Author Affiliations: Department of Epidemiology, Biostatistics, and Occupational Health, McGill University, Toronto, Ontario, Canada (firstname.lastname@example.org).
To the Editor: The study by Dr Canto and colleagues1 provides an example of an epidemiological phenomenon that deserves wider recognition. Differential selection from an underlying population cohort into a study data set can reverse the direction of observed associations, making a deleterious factor appear protective. It is well-known that conditioning on a variable that is affected by both an exposure and outcome can produce a distortion known as selection bias.2 Admission into the analysis data set in this study was a function of both the exposure (number of cardiovascular risk factors) and the outcome (all-cause, in-hospital, or 30-day mortality) because deaths occurring before hospitalization and patients with existing cardiovascular disease diagnoses were excluded. Approximately 30% of MIs lead to death prior to hospitalization.3 In the study, 75% of those who were admitted to the hospital after their MI were excluded (1.62 million of 2.16 million; Figure 1 in the article).
Banack HR, Harper S, Kaufman JS. Coronary Heart Disease Risk Factors and Mortality. JAMA. 2012;307(11):1137-1138. doi:10.1001/jama.2012.324