OBJECTIVES: To evaluate a new empirical algorithm for selecting disease-related medical billing codes associated with chronic lymphocytic leukemia (CLL).  METHODS: Patients in the SEER-Medicare database with a CLL diagnosis (2002 to 2010) were age/gender matched to a non-cancer sample.  A proprietary coding algorithm based on code frequency (sensitivity, specificity precision or accuracy) and cost was used to identify procedure (i.e., CPT and HCPCS) and diagnosis (ICD-9-CM) codes that differed between the CLL and non-cancer groups.  Summarized costs for claims with the empirically identified codes were compared to a traditional approach of identifying disease-related claims based on presence of a CLL diagnosis in the first diagnosis field.  The code set was applied to a sample from a prior CLL study conducted with commercial claims to assess generalizability.   RESULTS: The analysis evaluated 10,531 unique billing codes with total costs of >$1 billion (US; CLL 58.3%; non-cancer 41.7%) for 7,050 age and gender matched SEER-Medicare subjects per group.  The empirical algorithm found 333 codes that identified 25.0% of the CLL group costs as cancer-related. The traditional approach used claims that contained 2,001 codes and identified a much lower 14.6%.  Approximately 1% of costs were potentially misidentified in the non-cancer cohort, providing further confirmation of the codes selected by the empirical method.  Qualitative review of codes revealed stronger content validity with the empirical approach compared to the traditional approach.  Application of codes identified in the SEER-Medicare data to the commercial database provided validation by identifying almost twice as many costs as the traditional approach with a 1% error rate.  CONCLUSIONS: The traditional approach underreports costs and captures costs from procedure codes that do not appear to be cancer-related. Use of an empirical approach to identify disease-related diagnosis and procedure codes will increase content validity.