Validating Type 2 Diabetes Diagnoses Across the Madrid Hospital Network: Evidence for Secondary Use of the CMBD in the European Health Data Space

Abstract

Background/Objectives: Hospital discharge data are increasingly reused for research and within health data spaces, but their value depends on accurate diagnostic coding. The validity of type 2 diabetes mellitus (T2DM) coding in the Spanish hospital discharge database (CMBD) had not been assessed across a whole regional network. We aimed to quantify the precision and completeness of E11 coding across the public hospital network of the Region of Madrid. Methods: We conducted a retrospective validation study in the public hospitals of the Region of Madrid (2016–2025). In each of four hospital complexity groups, we selected admissions with and without an ICD-10-ES E11 code, matched by age and sex (787 and 781 analysed). The reference standard was the American Diabetes Association criteria, checked in health records. Predictive values were estimated directly; sensitivity, specificity and prevalence were reconstructed for the whole population, adjusting for sex and age. Results: Of 4,707,924 discharges, 918,797 (19.5%) had a T2DM code. The positive predictive value was 94.2% (95% CI: 92.1–96.0), the negative predictive value 94.0% (91.3–96.3), specificity 98.5% (98.0–99.0) and sensitivity 79.3% (72.4–86.2), a lower bound. The estimated prevalence of T2DM among discharges was 23.2%, against 19.5% coded; about 226,000 admissions with T2DM had no code. Only the positive predictive value differed between hospital groups (p = 0.006). Conclusions: The E11 code is highly precise but incomplete: it is reliable for building cohorts, whereas counts based on it underestimate the number of cases by about one in five.