A prognostic model integrating disease‑ and patient‑oriented factors identifies poor‑outcome patients with low‑risk MDS: A multicenter real‑world cohort study (KOTOSG)

Abstract

Abstract Background Low-risk myelodysplastic syndromes (LR-MDS) are generally associated with a lower risk of progressing to leukemia and tend to have a better outlook. However, about one-third of LR-MDS patients die within three years of diagnosis, revealing significant variability in prognosis. Recent research suggests that systemic inflammation and nutritional status are linked to poorer outcomes in LR-MDS. Therefore, we hypothesized that adding inflammatory and nutritional markers related to the patient to traditional disease variables might improve outcome predictions. To test this, we aimed to develop a comprehensive prognostic score. Methods We retrospectively examined adults newly diagnosed with LR-MDS across nine Kyoto Clinical Hematology Study Group institutions, from January 2012 to December 2023. LR-MDS was categorized as very-low-, low-, or intermediate-risk based on the Revised International Prognostic Scoring System (IPSS-R). Diagnoses and subtypes followed the 2017 WHO classification. Data collected from clinical records included age; sex; inflammatory/nutritional markers such as albumin (Alb), C-reactive protein (CRP), and total cholesterol (T-Cho); history of autoimmune disease; therapy-related status; presence of peripheral-blood blasts; and each IPSS-R component. Continuous variables were categorized a priori using institutional reference ranges or published cut-offs. To create a prognostic model, we used a LASSO-penalized Cox model for overall survival (OS) to identify relevant variables. First, the best penalty parameter (λ_min) was selected through 10-fold cross-validation. The model was then refitted to the full dataset using this λ_min, and variables with non-zero coefficients were tentatively chosen as potential components. This process was repeated in 1,000 bootstrap resamples; variables appearing in at least 80% of these resamples were included in the final model. Each retained factor was assigned one point, and patients received a cumulative risk score based on the total points. For defining risk groups, we tested all partitions from two to seven groups. The final grouping was based on selecting cut-points that maximized the Cox model's likelihood, ensuring each group had at least 10% of the cohort and achieved the highest Harrell's concordance index (C-index). Model performance was evaluated using the C-index and corrected for optimism with 1,000 bootstrap repetitions. Survival differences between groups were analyzed using Kaplan–Meier curves and log-rank tests. Results A total of 238 patients with complete data on all candidate variables were studied, with a median follow-up of 27.4 months. The distribution of IPSS-R risk categories was very-low (13%), low (56%), and intermediate (31%). Using a LASSO-penalized Cox model with 1,000 bootstrap resamples, seven variables that met an ≥80% retention threshold were identified: Alb ≤3.5 g/dL (26% of patients), T-Cho ≤141 mg/dL (35%), CRP ≥0.15 mg/dL (46%), platelets between 50–100×10^9/L (28%), the presence of peripheral blood blasts (7%), therapy-related disease (11%), and male sex (67%). Spearman's rank correlation showed no significant associations among these variables. We chose 3 partitions, which resulted in a Cox likelihood-ratio χ² of 67.7 and a C-index of 0.85. This stratification divided patients into good-risk (0–1 points, n=76), intermediate-risk (2–3 points, n=119), and poor-risk (≥4 points, n=43) groups, called the Kyoto Prognostic Scoring System for LR-MDS (KPSS-L). Median overall survival (OS) was 120.1, 51.2, and 15.7 months, respectively (p<0.001). Patients in the poor-risk group were mainly male, more often therapy-related, and had significantly lower Alb, T-Cho, and hemoglobin levels, along with higher CRP, compared to other groups. Leukemia was the leading cause of death (24%), slightly more common than in the other groups. Bootstrap validation indicated a higher optimism-corrected C-index for KPSS-L than for IPSS-R: 0.85 (95% CI, 0.80–0.91) versus 0.65 (95% CI, 0.56–0.74)). Conclusions We developed KPSS-L, a new prognostic score for LR-MDS that integrates inflammatory and nutritional markers with standard disease features. KPSS-L successfully stratifies patients by prognosis, outperforming IPSS-R. Those identified as high risk by KPSS-L showed significantly worse outcomes in LR-MDS, indicating a need for early treatment intervention.