OBJECTIVES: As clinical trial design is evolving, HTA methods need to adapt accordingly. Treatment switching (or crossover) occurs when patients in a control group switch to the experimental treatment during follow-up (or vice versa). Adjusting for crossover is crucial for interpreting outcomes which include the periods before and after the point of crossover. This study aimed to identify the statistical adjustment methods used to account for this in Technology Appraisals by the National Institute for Health and Care Excellence (NICE, England) and the associated commentary.

METHODS: NICE oncology appraisals (published between 2001 and May 2024) were reviewed, those using statistical methods to adjust for crossover when evaluating overall survival (OS) in were selected.

RESULTS: In total, 19 relevant appraisals were identified from 351 reviewed. NICE evaluated both unadjusted and adjusted results and welcomed the submission of multiple methods. The adjustment methods used were: the Rank Preserving Structural Failure Time (RPSFT) model, including its modified version; the Inverse Probability of Censoring Weights (IPCW) method; the Iterative Parametric Estimation (IPE) method, and the 2-stage estimation method. The RPSFT model emerged as the method (n=11), IPE was the least discussed. NICE acknowledged that each statistical method has specific assumptions and limitations, and the choice of model depends on various factors (e.g., patient switching rates, common treatment effects, maturity of OS data, covariate data availability). NICE evaluations generally noted that although uncertainty remained post-adjustments, impacting the decision-making process, these adjustments typically provided confidence in the presence of some benefit.

CONCLUSIONS: Selecting an appropriate adjustment method for trials with crossover is complex and requires case-by-case evaluation. Running multiple models and comparing their results helps to justify the chosen method and ensures appropriate assessment of treatment efficacy. Our results complement NICE DSU Technical Support Document 24 which was published while our research was ongoing.