Presentation (CTI)

OBJECTIVES: Health state utility values (HSUV) are key parameters of cost-effectiveness models in NICE technology appraisals (TAs) for oncology, where missing utility data in trials can compromise model validity. This study reviewed recent NICE oncology TAs to understand the reporting and handling of missing utility data in RCTs, as well as External Assessment Group (EAG) critique.
METHODS: A systematic review was conducted of oncology NICE TAs published between 1 June 2023 and 31 May 2025. Committee papers were screened for whether trial-based HSUVs were reported. Double data extraction was performed on reporting of HSUV missing data statistics, statistical methods applied to address missingness, and EAG commentary.
RESULTS: Of the 68 oncology TAs identified, 62 reported trial-derived HSUVs. Most appraisals reported the proportion of complete utility data, though many did not explicitly assess whether missingness was at random. Whilst mixed-effects modelling was the predominant statistical approach for handling missing utilities across TAs, it hinges on a missing-at-random assumption. Alternative trial analysis methods were rarely explored when that assumption appeared invalid. Time-to-death regression featured in some submissions but was often criticised for misalignment with the model’s health states. In more recent TAs imputation methods were adopted, or requested, including multiple imputation (e.g., TA1060, TA1063) and pattern mixture modelling (e.g., TA1059) to handle missingness, suggesting expectations for greater methodological rigour from EAG.
CONCLUSIONS: Missing utility data, which often stem from lack of post-progression utility measurements and low patient compliance, remain a common challenge in NICE oncology appraisals. Despite available statistical approaches when data are missing not at random, these are infrequently applied. EAGs play a crucial role in scrutinising assumptions behind missing data handling, where incorrectly specified missingness mechanism introduces bias. Enhanced guidance and routine reporting to validate statistical model assumptions, and application of techniques such as pattern mixture modelling may improve robustness of trial-based utility analysis.