OBJECTIVES

Analysis of real-world data commonly requires the development of clinical code sets to define particular conditions used in determining patient eligibility, outcomes and covariates of interest. However, little guidance exists on developing code sets and these can show wide discrepancies across studies of the same condition. Using the example of chronic obstructive pulmonary disease (COPD), we evaluated the impact of the choice of code set on study outcomes.

METHODS

Thirteen unique code sets for COPD using the International Classification of Diseases, 10^th Revision (ICD-10) were drawn from a previous literature review. These were mapped to ICD-10 Clinical Modification (ICD-10-CM) and used to select patients based on 2016 weighted national estimates from the US National Inpatient Sample (NIS). Demographics, costs and resource use were compared between the populations identified by each code set, both in full and after excluding codes common to all sets (J44*, "Other COPD").

RESULTS

Although code sets comprised 3–56 codes (identifying n=626,019–918,314 patients), the three J44* codes accounted for ≥98% of principal diagnoses in 10/13 analyses; therefore, age, sex, length-of-stay (LOS) and costs were similar across all code sets. Compared to patients with a principal diagnosis of J44*, patients with other principal diagnoses were typically younger (lowest total proportion of patients ≥65 years was 17.9% vs 59.8% for J44*), had higher costs (highest mean $29,593 vs $8,445 for J44*) and longer LOS (highest mean 14.1 vs 4.2 days for J44*).

CONCLUSIONS

The dominance of J44* as a principal diagnosis led to large overlaps between the 13 cohorts, so overall results were broadly similar among code sets. However, the additional patients identified by broader code sets had systematically different characteristics to the "core" cohort. These results highlight the importance of validating code sets, including careful consideration of the context in which specific codes are used in clinical practice and claims billing.