In October 2015, Congress again questioned the leaders of the Department of Defense (DoD) and Department of Veterans Affairs (VA) on the progress of making their Electronic Health Record (EHR) systems interoperable. This followed a Government Accountability Office (GAO) report suggesting that the goal of interoperability even by 2018 is overly optimistic. The struggles the DoD and VA face as they strive for interoperability provide valuable insights on the hidden barriers toward EHR systems interoperability in general. First, we need to explore history to establish a frame of reference.   

How the Interoperability Issue Emerged

An early call for exchanging EHR data between the DoD and VA came with the National Science and Technology Council, Presidential Review Directive 5 (PRD 5) in 1998. By that time, the VA had been working on the advancement of the Veterans Health Information Systems and Technology Architecture (VistA) system for well over ten years and the DoD had built the Composite Health Care System (CHCS) and was starting to build the Armed Forces Health Longitudinal Technology Application (AHLTA) on top of the CHCS foundation to create an updated user interface and to integrate the system’s back-end to be more of an enterprise solution.

The call for systems interoperability instead of just exchanging some patient data began in 2004 when the White House appointed the first National Coordinator for Health. As the DoD and VA communities started to think about how EHR data from one side can support the healthcare processes of the other side, the National Defense Authorization Act (NDAA) of 2008 directed DOD and VA to jointly develop and implement fully interoperable EHR capabilities. Then in 2009, President Obama directed the DoD and VA to “work together to define and build a seamless system of integration with a simple goal: when a member of the Armed Forces separates from the military, he or she will no longer have to walk paperwork from a DoD duty station to a local VA health center; their electronic records will transition along with them and remain with them forever.” By that time, the DoD had created the AHLTA Clinical Data Repository (CDR) which centralized patient records across the military services of the DoD under the Master Patient Index, and the VA had allowed the VistA system to remain evolving local deployments with potential variations across over 100 operating instances of VistA. Naturally for DoD, data intensive radiology records cannot be easily centralized because of bandwidth constraints. Naturally for VA, local data sets can still be collocated into regional data centers as was completed in the 2013 timeframe.        

In response to the 1998 PRD 5, the VA established a series of communication protocols known as the Bidirectional Health Information Exchange (BHIE). BHIE was effective at sending patient data from VistA instance to VistA instance. However, DoD CHCS could initially push only some information to VistA via BHIE, and significant differences between the data structures of CHCS and VistA limited effective integration. From 2004, the DoD worked with the VA on adding web service adapters to BHIE in an effort to create true bidirectional EHR exchanges between AHLTA/CHCS and VistA. Then in 2009, the departments began working on the Virtual Lifetime Electronic Record (VLER) initiative to enable service members and veterans to access their health records throughout their lives. In response to the 2008 NDAA and presidential guidance, the DOD and VA Secretaries committed in 2011 to developing a new common integrated EHR system with the goal of implementing it across the departments by 2017.

The Integrated EHR (iEHR) Program Office began with the ambitious plan of creating a new combined DoD and VA EHR system consisting of over 50 modules, such as pharmacy, immunology, and orders management, which were to be developed and released across eight capability sets. Cost estimates for the iEHR system grew to over $8 billion, and two years of attempts at collaboration between the DoD and VA led to minimal progress. In February 2013, the newly appointed Secretary of Defense, Chuck Hagel, and Secretary Eric K. Shinseki of the VA announced that the DoD and VA would pursue separate acquisition paths for their next generation EHR systems.

The VA released the VistA 4 Product Roadmap in March 2014, laying out an incremental modernization path for the VistA system. According to this roadmap, the VA would 1) converge the VistA code base to get a Standard version of VistA with 74 modular products and 42 core capabilities for redeployment across the instances, 2) create an enterprise Health Management Platform (eHMP) with open interfaces based on Health Level (HL) 7 standards to enable integration across legacy and new modules, and 3) develop new modules across four planned product releases extending to 2018. The new DoD Healthcare Management System Modernization (DHMSM) program released a Request for Proposal (RFP) in August of 2014 to execute an acquisition strategy of replacing the AHLTA/CHCS system, which is not very suitable for modular upgrades because of tight integration, with a commercial EHR system. The DoD awarded the DHMSM contract to the Leidos / Cerner Team in July 2015 after evaluating other EHR systems to include Epic, Allscripts, and a commercial version of VistA.

Looking Beyond the Fault of Leadership

As the DoD and VA pursue two vastly different EHR system modernization paths, Congress is rightfully concerned about whether interoperability objectives still within the plans of the DoD and VA will ever be implemented. The problem further becomes more of interoperability between all EHR systems with the entry of the Cerner commercial system into the debate. To say that the problem is one of DoD and VA leadership might be correct, but it does not provide insight regarding why generations of leaders have all stumbled in this endeavor. To delve deeper into the barriers toward DoD and VA EHR system interoperability, we need to first discuss a few myths.

Myth 1: Interoperability is so hard because of the immense number of eligible beneficiaries for medical service within the DoD and VA.

The DoD and VA do have a substantial number of eligible beneficiaries: more than 9.5 million for DoD and more than 22 million for the VA. These are large numbers, but if the patient records are very simple and standardized on both sides, the concern would just be one of operational capacity and bandwidth. If the data structures of DoD and VA electronic records could be cross mapped to allow transformation equations to convert data back and forth between the two data structures, then extracting, transforming, and loading data for one hundred patients and one million patients would be the same process. In fact, an enterprise service bus (ESB) could even synchronize the data given adequate capacity and bandwidth. In data migration or synchronization, there might be a bandwidth problem if a million health records all containing radiology (imaging) data needed to be transmitted at once. That is not the required level of daily health records integration between the DoD and VA. What large patient populations can do, however, is to increase the complexity of health records and magnify the inconsistency of data structures between groups of health records as in the case of the VA. This is connected with patient demographics, diversity of patient history, and treatment approaches / workflows in different medical communities. The challenges presented by these factors will be further explored later.      

Myth 2: Interoperability should be easy because the patient communities of the DoD and VA are similar.

Currently, the Military Health System supported by AHLTA / CHCS has a patient population of approximately 1.6 million active duty members of the U.S. armed forces, 2.4 million family members of active duty service members, and 5.5 million military retirees and their family members. The VA hospitals and clinics supported by VistA has a patient population of approximately 8.8 million veterans annually with service-connected disabilities, unemployable status due to service related conditions, special economic needs, special conditions of services, and other qualifying conditions. Although the central patient groups on both sides are affiliated with the military, these patient communities are by no means similar except for the small percentage of military retirees who decided to use VA facilities to treat service connected conditions instead of exercising Tricare benefits.

The active duty patients within the DoD are probably in better physical condition than the general population, but some of these patients require immediate trauma care from combat and training related injuries. The military retirees represent an older patient population, albeit one that has been able to complete a long military career while either having no disabilities or finding effective ways to overcome their disabilities. Finally, the family members of these two DoD groups are probably very reflective of the general population in terms of medical needs. They in many cases only come to DoD medical treatment facilities (MTFs) instead of commercial hospitals paid for by Tricare for convenience and quality of care.   

In contrast, patients come to VA clinics and hospitals, sometimes with long wait times, to seek treatment for the long-term consequences of serious disabilities and to get medical help because they lack better medical options. Many VA patients have chronic conditions that need continuous series of appointments and some patients walk in for urgent needs. The VA patient base is also an aging population, but it is a population with high levels of special needs such as treatment for post-traumatic stress disorder (PTSD). VA patients do get treatment from commercial hospitals, but they are assigned to external treatment through the VA scheduling process when internal wait times are too long. As explained above, there are some people who have exercised dual eligibility status for both DoD and VA medical facilities. This small group is not a counter balancing force to how differences between the patient groups impact the evolution of electronic health records.

Myth 3: Interoperability should be easy because of the common clinical processes within the DoD and VA

The DoD Defense Health Agency (DHA) oversees around 60 DoD hospitals with over 300 associated health clinics. The approximately 58,000 civilians and 86,000 military personnel in this system provide not only day-to-day care to eligible patients but also provide care for wounded warriors, service members in theater, and combat medicine. The Veterans Health Administration (VHA) oversees around 150 VA hospitals with over 800 associated health clinics. The over 270,000 personnel in this system provide on-site care and telehealth care for patients with chronic conditions.

Within these two vast medical communities, clinicians are continuously tracking advances in medical equipment, new drugs, combination therapies, and treatment procedures / techniques within the overall healthcare community. Then, they are responding to these advances in refining clinical workflows relative to 1) DHA and VHA mission priorities, 2) constraints and opportunities within their medical infrastructures, and 3) the needs of their patient base. Further, each community might be spearheading medical innovations and establishing best practices in areas such as traumatic brain injury and polytrauma care. Such internal medical advances could further create differences in workflow.  

In regards to mission priorities, the DoD must address the deployment readiness of troops, capabilities of the medical staff for combat medicine, and capacity of the medical workforce for handling a high number of causalities during conflict. The VA must in contrast address the efficient use of limited resources to serve a patient population with heavy needs. Being essentially a managed care medical organization, the VHA prioritization of procedures, specialists, and equipment may all have differences relative to the DHA. So despite the presence of best practice treatment pathways, there are differences in clinical and business processes between the DoD and VA. These differences drive variations in data structures and allow us to see that there will be operational impact if one side simply adopts the processes and data structures of the other side.

Why Is Interoperability So Hard?

At this point, DoD clinicians can see patient data in VistA and VA clinicians can see patient data in AHLTA. Equally, commercial hospitals within most of the fifty states can now pass health records information through the regional Health Information Exchanges (HIE). Finally through the VA’s Blue Button initiative, progressively more people will be able to download their own personal health records.

However, the true interoperability question moves beyond simple passing of data: it is the question of whether your information will work in my system and processes and whether my information will work in your system and processes. There are two brute force ways to integrate EHR data. The first way is to force each side to live with a common process and data structure that is the overlap of both sides. Overlaps can be identified by eliminating all the patient community and medical community specific process and data details on each side. Or, overlaps can be identified by creating summary levels of data where a common taxonomy can be established. With each of these solutions, both sides sacrifice either relevancy or fidelity. During the days of the iEHR program, the DoD and VA tried to established common requirements and tried to use the 3M Health Data Dictionary as a convergence point for their processes and data structures. Since no development of an actual capability module ever made it through the procurement process after two years, we can only speculate that clinical subject matter experts on both sides were not fully comfortable with the compromises that would have been required. And we cannot blame integration issues on legacy technologies in the case of the iEHR program because the program office had the mandate to create new technologies for the joint DoD and VA system. 

The second brute force way to achieve interoperability is for one side to compel the other side to adopt their process and data structures. This was the case when the VA tried to convince the DoD to procure the VistA system in the DHMSM program. What was perhaps not discussed was the outcome of DoD procuring VistA and evolving VistA within DoD to match its own process and data structures. Then, we end up with VistA systems on both sides that are not interoperable. With DoD now committed to using the Cerner commercial system, it remains to be seem as to whether DoD will start to adopt commercial processes and data structures or whether a version of Cerner will be established with process and data structures tailored for the DoD. Few will probably remember that the DoD once tried to use the commercial PeopleSoft human resource system to create a military personnel record system and ended-up cancelling the Defense Integrated Military Human Resources System (DIMHRS) program in 2010 after spending over $850 million. Assuming that DHMSM succeeds in implementing Cerner for DoD MTFs, the next question might be whether DoD will try to convince the VA to adopt Cerner.

If brute force integration approaches do not work, we should then perhaps step back and take a closer look at the nature of complexity in healthcare processes and data. This complexity clearly exits because we are endeavoring to understand and fix millions of incredibly sophisticated organic systems commonly known as human bodies. Modern medicine is telling us that the best medical treatments should be tailored to each individual even as medical communities are debating standards and common taxonomies. With each day, the scope, fidelity, interpretation, and interrelation of medical data advances. These descriptors of medical data are important because, while true information about the human body is a continuum, the ability of modern medicine to capture this information in the form of data is discrete and limited. For example, our blood pressure continuously fluctuates, but might only get measured once per day or once every several days. Then when two medical communities measure the same blood pressure at slightly different intervals or even at different times of the day, we have a simple data integration issue even though both sides are measuring exactly the same thing. It is therefore beneficial in any data integration and system interoperability discussion for all sides to explain what they see as their scope, fidelity, interpretation, and interrelations of data before debating about integration approaches.

Scope of Collected Data: This is the dimensionality of behaviors and factors being measured, the characteristics of each dimension, and the ranges in each dimension for collecting data. In medicine, dimensions might include physiological behaviors, mental behaviors, current environment factors, past environmental factors, bacterial and viral behaviors, family historical factors, and past medical history. At this level of discussion, we are not debating what data is actually being collected. Nevertheless, two medical communities might discover that they have substantial differences in characterization and measurement ranges even though the top level dimensions sound the same. Just within the current DoD EHR strategy, one obvious dimensional difference between the commercial Cerner system and the scope of DoD medicine is the coverage of combat related injuries and conditions. Thus, DHMSM program requirements compel Cerner to expand system and data scope. If two communities such as the DoD and VA cannot agree on scope and one side cannot compel the other side change scope, then the issue is one of fundamental misalignments in medical approach and philosophy.          

Fidelity of Collected Data: Assuming that medical communities can align their scope, the next item of discussion is how data is collected at the fundamental level. The fidelity of data is based on the quality of measurement instrumentation, the methods of measurement, the frequency of measurements, and the accuracy of medical workers. Variations in fidelity can result from differences in medical processes and procedures. However, much of the terminology of data at this level should be very similar because of the common application of technologies and medical best practices. The area where terminology can be complex is for data collected based on human observation. Observational data not associated with defined metrics, although valuable, can be hard to convert into computable data. Apart from select observational data, an analysis of processes and procedures should be conducted to align data fidelity between medical communities if we want to avoid brute force integration approaches and settling for the lowest fidelity common data elements. Adjustments in processes and procedures might be hard to culturally implement. However, it can establish common high fidelity measured data by advancing methods, increasing frequency of measurements, and improving worker accuracy.  

Interpretation of Collected Data: The application of medical data to support diagnostics and implementation of treatment approaches requires organization of data and generation of metadata to provide medical insight. This organizational and computational process can be defined as the interpretation of collected data. Treatment pathways established by the general medical community will specify how raw data should be organized and computed. Further, the totality of treatments within a specific medical community will create requirements for the structure of the EHR database. The challenge in converging the interpretation of medical data between medical communities is that treatment pathways are continuously being advanced in medicine and some areas of medicine do not have a single authoritative body for certifying treatment. Instead, communities such as the DoD and VA take into consideration the unique characteristics of their patient population as well as the limitations / opportunities in their individual medical facilities in determining treatment best practices. Converging the interpretation of collected data might be very hard for medical communities with different mission priorities. However, a full convergence might not be required if the scope and fidelity of data on both sides can be aligned. Such an alignment allows data from one side to be reinterpreted by the other side for system use. The simplest reinterpretation is a transform equation in extracting from the database of one system and loading into the database of the other side. If the interpretation is more complex, the receiving system might require another set of processes for how to compute loaded data from specific external sources in order to make such data useful in the system.    

Interrelations of Collected Data: Currently, the utilization of medical data processed by EHR systems is largely reliant on the skill and diagnostic experience of clinical professionals. In the near future, processed medical data can be used by clinical decision support systems to generate automated diagnostic recommendations specific to each patient. The simplest clinical decision support systems already in use are rules engines, with more advanced systems using techniques such as fuzzy logic, bayesian networks, and genetic algorithms being developed. These advancements create another layer of EHR data that describes the interrelations between collected medical data. As the goal of meaningful use standards established by the 2009 Health Information Technology for Economic and Clinical Health (HITECH) Act is to push medicine towards this future, we need to consider interoperability issues if smart EHR systems start to evolve along different paths in non-integrated medical communities. This concern places some urgency on medical communities to resolve system interoperability issues prior to the emergence of next generation technologies.

A Way Forward

This examination of EHR system interoperability has focused on data integration because there is no point in discussing other issues (such as legacy systems that cannot adapt to new data structures and commercial systems that refuse to expose databases) if we cannot figure out how to integrate data. After years of intense effort, the DoD and VA health information technology communities have probably resolved their data integration issues. Therefore, this article could be more about insights for future EHR system interoperability endeavors.    

By clearly defining the steps in exploring EHR data integration, medical communities can save time and avoid arduous paths. These steps also reveal that data integration efforts require a great deal of inductive analysis. While engineers and operations research analysts are well trained in inductive thinking, scientists and medical professionals are more trained in deductive (reductionist) thinking. So, having a room full of clinicians from divided medical communities debating data integration is probably not the best approach even though each side may have perfect understanding of their own data. Alternatively, engineers and operations research analysts without specific medical domain knowledge cannot figure out EHR data integration on their own.

The best way forward on overcoming interoperability barriers seems to be the formation of interdisciplinary teams where team members are willing to bidirectionally help one another gain medical knowledge and process integration understanding. This type of collaboration is still rare and can be even more difficult when the medical team members belong to different communities. However in a world where technology and medicine are becoming increasingly intertwined, medical communities cannot avoid the benefits and necessity of interdisciplinary and systematic approaches.