Comparing the prognostic value of risk stratifying models for patients with lower-risk myelodysplastic syndromes: Is one model better?

Some patients classified as having lower-risk (LR)-disease by the International Prognostic Scoring System (IPSS) fare more poorly than predicted. We examined the prognostic utility of IPSS, the MD Anderson LR-Prognostic System (LR-PSS), and the revised IPSS (IPSS-R) in a large cohort of patients classified as having IPSS LR-MDS in the MDS Clinical Research Consortium database. Actual overall survival (OS) was assessed in patients with IPSS LR-MDS (i.e. low and intermediate-1) using Kaplan-Meier methods. Harrell's c index (HCI) and Akaike information criteria (AIC) were used to compare the models. Median OS of 1,140 eligible patients was 47 months (95% CI, 44-52). Median follow-up was 62 months. HCI values indicating the discriminatory power of the models (higher is better) were better for LR-PSS (0.74, 95% CI, 0.70-0.78) than IPSS-R (0.64, 95% CI, 0.60-0.67) and IPSS (0.64, 95% CI, 0.60-0.68). Similarly, AIC values indicating the goodness of the fit were better for LR-PSS than IPSS-R and IPSS (8,110, 8,147, and 8,150, respectively, lower is better). LR-PSS assigned 25.1% and 37.4% of patients with IPSS LR-MDS into LR-PSS Category 3 and IPSS-R Categories ≥Intermediate, respectively. Of 291 patients (25.5%) who survived ≤24 months from diagnosis, only 37.1% and 45% were classified as LR-PSS category 3 and IPSS-R categories ≥Intermediate, respectively (P = 0.06). While both LR-PSS and IPSS-R distinguish groups with varied survival outcome among patients with IPSS LR-MDS, both tools fail to identify a significant subset with poor OS. Future studies should assess whether patients identified as at increased risk will benefit from earlier interventions with disease-modifying therapies.