The complexity of chronic disease at later ages: practical implications for prospective payment and data collection.

The diagnosis related groups system developed to pay for Medicare services under prospective payment has certain built-in constraints, owing to limitations in the data available at the time. To identify additional dimensions potentially relevant to case mix, we used a multivariate grouping model-"Grade of Membership"-on discharge records for patients 65 years of age and older hospitalized in Maryland in 1981 with diagnoses of breast cancer, leukemia, or lung cancer. We found that five dimensions are required to accurately type patients: prognosis, disease severity, the interaction effect of multiple illnesses, admission status, and treatment strategy. We also found that at least three possible stages of treatment can be identified and classified. We discuss the implications of our findings for DRG classification, administrative billing records, and the ICD-9 system.