Observational Studies of Treatment Effectiveness: Some Cautions

The last decade has seen an extraordinary proliferation of evidence that statins decrease mortality in patients who have, or are at risk for developing, cardiovascular disease (1, 2). The benefits extend to high-risk patients with normal cholesterol levels (2) and have led some to suggest that all individuals older than 55 years of age should receive statins as part of a “polypill” (3). In this issue, Spencer and colleagues (4) add to the evidence in favor of statins. They conclude that statin users with an acute coronary syndrome are less likely to present with ST-segment elevation or myocardial infarction than those not receiving statins (odds ratio, 0.78 [95% CI, 0.70 to 0.86]). More surprisingly, they also conclude that starting statin therapy after admission markedly decreases in-hospital mortality (odds ratio, 0.38 [CI, 0.30 to 0.48]).

Whereas randomized trials have provided much of the previous evidence supporting statins (1), Spencer and colleagues have conducted an observational study. This form of research design should always raise concern that bias in some form accounts for the findings. In observational studies, patients and their physicians select treatment on the basis of clinical need or preference, which can result in differences in clinical outcomes solely because of differences between those who do and do not receive a treatment. In contrast, random assignment makes it likely that the frequency of all patient characteristics, known and unknown, will be the same in the treatment groups. This powerful feature makes the results of randomized trials less susceptible to bias than observational studies. Biases associated with observational studies are common and have previously misinformed clinical practice. For example, …