1. Evaluation of Medicine Two types: –Societal level Economic evaluation –Individual level medical decision making

2. Economic Evaluation Comparison of costs and benefits Central question: How can we express the benefits of health care numerically?

3. Approaches Ignore health benefits - Cost minimization Express benefits as life-years gained Contingent valuation - CBA Express benefits as utilities - CUA

4. Medical Decision Making Optimal treatment selection Common approach: perform decision analysis expressing benefits in utility terms

5. QALYs Question: which utility model? Most common model: –Quality-Adjusted Life-Years (QALYs)

6. QALYs Additive model: Let (q 1,..., q T ) be health profile QALY model: U (q 1,..., q T ) =  V(q i )

7. Example Living for 10 years with asthma Suppose V(asthma) = 0.5 QALYs (10 years asthma) = 5

8. QALYs Advantages –intuitively appealing –easy to use in practice Disadvantages –may be too restrictive

9. Two questions QALY model: U (q 1,..., q T ) =  V(q i ) –How do we determine the utilities V(q i )? –Which are the assumptions underlying the QALY model?

10. Basic Ingredients Set of health states H Set of lotteries P over H Preference relation R over P Representing function V:P  IR such that V(P)  V(Q) iff x R y

11. Assumptions Health states are chronic: H = T  Q Health states are positive, i.e., preferred to death Expected utility holds

12. Expected Utility V ( (Q 1,T 1 ), p, (Q 2,T 2 ) ) = pU(Q 1,T 1 ) + (1  p)U(Q 2,T 2 ) U = a + bU, a real, b > 0

13. Chronic Health States Two attribute utility function U(Q,T) = V(Q)*T Q set of positive health states, T set of durations

14. First characterization Pliskin, Shepard & Weinstein (1980) 3 conditions –(mutual) utility independence –constant proportional trade-off –risk neutrality wrt life-years

15. Utility independence Quality of life is utility independent of duration if preferences over lotteries on quality of life holding duration fixed do not depend on the level at which duration is held fixed Duration is utility independent of quality of life if preferences over lotteries on duration holding quality of life fixed do not depend on the level at which quality of life is held fixed

16. Formally If ( p 1,(Q 1,T);….;p m,(Q m,T) ) R ( r 1,(Q 1,T);….;r m,(Q m,T) ) then ( p 1,(Q 1,T);….;p m,(Q m,T) ) R ( r 1,(Q 1,T);….;r m,(Q m,T) )

17. And If ( p 1,(Q,T 1 );….;p m,(Q,T m ) ) R ( r 1,(Q,T 1 );….;r m,(Q,T m ) ) then ( p 1,(Q,T 1 );….;p m,(Q,T m ) ) R ( r 1,(Q,T 1 );….;r m,(Q,T m ) )

18. Standard gamble (20y, Asthma)  ((20y, FH), 2/3, (20y, Death)) Then also (40y, Asthma)  ((40y, FH), 2/3, (40y, Death))

19. Intermediate result The following statements are equivalent: –utility independence holds –U is either additive, U(Q,T) = V(Q) + W(T), or multiplicative U(Q,T) = V(Q)*W(T)

20. Hence To arrive at the QALY model must (i) exclude the additive model and (ii) ensure linearity of W(T).

21. Constant proportional tradeoffs The preference relation satisfies constant proportional tradeoffs if (Q 1,T 1 )  (Q 2,T 2 ) iff (Q 1,  T 1 )  (Q 2,  T 2 ) for all Q 1, Q 2 in Q, T 1, T 2,  T 1,  T 2 in T and  nonnegative

22. Time Trade-off If (Q 1,T 1 )  (Q 2,T 2 ) then U(Q 1 ) = T 2 /T 1

23. Example (10 years asthma) ~ (8 years FH) U(asthma) = 0.8 (20 years asthma) ~ (16 years FH)

24. Exercise Show that CPT excludes the additive model Hence –U(Q,T) = V(Q)*W(T)

25. Risk neutrality Risk neutrality wrt life-years Risk neutrality for duration holds if for a fixed health status level all treatments with equal expected life duration are equivalent. (20y., FH) ~ ( (40y., FH), 0.5; (0y, FH) )

26. Implications W(T) linear Hence, have derived the QALY model

28. Theorem Under EU the following two statements are equivalent –The QALY model represents preferences for health –The preference relation satisfies utility independence, constant proportional tradeoffs and risk neutrality wrt life-years

29. Less restrictive result Take the opposite route Start with risk neutrality wrt life-years U(Q,T) is linear in life-years

30. Hence U(Q,T) = A(Q) + V(Q)*T Note: negative health states are allowed

31. Hence Have to get rid of term A(Q) to obtain QALY model. Assume zero condition: for duration zero all health states are equivalent

32. Exercise Show that the zero condition implies that A(Q) = 0 for all Q.

34. Theorem Under EU the following two statements are equivalent –The QALY model is representing –The preference relation satisfies risk neutrality wrt life-years and the zero condition

35. Hence In PSW representation can drop utility independence and can weaken CPT to zero condition

36. Empirical evidence Zero condition unobjectionable People are risk averse wrt life-years Hence, QALY model not descriptively valid Normative status risk neutrality?

37. More general model U(Q,T) = V(Q)*W(T) Miyamoto, Wakker, Bleichrodt & Peters (1998)

38. Standard Gamble Invariance For Q and Q unequal to death: (Q,T)  ( (Q,Y), p, (Q,Z) ) iff (Q,T)  ( (Q,Y), p, (Q,Z) )

39. Then U(Q,T) = V(Q)*W(T) + A(Q) Zero condition: A(Q) = 0

40. Theorem Under EU the following two statements are equivalent –The nonlinear QALY model is representing –The preference relation satisfies standard gamble invariance and the zero condition

41. One more result Under EU the following two statements are equivalent –U(Q,T) = V(Q)*T  –The preference relation satisfies standard gamble invariance and constant proportional trade-offs

42. Empirical evidence There is support for utility independence Miyamoto&Eraker Bleichrodt&Johannesson Guerrero Bleichrodt&Pinto and constant proportional tradeoffs Bleichrodt&Johannesson

43. However Maximal endurable time –violates utility independence Lexicographic preferences for low durations –violates constant proportional tradeoffs –more in line with increasing proportional tradeoffs

46. Alternative measures Mehrez & Gafni (1989): Healthy-years equivalent (HYEs) (q 1,..., q T )  (Full health, T´) HYEs (q 1,..., q T ) = T´

47. Claim ´´HYEs impose no assumptions on the utility function and are therefore entirely general´´ Q: Is this claim true?

48. 2-stage measurement procedure First stage: determine p such that (q 1,..., q T )  ((FH,T), p, death) Second stage: determine T´ such that ((FH,T), p, death)  (FH,T´)

49. Argument Two-stage gamble leads to same result as directly determining T´ from (q 1,..., q T )  (Full health, T´) Questions: –Is this argument correct? –If so, is it true that HYEs are exactly as restrictive as QALYs?