Positive Predictive Values of ICD-10-CM Diagnostic Codes in Identifying Diabetic Macular Edema in Claims Data: A Multi-Institutional Study in Taiwan.

Purpose: To evaluate the accuracy of ICD-10-CM diagnostic codes when identifying diabetic macular edema (DME) in claims data. Methods and Materials: We analyzed electronic medical records data from four hospitals in Taiwan, and included patients who received first-time ICD-10-CM DME diagnosis codes of E08-13.311, E08-13.321, E08-13.331, E08-13.341 or E08-13.351 during outpatient visits between 2017 and 2022. Two ophthalmologists independently reviewed medical records to confirm the DME diagnosis among a 15% simple random sample of patients assigned the ICD-10-CM DME diagnosis codes. The positive predictive value (PPV) was calculated as the proportion of true DME cases among patients identified by ICD-10-CM DME diagnosis codes. Additionally, we determined the PPVs to identify true DME cases when (1) the ICD-10-CM codes for DME were in the primary diagnosis position only, and (2) when ICD-10-CM DME coding was combined with first-line treatment (intravitreal anti-VEGF drugs covered by Taiwan's national health insurance program within 3 months following code assignment). Results: In total, 4,962 patients newly assigned ICD-10-CM codes for DME were identified, of which 745 patients (15%) (mean age: 62.7 years; females: 47.1%) were randomly selected for electronic medical records review. The PPV for the ICD-10-CM codes to identify true DME cases was 26.4% (197/745, 95% CI: 0.23-0.30). Among the 548 false-positive cases identified by ICD-10-CM DME codes, 516 (94.2%) were tentative diagnoses, 12 (2.2%) involved other retinal diseases, and 20 (3.6%) were remote DME. Among the 402 cases where ICD-10-CM DME diagnostic codes were present only in the primary diagnosis position, the PPV of ICD-10-CM codes for true DME cases remained similar (126/402=31.3%, 95% CI: 0.27-0.36). However, when coding was combined with intravitreal anti-VEGF treatment within 3 months following ICD-10-CM code assignment (55 cases), the PPV for true DME cases was high (55/55=100%, 95% CI: 0.94-1.00), and all confirmed cases were center-involved DME. Conclusion: ICD-10-CM codes alone showed suboptimal accuracy for identifying DME in Taiwan. However, when coding was combined with intravitreal anti-VEGF drug use, they reliably identified center-involved DME in routine care in Taiwan. These findings can inform future claims-based research requiring DME case identification and avoid misclassification bias.