Validation of the Medicare-Enhanced Laboratory and Demographics (MELD™) Dataset: A Comprehensive Psychometric, Epidemiologic, and Predictive-Utility Assessment of a 60-Million-Patient Real-World Evidence Resource

Background Real-world evidence studies increasingly rely on integrated data resources that combine Medicare fee-for-service claims with electronic medical records (EMRs), laboratory results, and patient-reported outcomes. The Medicare-Enhanced Laboratory and Demographics (MELD™) dataset, developed by Columbia Data Analytics, spans 2020 through 2024 and contains 60 million unique patients, roughly 1 billion visits, and more than 241 000 clinicians. Its data-quality properties must be characterized before regulatory-grade use. Objective To provide a preregistered, multidomain validation of MELD, quantifying internal consistency, concurrent validity against Centers for Medicare & Medicaid Services (CMS) benchmarks, construct validity, record-linkage quality, missing-data mechanisms, temporal stability, and predictive validity across 20 indication cohorts. Methods An analytic cohort of 32 118 604 patients with at least 12 months of continuous Medicare enrollment was derived. Internal consistency used Cronbach’s α and Kuder-Richardson KR-20. Concurrent validity compared MELD with CMS benchmarks using bias, mean absolute percentage error, Lin’s concordance correlation coefficient, and Bland-Altman limits of agreement. Construct validity used exploratory and confirmatory factor analysis. Record linkage used Fellegi-Sunter probabilistic matching. Predictive validity was assessed for 12-month mortality and 30-day readmission using the area under the receiver operating characteristic curve (AUROC), Brier score, and calibration slope/intercept. Results Internal consistency was high (Cronbach’s α, 0.87; KR-20, 0.84). MELD replicated CMS prevalence with mean bias, −1.32 per 1000; Lin’s concordance correlation coefficient, 0.993 (95% confidence interval, 0.986-0.997); and mean absolute percentage error, 3.1%. The 3-factor confirmatory model achieved excellent fit (comparative fit index, 0.962; root mean square error of approximation, 0.041; standardized root mean square residual, 0.036). Fellegi-Sunter linkage produced a 94.7% match rate with false-match probability <0.3%. Weighted κ between International Classification of Diseases, Tenth Revision claims and EMR problem lists ranged 0.66 to 0.91. AUROC was 0.821 for mortality and 0.783 for readmission, with calibration slopes 0.97 and 0.94. Conclusions MELD demonstrated strong psychometric, epidemiologic, and predictive properties, supporting use in health economics and outcomes research for Medicare-aged and adjacent populations. Limitations include residual missingness of race/ethnicity (26%) and EMR fragmentation outside networked provider groups.