1. Abra Jeffers VA Palo Alto Dept. of Mgmt Science & Engineering Stanford University Medical Decision Making

2. Decision Analysis Quantitative, systematic Identifies alternatives, outcomes, utility of outcomes Uses models Represent structural relationships Deterministic Probabilistic Compute expected outcomes of each alternative Balance feasibility v. reality / simplicity v. complexity Examine robustness 2

3. Identify Alternatives Mutually exclusive A, B mutually exclusive ⇒ P(A∩B) = 0 One and only one of the events must occur Collectively exhaustive A, B collectively exhaustive ⇒ P(A) + P(B) = 1 Events represent entire outcome space At least one event must occur 3

4. Identifying Outcomes Multi-dimensional Examples: Costs Infections averted Life years (LY) saved Quality adjusted life years (QALY) saved “Effectiveness” Ultimately utils – Howard school Cost-effectiveness Incremental cost to effectiveness ratio (ICER) 4

5. Cost-Effectiveness Plane ΔCΔC ΔEΔE 5 Increased Costs Increased Effectiveness Is ICER < WTP? Increased Costs Decreased Effectiveness Use status quo Decreased Costs Increased Effectiveness Use new strategy Decreased Costs Decreased Effectiveness Probably use status quo

6. Steps Determine problem Determine status quo intervention Determine alternative interventions Analysis Sensitivity Analysis 6

7. Example Hypothetical cohort with prevalent disorder Treatment Reduces probability of serious complications Long term side effects There is an imperfect test for disorder Alternatives: Treat no one Test all and treat positives Treat all 7

8. Decision Diagram Decision node Probabilistic node Terminal node 8

9. Example cont. Outcomes No disorder, no treatment No disorder, treatment Disorder, treatment, complications Disorder, treatment, no complications Disorder, no treatment, complications Disorder, no treatment, no complications For each outcome assume known costs and LE (clinical trial) Intermediate events Tests Side effects 9

10. Create Decision Node 10

11. Create Decision Node 11

12. Create Decision Node 12

13. Create Decision Node 13

14. 14

15. 15

16. 16

17. 17

18. 18

19. 19

20. 20

21. Strategy 1: Treat None 21

22. Strategy 1: Treat None 22

23. Strategy 1: Treat None 23

24. Strategy 1: Treat None 24

25. Strategy 1: Treat None 25

26. Select Subtree 26

27. Collapse Subtree 27

28. 28

29. Strategy 2 - Test 29

30. Strategy 2 - Test 30

31. Strategy 2 - Test 31

32. 32

33. 33

34. 34

35. 35

36. 36

37. 37

38. 38

39. Now What? Now we have our tree structure Lets start filling in probabilities 39

40. 40

41. 41

42. 42

43. # = 1- prev 43

44. 44

45. 45

46. 46

47. 47

48. 48

49. 49

50. 50

51. 51

52. 52

53. 53

54. 54

55. 55

56. 56

57. 57

58. 58

59. 59

60. 60

61. 61

62. Recap Thus far… Created decision tree Used descriptive variable names Used clones to take advantage of parallel structure Tracking variables Now What? Outcomes Assume we know from clinical trial Costs Life years 62

63. 63

64. Effectiveness Outcomes Normal – 15 life expectancy Disorder – 3 year detriment Complications – 10 year detriment Treatment – 0.05 year detriment 64

65. 65

66. 66

67. 67

68. 68

69. Add Complications Tracker 69

70. Add Complications Tracker 70

71. Add Complications Tracker 71

72. Add Disorder Tracker 72

73. 73

74. Payoff 1: LE_normal - if(t_disorder = 1; LEdet_disorder; 0) + - if(t_treatment= 1; LEdet_treatment; 0) - if(t_complications = 1; LEdet_complications; 0) 74

75. 75

76. 76

77. 77

78. Payoff 2: c_normal + if(t_disorder = 1; c_disorder; 0) + if(t_test = 1; c_test; 0) + + if(t_treatment = 1; c_treatment; 0) + if(t_complications = 1; c_complications; 0) 78

79. 79

80. 80

81. 81

82. 82

83. 83

84. 84

85. 85

86. Baseline = Status Quo 86

87. 87

88. 88

89. 89

90. 90

91. 91

92. Sensitivity Analysis 92

93. 93

94. 94

95. 95

96. 96

97. Caveats Assumed full knowledge of Future life expectancy Future lifetime costs Inappropriate if payoffs without full lifetime info Markov model 97

98. 98 Thank you!