Confidence Interval Estimation in Population Attributable Risk for a Case-Control Study

Abstract

Background: There exists increasing interest and need for constructing confidence intervals to demonstrate the precision of point estimates of population attributable risk. Methods: We propose some methods of substitution and bootstrapping which may provide "quick and easy" ways of obtaining confidence intervals of population attributable risk estimates in case-control studies. We compare our proposed methods with a previously published standard method, and demonstrate implementation using popular statistical software. Results: Illustrating the approach, we simulated data using different exposure proportions, odds ratios, and sample sizes. Simulated data were applied to proposed methods and the standard approach. In general, the confidence interval width decreased with increasing sample size, regardless of the approach or selected values of proportions and odds ratios. When the sample size reached 1000, the proposed methods performed well, with the difference in interval width ranging -1% to 3% of the standard method. Conclusions: The proposed procedures provide useful alternatives to the existing maximum likelihood estimator, especially when proportion and odds ratio estimates are obtained from separate sources of data. Another major advantage of the proposed approaches is the ability to provide confidence intervals for adjusted population attributable risk, even when some of the adjusted covariates are continuous in nature.