Simulation study of the effect of the early mortality exclusion on confounding of the exposure-mortality relation by preexisting disease.

The authors conducted a simulation study to evaluate whether exclusion of the early mortality (deaths occurring during a prespecified period immediately after baseline) reduces confounding of the exposure-mortality relation by preexisting disease. The simulation specified an exposure that decreased mortality risk in the absence of confounding and then introduced confounding by preexisting disease that biased the "true" protective effect of exposure towards greater risk. In 2,000 cohorts, exclusion of the early mortality (deaths occurring during the first 25 months of a 60-month follow-up period) did not alter the mean hazard ratio for exposure under conditions of confounding by preexisting disease that produced a constant, threefold increase in mortality risk during follow-up (the mean hazard ratio was 1.72 for all subjects and 1.72 after exclusion of the early mortality). However, when the authors specified confounding by preexisting disease which produced a threefold increase in mortality risk that attenuated over time, exclusion of the early mortality consistently identified the "true" protective effect of exposure (the mean hazard ratio was 1.07 for all subjects and 0.31 after exclusion of the early mortality). Thus, under conditions of confounding by preexisting disease which produces an increase in mortality risk that attenuates over time--an effect that does have empirical support-the early mortality exclusion can be very effective in revealing the "true" exposure-mortality relation.