Keeping It Clean

By Laurie Gelb, July 7, 2011

Recently I became aware that my husband’s national pharmacy record contains not only his own data, but that of another patient.

The root cause: the other patient, in another state, with a different payor, was never asked to confirm his address when he picked up his meds. He has the same DOB, first and last name as my husband, though their middle initials, and, of course addresses, are different. Naturally, a "boomer" generation yields birth date clusters. Had anyone ever asked him one simple question at each pickup, “What is your address?” and compared it with the primary address on screen, I wouldn't be writing this.

Between the pharmacy that repeatedly pulled up the wrong record, me, the health plan, the Web team and the pharmacy’s HQ, there have been about 20 phone calls + a series of logins to investigate and re-separate the records of these two patients. I have participated in about half of those. I was told that “one other case” has occurred in memory, meaning probably thousands as yet undetected or unresolved.

A month since I first notified the pharmacy of the issue, the incorrect data are still mingled, though not for lack of trying. As a last resort, my husband’s record has been deleted and re-created, so far with two sets of login credentials for the new record, neither of which works.

Now the question for your EHR vendor: what automated internal validity checks are run on the data populating the record associated with a single MR number, other than obvious single-field validations like date formats? We've already seen the error rate in e-Rx. An EHR selling point is medication alerts. We would expect a pharmacy record to do as well, no? However,  for months now, no edits or alerts have popped up, though the combination of my husband and his counterpart results in a patient who has been on two macrolides, a steroid, warfarin, rx NSAID, ED drug, opiate and four antihypertensives, among other things.

So to your knowledge, do clinicians entering EHR data routinely verify anything other than name? Or do they simply presume the applicability of a paper chart that someone else pulled, or the EHR that they just opened? We certainly can’t tell ourselves that they would always spot internal inconsistencies.

No doubt accidental merges have already occurred in EHRs. And clearly the structure underlying most EHRs (if not all) would have a difficult time backing out a large quantity of data and re-associating it with a second identifier, until we really standardize import/export formats. 
And is it the clinician's job to reassign data into the appropriate records?

Certainly there are HIPAA implications as accounting of disclosures becomes more robust. With an increasingly cloud-based environment but no über-record, contradictory information will find its way into multiple databases, with little impetus or procedure for reconciliation.  I don't see the words “cleaning” or “data validation” anywhere in the PCAST HIT report, or in too many near-term HIT agendas. 

My pharmacy woes don't bode well for the far more complex EHR. As it happens, I've seen errors in every personally-verifiable EHR I've ever skimmed, including at sites used as Federal models. And if I weren’t ordering refills via mail order, I would never have seen the merge, with unforeseeable consequences.

For example, my husband wears a MedicAlert bracelet, linked to an accurate drug list. In an ER, which list would a physician believe: the one from a pharmacy, that a fully functional EHR will link in, or the “self-reported” one? But the former would be dangerously misleading, and, in fact, would also call the list of conditions on my husband’s wrist and his PHR into question. Then what?

Probability of 100% human verification in the next decade? Zero, unless you design systems that require it. Of course, there are many solutions for positive ID, from biometrics to unique credentials. All require time and money. Whose?

In the zero payment for errors mindset, as EHRs become the go-to reference, who does the cleanup and how? And why -- what are the incentives for doing so? The answers to these questions may influence your cost trend over the next few years than we yet know.