Although we aspire to create learning health systems that can deliver precision care, too often patients move faster and further than their health information. Among the many challenges, health IT system interoperability, getting medical record systems to "talk" to each other, is top of mind for many.
Biomedical informaticians have seen interoperability as a wicked problem for a long time, but the 21st Century Cures Act sharpened our focus on it with its provisions for certification that information be “accessed, exchanged and used without special effort” through technologies such as open application programming interfaces (API). The language in the Cures Act is quite broad, and the thought of trying to make all data elements for all purposes is enough to make many want to give up. Now seems a good time to remind ourselves that when it comes to data standardization, not all juice is worth the squeeze.
As health IT systems are increasingly making data available through open APIs based on the FHIR standard, we also face the formidable challenge of representing health data elements with common vocabulary standards so that different applications can understand the content. Even within a common type of health data, such as laboratory test results, the effort to standardize every last test is extensive. And there are diminishing returns.
When we studied standardized test results within the Indiana Network for Patient Care, the largest inter-organizational clinical data repository in the country, there was a Pareto type distribution where a few tests accounted for most of the result volume. Indeed, less than 20% of the tests accounted for more than 99% of the volume, and that same set of tests accounted for all of the results for 99% of patients. Further, the effort to standardize all data elements is disproportional; the Canada Health Infoway has noted that about 70% of laboratory tests take about 30% of the total mapping time (the “easy” ones), and the remaining 30% (the “hard” ones) will take the last 70% of time.
Both nationally and locally, our standardization efforts benefit from a clear picture of what we trying to accomplish and which data are needed for those purposes. Some fruit is always lower than others. In my experience, the data elements ripest for standardization are those that a) are already available in a discrete electronic format, b) there is a business purpose for leveraging the standardized data, and c) clinicians would rejoice if they were aggregated from many sources and available without extra input burden.
We do need to be mindful that different end goals will require different kinds of standardization. In another study, we described how limiting your upfront mapping efforts to only common tests might hinder secondary uses of laboratory test data like public health reporting, quality measure calculations, etc.
The ONC’s specification of a Common Clinical Data Set was a step in this direction, and the FHIR US Core Implementation Guide (Release 1) provides further clarity on how to implement that set with a standard, open API. The FHIR US Core’s Vital Signs profile provides a good example of picking a high value target and being precise about the standardized structure and codes (semantics) we should use for these data. Unfortunately, there are other high value targets, such as radiology procedures, that at a national level we haven’t yet pressed for standardizing.
As we expand the breadth and depth of health data our applications use, especially in emerging areas such as social determinants of health, a key consideration should be understanding which variables are worth the standardization squeeze.
About the Contributor
Daniel J. Vreeman, PT, DPT, MS is Director, LOINC and Health Data Standards at the Regenstrief Center for Biomedical Informatics and the Regenstrief-McDonald Scholar in Data Standards at the Indiana University School of Medicine. Dr. Vreeman’s primary research focus is on the role of standardized clinical vocabularies and standards to support electronic health information exchange.
LOINC is the world's most commonly used universal standard for identifying health measurements, observations, and documents. With support from the National Library of Medicine, the Regenstrief Institute, the Regenstrief Foundation and other organizations, LOINC is an open, freely available standard.