John R. Zaleski, PhD, CPHIMS

When I began conducting my own research ~20 years ago, I envisioned interaction among large numbers of databases from which I could automatically draw needed information to assemble the models I required to verify performance of one patient in comparison with a population of patients having the same specific characteristics. The comparison of a test patient with a population was not only the basis for my predictive methodology but a basic tenet of Bayes’ Theorem and conditional probability. I envisioned back then a day would come when all the data I was attempting to cobble together by hand and by computer would be readily available… perhaps 10 years down the road.

Fast forward to today.

In the AHRQ’s opening FAQ about what comparative effectiveness research (CER) is, they state that, they state that “Comparative effectiveness research is designed to inform healthcare decisions by providing evidence on the effectiveness, benefits, and harms of different treatment options.”

Physicians rely on evidence, in the form of corroborating (and some contradictory) studies to validate and verify hypotheses related to treatment. Accepted treatments and approaches for care are those most usually associated with corroborating evidence–a great deal of corroborating evidence. Not all data arrive in the form of simple measurements. Data take on the form of studies, published research papers, laboratory results, films, guidelines, textbooks, and from bedside medical devices.

To a degree, the electronic health record (EHR) was seen as a key element in achieving the type of data richness that was necessary to support comparative effectiveness research. The EHR and its adoption are necessary but (yet) insufficient in terms of the CER enterprise. In “Optimizing Health Information Technology’s Role in Enabling Comparative Effectiveness Research“, Navathe and Conway note that “…[f]or CER to be conducted, …EHRs must be connected to data networks enabling access to at least portions of their captured data. It will be challenging to implement EHRs on a large scale and to develop electronic networks substantial enough to produce observational data that alter clinician, patient, and other decisions.”

As Navathe and Conway state, healthcare data is not just personal, it is a strategic asset upon which basic research depends. Despite visions of a unified, singular database “Borg” where all data would go to be mined endlessly by large mainframes, the facts most probably near term and into the future are that data will be federated. They will reside in many different pockets “owned” by many separate and disparate institutions large and small. The key, then, to my mind, is not figuring out how to assemble them into some common structure, but enabling their inter operative interaction, much the way the Internet operates today. When we type in a Google query, we are not interacting with some great big “WOPR” computer (references to the 1983 movie WarGames), we are interacting with millions of computers around the world.

What I did not realize 20 years ago that I have finally come to realize is that I thought then the “hard” problem was solving the predictive model for the patients I was monitoring, when the really hard problem was assembling all of the data together from the many disparate sources. Through all of the standards committees and claims of interoperability, we are really a long way off in terms of enabling the one thing that would allow us to do what I had started out almost 20 years ago doing: automating the data collection, from any source, that would allow us to focus on the real business of health care research, and would enable the vision of comparative effectiveness research.