Abstract
The International Prognostic Scoring System–Molecular (IPSS-M) remains the benchmark for risk stratification in myelodysplastic neoplasms (MDS), incorporating disease-specific variables such as cytogenetics, mutational burden, bone marrow blasts, and degree of cytopenias. Yet, in clinical practice, most MDS patients are elderly with substantial comorbidities, including cardiovascular disease, diabetes, prior malignancies, chronic renal insufficiency, or pulmonary impairment, that can affect hematologic responses, worsen heart-failure symptoms, and ultimately dictate survival and treatment tolerance. Consequently, relying only on molecular tools for risk prognostication may yield an incomplete risk stratification in MDS. To address this gap, we developed an Enhanced IPSS-M model that integrates the standard IPSS-M framework with established comorbidity metrics, such as the Charlson Comorbidity Index (CCI) and ECOG performance status, to provide more personalized prognostication that reflects the clinical complexity of contemporary MDS patients.
We constructed a training cohort using MDS patients treated at Karmanos Cancer Institute (KCI), each with extensive clinicogenomic profiles, documented comorbidities, and complete survival outcomes. An independent fine-tuning cohort was derived from the MDS IWG dataset available through cBioPortal (Cerami et al., 2012). We integrated validated comorbidity assessments (CCI and ECOG) into the open-source IPSS-M framework and recalibrated the model based on the combined impact of comorbidities and disease-specific factors on overall survival. The primary objective was to determine how patient comorbidities modulate the prognostic significance of established disease-specific factors and to quantify their independent contribution to risk stratification accuracy in the Enhanced IPSS-M framework.
We developed an Enhanced IPSS-M prognostic model using a real-life cohort of 236 MDS patients (KCI) with in-depth clinical, molecular, and comorbidity data as the training dataset, with model benchmarking and performance validated against 3,323 patients from the MDS-IWG cohort. Patients in the training cohort were younger (67 vs 72 yrs; p=<0.01) compared to the MDS-IWG cohort, with predominantly good-risk IPSS-R cytogenetic category (79% vs 67%; p=<0.01), comparable median bone marrow blasts (2% vs 2.5%), and lower incidence of poor-risk mutations including TP53 (2% vs 11.3%; p=<0.01) and ASXL1 (3% vs 15%; p=<0.001). A machine learning algorithm was trained to faithfully recapitulate the open-source IPSS-M model, with concordance index (c-index) performance validated against the MDS-IWG cohort to ensure accurate baseline reproduction. The IPSS-M risk labels were more evenly split in the MDS-IWG cohort; however, the training cohort was enriched with low-risk patients (27% vs 42%; p=<0.01). The most frequent comorbidities were hypertension and diabetes mellitus type II (each ~45%), followed by COPD (24%), and nearly half of patients (46%) demonstrated reduced performance status (ECOG >2). The Enhanced IPSS-M model achieved superior prognostic discrimination with a concordance index of 0.692 compared to baseline IPSS-M performance (c-index of 0.595 for training and 0.598 for validation, respectively), representing a statistically significant and clinically meaningful improvement (p<0.001). Critically, the enhanced model refined risk stratification for the majority of patients: While 48% remained in their original risk category, 51% were reclassified to higher-risk strata, potentially enabling a realistic prognostication by factoring in comorbidities rather than by molecular criteria alone and potentially identifying patients requiring more intensive monitoring or treatment.
The Enhanced IPSS-M model successfully integrates patient comorbidities with established molecular and disease-specific factors, achieving a clinically meaningful improvement in prognostic accuracy. By incorporating real-world comorbidity data (CCI and ECOG performance status), this enhanced framework addresses an unmet need in the current MDS risk stratification, with over half of patients being appropriately reclassified to higher-risk categories. Ongoing work is focused on using more sophisticated AI-driven risk stratification that allows for personalized MDS-prognostication.
