A Refined Comorbidity Measurement Algorithm for Claims-Based Studies of Breast, Prostate, Colorectal, and Lung Cancer Patients

Purpose We evaluated (i) how combining comorbid conditions identified from Medicare inpatient or physician claims into a single comorbidity index compared with three other comorbidity indices and (ii) the need for comorbid condition weights that are specific to different cancer sites. Methods This observational study used the SEER-Medicare linked database, from which four cohorts of cancer patients were derived: breast ( n = 26,377), prostate ( n = 53,503), colorectal ( n = 26,460), and lung ( n = 33,975). We calculated two established (Charlson; NCI) and two new (NCI Combined; Uniform Weights) comorbidity indices, and used Cox proportional hazards models to assess their predictive ability. We also used a pooled dataset to examine the inclusion of cancer site-specific condition weights. Results The four comorbidity indices all significantly predicted mortality, but the NCI and new NCI Combined indices showed the greatest contribution to model fit. The new NCI Combined index is simpler to use and statistically more efficient than the NCI index. Modeling further demonstrated the utility of cancer site-specific weights. Conclusions Our results support the need for cancer site-specific comorbidity measures that employ empirically-derived condition weights. The new NCI Combined index is a refined, easier to implement comorbidity measurement algorithm appropriate for investigators using administrative claims databases to study four commonly-occurring cancers.