The Boston Globe ran a piece this morning about the challenges of using insurance claims data to populate PHRs. It’s nicely done, and highlights what a shame it is that the Boston Globe is teetering on the verge of bankruptcy.

Electronic health records raise doubt – The Boston Globe.

Dave deBronkart, the patient involved, has been an active blogger for several years at e-Patient Dave.

Claims data is very challenging – it’s messy and not particularly precise.  I always envisioned claims information as being useful primarily as a directory, rather than a source of actual content for PHRs. The insurance company knows where you’ve gotten treatment, and that information can be used by a PHR platform to seek out information from clinical systems. Of course, that requires a lot more interoperability in the entire system than we have today.

Another approach is to work backwards – rather than including every claim from an insurance company in the PHR, only include the unambigous subset. A flu shot is a flu shot. But some data is just ambigous – like the cancer in the story, which seems tohave used a broad billing code. And some, of course, is intentional – physicians who file a misleading diagnosis so that a patient can be reimbursed for an off-label use of an expensive drug.  So for these cases – give the user an interface that allows them to review potentially problematic data, and explain clearly that some of it might not be entirely accurate, and there’s a good reason for that.

Finally, consumer controlled health records require very clear statements of data provenance. This was something our group paid a lot of attention to back when we were working on active development in the area, and Microsoft does a pretty nice job in HealthVault as well. I don’t know what Google’s design rationale was in not choosing to surface this information more prominently, but to my mind it was a mistake. Ten years from now we may have clinical data flowing so effortlessly that we can simply sit back and trust in the validity of any automatically sourced data. But that day is not today.