1. 1 A Pay-For-Performance Program for Diabetes Care T.T. Lee, S.H. Cheng* Institute of Health Policy and Management, National Taiwan University, Taiwan, ROC Department of Health, Taiwan, ROC ASHE 6/24/2008 A Preliminary Assessment in Taiwan

2. 2 Introduction 1 Pay-for-performance (P4P) is a healthcare management strategy adopted in many countries to link the payment for services to desirable health outcome. (Rosenthal et al 2004, Doran et al 2008) Financial incentives of P4P program Outcome based or process based At clinician level or hospital level (Petersen 2006)

3. 3 Introduction 2 Studies evaluating the effects of P4P programs have been inconclusive. Improved outcome (Christensen et al 2000, Lindenauer et al 2007) Mixed results (Rosenthal et al 2005, Glickman et al 2007) No significant changes (Hillman et al 1998, Hillman et al 1999) Unintended consequences (Shen 2003) The appropriateness of providing financial incentives for high- quality services continues to be debated. (Wodchis et al 2003, Mannion & Davies, 2008)

4. 4 Introduction 3 Taiwan’s National Health Insurance National Health Insurance implemented in 1995. Most of services are reimbursed on a Fee-for-services basis 53 procedures are reimbursed by fixed payment schedules Taiwan’s P4P programs since 2001 Diabetes, tuberculosis, breast cancer, cervical cancer, asthma Encouraging healthcare providers to increases the monitoring and follow-up care for patients. Financial incentives are for process-based health care services (Cheng 2003)

5. 5 Introduction 4 Diabetes mellitus is the forth leading cause of death in recent years in Taiwan (42.46 death / 100,000) (DOH, 2007) Diabetic patients do not receive adequate care, especially insufficient annual check-ups. (BNHI, 2006) Taiwan’s P4P program for diabetes care Providing financial incentives to increase follow-up visits “increased physician fee” and “case management fees” were provided in addition to regular FFS reimbursement

6. 6 Objectives To assess the effects of a P4P program for diabetes care under Taiwan’s National Health Insurance (NHI) system.

7. 7 Methods 1 Data source, subjects and study design Data source NHI claim database in 2005 and 2006 Subjects: diabetic patients was identified by Diabetes-related diagnoses 250 or A181 Prescription drug for diabetes at least for 3 months/ year Study design: natural experimental design Intervention group: patients enrolled in the program in 2006 Control group: patients had never joined the program, randomly selected (with 1:2 in the 2 groups)

8. 8 Methods 2 Healthcare utilization Number of essential 7 exams/tests Laboratory tests for blood glucose, HbA1C, lipid profile, serum creatinine, SGPT/ALT Urinalysis Dilated eyeground examination/ ophthalmic photograph Diabetes-related utilization Number of diabetes-related physician visits Number of diabetes-related hospitalizations

9. 9 Methods 3 Healthcare expenses Diabetes-related expenses Expense for diabetes-related physician visits Expense for diabetes-related inpatient services Expense for all diabetes-related health services Total healthcare expenses Expense for physician visits Expense for inpatient services Expense for all health services

10. 10 Methods 4 Statistical analysis Difference-in-difference regression model (Efron 1993) Means of predicted values for the intervention group and control groups, before and after the program Bootstrapped SEs for the differences and the difference-in- differences Generalized Estimating Equations (GEEs) (Diggle et al 2002) Taking into account the correlation within in the same patient before and after enrolled in the program

11. 11 Results 1 The participants CharacteristicsTotalIntervention groupControl group (n=38,671)(n=12,499)(n=26,172) Sex, n(%) Male18,633(48.2)5,799(46.4)12,834(49.0) Female20,038(51.8)6,700(53.6)13,338(51.0) Age, years, n(%) ≦ 55 10,810(28.0)3,788(30.3)7,022(26.8) 56-7016,698(43.2)5,707(45.7)10,991(42.0) ≧ 71 11,163(28.9)3,004(24.0)8,159(31.2) mean62.861.563.4 Charlson Comorbidity Index, n(%) 01,066(2.8)290(2.3)776(3.0) 118,070(46.7)5,454(43.6)12,616(48.2) ≧ 2 19,535(50.5)6,755(54.0)12,780(48.8)

12. 12 Results 2 The number of exams/tests and utilization Pre-programPost-programDifferenceP-value (2005)(2006)Post-PreSE No. of essential exams/tests a Intervention group3.7966.3772.5810.016<.001 Control group3.4963.6260.1310.011<.001 Difference0.3002.7512.4500.019<.001 No. of diabetes-related physicians visits b Intervention group14.97417.4992.5260.060<.001 Control group14.76815.2840.5150.033<.001 Difference0.2062.2162.0100.069<.001 No. of diabetes-related hospitalizations b Intervention group0.2340.2480.0140.0080.076 Control group0.2640.3050.0410.005<.001 Difference-0.030-0.057-0.0270.0090.003 a Predicted values obtained from GEE models with Poisson distribution b Predicted values obtained from GEE models with negative binominal distribution

13. 13 Results 3 Diabetes-related healthcare expenses Pre-programPost-programDifferenceP-value (2005)(2006)Post-PreSE Expense for diabetes-related physician visits Intervention group8171,0992825<.001 Control group853895424<.001 Difference-362042407<.001 Expense for diabetes-related inpatient services Intervention group368359-8180.603 Control group46258312116<.001 Difference-94-223-12924<.001 Expense for all diabetes-related health services Intervention group1,1851,45827418<.001 Control group1,3151,47816317<.001 Difference-130-2011025<.001 a Predicted values obtained from GEE models with normal distribution

14. 14 Results 4 Total health expenses Pre-programPost-programDifferenceP-value (2005)(2006)Post-PreSE Expense for physician visits Intervention group1,4201,7563369<.001 Control group1,6601,8161569<.001 Difference-241-6018113<.001 Expense for inpatient services Intervention group414399-15210.449 Control group54670716121<.001 Difference-132-308-17629<.001 Expense for all health services Intervention group1,8332,15532222<.001 Control group2,2062,52331723<.001 Difference-373-3685320.842 a Predicted values obtained from GEE models with normal distribution

15. 15 Discussion 1 The number of exams/tests and utilization Patients participating in the P4P program received nearly all essential exams/tests (6.38 out of 7) The program increased essential exams/tests which might have improved the monitoring of patient’s condition The number of diabetes-related physician visits in the P4P program grew by 2 visits Suggesting there was more attentive follow-up care Patients participating in the P4P program were less likely to be hospitalized The lower hospitalization rate might reflect better ambulatory care (Diabetes is a typical ACSC)

16. 16 Discussion 2 Health care expenditures The net increase for diabetes-related physician visits was 240 US dollars between the 2 groups Only a small portion (42 US dollars) was due to the program’s management fees. A large portion was attributed to additional expenses under the fee- for-service payment system. We found possible spill-over effect of the P4P program for diabetes care The increase expenses due to physician visits were almost offset by the decreased expense for inpatient services, the net increase of overall healthcare expenses was minimal. Better diabetes control via enhanced physician visits might have reduced the overall use of hospital services.

17. 17 Conclusions P4P program for diabetes care has increased the number of essential exams/tests and follow-up physician visits. The increased diabetes-related ambulatory care might have reduced the likelihood of hospital admissions and resulted in reduced cost for overall hospitalizations. Further study is needed to evaluate the cost- effectiveness and its long-term effects of P4P programs on quality or expenditure.

18. 18 Thank you very much