Background: The Molecular International Prognostic Scoring System (IPSS-M) has improved the prediction of clinical outcomes for myelodysplastic syndromes (MDS). The Artificial Intelligence Prognostic Scoring System for MDS (AIPSS-MDS), based on classical clinical parameters, has outperformed the IPSS, revised version (IPSS-R). For the first time, we validated the IPSS-M and other molecular prognostic models and compared them with the established IPSS-R and AIPSS-MDS models using data from South American patients.Methods: Molecular and clinical data from 145 patients with MDS and 37 patients with MDS/myeloproliferative neoplasms were retrospectively analyzed.Results: Prognostic power evaluation revealed that the IPSS-M (Harrell's concordance [C]index: 0.75, area under the receiver operating characteristic curve [AUC]: 0.68) predicted overall survival better than the European MDS (EuroMDS; C-index: 0.72, AUC: 0.68) and Munich Leukemia Laboratory (MLL) (C-index: 0.70, AUC: 0.64) models. The IPSS-M prognostic discrimination was similar to that of the AIPSS-MDS model (C-index: 0.74, AUC: 0.66) and outperformed the IPSS-R model (C-index: 0.70, AUC: 0.61). Considering simplified low-and high-risk groups for clinical management, after restratifying from IPSS-R (57% and 32%, respectively, hazard ratio [HR]: 2.8; P = 0.002) to IPSS-M, 12.6% of patients were upstaged, and 5% were downstaged (HR: 2.9; P = 0.001). The AIPSS-MDS recategorized 51% of the low-risk cohort as high-risk, with no patients being downstaged (HR: 5.6; P < 0.001), consistent with most patients requiring disease-modifying therapy.
Conclusions:The IPSS-M and AIPSS-MDS models provide more accurate survival prognoses than the IPSS-R, EuroMDS, and MLL models. The AIPSS-MDS model is a valid option for assessing risks for all patients with MDS, especially in resource-limited centers where molecular testing is not currently a standard clinical practice.