Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
6 Potential Opportunities to Enhance Preparedness Through Health Information Exchanges and Predictive Analytics Highlights of Main Points Made by Individual Speakers1 â¢ Health information exchanges enable the sharing of data across disciplines, but standards would allow for optimal interoperability. â¢ Some exchanges are not interoperable between states, even though patients often seek care outside their home state. Unless states and regions plan ahead for the ability to access each otherâs health information exchanges, the exchanges will not achieve their full potential for data sharing. â¢ Improved surveillance and threat detection involves automated, real- time access to electronic health records to collect laboratory and syndromic surveillance data and conduct ad hoc queries (including the flexibility to query for elements that may arise in a disaster). â¢ Predictive analytics, artificial intelligence, and natural language processing technologies could help to better direct resources intelligently during disaster situations. â¢ The fluctuating and uncertain demand process for jurisdictions is experienced by every state and needs to be taken into account when planning distribution logistics. â¢ Nathaniel Hupert recommended that participants look to industry and to the National Oceanic and Atmospheric Administration for models that could be the base for public health information technology systems to incorporate into everyday decision support. 1 This list is the rapporteursâ summary of the main points made by individual speakers and participants, and does not reflect any consensus among workshop participants. 61
62 IMPACTS OF THE ACA ON PREPAREDNESS Opportunities to enhance preparedness, response, and resilience through health information technology are numerous. With the passage of the Affordable Care Act (ACA) more health information exchanges (HIEs) are being created, which can help to enable sharing of data, but as mentioned in the previous chapter, barriers still exist that keep the process from being as streamlined as it could be. However, with the proliferation of health data and possibilities for sharing, there is also potential for predictive modeling and analytics that can support decision making for authorities during public health emergencies, especially a pandemic or emergency requiring rapid medication distribution. This chapter explores some of these opportunities and the challenges that are still present as provisions from the ACA begin to be implemented. HEALTH INFORMATION EXCHANGES As noted above, one approach to integrating health information across disparate systems is the development of HIEs. The speakers from Kansas and New York described examples from their states and their applications to public health preparedness. Kansas Health Information Network (KHIN) Michelle McGuire, senior project manager at KHIN, described KHIN as an example of a multi-functional health information exchange. This provider-led HIE is a publicâprivate partnership in association with the Kansas Department of Health and Environment. KHIN is currently working to be able to connect to other state HIEs. Participants in the HIE across the state include hospitals, clinics, and federally qualified health centers, as well as physician practices, dental clinics, optometrists, substance abuse centers, community mental health centers, home health organizations, safety net clinics, pharmacies, hosp- ices, long-term care facilities, laboratories, behavioral health providers, public health departments, and schools. Currently there are 367 KHIN members that are, or will soon be, sending data for more than 1 million unique patients to the exchange. KHIN is also transmitting data to the public health department for syndromic surveillance (see Figure 6-1). Thus far more than 900,000 records have been sent, as well as data on more than 20,000 immunizations.
HIES AN ND PREDICTIV VE ANALYTICS 63 FIGUR RE 6-1 Descrription of thee two-way com mmunication involved in tthe Kansass Health Inform mation Network k. NOTE: IDNs = integ grated delivery y networks; KDDHE = Kansaas Department of Health and Environm ment; RHIN = rural health iinformation neetwork; WHIE E = Wisconnsin Health Infformation Exchhange. SOURC CE: McGuire presentation, p November N 19, 22013. McGuire descrribed several of the servicces KHIN haas available, ffor exampple, secure cliinical messag ging for comm munication aand information exchannge among prroviders, heallth informatioon exchanges, and electronnic health record vend dors. The maain use of KH HIN is the P Provider Porttal, which allows proviiders to query y for a patienntâs records ffrom any of tthe other participating p hospitals or clinical entitties. In this rregard, the H HIE can im mprove patientt protection, helping h in thee continuity oof care from oone health care facilityy to another by providinng a single llocation for all patientt records. Th here are state--level interfacces with the cancer registtry and with w a new in nfectious dissease registryy, as well ass reporting ffor immun nizations, synndromic surv veillance, andd reportable ddiseases. KHIIN also haas functionaliity for data ex xtraction from m multiple soources. KHIN N is now offering o a perrsonal health record, fosteering patient engagement bby allowing patients to access th heir own heaalth informattion and shaare
64 IMPACTS OF THE ACA ON PREPAREDNESS information with providers. This would be extremely helpful in disasters to help providers, possibly out of state, to treat the âwhole person.â An HIE is not traditionally considered a disaster planning tool but could be of great help in public preparedness, McGuire said. For treatment of patients, an HIE can provide access to health records (including immunizations, medication, recent laboratory result, diag- noses, allergies, provider names, and contact information) no matter where the patient is transported for care. HIEs can also be used to locate patients during the disaster (as was done following the Boston Marathon bombings). As soon as a patient is registered in a hospital system, there would be a record in the HIE, associated with current records. Offsite or out-of-state/region data storage is also advantageous for disaster recovery. Data captured during a disaster can also inform the response. For example, information entered into the Infectious Disease Registry can provide the ability to contact the infected patient sooner and to reduce disease investigation time. Current Barriers There are several barriers to effective HIE data capture, McGuire noted. The different EHR vendors have variations and limitations in their capabilities. In addition, the Centers for Disease Control and Prevention (CDC) does not require the use of any particular vendor, so for example, Kansas City spans the Kansas-Missouri border, and each state uses a different vendor for sending syndromic surveillance information to CDC. Interfaces are costly and time consuming, and hospitals in Kansas, for example, currently need to pay for and institute both HL72 interfaces to send data to public health, and Continuity of Care Document interfaces for Meaningful Use compliance. Often, HIEs do not cross state lines, yet patients are transported across state lines for care both routinely and in disasters. Of course, not all hospitals or facilities are participants yet. Unless states and regions plan ahead for the ability to access each otherâs exchanges, the HIEs will not achieve their full potential for data sharing, McGuire said. She also encouraged participants to include HIEs in disaster planning. In closing, she referred participants to the HIMSS Dashboard, a website tracking the different HIEs across the country,3 adding that some larger states have multiple HIEs. 2 Health Level Seven International (HL7) is a nonprofit organization that develops standards for integrating, sharing, and retrieving electronic health information. 3 Available at http://apps.himss.org/StateDashboard (accessed June 8, 2014).
HIES AND PREDICTIVE ANALYTICS 65 Surveillance and Threat Detection: The State Health Information Network New York (SHIN-NY) Gus Birkhead, deputy commissioner and director of Public Health Programs at the New York State Department of Health, provided a state health department perspective on surveillance and threat identification. State health departments have a statutory mandate to collect surveillance data, including vital records data and other health data. In New York, the state health department is the health system regulator, operates the state Medicaid program and the state health exchange, is the shared lead agency in health emergencies, and is a source of expert medical guidance to the health care community. The state department of health is at the crux of all the issues that relate to surveillance, threat detection, and response communication. Electronic Surveillance Birkhead described New Yorkâs current electronic surveillance capabilities across a variety of venues, including clinical laboratory reporting, discharge data, vital records, poison control calls, emergency medical services (EMS), Medicaid pharmacy claims, influenza-like illness (ILI) sentinel surveillance system, and others. Much of the communicable disease, cancer, and chronic disease surveillance systems have been built on the electronic clinical laboratory reporting system. Using influenza as an example, Birkhead described how all of the different surveillance systems that are used on a regular basis can come into play for threat identification and response. This is another area highlighting the advantages for preparedness to be built into everyday care to be successful. For example, the Health Care Emergency Response Data System (HERD) provides laboratory reports on positive influenza samples at clinical laboratories by type of flu (A, B, or unspecified) and can compare the data with previous flu seasons. The ILI sentinel surveillance system is an ad hoc system of about 100 physician practices around the state who report weekly to the state department of health on the percentage of their patients with ILI. Another ad hoc data collection mechanism is ILI in emergency department (ED) reports. In addition, hospitals enter data each week on their positive flu hospitalizations. These ad hoc systems are labor intensive and slow, and Birkhead noted that this type of surveillance would have much greater potential if the state could tap directly into electronic health records (EHRs) and pull
66 IMPACTS OF THE ACA ON PREPAREDNESS information on flu-like illness visits. In theory, this could allow development of a statewide picture on a daily basis, he said, but flu is just one example of such monitoring. Following Hurricane Sandy, for example, the state was able to monitor respiratory illness, gastrointestinal illness, carbon monoxide poisoning, and other issues that are prevalent after many kinds of disasters to help guide citizens through recovery. Birkhead described SHIN-NY as a platform on which various applications and tools can be built (e.g., physician alerts, patient engagement, care plan management) to enable bidirectional flow of information. For example, for vaccinations given in a pharmacy (i.e., not at the patientâs medical home), the immunization registry can push data out to EHRs. There is also functionality for public health to query any of the participating systems through a Universal Public Health Node. Future Directions Going forward, the intent is to develop a unified, statewide, standards-based health information network that taps directly into EHRs and other data sources to provide access to data whenever needed, without putting a burden on the reporting sources. To this end, New York State is the public partner in a publicâprivate partnership that is developing SHIN-NY with a private partner, the New York E-Health Collaborative. The network includes 12 regional health information organizations that provide geographic coverage for the state, each using a core set of data standards to collect and share health data from participants (e.g., hospitals, providers, laboratories, pharmacies, long- term care, health plans, public health officials, patients). The goal is to improve quality of care, efficiency, and patient satisfaction using information technology tools to enable collaboration among patients, providers, public health, and payers, while safeguarding privacy, confidentiality, and data security. No single entity can deliver on this goal alone, Birkhead said. He noted that New York was able to use available state and federal Health Information Technology for Economic and Clinical Health (HITECH) Act funding for the development of SHIN-NY. Going forward, Birkhead said that Medicaid Meaningful Use Matching Funds could cover about one-quarter of the statewide network costs, with the balance from state Health Care Reform Act funds. Improved surveillance and threat detection involves automated, real- time access to EHRs to collect laboratory and syndromic surveillance data and conduct ad hoc queries (including the flexibility to query for
HIES AND PREDICTIVE ANALYTICS 67 elements that may arise in a disaster). Monitoring outcomes for diseases of public health interest includes developing an all-payer database, Birkhead said, adding that just leveraging publicly run health insurance programs (Medicaid, the exchange, state and local employees insurance plans run by the state) would include slightly more than half of the population in the state. If timely, then this could be a threat detection device, but more likely it will be used to establish the baseline in terms of services during an event. There needs to be bi-directional information exchange and real-time guidance and decision support for conditions of public health interest, he concluded. PREDICTIVE ANALYTICS IN PUBLIC HEALTH PREPAREDNESS The ACA provides a very important opportunity to improve the uptake of EHRs and the ability to use the data in them; to improve surveillance methods and use these data for improved models, analysis, and decision making; and to improve service delivery through public and private partnerships, said Brandon Dean, staff analyst for the Los Angeles County Department of Public Health. Public health interventions are focused upstream, he continued. Effective planning and execution before an event can lead to a delay in peak impact, decreased burden on hospitals and infrastructure, and dim- inished overall health impacts. Public health has a complex relationship with health care, he said, and it is important that we continue to think of this as a âsystems of systemsâ and understand how they integrate. Modeling Dispensing and Surge Capacity for Emergency Planning in Los Angeles County Models and analytic projects can help manage the complexities of integrating health care and public health, Dean said, and he offered several examples from his work in Los Angeles. There are about 11 million citizens throughout 88 cities in Los Angeles (LA) County, which Dean noted means there are 88 mayors with whom county public health has to coordinate. LA County has a very robust surveillance system that
68 IMPACTS OF THE ACA ON PREPAREDNESS includes 14 separate data systems that feed into the main system.4 As of 2009, about 17 percent of the population of LA County was uninsured5 (prior to the ACA), about 54,000 individuals are homeless,6 and 36 percent of county residents are foreign born. Pandemic Modeling Dean described two models that were explicitly designed for local health departments to understand the spread of pandemic influenza within the community and the effect on the hospital systems (see Figure 6-2), and to use that information to drive local planning and policy development. A community mitigation model was developed in collaboration with the Longini modeling group and the National Institutes of Healthâfunded Models of Infectious Disease Agent Study (MIDAS), and a surge model that used the output of the community mitigation model was developed in collaboration with the Hospital Association for Southern California. During the 2009 H1N1 pandemic, the community mitigation model was used to predict the effects of vaccine coverage on community attack rate. The model suggested that if nothing was done, the average attack rate would be 36 percent of any population, translating to about 3.5 million people in LA County. If 30 percent of the population could be vaccinated, then 18.7 percent would be affected, and if 50 percent of the population could be vaccinated, then it would drop to only 0.8 percent affected. As a result of the modeling, the health department set the target of administering between 3 and 5 million courses of vaccine as quickly as possible. The fall 2009 vaccination campaign, carried out by 110 points of distribution (PODs) over a 6-week period, was able to administer 280,000 courses of vaccine. Of the 28,000 physicians in LA County, about 6,000 (mostly pediatricians and obstetrician/gynecologists) gave the vaccine to their patients. 4 The feeds are primarily banded into eight categories, and include: ED visits, nurse call data, poison control, over-the-counter medication sales, ReddiNet (two-way hospital re- porting system), veterinary, 911 calls, and coroner. 5 See http://www.chcf.org/~/media/MEDIA%20LIBRARY%20Files/PDF/A/PDF%20 AlmanacRegMktQRGSixCombined13.pdf (accessed May 20, 2014). 6 See http://documents.lahsa.org/planning/homelesscount/2013/HC13-Results-by-SPA- and-SD-Nov2013.pdf (accessed May 20, 2014).
HIES AN ND PREDICTIV VE ANALYTICS 69 FIGUR RE 6-2 Integraation of modelss to inform planns and policiess in LA Countyy. SOURC CE: Dean presentation, Noveember 19, 20133. In asssessing the campaign, Dean D said thhere was noo metric or a mechaanism to dettermine how w much vacccine was addministered bby providders. A researrch corporatio on was hired to conduct ssampling for 10 weeks, and it was determined that t about 3, 335,000 courrses of vacciine were given, g just reaaching the low w end of the taarget. Th he surge mod del was first developed d in 22003 to prediict the effect of an annticipated clo osing of sev veral governm ment-run hoospitals on tthe hospitaal system. Th he model was modified in 2008 to lookk at the effect of an inflluenza pandeemic on the hospital h systemm as a wholee. Using unm met need (i.e., ( beds) ass the primary output, the rresults stresseed the need ffor upstreaam public health h interveention to allleviate the bburden on tthe hospitaal system. Itt was also cllear there woould be uneqqual impacts in s of the hoospital, local ppopulation, annd differeent regions, based on the size other factors. f Anthra ax Modeling An nother examp ple described d by Dean involved meedical counteer- measuures for anthrax exposure. Current planns require thee dispensing of prophyylactic medication to 10 million m peoplee within 48 hhours. There aare numerrous PODs, he h said, but more m capacityy is needed. Modeling w was used too study the efffect of additiional dispensiing partners, bboth public annd ults suggested that not ddistributing anny intervention privatee. Initial resu
70 IMPACTS OF THE ACA ON PREPAREDNESS could lead to 400,000 cases of anthrax. With LA Countyâs current capacity, the number of cases could be reduced to about 142,000 by dispensing within 48 hours post exposure. If dispensing capacity could be extended, through additional POD sites or engaging partners such as pharmacies, then the number of cases could be reduced to 124,000. Better planning and allocation of resources in a more systematic way could cut the number of cases further. Prescription Drug Use The last example shared by Dean is a pilot program to track prescription drug use by the most vulnerable residents (by both demographics and geography) prior to a catastrophic event. By 2014, about 1 million uninsured residents are expected to have acquired insurance under the ACA, the majority of them through the L.A. Care Health Plan. Using de-identified data from health records and insurance records, the goal of the tracking program is to build profiles of what these particular groups will need in an emergency and to work with the public and private providers to coordinate pharmaceutical care services for these individuals during a crisis to prevent the ad hoc provision of medications that often occurs at the local level currently at various emergency shelters. Ideally, this could help to inform real situations that Dean described in his simulations. This is useful information to inform citiesâ planning, but not every city has these technologic capabilities. As the ACA progresses, and more health care systems increase EHR use and Meaningful Use requirements (enabled through the American Re-investment and Recovery Act), it may be easier and ideally more routine to have better predictive modeling at the local level that can target a range of needs. Modeling and Planning for Pandemic Dispensing: Integrating Real- Time Information for Decision Support Nathaniel Hupert, associate professor of public health and medicine at Weill Cornell Medical College, shared a photograph of New Yorkers standing in long lines for smallpox vaccinations in 1947 and noted that people still stand in long lines during vaccination exercises in stadiums and other large venues. Continuing at this rate, it will take an extremely long time to get countermeasures to all 8 million residents of New York City, he said, and the city set out to model alternate ways of reaching the community.
HIES AND PREDICTIVE ANALYTICS 71 One of the approaches that has been modeled and exercised, and was used in 2009 and 2010 during the H1N1 influenza pandemic, is involving retail pharmacies in public health dispensing. He cited a study of the distribution of antiviral drugs in California in 2009. The study found that key challenges for local health departments were access to information on private retail supply, confusion regarding the use of public versus private sources of antivirals, and tracking of antiviral use (Hunter et al., 2012). The key difference between an optimal system for a given set of conditions, and a resilient system that can respond to many different conditions, he said, is creating systems that are designed for sharing information. Hupert and colleagues built a system to model pandemic influenza outbreaks using the 62,000 retail pharmacies across the country to measure potential capacities. Hupert noted the challenges of modeling such scenarios, and the many assumptions that must be incorporated (e.g., size and type of pharmacies, customer volume at a given pharmacy, volume of prescriptions a pharmacist could fill). The model suggested that hypothetically, there is adequate pharmacy capacity to provide antiviral prescriptions to people as the flu hits its theoretical peak in a given county. However, a logistics model of the weekly requests for antiviral drugs from the strategic national stockpile (SNS), and the daily delivery of new antivirals to all 62,000 pharmacies, predicted that the SNS would run out of pediatric antivirals. In addition, even though there were theoretically enough adult antiviral doses in the SNS in the hypothetical example, there was still unmet demand at the local level (e.g., people arriving at the pharmacy after the last dose had been given and the next shipment had not yet arrived). With many critical products including pharmaceuticals, medical equipment, and supplies manufactured overseas and delivered to hospitals, businesses, and homes âjust in timeâ there is the potential for limited or no surge production and delivery capacity. A glitch in supply, production, or transportation thus could become a supply problem at the pharmacies themselves. This model shows, Hupert concluded, that adequate supply is not a guarantee of a high system fill rate or ability for the inventory to meet demand. The fluctuating and uncertain demand process is experienced by every state. Careful design of inventory allocation rules is essential to ensure that fill rates are as high as possible, and effective mass prophylaxis requires that inventories are available at the right place at the right time. Citing the work of John A. Muckstadt on information systems
72 IMPACTS OF THE ACA ON PREPAREDNESS for supply chain management for large corporations, Hupert offered an approach to improved response involving information systems, decision support, and response strategies. Planned response strategies can be accomplished with real-time information that involves collaborative decision support, for example, between the federal government and retail (see Figure 6-3). Decision Governmentï§ï¨Retail Support Collabora ve Coordinated Communicated Informa on Response Sta c Unan cipated Systems Strategies Updated An cipated Real- me Planned FIGURE 6-3 Approach to improved response combining real-time information systems and planned response strategies for collaborative decision support between the government and the private sector. SOURCE: Hupert presentation, November 19, 2013. Public- and Private-Sector Integration Hupert noted that he had not found provisions in the ACA addressing the integration of private health care sector information systems into public health response, and this will need to be considered if public authorities are going to maximize the capability at the local level of providing countermeasures in an emergency scenario, as previously described. He also observed that, with regard to public health emerg- encies and the ACA, there is a divide between emergency care and health information technology. For example, two sections of the ACA, Section
HIES AND PREDICTIVE ANALYTICS 73 1104 on Administrative Simplification: Operating Rules (regarding Health Insurance Portability and Accountability Act [HIPAA] trans- actions) and Section 3504 on Regionalized Systems for Emergency Care (regarding trauma systems) do not coordinate at all. We must have information systems with HIPAA-protected information about patients at the pharmacy that is available to emergency care, he said. Hupert recommended that participants look to industry and to the National Oceanic and Atmospheric Administration (NOAA) for models that could be the base for public health information technology systems. NOAA, for example, has a distributive system of data sensors with layers of analytic and security wrapping that could be used in this case to support data sharing and coordination but also meet HIPAA requirements. As the ACA is being implemented, it is an opportune time to bring together the people who run the various health information systems and make sure that public health is part of the system, he concluded. Incorporating Modeling into Everyday Health Decision Support Many participants also discussed how health care and public health decision makers, with numerous competing priorities for their time and resources, could use existing HIEs for decision support without the need to build or rebuild models for each new situation. Hupert recommended that health care look to other sectors in which modeling has achieved successful results, such as manufacturing systems or software development. How does the model work under ideal circumstances, and what are the tactical models for how to redirect when things go wrong? One view of how modeling should integrate into public health is that users should not need to think about the model itself. Hupert added that the modelers need to address the âright questionsâ to derive useful information that can be passed on to the decision makers (i.e., without the model itself being forefront for the decision maker). Cairns pointed out how people are very comfortable with the models used by the National Weather Service. That modeling system has been incorporated into daily life, with dissemination tools (e.g., cell phone apps for weather) reaching across the world. A question is how can public health provide something that has functional reality and importance to people in their daily lives. Hubert said the National Weather Service is able to do this because hundreds of millions of dollars have been invested over the past 50 years, to the point where it is part of
74 IMPACTS OF THE ACA ON PREPAREDNESS peopleâs daily lives. People are not concerned with the technology behind the tornado warning, but they know what action to take when there is a warning. A challenge for public health preparedness is that modelers are dealing with such potentially rare events that would not impact peopleâs daily lives like weather does. Cairns countered that North Carolina is tracking emergency health records across entire populations, collecting data on health care encounters every day. Essential to the National Weather Service modeling are timely data and good distribution of sampling. As more people become covered and interact with the health care system and data points are entered into health information systems in real time, the question is how to take that comprehensive sampling and turn it into information of value in a timely manner. Cairns added that many of the predictive analytics used in the detection of emerging health threats (e.g., North Carolina Bio- Preparedness Collaborative; see Chapter 8) were not developed for health care, but for intelligence, fraud detection, financial services, and other venues. He suggested that trying to match patients with health resources (e.g., available vaccine) is similar to just-in-time supply chain management principles. We need to think differently about what we are doing and embrace technologies and tools, he said.