The workshop began with presenters providing stage-setting definitions and background information, which are summarized in the first half of this chapter. Richard Platt, professor and chair of the Department of Population Medicine at the Harvard Medical School and Sue Bakken, alumni professor of nursing and professor of biomedical informatics at Columbia University, provided definitions and general background on electronic health data, big data, and data science. Lushniak and Perren Cobb, director of Surgical Critical Care Institute and clinical professor of surgery and anesthesiology at the Keck School of Medicine, University of Southern California, provided definitions and general background on operations for response and clinical networks, respectively.
To set the stage for the workshop discussions, Yon Yu, associate director, Regulatory Affairs, National Center for Emerging and Zoonotic Infectious Diseases, CDC, HHS, asked a panel of stakeholders to consider the primary data sources discussed during the lightning terminology presentations and share perspectives on how their sector might contribute to a coordinated MCM monitoring and assessment effort during a PHE, which is presented in the second half of this chapter. Levy discussed her perspective as a local public health official. Theresa Cullen, associate director of the Global Health Informatics Program at Regenstrief Institute, Inc., offered perspectives both from health information technology (IT) and informatics and of a health care provider in the community. Patel provided a federal-level perspective, and Paul Petersen, director of the Emergency Preparedness Program of the Tennessee Department of Health, spoke from the perspective of a state health department. An industry perspective was provided by
Quazi Ataher, senior director of epidemiology in Worldwide Safety Strategy at Pfizer Inc. Finally, Adam Wilcox, chief analytics officer at the University of Washington, shared his perspective as an academic researcher.
Electronic Health Data
Platt offered a framework for thinking about electronic health data in which he highlighted five key types of electronic health data that are developed principally during the delivery of health care: EHR, electronic
laboratory data, administrative and claims data, prescription drug dispensing records, and public health registries (see Box 2-1). These types of health data exist in electronic form for substantial portions of the population, he said.
Platt expressed optimism that, in the future, data sources will be interoperable and information will be available about whole populations in real time. At present, however, the task is to consider how to assemble a patchwork of data sources into a comprehensive surveillance system for MCMs that can be activated rapidly, at any location, and covering a large fraction of the population. One consideration is that existing data sources are often part of the conventional care delivery system, while MCM dispensing often takes place in non-traditional settings.
Substantial technical challenges must be overcome to make incompatible data systems work with one another and work more rapidly than during routine surveillance, Platt said. Legal, propriety, and governance barriers present an even greater challenge to interoperability of systems. It will take effort and good will to find ways to make these systems work together for purposes other than that for which they were primarily designed and in a way that is still consistent with their original missions, Platt said.
There are a variety of potential solutions to increasing EHR interoperability, including standardization. Platt observed that EHRs were designed for use in individual health care settings, making it difficult to extract and analyze compatible population data for surveillance efforts. This is not an informatics problem, he continued, but an information problem.
Bakken described the four dimensions of big data as volume (scale of data), velocity (analysis of streaming data), variety (different types of data), and veracity (uncertainty of data). She highlighted several considerations for best leveraging big data:
- Promoting data models, sharing, and standardization;
- Ensuring trustworthiness, security, and privacy of the data;
- Maintaining and distributing knowledge derived from data; and
- Appreciating the importance of learning organizations and organizational learning.
In conjunction with traditional data sources, an increasing number of novel big data sources can be useful for detection, surveillance, management, and evaluation of MCM use, said Bakken. However, integration is necessary to take advantage of this wide variety of data sources, she said, and can occur at different time points. In some systems, data are integrated early on by standardizing multiple data sources at input. In other instances, integration might occur at an intermediate time point by mapping multiple data sources to a common data model for analysis. Late integration can occur after data sources have been separately analyzed and are brought together to create a larger analytical dataset.
The ability to analyze vast quantities of data, rather than smaller datasets, could enable population-level analysis during PHEs, said Bakken. She emphasized that working with big data requires a willingness to embrace the “real-world messiness” of the data and accept the usefulness of identifying correlations rather than causation.
In concluding her remarks, Bakken posed several questions to foster discussion: How do we harness big data and implement the right infrastruc-
ture for generating actionable insights? What are the privacy and security considerations that need to be addressed? What are the strategies for normalization and integration for unstructured data? What are the potential roles for crowdsourcing and mining of non-traditional sources?
Concept of Operations for Threat Response
Concept of operations (CONOPS) for threat response is a description of how a set of capabilities can be employed to achieve desired objectives or an end state, Lushniak said. It is a document that describes the characteristics of a proposed system from the viewpoint of an individual who will use that system. CONOPS are used to communicate quantitative and qualitative system characteristics to all stakeholders. It is widely used in the military, governmental services, and other fields, Lushniak noted, and the term CONOPS has multiple uses and multiple definitions within those systems. A CONOPS includes
- A statement of the goals and objectives of the system;
- Strategies, tactics, policies, and constraints affecting the system;
- Organizations, activities, and interactions amongst participants and stakeholders;
- A clear statement of responsibilities and authorities; and
- Specific operational processes for fielding the system and processes for initiating, developing, maintaining, and retiring the system.
There are CONOPS for many different threat response capabilities, Lushniak explained, and the task of this workshop is to discuss how to build big data, electronic health data, and clinical networks for MCM monitoring and assessment into these operations. Coordination of operations across local, state, and national levels and across all stakeholders is needed, he said.
A CONOPS is developed to provide a narrative of the process to be followed in implementing a system; define the roles of the stakeholders involved; and offer a clear methodology to realize the goals and objectives of the system. PHE responses are all-hazards approaches, Lushniak said, with an overarching goal being an active MCM monitoring and assessment capability that will allow the pooling, analysis, and sharing of information to guide MCM use during the same event or future events. Lushniak emphasized the importance of building this system into a CONOPS.
Cobb began his discussion of clinical networks with the following quotation from a New England Journal of Medicine article discussing research as a part of PHE response:
Although responses to recent events have typically used the best available science at the time, additional research, done in parallel with and after the response itself, is often essential to address the most pressing knowledge gaps presented by public health emergencies, and to ensure that they are addressed by the time another similar disaster strikes. Recent events, however, have illustrated gaps in the planning for, and rapidly executing, scientific research in the context of the disaster response. (Lurie et al., 2013, p. 1251)
In a follow-up to this article, Nicole Lurie, former ASPR, charged the U.S. Critical Illness and Injury Trials Group to bring together individuals and subject-matter experts from the federal government, academia, industry, and the community to identify key clinical questions that need to be answered in response to any type of PHE, across all hazards. The group developed a list of six questions they believed would be useful to clinicians and researchers and persons responsible for systems and operational evaluations (Murphy et al., 2015):
- Clinician end-users and researchers
- What was the nature of the insult and the resulting phenotype?
- As a responder, what, if anything, did you have to do differently?
- Did diagnostics, countermeasures, and therapies work as expected?
- What was the impact on the patient and care setting?
- Systems and operational evaluations
- Was there anything essential needed that you did not get?
- What is the best/worst case that could happen next time?
Using these questions as a starting point, stakeholders must next determine what data and infrastructure are needed to provide answers to these questions. A significant amount of work and funding has been under the general umbrella of PHEMCE, Cobb said, to build an infrastructure that supports a “network of networks,” including
- A national network of acute and critical care research organizations of academic and community hospitals for adults and children, across the care continuum, from pre-hospital through rehabilitation;
- A rapid communication network with quarterly queries to assess national health system stress;
- Infrastructure for prospective trials for national PHEs, such as influenza and anthrax;
- National data coordinating centers;
- Human subjects research review with local and national IRBs, through the NIH Public Health Emergency Review Board; and
- Coordination of efforts with international organizations and clinical trials groups.
Overlaying a series of maps,1 Cobb showed how this network of networks is geographically distributed across the country. Networks shown included sites of the U.S. Critical Illness and Injury Trials Group (recently renamed the Discovery Research Network of the Society of Critical Care Medicine); CDC National Ebola Training and Education Center; sites of the Johns Hopkins University Research Network; Pediatric Emergency Care and Research Network sites; Pediatric Acute Lung Injury and Sepsis Network Investigators; and the National Heart, Lung, and Blood Institute (NHLBI)-funded Prevention and Early Treatment of Acute Lung Injury Research Network. The United States is well covered by existing clinical trial networks, Cobb observed, although there are opportunities for better geographic representation. Importantly, he said, large numbers of centers are interested in clinical research during PHE response. However, he continued, there is not an opportunity for the leadership from these various networks to share information (e.g., lessons learned or efficient use of available resources) and coordinate efforts. In closing, Cobb referred workshop participants to a list of the key topic areas discussed at a 2015 Institute of Medicine (IOM) workshop, Enabling Rapid and Sustainable Public Health Research During Disasters (IOM, 2015). What is needed now, Cobb concluded, is an action plan that identifies metrics of success for these priorities and outlines exercises that test capabilities and capacity.
Local Public Health Perspective
Levy described the role of her department, Public Health–Seattle and King County, as carrying out the local response to a PHE and ensuring that the measures implemented meet the needs of the local population. She
1 The maps from Cobb’s presentation are available at www.nationalacademies.org/hmd/~/media/Files/Activity%20Files/PublicHealth/MedPrep/4_Cobb.pdf (accessed October 14, 2017).
emphasized a need for equity and an awareness of which members of the local community may not be represented in the datasets discussed throughout the workshop. Information gathered during a PHE is primarily needed at the local level to understand more about the incident that occurred; the types of MCMs available for both prophylaxis and treatment; what MCMs are accessible in the supply chain that will influence the response; and what is known about the effectiveness of the MCMs being used.
Local health departments face a variety of challenges in communicating to the public during a rapidly evolving situation. Levy noted the struggle among MCM sponsors and developers to balance obtaining “perfect data” for research and development and approval with getting data out to local residents quickly. Similarly, she said, local health departments struggle with risk calculations regarding messaging for promoting healthy behavior in the public during a PHE. Considerations for this calculation include noting that there are many different ways people prefer to receive information, differing functional and access needs, and language barriers. Risk communication needs to be succinct, she added, by being simplified to a few key messages that promote healthy behavior by the public during the event.
In the short term, policy and decision making are influenced by several factors: who was affected (e.g., closed or open population), whether other areas were affected, and whether the exposure was accidental or intentional, said Levy. She challenged the commonly held notion that all incidents are local by stating that “all incidents are regional,” and accordingly, cross-jurisdictional coordination is essential. Potentially affected populations cross county boundaries every day for work and for health care, she noted, and an uncoordinated message across the region could confuse the public on the proper course of action. Yu added that the public’s perception and acceptance of their role in contributing to data collection on MCM use is an important part of that messaging, as well.
Community Provider Perspective
Cullen shared her perspective as a family medicine doctor who works shifts in a local hospital emergency department, the prior chief information officer of the Indian Health Service, and the prior chief medical informatics officer of the U.S. Department of Veterans Affairs; she is presently in health informatics and health research at Regenstrief. Health care providers and responders are dealing with the individual patient, an N-of-1, and data collected for that N-of-1 form a narrative that stems from the many different questions the provider asks when trying to establish a diagnosis, said Cullen. The narrative that evolves at the point of care provides an opportunity for early recognition as part of the public health and MCM response to an emerging event. Currently, she said, this un-
structured narrative, oftentimes captured within EHRs, is not accurately leveraged in PHE responses.
Federal Health Perspective
Patel remarked that stakeholders should consider the timing of events during a PHE and how data needs might change as a response evolves. Often, data needs change throughout a response, and being clear about what is needed on the front-end and how needs may change over time is important. Patel noted that the very early stages of a PHE response are critical for the federal government to be able to mobilize emergency operation centers, which streamline coordination efforts and allow cross-talk among government agencies. Furthermore, a key concern at the onset of a response is the identification of regulatory and policy needs, because much of what happens at the local level is contingent on these set rules and regulations, she said. Information gathered in the early stages of a PHE could influence decisions regarding declaration of states of emergency (e.g., invoking the Public Readiness and Emergency Preparedness Act Declaration2) and other regulatory mechanisms, such as EUA for MCMs. In response to a question on how coordination across the federal government could be done more effectively, Patel noted that the more stakeholders and their systems are exercised and tested in situations in which data sharing is essential and flows in different ways, the better and more coordinated responses become.
State Health Department Perspective
Petersen shared his viewpoint on data needs at the state level, including three major takeaways: the importance of timely risk communication to affected populations, clarification on the scale and scope of a PHE, and identification of the resource needs (including MCMs) that are not readily available for response efforts. He emphasized the importance of providing action steps that the public can take for themselves, and being transparent regarding any prioritization for MCM dispensing. Communicating MCM prioritization helps to foster public trust and adherence with any MCM dispensed. All emergencies start locally, Petersen added, and decisions (e.g., policy development, data collection requests) made at the federal and state levels can have significant impact on those working at the local level. Furthermore, Petersen emphasized that data sharing among stakeholders,
2 For information on the Public Readiness and Emergency Preparedness (PREP) Act, see https://www.phe.gov/Preparedness/legal/prepact/Pages/default.aspx (accessed August 23, 2017).
including between state and local health care providers, can significantly improve monitoring efforts during a PHE (see Box 2-2).
Pharmaceutical Industry Perspective
Ataher said the pharmaceutical industry is always concerned with the benefit–risk balance of its products. During the usual drug approval process, there is ample time to assess the benefit and risk profile of a product, and post-approval studies, in which large datasets are gathered to study effectiveness, may last up to 5 years. In a PHE setting, however, the time for such assessment is limited. The key to success, Ataher suggested, is to plan for the unexpected by understanding what data will be needed in order to develop predefined clinical protocols that can be deployed for individual products in an emergency setting.
One difficulty in conducting clinical research during PHEs is the inability to conduct randomized clinical trials in defined populations; rather, products are administered to a generalized group of patients with limited background health data at the time of MCM administration, said Ataher. Furthermore, he remarked, from an epidemiological standpoint, confounding variables increase the difficulty in analyzing data from this research.
Ataher suggested that potential solutions to these issues could include having a predefined set of variables on what types of data should be collected, developing predefined study protocols, and identifying target sites and an accompanying “SWAT team for data collection” that could be readily activated to collect data without interfering with the administration of the MCM.
Academic Research Perspective
Wilcox noted that academic researchers have an important role to play during the course of a PHE response, parallel to the role of first responders, by providing observational research and analytical capabilities. For example, Wilcox noted that during the 2009 H1N1 influenza pandemic, a research team at Columbia University quickly assessed incoming data from CDC to assess the origins, diversification, and spread of the virus. Using innovative bioinformatics techniques, the team was able to decipher the genetic origins of the virus and determine that in the recent past, the virus was endemic to pigs—that is, swine flu—and not birds, as originally thought. This information was critical to monitoring the spread of the disease because it prompted public health officials to increase surveillance efforts in pigs and provided valuable information for studying and developing vaccinations for flu viruses of related structure in the future.
Wilcox concurred with Levy about the importance of understanding which populations are represented (or not represented) in a given data sample. Traditionally, research and health care delivery has focused on the population that happened to be coming in for care, he said. Population health initiatives throughout the country are beginning to take into consideration the populations that are not accessing care, however, and this shift has affected the way data systems are designed and increased attention to using the correct methodologies to link the data (see Box 2-3 for a case study on this issue).
FDA Perspective: Data for Regulatory Decision Making
Maher described what information FDA needs to make regulatory decisions related to MCM use. For PHEs, there is a general idea of the types of questions that should be asked, she said, as listed by Patel. However, it is difficult to define questions with more granularity in advance of a PHE. One challenge, also noted by Wilcox, is that data are used to answer the questions, but they also inform the development of the questions. It is a cycle of researchers asking for the data and data holders asking what questions the researchers are trying to answer, said Maher.
One piece that is missing, she said, is the conversation about the many different places data might be, and whether and how that data might be accessed and used in a way that is not disruptive to the ongoing response operations. Maher raised several questions from a regulatory standpoint: How can all data, regardless of its location, be harvested and leveraged to answer not only regulatory questions, but also response-related questions? How can systems already in place be used and leveraged to answer questions? How can disconnected data sources be connected? Maher noted that a key question for monitoring and assessing MCM use is whether the MCM is providing a benefit (i.e., acting as expected) or causing more harm, and she pointed out another important question: How quickly can targeted studies be designed, using the detected safety and efficacy signals, to conduct research during the emergency to answer this question?
Ben Eloff, deputy director of epidemiology in the Center for Devices and Radiological Health at FDA, added that with regard to devices, regulatory decision making would be better informed if there were more detailed identification of the medical product and the exposure to that medical product. Drugs and biologics are relatively easy to identify, Eloff said, and he referred to systems such as the National Drug Code, which clearly identifies all drug products by an exclusive number. Medical devices span the gamut from tongue depressors and stethoscopes to implanted devices and defibrillators. It is more difficult to understand what role a given device might have
in the identification of or response to an emergency. Having more granular information about the specific devices used would be of value, he said. For example, if one manufacturer’s meter has an error in their algorithm that is providing erroneous results, and other meters are not, this error can only be detected if the specific device is known.
This page intentionally left blank.