The Average is for the Average

In the 19th century, Adolph Quetelet developed the notion of a physically average human “who is characterized by the mean values of measured variables that follow a normal distribution.” According to him, “the determination of the average man is not not merely a matter of speculative curiosity; it may represent the most important service to the science of man and social service.”  Ever since then science and medicine has largely focused on the normal distribution with the mean of that distribution as the primary indicator of response and the related p value as a measure of significance of the response. In the current era of escalating health care costs with diminishing returns there is an effort to bring standardization and predictability to  healthcare under the stimulus of the federal government. Concepts such as population health, clinical practice guidelines, value based care/purchasing are all interrelated solutions to standardize health care.

However, medicine is a vast and heterogeneous domain and attempts to standardize care is important but should be pursued carefully. In our haste to decrease cost and increase efficiencies, guidelines are being developed at a rapidly increasing rate. For example, the number of clinical practice guidelines has increased from a mere 72 to a number greater than 7300. However, it is important to note that these guidelines and quality of care measures are largely developed by experts who synthesizes results from randomized controlled trials (RCTs) looking at a single disease. RCTs provide estimates of how well a treatment works on an average patient in the trial but cannot predict the response of treatment of an individual patient. There is inherent variability in the response with some patients who derive substantial benefits, some who derive a little benefit, and some who are harmed. The results are compiled as the mean response with the p value as a measure of significance. However, it can be dangerous to misapply the results of one or even a synthesized group of RCTs to an individual patient because patients seen in a clinical setting are not average and do not fit in the Gaussian model of the RCT. There are factors internal and external to the patient that make them decidedly “abnormal.” Patients have social challenges that hamper medication adherence, they have genetic variations that render medications ineffective, their disease process is more or less severe than the average and consequently, they derive limited benefit from the medication, they have multiple medical conditions that adversely or incompletely interacts with the treatment offered. Randomized controlled trials do not capture any of this clinical diversity and thus,  synthesized clinical trials poorly predict how an individual patient will respond to treatment in that particular context of patient care. It is no surprise that the heterogeneity of responses might actually be the rule rather than the exception. According to an article in Nature, millions of people routinely take drugs that are ineffective for them. This alarming number ranges anywhere from 5 to 25%. Commonly prescribed drugs such as statins benefit as few as 2% of patients. However, PQRS 438 “measures the percentage of patients who were prescribed or were on statins” or PQRS 002 measures the “percentage of patients aged 18-75 years of age with diabetes whose LDL-C was adequately controlled.” So, how applicable is a quality measure that might only be effective for 2% of the population?

In all industries, standardization is important to maximize compatibility, interoperability, safety, repeatability, and quality. The difference in health care is that the stakes are potentially higher as the “widget” is the health of a person. Consequently, although attempts at standardization via clinical practice guidelines and quality metrics are important as they can potentially provide foundational guidance for busy frontline providers, we might be prematurely pushing guidelines and metrics that do not accurately reflect the population under consideration – an aging, multimorbid patient on multiple medications with a variable social support system. As stated in Goodhart’s Law, when a measure becomes the target, it ceases to to be a good measure. This in turn not only  decreases creativity and out of the box thinking on the part of the practitioner but eventually leads to more poor outcomes than would be predicted by a normal distribution. Clinicians should be very wary of falling into the trap of casual compliance with practice guidelines because misrepresenting the “unaverage” for the average in medicine has potentially dangerous consequences for the patient.