Abstract
Introduction. Multiple Myeloma (MM) represents a malignant hematological disease characterized by significant inter-patient and intra-patient biological and clinical heterogeneity responsible for extreme variability in outcome measures. Capturing this heterogeneity has been, and remains, one of the major challenges in personalizing therapy. Current staging systems such as ISS, R-ISS and R2-ISS describe the prognosis of groups of patients with similar characteristics but fail to predict the prognosis of an individual patient, especially those in intermediate stages of the scores. This occurs mainly because, within these systems, variables are used in discrete categories rather than along a continuum. It is, in fact, counterintuitive to consider that patients with high-end values of a given variable might share the same prognosis as those with low-end values within the same prognostic group.
Methods. The aim of this study was to identify demographic, clinical, laboratory variables, used in a continuous manner, and cytogenetic features associated with OS and PFS to construct a personalized prognostic score with these variables for newly diagnosed MM patients. Patients came from the following studies: phase III EMN-01 trial (ClinicalTrials.gov: NCT01093196) and phase II IST-CAR506 study (ClinicalTrials.gov: NCT01346787) enrolling patients ineligible for ASCT, phase III EMN02-HO95 trial (ClinicalTrials.gov: NCT01208766) including patients eligible for ASCT. OS and PFS were analysed using Cox proportional hazards models, prioritizing continuous variables to preserve their full prognostic value. Initial models included demographic, clinical, and cytogenetic features, followed by model reduction via likelihood ratio tests. The final model was validated through multiple performance metrics including Harrell's C-index, Brier score, and Net Reclassification Improvement (NRI), and was compared to the R-ISS. Model calibration was assessed through Cox-Snell residuals, after assessed proportional hazards assumptions. In addition, time-dependent ROC analysis was used to determine an optimal cut-off within the R-ISS II subgroup, allowing finer stratification into intermediate-high and intermediate-low risk categories.
Conclusions. Through the integration of continuously distributed variables, our prognostic model demonstrates superior predictive power for OS and PFS compared to the R-ISS, particularly within the intermediate-risk cohort. Moreover, the model enables individualized estimation of OS and PFS probabilities at predefined time points and could be implemented through incorporation of additional parameters, including the proportion of circulating plasma cells, presence of extramedullary disease, and additional cytogenetic aberrations, quantified either as binary features (presence/absence) or as the percentage of plasma cells harboring specific abnormalities. Validation in larger patient cohorts, especially those receiving contemporary therapeutic regimens, is warranted to confirm its clinical utility and generalizability.
