Derivation and validation of the SEER-Medicare MDS risk score (SMMRS).

Abstract

7128 Background: The Surveillance Epidemiology and End Results (SEER) program has been collecting data on myelodysplastic syndromes (MDS) since 2001. These data, when linked to Medicare claims (SEER-Medicare), are an outstanding resource for MDS-related outcomes research. Unfortunately, in terms of prognostic factors at diagnosis, only morphologic subtype is available, and in about half of the registry cases, this is unclassifiable (MDS-U). We aimed to devise a prognostic risk score for use in the SEER-Medicare dataset, and compare it to the standard International Prognostic Scoring System (IPSS) using a granular MDS patient database at our institution (the DFCI/MDS CRIS). Methods: Borrowing from several validated MDS risk scoring systems, a set of candidate predictors that could be determined from claims was pre-specified. Cox proportional hazards models were then built for overall survival, using the 2001-2007 SEER-Medicare MDS dataset (n=9820). Different categorizations of each predictor were tested so that the final model would achieve the best predictive performance. The model was then validated independently using DFCI/MDS CRIS patient data (n=328). C-statistics were calculated to compare the performance of the new scoring system (SMMRS) with the IPSS. Results: The final model included cytopenias, MDS morphologic category, age at diagnosis, hospitalization at diagnosis, red cell or platelet transfusion dependence, and Charlson comorbidity score. The C-statistic for the final model in SEER-Medicare was 0.688. The table shows how the SMMRS performed in the DF/MDS CRIS database compared to, and in combination with, the IPSS. When the analyses were restricted to DF/MDS CRIS patients 65 or older (n=164), similar results were observed. Conclusions: Although missing some important clinical characteristics (eg, cytogenetics), the SMMRS can risk-stratify SEER-Medicare MDS patients with a precision similar to that of the IPSS. The SMMRS thus promises to make the large SEER-Medicare dataset much more useful for MDS clinical research. [Table: see text]