A scalable infrastructure to capture real-world clinical experiments
Last September, I argued in this blog for a data infrastructure, which would be supported by a public-private partnership, capable of capturing real-world, “N of 1″ clinical experiments. Such experiments collectively represent a huge potential repository of clinical scientific evidence. Today, experimental evidence coming from routine clinic encounters is underutilized. Case reports, case series, and retrospective chart reviews, have gained a reputation as weak evidence of causation (typically described as anecdotal evidence) and have thus fallen into disfavor. Too bad, because such anecdotes have historically frequently been the impetus for exploration of a new use for a drug or a signal of an adverse effect that was previously unseen or unrecognized as therapy-related. In contrast, with the rise of managed care and the need for drug and device manufacturers to demonstrate a favorable benefit to risk balance for new therapies, studies that make use of databases linking diagnoses, outcomes and therapies are becoming more common. Such studies are likely to become increasingly important for delivering improved safety surveillance of therapies and would benefit from larger, more demographically diverse patient pools. They would also benefit from a larger amount of clean contextual metadata (data about data). For instance, researchers might now be able to tell when a prescription was first filled by searching PBM or insurance records, but it would be more difficult to ascertain the day the prescribed drug was stopped. However, perhaps the prescribing physician noted in the patient chart (the Electronic Health Record, EHR, in my utopican example) that the drug was stopped 20 days after filling the prescription because of a rash that developed and was restarted after a 30-day off period without reoccurrence of a rash. Charts may be abstracted for smallish studies, but you can’t manually abstract charts on a million patients. Plus, today’s abstractions (e.g. CPT coding) are lossy mechanisms for transforming information into a usable form.
So, what’s needed for both the small case studies and the large epidemiological studies is a scalable infrastructure capable of pooling information from disparate sources and storing it in a way that makes it accessible to a variety of information systems, ideally without significant information loss. I imagined that such a system was far off, but maybe I was being pessimistic. In the June issue of BIO-IT World, Mike May reports on the Mayo Clinic’s efforts towards making this vision a reality. They call their system LexGrid, or Lexical Grid. LexGrid is described by Mayo as “a way to bridge terminologies and ontologies with a common set of tools, formats and update mechanisms.” Basically, as I understand it, LexGrid takes these variously coded ontologies from all sorts of information systems (pharmacy, EHR, lab, etc) and links them together, representing them by a common model. So, local data elements become common data elements, local descriptions become common terminologies, local classifications (like CPT) become defined by the common terminology structure. Pretty cool idea.
Even cooler is that it’s being done in clinical medicine first, and the system’s designer Christopher Chute has some grand visions for how it can be used: “every [person] on the planet would have every health encounter…stored as a structured representation of that event, so that you’d be able to ask useful questions across nine billion people.” Now that’s what I’m talkin’ about. How about pharma getting behind Dr. Chute and his colleagues on this project?
