REQUIREMENTS TO MEET THE INFORMATION NEEDS OF HEALTH PROVIDERS

Getting the right health information to the right health providers at the right time also affects how health organizations coordinate with one another. Although response agencies (eg, OCHA, FEMA) broadly support and advocate for improved coordination in the aftermath of disasters, challenges remain and criticism continues. Evaluation and “lessons learned” reports document the need for improved information sources, which play a role in strengthening the coordination of organizations within PHS in these dynamic environments. ^{Reference Goyet, Sarmiento and Grünewald9,Reference Waugaman and Fast11,Reference Bennett, Bertrand and Harkin18} The Tsunami Evaluation Coalition Report evaluated the global humanitarian response to the 2004 Indian tsunami disaster and identified multi-level coordination challenges between the United Nations, international NGOs, and governments where duplicative service efforts occurred at various locations. Medical team deployments did not match the health needs in numerous affected locations, and, at times, incoming and temporary medical teams displaced the local functioning health systems. ¹⁹ In the US disaster context, delayed evacuations of patients from health facilities continue to be a persistent problem. For example, during Hurricane Katrina, Superstorm Sandy, and Hurricane Irma, delayed evacuations have resulted in morbidity and mortality. ^{3,Reference Redlener and Reilly4,20,Reference Rabin21,Reference Dosa, Hyer and Brown22} In the aftermath of Hurricane Irma, 12 individuals at a rehabilitation center died due to the lack of temperature control caused by power outages. The acute needs of this vulnerable population were not quickly identified by the PHS. ^{Reference Wright23} Crucial and related emergency infrastructure activities such as power, water, and road infrastructure are examples of the complex response environment for which health agencies and responders need to make timely decisions. These examples indicate a need for an improved way to achieve health situational awareness during disasters. With a more accurate and timely understanding of evolving situations, health care coordinating bodies can better direct resources and services to hospitals and communities with the greatest need, while using limited resources in the most effective way. ^{Reference Goyet, Sarmiento and Grünewald9,Reference Powell, Hanfling and Gostin17,Reference Hullah, Llewellyn and Aryal24‐Reference Gates26}

THE PARADOX OF INFORMATION SCARCITY AND OVERLOAD

During disasters, gathering both timely and relevant health information is a challenge due to the paradox of information scarcity and information overload. In some humanitarian environments where protracted conflict or restrictive settings predominate, limited health information is common. Health information from North Korea has been limited to a few organizations with access to health information (eg, UNICEF, World Health Organization [WHO], World Bank). ^{Reference Grundy, Biggs and Hipgrave27} Mortality studies in Iraq also exemplify the challenges that researchers and health organizations face in quantifying the impact of complex humanitarian crises on the lives of those affected during war and conflict. ^{Reference Tapp, Burkle and Wilson28‐Reference Burnham, Lafta, Doocy and Roberts30} Limited access to mortality data due to a paucity of comprehensive or updated information can result in a limited situational awareness of the greatest needs for health services and the most vulnerable groups at risk for disease and injury. US disasters have faced similar challenges of information scarcity contributing to not only challenges, but also controversy in determining the number of deaths due to a natural disaster. Early reports of death counts after Hurricane Maria in Puerto Rico varied from 64 to estimates of up to 8498, due to difficulty in accessing reliable information about deaths after the disaster. Final government estimates of 2975 were released almost 1 year after the hurricane. ^{Reference Kishore, Marqués and Mahmud31,Reference Santos-Burgoa, Sandberg and Suárez32,Reference Robles, Kenan, Fink and Almukhtar33}

In contrast, the presence of information often does not equate to an improved awareness of the disaster environment. ^{Reference Waugaman and Fast11,Reference Crowley and Chan34,Reference Altay and Labonte35} In the past decade, digitally enabled information channels have grown exponentially with access to the Internet, mobile phones, and other digital devices. As a result, the post-disaster environment is frequently affected by information overload. During the aftermath of the 2010 Haiti earthquake, for which both global and US disaster response systems activated, the growth of digital information in the form of mobile communications, web-based information systems, geographic information, and digital volunteer activity resulted in not only unprecedented large volumes of information, but also the recognition by the humanitarian community of the information ecosystem and challenges that lie ahead. ^{Reference Crowley and Chan34,36} In this complex health information environment, achieving public health situational awareness with both information overload and scarcity depends on small and large-scale human and automated efforts to collect, filter, and organize conventional and unconventional forms of health information. Social media such as Twitter streams, considered by many as an unconventional health information stream, provide an abundance of big data and new opportunities. For example, more than 20 million Twitter posts (tweets) were generated within a week during Hurricane Sandy, which overwhelmed US response teams in filtering relevant signals (eg, updates on medical facilities) from the noise (eg, prayers, jokes). ^{Reference Purohit, Castillo and Diaz37}

OPPORTUNITY: COLLECTING UNCONVENTIONAL INFORMATION STREAMS

There are a variety of unconventional digital data streams available for consumption in both US and global contexts, including social media (eg, Twitter, Facebook), news, RSS feeds, blogs, and official agencies’ situation reports. These new information streams can potentially improve situational awareness by providing enhanced, aggregated overviews of common information from the diverse streaming sources. As a result, it opens up opportunities to help explain the emerging needs and potentially support decision-making among emergency response teams. ^{Reference Van de Walle, Brugghemans and Comes38}

Social media and Web data streams can provide an unconventional, alternative source of information for health response activities in a disaster (see examples in Table 2). It also opens up novel opportunities to addressing some of the challenges in collecting the right information at the right time. By collecting relevant information about evolving needs for medical supplies, patient transport services, or health concerns, improved health-focused situational awareness can potentially impact the decision-making of public health and medical responders. Big data from social media present opportunities to extract valuable information (ie, defined as content that can influence disaster response planning and operations). It is often dependent upon computing techniques that can process large-scale data streams (ie, information overload) to filter and extract relevant information to address the existing challenge of low signal to noise ratios. The resultant relevant information may be of value to the health responders (eg, doctors, national and international emergency responders) who are coordinating health services (eg, medical supplies) and also coordinating with local and national public health offices.

TABLE 2.

Examples of social media messages from recent disasters and their implication for health services.

Social media platforms provide application programming interface (API), which facilitate the collection of these large public data streams so that they can be interpreted and analyzed for various purposes. Because there is such a large volume of data that can be collected during disasters, there are 2 major computational methods for filtering event-relevant data streams. Geo-fencing filtering is a method in which a specific disaster area is defined by a boundary for which the data are filtered. Keyword-based filtering is a method in which a given set of relevant keywords is used to narrow down the volume, such as the name of the disaster (eg, “Nepal earthquake,” “Hurricane Sandy”) or the sector of response (eg, “health,” “hospital”). Additional filtering can be done using metadata of social media posts, where information is collected for each unit of social media communications (ie, Tweet, Facebook post) such as author and time. Depending on the application, further task-specific information extraction and filtering methods are required. All of these filtering techniques can begin to address the challenges of information overload seen in social media-based big data during disasters.

CURRENT USE IN CONTEXT

The adoption of unconventional social media information into organizational practice has lagged behind the adoption of digital data collection tools such as electronic medical records and telemedicine, but remains an opportunity for further development. Currently, research groups are mining large volumes of social media information for epidemiologic surveillance while few organizations are monitoring social media feeds to help inform their situational awareness during responses.

The most studied use of social media information in public health practice lies with investigating the role that social media can play in epidemiology, public health surveillance, and detection of disease outbreaks. For instance, the HealthMap tool has primarily focused on epidemiologic surveillance by mining large volumes of health-related information from Web sources, including Twitter, to track and monitor emerging infectious diseases, including influenza. ^{Reference Paul, Sarker, Brownstein, Altman, Dunker and Hunter46‐Reference Bennett, Olsen and Harris48} Despite these advances in mining social media for health purposes, future research is needed beyond epidemiologic surveillance research. In disaster settings, epidemiologic surveillance for emerging outbreaks of disease and PHS is interrelated but represent a subset of essential services for a comprehensive health response. According to the Sphere Handbook and the WHO Emergency Response Plan, control of communicable diseases is one component of overall essential health services and is interrelated with broader PHS. ^6,49 Expanding and further developing computation techniques to mine social media data beyond epidemiologic usage to essential health services, including health logistics and health coordination, are areas of important research exploration.

More broadly, existing techniques and tools (eg, Hootsuite, TweetDeck) for basic social media filtering have been adopted in the operational environments of response organizations. ^{Reference Reuter, Ludwig, Kaufhold and Spielhofer50‐Reference Houston, Hawthorne and Perreault52} Disaster management organizations, both globally and at the national, state, and local health levels, are increasingly engaging in social media to communicate with the public during disasters and explore ways in which information can be extracted to influence their response activities. ^{Reference Markenson and Howe53‐Reference Sutton, League, Sellnow and Sellnow55} The United States Health and Human Services investigated the role of tweets during the 2012 Hurricane Isaac disaster, using key word searches, and determined that non-traditional data sources can provide valuable health information. ^{Reference Bennett, Olsen and Harris48} During the Nepal earthquake response, ACAPS, a humanitarian organization focused upon providing situational awareness to responding agencies during crises, evaluated the use of social media. ^{Reference Luge51} The evaluation found that social media could help organizations gain a better understanding of the public’s reaction to media reports but found limited use in understanding the needs of those affected.

However, such tools, experiences, and early research are highly limited to the basic key-word-driven searches for accessing relevant content, as well as basic descriptive and trend analyses. Incorporating timely information in response coordination tasks such as caring for the rescued patients from a disaster zone, as well as responding to help calls seeking medical help, is primarily manual and labor-intensive. ^{Reference Morrow, Mock, Papendieck and Kocmich56} There is a need to design automatic techniques for real-time filtering of relevant social media content and open web streams using health terms and public health knowledge to drive the way computers filter out noise and capture important information that can be operationally relevant for health responders.

FUTURE RESEARCH CHALLENGES TO EXTRACT RELEVANT INFORMATION FOR HEALTH PROVIDERS

To enable a more health-specific approach to social media analysis during a crisis that can further improve how non-traditional data can be consumed by health responders, 2 major challenges should be addressed: (1) advancing mining methodologies that can extract health meaning and creating relevant knowledge bases focused on US response and global response systems for action, and (2) expanding current research design and approaches to focus on public health and health activities, specifically operational response.