Data Sources for Measuring Comorbidity

Publisher Website
Google Scholar
Abstract
Identifying appropriate comorbidity data sources is a key consideration in health services and outcomes research. Using cancer patients as an example, we compared comorbid conditions identified: 1) on the discharge facesheet versus full hospital medical record and 2) in the hospital record versus Medicare claims, both precancer diagnosis and associated with a cancer treatment-related index hospitalization. We used data from 1995 Surveillance, Epidemiology and End Results patterns of care studies for 1,382 patients. Comorbid conditions were ascertained from the hospital record associated with the most definitive cancer treatment and Medicare claims. We calculated the prevalence for and assessed concordances among 12 conditions derived from the hospital record facesheet; full hospital record; Medicare claims precancer diagnosis, with and without a rule-out algorithm applied; and Medicare claims associated with an index hospitalization. The proportion of patients with one or more comorbid conditions varied by data source, from 21% for the facesheet to 85% for prediagnosis Medicare claims without the rule-out algorithm. Condition prevalences were substantially lower for the facesheet compared with the full hospital record. For prediagnosis Medicare claims, condition prevalences were more than 1.7 times greater in the absence of an algorithm to screen for rule-out diagnoses. Measures assessing concordance between the full hospital record and prediagnosis Medicare claims (rule-out algorithm applied) showed modest agreement. The hospital record and Medicare claims are complementary data sources for identifying comorbid conditions. Comorbidity is greatly underascertained when derived only from the facesheet of the hospital record. Investigators using Part B Medicare claims to measure comorbidity should remove conditions that are listed for purposes of generating bills but are not true comorbidities.