that inclusion criteria were met. Included abstracts were coded for event type, timeframe, region, subject of focus, and vulnerable population focus. A total of 558 cyclone articles were retained for article review; 193 articles focusing on the impacts of cyclones on human populations in terms of mortality, injury, and displacement were prioritized for abstraction. Upon full review, 49 articles were retained including 48 that underwent dual review, standard data abstraction and one that was identified as a review article (Figure 1). The additional review then identified eleven articles through October 2012 that met the inclusion criteria for abstraction in the mortality and injury review. A summary of the final 58 abstracted articles is presented in Table 1.