OBJECTIVES: Landmark-response modelling has been suggested as one of several flexible techniques that may capture the distinctive characteristics of immuno-oncology and CAR-T therapies, including delayed onset of treatment effects and the potential for long-term survival, better than standard parametric survival models. Our aim was to investigate the use and acceptance of landmark-response models in technology appraisals (TAs) conducted by NICE.

METHODS: The NICE website was searched for relevant keywords related to landmark-response models using the website’s own and the Google Advanced search engines. Manufacturer submissions, External Assessment Group (EAG) reports, and Committee documents related to retained oncology TAs underwent two rounds of screening.

RESULTS: We identified only two TAs in which manufacturers used a landmark-response approach to model the impact of immunotherapy on survival endpoints directly. In TA530, the EAG and Committee concluded that the need for a response-based approach was inadequately justified and preferred the use of standard parametric distributions. In TA650, the EAG and Committee did not provide comments on the landmark-response model included by the manufacturer as a scenario analysis. Models used in several other TAs showed conceptual similarities to the landmark-response approach. In TA642, TA763 and TA813, models differentiated patient response in relation to a potential stem cell transplant. In TA489 and TA644, non-responder data from single-arm pivotal studies were used as a proxy for patients not receiving active treatment to generate a comparator. In TA421, the EAG’s landmark model applied a common (rather than response-based) survival trend from the landmark point onwards. As a landmark point was not explicitly mentioned in TA644, TA813 and TA939, these models may have used a response-based approach from baseline.

CONCLUSIONS: Despite publication of guidance on flexible models for survival analysis by the NICE Decision Support Unit in 2021, landmark-response models have hardly ever been used in NICE TAs.