1. Agenda Medicare Dialysis Model

2. Medicare Established 1965 –President Johnson Who’s covered? –65+ and legal and paid Medicare taxes for +10 years –Social Security disability for +2 years –Social Security disability and ALS –On dialysis or need kidney transplant Part A –Hospital stays +1 night –Skilled nursing facilities (short term) Part B: –Most medical care Part C: Medicare Advantage –Established 1997. Complicated –22% of Medicare population –A+B through private providers Part D: –Established 2003. Complicated –Private plans that cover drugs

3. Medicare Insurance Premium: $96.40/mo. for Part B –Higher for higher incomes Deductibles –$1069 for hospital stays (Part A) –$135 for Part B Co-Pays for Part B –20% for most –0% for lab work Out of pocket expenses can be covered by –Medicaid for poor –Private insurance (Medigap) –Except “donut hole” for drug coverage

5. for Part Afor Parts B & D

8. Medicare Reimbursement Fee for service Sets rates –Lower than private health insurance –Sometimes using Average Sales Price (ASP) –Does not negotiate drug prices for Part D Moving towards “pay for performance” –Paper looks at optimal contract for dialysis

9. Agenda Medicare Dialysis Model

10. Renal = kidney related Produce urine Remove toxins from blood Homeostasis = regulate –Electrolytes (salts) –pH –Produces renin regulating blood pressure –Absorbs glucose and amio acids –Metabolizes vitamin D into calcitrol (calcium balance) –Erythropoietin (EPO) production (hormone for red blood cell production) Kidneys

11. Kidney Function estimated glomerular filtration rate (eGFR) +90% normal +60% hardly noticeable < 60% Chronic kidney disease (CKD) 30-59% anemia + weak bones ≤ 20% causes serious health problems ≤ 10%, 15% End Stage Renal Disease (ESRD) –Need dialysis or transplant (long waitlist)

12. Chronic Kidney Disease (CKD) Chronic = deterioration over time ≠ acute Most diseases attack both kidneys 0.2% prevalence Common causes –Diabetes –High blood pressure Treatment can slow progression 10-20 years until ESRD

13. Dialysis Hemodialysis (hemo = blood) –3x week, 3-4 hr sessions in clinic –Alternatively at home more frequently –Vein in hand/arm –Most common (focus of paper) Peritoneal dialysis –Pump fluid into peritoneal cavity –Exchange through peritoneal membrane –Permanent tube in abdomen –4-5x day, less equipment Also inject drugs

14. What can go wrong? Hospitalized ~ 30% of the year Causes –Heart problems –Fluid build-up –Infection Dosage = Urea Reduction Ratio (URR) –Adequate = +65% Anemia = Hematocrit level (red blood count) –Optimal = 33-36%

15. Drugs billed separately (40% of revenue) Lab work billed separately New rule would bundle them (9/15/2009)

16. Stylized Medicare Payments $130/session When hospitalized –No payment to provider –Costs Medicare $30,600 / year

17. Evidence-Based Incentive Systems for Medicare Dialysis Payments Incentives matter Optimal contract design With data! Dialysis is a good example.

18. Agenda Medicare Dialysis Model

19. Principal Agent Model 2 player game –Principal = Medicare –Agent = Dialysis provider Sequential game 1. Principal announces contract  2. Agent takes hidden action e 3. Outcome o(e) observed Principal receives E[U(o,-  (o))] Agent receives E[u(e,  (o))]

20. Principal Agent Model Agent optimality: e*(  ) in arg max e E[u(e,  (o(e)))] Principal optimality:  * in arg max  E[U(o(e*),-  (o(e*)))] s.t. Agent participation constraint holds U 0 ≤ E[u(e*,  (o(e*)))] 1. Principal announces contract  2. Agent takes hidden action e 3. Outcome o(e) observed Principal receives E[U(o,-  (o))] Agent receives E[u(e,  (o))]

21. Intermediate and Downstream int = Intermediate, ds = downstream (final) Outcome a vector: o = (o int,o ds ) Action a vector: e = (e int,e ds ) o(e) = simple function + correlated noise –o int = e int +  int –o ds = o int +  ´ ds = e int + e ds +  ds –noise mean 0 and  = Cov (  int,  ds ) E[o int ] = e int, E[o ds ] = e int + e ds

22. Simplifications Affine contract:  (o) =  0 +  int o int +  ds o ds Aligning incentives: o int = E[o ds | o int ] Action/effort has cost g(e) = c T e+0.5 e T Q e –Increasing costs to effort Agent has exponential utility –u(x) = -exp (-r x) –Constant absolute risk aversion –u(e,  (o)) = - exp (-r [  (o) - g(e)]) Principal risk neutral –E[U(o,-  (o))]= v o ds -  (o)

23. Dialysis Application Outcomes o = (o int,o ds ) –o ds = fraction of hospital free days in year –o int = f(DOSAGE,ANEMIA) DOSAGE = % of treatments URR ≥ 65% ANEMIA = % of treatments hematocrit in [33%,36%] Current payment scheme:  (o) =  current o ds Reservation utility U 0 set by current payment scheme

24. Risk Adjustment Principal able to observe patient characteristics (part of the noise) –  int   int,i + h int (PAT i ) –  ds   ds,i + h ds (PAT i ) Payment scheme is risk adjusted –  (o) =  0 +  int (o int -h int (PAT i )) +  ds (o ds - h ds (PAT i )) –Similar to adjustment for case-mix in current scheme

25. Parameters r unknown, baseline 2·10 -5 –paying $10 ~ 50-50 chance of winning/losing $1k v = $30,600 / year hospital free g(e), , f(DOSAGE,ANEMIA) fit from data –g(e) adjusted R 2 = 0.034

26. Results Current payment scheme  ds = $27,900/year close to optimal for  int = 0 Optimal scheme:  (o) = $27,700o int + $400 o ds $2,140 increase in Medicare payments to provider +27 hospital free days $123 savings for Medicare Reward (and risk) increased for provider 266k Medicare patients on dialysis +20k hospital-free life years, $32M savings

27. Sensitivity Higher risk aversion leads to small  0 Diminishing returns for increasing v