Predicting Pharmacy and Other Health Care Costs Arlene S. Ash, PhD Boston University School of Medicine & DxCG, Inc. Academy Health Annual Meeting San Diego, CA June 6, 2004 1 Predicting Drug and Other Costs from Administrative Data • Use various “profiles” – Rx – Dx – Both • To predict next year’s costs – Total $ – Non-pharmacy $ – Pharmacy $ 2 Data • 1998-1999 “Commercial Claims and Encounters” Medstat MarketScan • N ~ 1.3 million – Mean age: 33 yrs – Percent female: 51% • Diagnoses: ICD-9-CM codes • Pharmacy: NDC codes • Costs (incl. deductibles, copays, COB) 3 DCG Model Structure • Diagnoses drive prediction (Risk Score, or RS) – – – – ~15000 Diagnoses group 781 Disease Groups 184 Condition Categories (CCs) Hierarchies imposed 184 HCCs • Model – Predicts from age, sex and (hierarchical) “CC profile” – One person can have 0, 1, 2 or many (H)CCs – Risks from HCCs add to create a summary RS 5 Sample DCG/HCC Year-2 Prediction Prediction for Year 2 $805 48 year old male $3,512 HCC16: Diabetes w neurologic or peripheral circulatory manifestation $1,903 HCC20: Type I Diabetes $266 HCC24: Other endocrine/metabolic/nutritional disorders $455 HCC43: Other musculoskeletal & connective tissue disorders _____ $6,941 FINAL PREDICTION (RS) 6 Pharmacy Model Structure • 80,000+ NDC codes 155 RxGroups • Hierarchies imposed – E.g., insulin dominates oral diabetic meds • Relevant coefficients add to create a risk score for each person 7 Rx Classification System NDC codes (n ~ 82,000+) RxGroups (n = 155) Aggregated Rx Categories (ARCs) (n = 17) 8 Sample RxGroup Year-2 Prediction $3,352 79-year old male $1,332 RxGroup 23: Anticoagulants (warfarin ) $1,314 RxGroup 42: Antianginal agents $1,538 ______ $7,536 RxGroup 116: Oral diabetic agents FINAL PREDICTION 9 Year-1 Dx and Rx Prevalence • Diagnoses – 74% have at least one valid ICD-9 code – Mean # of HCCs per person: 2.5 • Pharmacy – 66% have at least one prescription – Mean # of RxGroups per person: 2.5 10 Year-2 Costs • Total Cost (incl., inpatient, outpatient and pharmacy) – Mean: $2,053 – CV: 386 • Non-Pharmacy Cost – Mean: $1,601 – CV: 471 • Pharmacy Cost – Mean: $452 – CV: 278 11 Predictive Power of Models (Validated R2) Predictors Total $ Non-Pharm $ Pharmacy $ Rx 11.6% 7.1% 48.2% Dx 14.6% 11.6% 22.5% Rx & Dx 16.8% 12.4% 49.3% 12 Validated Predictive Ratios (E/O) Rx Asthma Dx Dx Rx & Dx (n=38,000) 0.90 0.98 1.00 Asthma/COPD Rx (84,000) 0.95 0.86 0.95 Depression Dx (49,000) 0.85 1.01 1.01 Antidepressant Rx (90,000) 0.98 0.82 0.99 Diabetes Dx (33,000) 0.84 1.02 1.03 Diabetes Rx (23,000) 1.01 0.90 1.03 13 Take Home Lessons • Predicting next year’s cost is easiest for Rx $, hardest for Non-Rx$ • Both kinds of data predict well – Dx predicts other costs better – Rx predicts Rx$ much better than Dx – Both together are extremely accurate • The high predictabiity of Rx$ from Rx data bodes ill for the viability of the new Medicare drug insurance product 14