Key result
Use of time-fixed Cox models that excluded immortal time resulted in a hazard ratio of 0.89 for cardiovascular disease associated with RAS inhibitors, avoiding the substantial bias introduced by time-dependent models.
Why the study?
Which statistical method best controls for immortal time bias when evaluating the effect of RAS inhibitors on cardiovascular disease in patients with type 2 diabetes?
Cohort (n=3,928)
Yes
Which statistical method best controls for immortal time bias when evaluating the effect of RAS inhibitors on cardiovascular disease in patients with type 2 diabetes?
Hazard Ratio: 0.89 (95% CI 0.68–1.17)
p-value: p=0.408
In pharmacoepidemiologic studies, time-dependent drug exposure models and models moving immortal time from users to nonusers can introduce substantial bias, whereas time-fixed models excluding immortal time better approximate RCT findings.
Time-fixed models excluding immortal time may yield less biased RAS inhibitor estimates in T2D; leaves open whether findings alter practice pending RCTs.
BACKGROUND: Pharmacoepidemiologic analysis can confirm whether drug efficacy in a randomized controlled trial (RCT) translates to effectiveness in real settings. We examined methods used to control for immortal time bias in an analysis of renin-angiotensin system (RAS) inhibitors as the reference cardioprotective drug. METHODS: We analyzed data from 3928 patients with type 2 diabetes who were recruited into the Hong Kong Diabetes Registry between 1996 and 2005 and followed up to July 30, 2005. Different Cox models were used to obtain hazard ratios (HRs) for cardiovascular disease (CVD) associated with RAS inhibitors. These HRs were then compared to the HR of 0.92 reported in a recent meta-analysis of RCTs. RESULTS: During a median follow-up period of 5.45 years, 7.23% (n = 284) patients developed CVD and 38.7% (n = 1519) were started on RAS inhibitors, with 39.1% of immortal time among the users. In multivariable analysis, time-dependent drug-exposure Cox models and Cox models that moved immortal time from users to nonusers both severely inflated the HR, and time-fixed models that included immortal time deflated the HR. Use of time-fixed Cox models that excluded immortal time resulted in a HR of only 0.89 (95% CI, 0.68-1.17) for CVD associated with RAS inhibitors, which is closer to the values reported in RCTs. CONCLUSIONS: In pharmacoepidemiologic analysis, time-dependent drug exposure models and models that move immortal time from users to nonusers may introduce substantial bias in investigations of the effects of RAS inhibitors on CVD in type 2 diabetes.
No takes yet. Share an insight, caveat, or question.
Yang et al. (2014) conducted a cohort in Type 2 diabetes (n=3,928). Renin-angiotensin system (RAS) inhibitors vs. Nonusers of RAS inhibitors was evaluated on Cardiovascular disease (CVD) (HR 0.89, 95% CI 0.68-1.17, p=0.408). Use of time-fixed Cox models that excluded immortal time resulted in a hazard ratio of 0.89 for cardiovascular disease associated with RAS inhibitors, avoiding the substantial bias introduced by time-dependent models.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: