Premium
Estimating the effect of immortal‐time bias in urological research: a case example of testosterone‐replacement therapy
Author(s) -
Wallis Christopher J.D.,
Saskin Refik,
Narod Steven A.,
Law Calvin,
Kulkarni Girish S.,
Seth Arun,
Nam Robert K.
Publication year - 2017
Publication title -
bju international
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.773
H-Index - 148
eISSN - 1464-410X
pISSN - 1464-4096
DOI - 10.1111/bju.13918
Subject(s) - medicine , hazard ratio , confidence interval , observational study , cohort study , population , proportional hazards model , cohort , survival analysis , meta analysis , demography , testosterone (patch) , retrospective cohort study , oncology , environmental health , sociology
Objective To quantify the effect of immortal‐time bias in an observational study examining the effect of cumulative testosterone exposure on mortality. Patients and Methods We used a population‐based matched cohort study of men aged ≥66 years, newly treated with testosterone‐replacement therapy ( TRT ), and matched‐controls from 2007 to 2012 in Ontario, Canada to quantify the effects of immortal‐time bias. We used generalised estimating equations to determine the association between cumulative TRT exposure and mortality. Results produced by models using time‐fixed and time‐varying exposures were compared. Further, we undertook a systematic review of PubMed to identify studies addressing immortal‐time bias or time‐varying exposures in the urological literature and qualitatively summated these. Results Among 10 311 TRT ‐exposed men and 28 029 controls, the use of a time‐varying exposure resulted in the attenuation of treatment effects compared with an analysis that did not account for immortal‐time bias. While both analyses showed a decreased risk of death for patients in the highest tertile of TRT exposure, the effect was overestimated when using a time‐fixed analysis (adjusted hazard ratio [ aHR ] 0.56, 95% confidence interval [ CI ]: 0.52–0.61) when compared to a time‐varying analysis ( aHR 0.67, 95% CI : 0.62–0.73). Of the 1 241 studies employing survival analysis identified in the literature, nine manuscripts met criteria for inclusion. Of these, five used a time‐varying analytical method. Each of these was a large, population‐based retrospective cohort study assessing potential harms of pharmacological agents. Conclusions Where exposures vary over time, a time‐varying exposure is necessary to draw meaningful conclusions. Failure to use a time‐varying analysis will result in overestimation of a beneficial effect. However, time‐varying exposures are uncommonly utilised among manuscripts published in prominent urological journals.