Treatment-Specific Prediction Models in Multiple Myeloma: A Critical Review of Current Evidence and Future Directions

Abstract

Background and Objectives: Multiple myeloma (MM) is characterized by substantial clinical heterogeneity, leading to wide variability in treatment response and toxicity. Although numerous prognostic tools exist, relatively few models estimate outcomes conditional on a specific therapeutic regimen. Treatment-specific prediction models are an important step toward individualized therapy selection. This review synthesizes the current landscape of treatment-specific clinical prediction models in MM. Methods: A structured search of PubMed and Embase/Scopus identified multivariable clinical prediction models developed within a static treatment framework, evaluating treatment-specific therapeutic or toxicity-related outcomes in MM. Information was extracted on treatment regimens, predictors, modeling methods, validation strategies, and reporting of clinical utility. Results: Thirteen models were identified, evaluating therapeutic (n = 10) or toxicity-related (n = 3) outcomes across regimens including bortezomib-based induction, daratumumab-containing combinations, ixazomib-based triplets, and CAR-T therapy. Predictors were mainly routine clinical and laboratory variables, with limited integration of cytogenetics or patient-reported outcomes. Most models used traditional regression methods; calibration was inconsistently reported, and external validation was performed in seven studies. Decision curve analysis was included in only two models. Conclusions: Methodological and translational gaps remain, including limited transparency, scarce external validation, and lack of patient-reported or longitudinal predictors. None of the models have been implemented as online calculators or integrated into electronic decision-support systems, limiting real-world uptake. Addressing these gaps is essential for developing clinically meaningful prediction tools to support personalized treatment in MM.