There are a lot of initiatives underway in health care, all supposedly to improve quality and control spending. One that has been around for several years is measuring provider performance, on ultimate outcomes, processes of care, patient satisfaction and spending. In the latest phase of these initiatives, reimbursement is partially or completely dependent on performance on these measures. The stakes have become very high, with not just financial consequences, but reputational ones for providers. And if these performance measuring and payment initiatives have unintended consequences, they could actually make care worse. A Perspective in the Journal of the American Medical Association illustrates the potential dangers of one kind of error in performance measurement. (JAMA Article)
The error is called surveillance bias. The underlying cause is from one group of patients being followed and treated more closely than others, leading to “the more you look the more you find” phenomenon. When surveillance bias is present, what appear to be differences in prevalence or outcomes may just be differences in how closely patients or providers are examined. If the same level of surveillance is not routinely used for everyone in a similar situation, you are likely to have surveillance bias affect your results. An example used in the article is deep vein thrombosis, a very serious complication for trauma patients. Some providers use a tool called duplex ultrasound to screen asymptomatic patients, but others don’t, believing that it is not cost-effective. Providers who do the screening find more DVT, even though it may not be clinically significant. So if they are measured on their DVT rates, they either look worse if the rate is supposed to be lower, or better if they are supposed to find and treat more of it. Either way, it is not clear that more screening leads to better patient outcomes.
The authors point out the unintended consequences of surveillance bias. Providers may avoid testing to minimize reported complications. DVT is a never event for some payers, meaning a provider would be incented to avoid finding and/or reporting it. To avoid these problems, surveillance should be clearly defined and standardized across providers, so that everyone is looking for the same things in the same patients in the same way. Measures for which this is difficult probably shouldn’t be used. The consequences of measurement error and measurement comparison error across providers are too great under pay-for-performance programs and it is unfair to providers to subject them to uncertainty which has financial and reputational consequences. But the biggest risk of all is that defects in program design and operation undermine confidence in the notion of performance measurement and consequences. These programs can be a critical component of both improving quality and lowering cost.