Methodologic Approach to Defining Comorbidities in a Cohort of Patients With Cancer: An Example in the Optimal Breast Cancer Chemotherapy Dosing Study.

Abstract

Purpose: We evaluated the definitions of five comorbidities (renal and hepatic impairments, anemia, neutropenia, and thrombocytopenia) in women with breast cancer using data from electronic health records.

Methods: In 11,097 women receiving adjuvant chemotherapy for stage I-IIIA breast cancer at Kaiser Permanente Northern California or Kaiser Permanente Washington, we assessed comorbidity definitions in two commonly used lookback windows (1 year before diagnosis, T1; and extending until chemotherapy initiation, T1-2). Within each, we assessed data availability and agreement between International Classification of Diseases (ICD)-defined and lab-defined comorbidities of varying severity. To assess how different pieces of information may affect providers in making treatment decisions, we used multivariable logistic regression to evaluate four-category (with comorbidity indicated by both ICD and lab, ICD-only, lab-only, or neither) and collapsed binary (comorbidity indicated by either ICD or lab v neither) definitions in relation to first cycle chemotherapy dose reduction (FCDR).

Results: Extending the lookback period to chemotherapy initiation increased laboratory data availability (missingness ≤4.1% in T1-2 v >40% in T1). Assessment of agreement guided selection of laboratory cutpoints. In both time periods, the four-category and binary comorbidity variables were associated with use of FCDR, but binary variables had larger cell sizes and more stability of regression models. Ultimately, the comorbidity variables in T1 were defined by a combination of either ICD/qualifying laboratory values. Results were similar in T1-2, except laboratory data alone outperformed the combination variable for renal and hepatic comorbidity.

Conclusion: Leveraging both ICD and lab data and extending the lookback period to include postdiagnosis but prechemotherapy initiation may provide better capture of comorbidity. Future studies may consider validating our findings with a gold-standard comorbidity status and in other community health care settings.