Validating ICD code case definitions for condition case ascertainment in multimorbidity measurement: a retrospective chart review

BACKGROUND: Accurate case ascertainment of chronic conditions is critical for research, clinical decision-making, and population health management, especially for older adults with high multimorbidity. However, validation of ICD-based case definitions in the electronic health record (EHR) data remains limited. We sought to determine whether a single ICD code is sufficient for accurate identification of chronic conditions in the EHR, or whether multiple codes improve validity. METHODS: Population-based retrospective chart review, 2013-2019, at large academic tertiary and quaternary care health system. Our sample included adults aged ≥18 years with ≥2 encounters within a 2-year period. We conducted gold-standard chart review to determine validity of using ≥1 or ≥ 2 ICD codes to identify 23 chronic conditions included in the validated multimorbidity-weighted index, specifically those lacking robust case definitions in EHR data. Validation statistics included positive predictive value (PPV), negative predictive value (NPV), Cohen's kappa, sensitivity, specificity, and percent sample size loss associated with stricter ≥2 ICD code case definitions. RESULTS: The final analytic sample included 780,873 adults (mean (SD) age 47.7 (18.0) years, 56.9% female). For 22 of 23 conditions, use of ≥ 1 ICD code yielded a PPV ≥0.70. Requiring ≥2 ICD codes yielded minimal improvement in PPV but with substantial sample size loss that ranged from 16.8% to 64.5%. CONCLUSIONS: ≥1 ICD code was sufficient to accurately identify most chronic conditions with high accuracy in the EHR over a multi-year timeframe. However, condition-specific validation remains essential, as performance varies by condition.