Xuan, Jingyi; (2025) Improving the use of inverse probability of censoring weighting in clinical trials with intercurrent events. Doctoral thesis (Ph.D), UCL (University College London).

Preview

Text Thesis.pdf - Accepted Version Download (13MB) | Preview

Abstract

Randomised controlled trials (RCTs) are the gold standard for evaluating therapeutic interventions, as randomisation facilitates balanced treatment arms for comparison. Intention-to-treat (ITT) analysis is the standard approach but may not address research questions involving loss to follow-up or intercurrent events (ICEs). Per-protocol (PP) analysis by excluding affected observations often lacks clarity in its targeted research question and is prone to selection bias when ICEs are associated with confounders. Hypothetical estimands, as outlined in the ICH E9 (R1) addendum, articulate what would be observed if certain ICEs had not occurred. To estimate a hypothetical estimand, Inverse probability of censoring weighting (IPCW) extends PP by removing the selection bias through adjustment for relevant confounders. IPCW relies on accurate model specifications and comprehensive data on covariates and ICEs. Using a simulation study based on an HIV trial ODYSSEY, I compared IPCW and PP regarding bias under varying conditions. Further simulations explored unstabilised weights, stabilised weights with time adjustments, and stabilised weights incorporating time and baseline covariates under both correctly specified and mis-specified outcome models. In STAMPEDE2, a complex platform study with sequential randomisation, following specifications of estimands and estimators, IPCW was applied and evaluated in a simulation emulating real-world scenarios. The results demonstrate that IPCW is a reliable alternative to PP analysis for the estimate of a hypothetical estimand. Stabilised weights improve precision but may introduce bias under model misspecification. Simulations of STAMPEDE2 indicate that ITT may perform poorly in some settings, while IPCW maintains validity with relevant covariates well-collected. However, IPCW struggles with omitted confounders or deterministic ICE occurrences, highlighting the need for careful analysis plans. In conclusion, IPCW extends and outperforms PP for estimating hypothetical estimands with comprehensive data on confounders and ICEs and precise model specifications. Improving IPCW requires thoughtful stabilisation choices and well-designed analysis plans that integrate IPCW effectively.

Download activity - last month

Export as JSONCSVXML

Download activity - last 12 months

Export as JSONCSVXML

Downloads by country - last 12 months

Export as XMLCSVJSON

Archive Staff Only

View Item