|
Abstract:
|
We evaluate population associations from case-control data via simulation of the reconstructed population method (RPM) (Sommerfelt, et al., Clinical Infectious Diseases, 2012), which makes post-hoc adjustments via weights. The weighted case-control data should reasonably represent the population from which the cases were drawn, thus allowing for evaluation of outcomes that were not the original case definition. We simulate a population of 50,000 children, from which 1,000 cases and 1,000 controls are drawn at random. A random sample of 5,000 is also taken, from which weights for the RPM will be determined from an estimate of the probability of the event that defines a case. A case is defined by diarrhea with onset within the last seven days, and the outcome of interest is mortality. We assume a pathogen is associated with both diarrhea and mortality, and use RPM to evaluate the population association of the pathogen with mortality. We compare the odds ratio estimate from weighted logistic regression using RPM with results from analyses of controls only and of cases and controls combined, both of which will in general produce biased estimates.
|
