Premium
Accounting for length of hospital stay in regression models in clinical epidemiology
Author(s) -
Weber Susanne,
Wolkewitz Martin
Publication year - 2020
Publication title -
statistica neerlandica
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.52
H-Index - 39
eISSN - 1467-9574
pISSN - 0039-0402
DOI - 10.1111/stan.12187
Subject(s) - logistic regression , epidemiology , hazard , medicine , proportional hazards model , regression analysis , dependency (uml) , statistics , emergency medicine , intensive care medicine , computer science , mathematics , artificial intelligence , biology , ecology
In hospital epidemiology, logistic regression is a popular model to study risk factors of hospital‐acquired infections. One key issue in this analysis is how to incorporate the time dependency of acquiring an infection during the hospital stay. In the applied literature, researchers often simply adjust for the entire length of hospital stay, which also includes the time after infection. A further issue is that discharge and death are competing events for hospital‐acquired infections. After discussing the limitations of logistic regression adjusted for length of stay in this setting, we compare this approach with appropriate analyses incorporating competing risks and with an illness–death model with hospital‐acquired infection as an intermediate event. The cumulative incidence function, cause‐specific hazard ratios, and subdistribution hazard ratios are considered as reference measures. Real‐life and simulated data are used to demonstrate biases and limitations associated with logistic regression adjusted for length of stay. We conclude that logistic regression adjusted for length of stay should not be used when investigating hospital‐acquired infections and that appropriate methods involving the use of multistate models should be used to capture the time dependency in time‐to‐event settings, especially in the presence of competing events.