Abstract
Background：Despite therapeutic advances, early mortality remains a major determinant of poor outcomes in patients with multiple myeloma (MM) of renal impairment (RI). Currently, no validated prognostic models are specifically available for predicting early mortality risk within this high-risk subgroup.
Aim：This multicenter study was designed to develop and clinically validate a risk prediction model for early death (<12 months), utilizing accessible clinical parameters to identify patients at greatest risk for prompt intervention.
Methods：This multicenter retrospective study analyzed newly diagnosed adult MM patients with RI treated between 2007 and 2020. From an initial pool of 334 previously reported cases, we excluded patients lacking baseline bone marrow plasma cell (BMPC) percentages or follow-up hematologic/renal response data. The final cohort comprised 222 patients from Beijing Chao-Yang Hospital (training cohort) and 63 patients from two independent centers (validation cohort).Candidate predictors were assessed using logistic regression with backward selection. The final model was presented as a nomogram and rigorously validated through Harrell's C-index, time-dependent ROC analysis, and resamples for internal validation and optimism correction.
Results: The cohorts demonstrated balanced demographics, high-risk disease, comparable treatments, and similar hematologic or renal response rates, with renal parameter variations, confirming cohort heterogeneity suitable for prognostic model validation. In the multivariate logistic regression analysis, five independent predictors of early mortality were identified: age (> 55 years; OR 3.48, 95% CI 1.10-11.01), Serum calcium level (≥ 2.5 mmol/L; OR 3.29, 95% CI 1.30-8.29), bone marrow plasma cell（BMPC）percentage (≥40%; OR 2.90, 95% CI 1.24-6.77), renal response (≥PR; OR 0.27, 95% CI 0.11-0.65), and hematologic response (<CR; OR 20.93, 95% CI 4.61-95.10). Based on these predictors, a nomogram for predicting early death risk was developed with points allocated proportionally to each variable's regression coefficient. The model showing excellent discrimination (AUC 0.858, p<0.001) and calibration (χ²=6.32, p=0.612) that persisted in validation (AUC=0.747, p=0.010) with acceptable performance (calibration intercept=-0.07, slope=0.94), confirming both the critical prognostic value of treatment response and the model's clinical utility. Each adverse prognostic factor was assigned 1 point for its absence and 2 points for its presence. A cumulative score was calculated by summing points across all five prognostic factors. The risk score categorized 285 patients into three prognostically distinct groups: Low Risk group (5–6 points, n=61; early mortality 1.6%, estimated median OS 76 months), Mid Risk group (7–8 points, n=158; early mortality 13.9%, estimated median OS 46 months), and High Risk group (9–10 points, n=66; early mortality 43.9%, estimated median OS 14 months). Statistically significant differences were observed in early mortality rates (p<0.001) and survival curves (p<0.001).
Conclusion：This validated prognostic model provides a practical tool for early mortality risk stratification in MM patients with RI. The combination of age, serum calcium level，renal response, hematologic response and BMPC offers clinicians a reliable method to identify patients who may benefit from more intensive monitoring and therapeutic interventions. Given the insufficient cytogenetic data and missing serum free light chain results, further prospective validation in diverse populations is warranted to confirm generalizability.
