Health care data and hospital quality:Is "not dead" good enough?

Dueling interpretations of health care quality data in the UK is leading to a review of the use of death rates as indicators of hospital quality. I recently discussed the implementation of a health care quality watchdog in the UK that uses Hospital Standardized Mortality Rates (HSMRs) to help establish whether the number of patients dying in the hospital is within the expected, acceptable range. (HSMRs were developed and validated at the Institute for Healthcare Improvement, the think tank and quality improvement enterprise run by the newly designated Medicare-director Don Berwick. It's a small health care world!)

But now the shortcomings of available data is causing the British Department of Health to reconsider its use of HSMRs. The concern was originally unearthed when a British health trust simultaneously received ratings of "appalling" and "top ten" from two respected assessors of health care quality. While number crunchers are still flexing their grey matter to discover the underlying reasons for these conflicting analyses, one potentially disturbing trend has been discovered.

The diagnostic codes that hospitals assign to procedures and patients turns out to be exceedingly important on eventual measures of quality. At the same time, attributing codes to procedures may be more art than science. Patients who receive coding for palliative care, for example, are assumed to be entering the hospital in ordered to die, so their death is not considered unexpected. And according to a review of hospital records, the number of patients entering hospital with this code has increased some 300% over the past 5 years. Since multiple codes could potentially be assigned to many patients, it's raised the possibility that hospitals may be gaming the system: Lowering their HSMRs, and hence boosting their quality scores, by favoring certain diagnostic codes over others.

This is of course one of the problems with using administrative data, rather than actual clinical data. When you review a medical chart (clinical data) it becomes relatively clear what the problem was, and whether care was appropriate. When you review a medical bill (administrative data), it's a lot harder to tell why someone did something to you, and whether they should have done something else instead. Before we reform and improve health care quality we need good data so we can see where we are, where we are going, and when we get there. Unfortunately, if you think health reform is challenging, try reform the gathering and sharing of health care data. It appears we have a ways to go.