A mortality model for relapsed or refractory multiple myeloma: emulation of subjective life expectancy criteria from a clinical trial in real-world data

Abstract

Patients with limited life expectancy (LE) are often excluded from oncology trials based on subjective judgment, posing a challenge when contextualizing trial findings with real-world data (RWD). This study evaluated the feasibility of modeling mortality in RWD for relapsed/refractory multiple myeloma (RRMM) to replicate patient exclusion (LE < 6 months) for application in an external control arm study.

A targeted literature review (TLR, January 2016–March 2022) was performed to identify mortality prediction models in RRMM. Because no suitable model was found, a de novo mortality model was developed using the COTA database.

Twenty-one prognostic factors identified via literature review and clinical experts were included in the models. Cox proportional hazards models with bootstrap validation to correct for optimism were used to identify the best-performing model. A threshold of ≥50% predicted mortality probability within 6 months of the start of the line of therapy (LOT) was used for sensitivity and specificity estimates.

Patients (N = 1,112) contributed 2,626 LOTs. A model with the top 20 prognostic factors maximized concordance and accuracy while achieving pre-specified performance thresholds, except for the F1 score. The predicted and observed 6-month mortality probability were both 26%.

Mortality prediction for RRMM remains challenging with RWD. The de novo model exhibited strong discrimination performance but low F1 score, suggesting it may be useful to identify patients at highest risk who may be least likely to meet trial LE criteria.