gms | German Medical Science

Introduction: A competing risks bias is a common pitfall in analyzing the influence of risk factors on acquiring nosocomial infections in a hospital setting. Ignoring the effect on events that preclude nosocomial infection (for example discharge) can lead to skewed results or even completely false interpretations [1]. Methods have been developed that modify a Nested Case-Control (NCC) design for the analysis of competing risks [2], however, these methods can be inadequate in a hospital setting. We demonstrate how an NCC design can be improved to take into account competing risks and thus provide both etiology and prediction analysis for nosocomial infections.

Methods: A traditional NCC design employs time-dynamic sampling to select controls at the time of infection. Our extended method reuses these controls for the analysis of being discharged/dying in an intensive care unit. The controls are weighted with the inverse of their inclusion probabilities for use in a weighted Cox and Log-Binomial model. The inclusion probabilities are calculated with data routinely collected in hospitals, thus our improvements require little additional effort and information in comparison with traditional methods.

Results: We demonstrate this extended method on real data from two Spanish intensive care units as well as simulated data. The estimates for competing events based on reduced cohorts are in good agreement with the full cohort estimates. Due to control selection with respect to a single event of interest, our adapted method is not only appropriate for future studies but also suitable for the reanalysis and improvement of NCC studies that have already been conducted.

The authors declare that they have no competing interests.

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