5B-06 - Comparing Intention-to-Treat and As-Treated Analyses with Informative Censoring Bias Adjustment in Observational Data

Background: Observational comparative studies can be analyzed using intention-to-treat (ITT) (i.e., initial-treatment) or as-treated (AT) (i.e., per-protocol) approaches to estimate distinct treatment effects. While AT analyses are vulnerable to selection bias introduced by informative censoring, inverse probability of censoring weights (IPCW) address this by creating a pseudo-population where uncensored individuals similar to those who were censored are given more influence. However, little attention is given to the fact that for interval-censored data these weights can either reweight the cohort at the start of the interval (i.e., non-lagged weights) or at the end of the interval (i.e., lagged weights), leading to under- or over-representation of censored individuals.

Objectives: Our objective was to compare ITT and AT analyses (where patients are censored at treatment discontinuation) and the influence of IPCW adjustment timing on AT analyses.

Methods: We compared marginal hazard ratios of all-cause mortality from ITT and AT analyses between new users of selective serotonin reuptake inhibitors (SSRI) and serotonin-norepinephrine reuptake inhibitors (SNRI) from 2019 to 2022 using prescriptions from the Clinical Practice Research Data Link linked to the Office of National Statistics for mortality. We used inverse probability of treatment weighs and created IPCW using A) non-lagged and B) lagged models to adjust for informative censoring in the AT analyses at deciles of patient follow-up. Analyses were replicated changing the exposure to new use of antihypertensives to assess the impact of IPCW in a different context.

Results: We identified 335,469 SSRI initiators and 24,318 SNRI initiators. After inverse probability of treatment weighting, AT estimates (HR: 1.50, 95% CI: 1.30, 1.74) were further from the null than ITT estimates (HR: 1.22, 95% CI: 1.12, 1.32). Applying IPCW attenuated AT estimates with both lagged and non-lagged models (lagged HR: 1.24, 95% CI: 1.08, 1.44; non-lagged HR: 1.16, 95% CI: 1.00, 1.33). Censoring was associated with age, as younger patients were more likely to discontinue SSRIs than SNRIs. This resulted in an accruement of older patients on SSRIs over time, making AT estimate more representative of effects in older patients, which IPCW helped address. In the 337,981 antihypertensive initiators, covariate-dependent censoring was minimal and IPCW did not influence AT estimates.

Conclusions: In populations with high rates of covariate-associated treatment discontinuation, IPCW can help address informative censoring that makes groups less comparable over time. When the population is too large to calculate weights at every censoring time, both lagged and non-lagged IPCW applied to intervals of follow-up yielded similar AT estimates.