By Dave Lareau, CEO of Medicomp Systems
Thanks to the wider use and acceptance of FHIR, CDA and other standard terminologies, the healthcare industry is closer than ever before to gaining a technical infrastructure for managing the sending of “stuff” back and forth between systems.
But sending information from one system to another is of limited benefit to clinicians without tools that quickly assimilate incoming information and make it actionable at the point of care. Merely possessing the technical means to exchange information using standard formats, including the hot new FHIR standard, is not enough. Clinicians need the ability to receive information, make sense of it, and then view the unique data points that are related to a specific clinical problem. As long as the information is not organized in a usable format, the healthcare industry still has not achieved true clinical interoperability. Instead, we’ve simply punted over to clinicians a bunch of “stuff” that’s no more valuable to patient care than a dumpster fire.
Actually, it is worse than that.
For the last couple of months, we have been working with FHIR APIs from some of the big EHR vendors – and have made some troubling discoveries. One of the two largest vendors has very few lab results coded to a usable standard, such as LOINC or SNOMED. Most EHR vendors store encounter information or “progress notes” in a manner that doesn’t support the bulk-loading of notes, but instead requires users to retrieve encounter notes one-at-a-time. These cumbersome encounter-by-encounter designs made much more sense when EHRs were primarily used to facilitate billing. However, this design fails the usability test when managing a patient’s health over time or tracking the progression of chronic conditions.
It gets worse.
EHRs, particularly those that incorporate dictation or speech recognition technologies, typically contain a significant amount of free text within their encounter progress notes. The industry has long hoped that natural language processing (NLP) would eventually mature enough to convert all that free text into codified data. Unfortunately, the error rate for converting speech-to-text-to-coded data is still at least 10 percent, rendering the technology too flawed for clinical decision support and analytics applications, and unacceptable for the creation of trustworthy insights based on artificial intelligence algorithms.
The good news is that through wider use of FHIR, CDA and terminology standards, the healthcare industry is creating a better technical infrastructure for sending data. The bad news is that we risk overwhelming the clinician on the receiving end with imprecise or unusable information.
There is a better approach.
The terminology standards of SNOMED, LOINC, RxNorm, ICD-10, CPT, UNII, CVX, DSM-5, and others make it possible to exchange clinical data between systems; but, the usefulness of these standards will be limited until EHR vendors provide a means of presenting this data in a usable format to clinicians at the point of care.
At the same time, we need tools for getting more precise clinical data into healthcare information systems, and for making specific data available to address the unique clinical issues of individual patients. By addressing these requirements, we will be well-positioned to provide true clinical interoperability in a manner that leads to better coordination of care and optimal patient outcomes.