Over the past several decades, 2 separate but emerging trends have converged: an increase in the number of disasters of all varieties and the rise of big data as an analytical tool. In the United States (U.S.) and around the world, disasters have been occurring with increasing frequency. ^{1–Reference Greenough, McGeehin and Bernard3} A disaster is commonly defined as “a sudden, calamitous event that seriously disrupts the functioning of a community or society and causes human, material, and economic, or environmental losses that exceed the community’s or society’s ability to cope using its own resources.” ⁴ From kinetic natural disasters such as hurricanes and earthquakes, to pandemics such the 2009 H1N1 and coronavirus disease 2019 (COVID-19), and terrorism, the increasing incidence of disasters has created several public health preparedness challenges. Since 2000, more than 1 million people have died globally from natural disasters, with an associated economic impact exceeding $1.7 trillion. ^{Reference Thomas and López5} In the United States alone, there were 119 major natural disasters between 2010 and 2019 which caused $810.5 billion in expenses and killed 5217 people. ⁶ While the United States has borne a significant impact from such events, from 1970 to 2008 more than 95% of natural disaster-associated deaths were in developing nations. ^{Reference Thomas and López5} Climate-related disasters, in particular, have increased nearly 35% since the 1990s. ¹

In the United States, the Federal Emergency Management Agency’s (FEMA’s) National Response Framework outlines 5 core capabilities within the National Preparedness Goal: prevention, protection, mitigation, response, and recovery. ⁷ These phases, which comprise the disaster life-cycle, culminate in recovery-phase activities that necessitate addressing longer-term physical health and psychosocial sequelae along with infrastructural impacts. ^{Reference Fitzpatrick, Willis and Spialek8–12} These long-term needs form the basis of the recovery phase, in which the goal is to reconstitute the “livelihoods and health, as well as economic, physical, social, cultural and environmental assets, systems and activities, of a disaster-affected community or society.” ¹³ Such recovery efforts are consonant with the United Nations Office for Disaster Risk Reduction’s concept of “Build Back Better,” ¹² and with FEMA’s National Disaster Recovery Framework. ¹¹ As disasters are increasing in frequency and severity, it is critical to recognize that many areas hit by prior disasters are also at higher risk for future disasters, which adds urgency to the need to recover quickly and effectively. Areas still recovering from prior events, or those having not yet adequately addressed physical or social vulnerabilities, may experience even higher morbidity and mortality along with reduced resiliency from disasters. ^{Reference Ferris, Petz and Stark14}

Against the backdrop of increasingly frequent and severe disasters, the role of “big data” to aid recovery-phase activities requires more explicit ascertainment. In the 21st century, use of data has become a key feature of nearly every industry. While business and private industries, from banking to health care, have made great use of big data for improved efficiency, opportunities exist for enhanced use of big data in disaster management. ^{Reference Yu, Yang and Li15,Reference Elichai16} This lag, or lack of use, is often attributed to the highly dynamic, disparate and diverse nature of data in disasters, yet analytic approaches aimed at big data are actually designed to deal with each of these considerations and more. ^{Reference Yu, Yang and Li15} As analytic approaches to big data have been further developed and validated, their use across business, government, and academia have grown. ^{Reference Wamba, Akter and Edwards17}

While the term “big data”, and the analytic approaches to manage it, is not always clearly defined, there is general consensus that big data should be defined in terms of Volume, Velocity, and Variety. ^{Reference Wamba, Akter and Edwards17,18} Volume refers to the quantity of data, which in a disaster setting may come from established steady-state data sources such as emergency department reporting, ^{Reference Vollmer, Glampson and Mellan19} responders and other aid workers actively collecting information in the field, Internet of things (IoT) devices, ^{Reference Asadzadeh, Pakkhoo and Saeidabad20} social media, ^{Reference Shoyama, Cui and Hanashima21,Reference Yeo, Knox and Hu22} industrial equipment, geographic information systems (GIS) technology, overhead imagery, and more. Velocity refers to the rate at which data are received or processed. Variety refers to the mixed structure and format of the data, especially when a mix of sources are being aggregated in real time.

To date, there have been limited attempts to systematically study and apply the technological advances of big data to disaster management, let alone to recovery-phase efforts therein. Traditionally, most of the limited focus has been placed on response-phase activities in which responders attempt to save lives and property. ^{Reference Freeman, Blacker and Hatt23} This focus has extended from more traditional academic literature reviews ^{Reference Freeman, Blacker and Hatt23} to startup companies working on machine learning and artificial intelligence to find vulnerable populations during disasters. ^{Reference Flavelle24,Reference Schwab25} The application of these modern technologies to process big data into usable metrics present promising opportunities for more efficient disaster recovery. As the Volume, Velocity, and Variety of data continue to challenge the disaster management community, and with the pace of disasters increasing globally, enabling adoption and use of big data across all phases of the disaster life cycle, including recovery, is more critical than ever. To that end, here, we present an integrative literature review focusing on big data concepts as applied to the recovery phase of the disaster life cycle.

Methods

To assess the existing knowledge base on disaster recovery using big data, we performed an integrative literature review. ^{Reference Torraco26} The literature search was conducted using search terms developed iteratively a priori by the research team. On March 31, 2020, we conducted the final search of PubMed, Embase, CINAHL, LILACS, Web of Science, Scopus, Biosis Citation Index, Compendex, Inspec, NTIS, and GeoBase. All results from each database’s inception to the search date were included in the review process. Search terms were developed iteratively a priori. Full search terms and scope were adapted for each database and are available in Supplementary Table 1.

Using the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) Checklist ^{Reference Moher, Liberati and Tetzlaff27} as a guide throughout, all citations were imported into EndNote, a citation management tool (Clarivate Analytics, Philadelphia, PA), and duplicates were removed. The unique records were then imported into Covidence (Veritas Health Innovation, Melbourne, Australia), a systematic review software program for initial title/abstract and follow-on full-text review. At each stage, 2 reviewers screened the articles with a third reviewer resolving conflicts. At the title/abstract phase, each reviewer was blinded to the other reviewer’s decision.

Final inclusion and exclusion criteria are listed in Table 1. All articles required, at a minimum, a real-world/nontheoretical disaster research publication during the recovery phase with some implementation of a big data component. Big data can be a nebulous concept, so for the purpose of this research the definitions from Gandomi and Haider ^{Reference Gandomi and Haider28} were used. Broadly, the tool must include some component of high-volume data, high-velocity data or highly variable data where “traditional data management and analysis technologies (would) [be] inadequate for deriving timely intelligence”. ^{Reference Gandomi and Haider28} Articles were further screened by the reviewers for methodological rigor, including a clearly logical research process, no obvious technical concerns and publications in a peer-reviewed source. Each article included in the final analysis was abstracted for basic features, including the publication data (eg, year, first author), type of disaster (eg, hurricane), study type (eg, case study), and, by consensus, qualitative major themes.

Table 1. Inclusion and exclusion criteria for final review

Results

On final search, 51,504 articles were initially identified. After removing duplicates, 25,417 articles remained from the 11 databases used. Blinded dual title/abstract review led to 238 articles being advanced to the full-text review. After full text review, 18 articles ^{Reference Akter and Wamba29–Reference Wen and Lin46} were ultimately included in the final analysis. Figure 1 displays the final PRISMA ^{Reference Moher, Liberati and Tetzlaff27} diagram.

Figure 1. Big Data in Disaster Recovery PRISMA Flow Diagram.

Geographically, 8/18 (44%) of the included articles focused on disasters impacting the United States, with the remaining including Canada, France, India, Italy, Japan, Pakistan, and Turkey. While there was no left-sided boundary on publication dates, the earliest identified publication was in 2007. Median year of publication was between 2015 and 2016. Overall, disasters studied focused on hurricanes 39% of the time, followed by earthquakes (22%). All but 2 identified studies were classified as a case study or case series (89%), with the outliers being a systematic review and a literature review. Thematically, social media and GIS were the most commonly identified modalities, at 44% and 33%, respectively. Full results are displayed in Table 2.

Table 2. Characteristics of included studies (n = 18)

Abbreviations: GIS, Geographic Information Systems; ICT, information communication technology; her, electronic health records.