OBJECTIVES : In oncology trials with progression-free survival (PFS) as primary end point, the survival data may be immature at time of health technology appraisal. This might lead to uncertainty around estimation of incremental cost-utility ratio in cost-effectiveness studies. Response has been demonstrated as a predictive variable for overall survival (OS) in multiple myeloma and treatments have been found to impact response distribution, and therefore a response-based partition survival model was considered. This study aims to investigate the applicability of a response-based model in newly diagnosed multiple myeloma (ndMM) patients who are transplant ineligible.

METHODS : Using the patient-level data from ALCYONE and VISTA, we assessed the key possible assumptions in a response-based model: 1) whether PFS/OS differ by response; 2) whether there is treatment effect within a certain response. Four responses were categorized: complete response, very good partial response, partial response and no response. Additionally, we assessed as internal validation whether we could predict PFS/OS based on the fitted curves by response and the response distributions for ALCYONE, VISTA and one trial as external validation where patient-level data were unavailable (i.e. FIRST trial).

RESULTS : Log-rank test and cox regression confirmed that PFS/OS statistically differ by type of response. Within a certain response category, a distinction in PFS/OS was observed between continuous treatments and finite treatments. Based on the fitted curves by response for continuous and finite treatments separately, the PFS/OS of ALCYONE, VISTA and FIRST were well predicted.

CONCLUSIONS : By distinguishing between continuous treatments and finite treatments, the response-based model predicts PFS/OS well. This suggests the plausibility of using response-based model in ndMM transplant ineligible patients. However, as we only have patient-level data from ALCYONE and VISTA, whether using patient-level data from other ndMM trials also results in good response-based survival predictions is uncertain.