gms | German Medical Science

Background: In the realm of observational studies, the prevalence of post-selection inference issues is quite substantial. Nevertheless, these issues tend to be overlooked and the results of the statistical modeling are often reported and interpreted without considering their inherent uncertainty. A recent proposal by Efron introduced a bootstrap approach to address the issue of incorporating uncertainty in statistical inference following model selection. Nonetheless, the performance of Efron’s approach remains uncertain

in the context of (matched) case-control sampling designs that involve a binary exposure variable and a mixture of confounding variables.

Objectives: The objectives of the TORONTO Monte Carlo simulation study were twofold: firstly, to evaluatethe precision and accuracy of the bagged estimator employed in Efron’s approach; and secondly, to compare and ass ess the coverage probability of the confidence interval estimators (specifically, smoothed (bias-corrected), percentile and standard) proposed by Efron within (matched) case-control sampling designs using a regularized procedure in the logit model and non-parametric bootstrapping.

Methods: A comprehensive Monte Carlo simulation study was conducted to evaluate the operating characteristics of Efron’s approach in the context of (matched) case-control sampling designs. The data-generating process incorporated random variables with different prespecified probability distributions. Pseudo-random samples were generated from a hypothetical population consisting of 100,000 units of observation in order to emulate a case-control sampling design for the prespecified scenarios. The matching for matched case-control sampling designs was accomplished through propensity score matching.

Findings: The key findings of the TORONTO Monte Carlo simulation study will be summarized in this talk.

The authors declare that they have no competing interests.

The authors declare that an ethics committee vote is not required.