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

Area of Science:

• Health Informatics
• Clinical Epidemiology
• Chronic Disease Management

Background:

• Accurate identification of chronic conditions in electronic health records (EHRs) is crucial for research and clinical decisions, especially in older adults with multimorbidity.
• Validation of International Classification of Diseases (ICD)-based case definitions in EHR data is limited.
• The study addresses whether single or multiple ICD codes are sufficient for accurate chronic condition identification.

Purpose of the Study:

• To determine if a single ICD code is sufficient for accurate chronic condition identification in EHR data.
• To assess if using multiple ICD codes improves the validity of chronic condition ascertainment.
• To evaluate the trade-off between case definition stringency and sample size loss.

Main Methods:

• Retrospective chart review of a large, population-based sample of adults (≥18 years) with at least two encounters within two years.
• Validation of ≥1 vs. ≥2 ICD codes for 23 chronic conditions lacking robust EHR definitions.
• Calculation of validation statistics including positive predictive value (PPV), negative predictive value (NPV), and Cohen's kappa.

Main Results:

• The study analyzed 780,873 adults, with a mean age of 47.7 years and 56.9% female.
• For 22 out of 23 conditions, using ≥1 ICD code achieved a PPV of ≥0.70.
• Requiring ≥2 ICD codes showed minimal PPV improvement but resulted in substantial sample size loss (16.8%–64.5%).

Conclusions:

• A single ICD code is generally sufficient for accurate chronic condition identification in EHR data over multiple years.
• Condition-specific validation is essential as the performance of ICD codes varies.
• The findings support the efficiency of using single ICD codes for large-scale chronic disease research while acknowledging the need for careful validation.