Health care decision making Dr. Giampiero Favato presented at the University Program in Health Economics Ragusa, 26-28 June 2008
Health care decision making Introduction to cost-effectiveness analysis Combining costs and effects Incremental ratios and decision rules Beyond the ICER Information for decision making Trials vs. models Introduction to decision analysis Incorporating uncertainty
Forms of economic evaluation Difference
Structure of economic evaluation Standard treatment New intervention Resource use Health outcomes Health outcomes Resource use Physical quantities, QALYs, Monetary value Total cost = resource use * unit cost Physical quantities, QALYs, Monetary value Total cost = resource use * unit cost Benefit with standard treatment Cost associated with standard treatment Patient-specific benefit with new intervention Patient-specific cost under new intervention Cost-effectiveness analysis
Cost-effectiveness analysis Mutually exclusive programmes Incremental cost-effectiveness ratios = ΔC = Cost new treatment – cost current treatment ΔE Effect new treatment – effect current treatment Decision rules Independent programmes
(Strong) Dominance Programme Costs Effects Management of angina A B C 20 30 50 60 110 8 4 19 23 Dominated: A has lower effects and higher cost than A Management of angina
Average vs. incremental cost-effectiveness ratios Programme Costs Effects A B C D E Breast screening 110 120 150 190 240 20 29 50 60 70 C/E ΔC/ΔE 5.50 4.14 3.00 3.17 3.42 - 1.11 1.43 4.00 5.00 Average ratios have no role in decision making
Incremental cost-effectiveness plane New treatment less effective New treatment more effective New treatment more costly New treatment less costly New treatment dominates Old treatment dominates New treatment more costly and more effective New treatment less costly and less effective
Maximum acceptable ratio New treatment less effective New treatment more effective New treatment more costly New treatment less costly Maximum ICER
Cost analysis decision rule Choose new technology (n) if: ICER = Δ Costs < l Δ Effects
Cost-effectiveness frontier – management of HIV Difference in effects Difference in costs A B D E
The cost-effectiveness plane
Maximum acceptable ratio New treatment less effective New treatment more effective New treatment more costly New treatment less costly Maximum ICER
Maximum acceptable ratio When intervention more/less costly and more/less effective than comparator, cannot determine whether cost-effective unless use data from outside study maximum acceptable ratio Set by budget constraint Set by maximum willingness to pay per unit of effect Administrative ‘rule of thumb’ Empirically based
Cost effectiveness league tables In recent years it has become fashionable to compare health care interventions in terms of their relative cost-effectiveness (incremental cost per life-year or cost per quality-adjusted life-year gained). There are two, quite distinct, motivations behind the league table approach: 1. Analysts undertaking an evaluation of a particular health treatment or programme often seek, quite appropriately, to place their findings in a broader context. 2. Some analysts seek to inform decisions about the allocation of health care resources between alternative programmes. Most of the criticisms of league tables are directed at the second of these two potential motivations.
League table: an example
Grades of recommendation for adoption of new technologies A: Compelling evidence for adoption New technology is as effective, or more effective, and less costly B: Strong evidence for adoption New technology more effective, ICER ≤ $20,000/QALY C: Moderate evidence for adoption New technology more effective, ICER ≤ $100,000/QALY D: Weak evidence for adoption New technology more effective, ICER > $100,000/QALY E: Compelling evidence for rejection New technology is less effective, or as effective, and more costly
Grades of recommendation for adoption of new technologies II New treatment less effective New treatment more effective New treatment more costly New treatment less costly A B C D E
Trials and economic evaluation Internal validity External validity Relevance Inappropriate comparators Limited follow-up Surrogate/intermediate endpoints Information synthesis Uncertainty
Contrasting paradigms Measurement Testing hypotheses about individual parameters Relatively few parameters of interest Primary role for trials and systematic review Focus on parameter uncertainty Decision making What do we do now based on all sources of knowledge? Decisions cannot be avoided A decision is always taken under conditions of uncertainty Decision making involves synthesis Can be based on implicit or explicit analysis
What is a decision model? Mathematical prediction of health-related events Usually comparison of mutually exclusive interventions for a specific patient group Events are linked to costs and health outcomes Synthesise data from various sources Uncertainty in data inputs Focus on appropriate decision Clinical versus economic
Key elements of models Models are simplified versions of reality As simple/complex as required without losing credibility Allow Comparison of all feasible alternative interventions/strategies Exploration of the full range of clinical policies For range of patient sub groups Systematic combination of evidence from variety sources
Data sources for modelling Baseline event rates Relative treatment effects Long-term prognosis Resource use Quality of life weights (utilities) Observational studies/trials Trials Longitudinal observational studies Cross sectional surveys/trials Type of parameter Source
SIMPLE DECISION TREE Chance node Decision node ICER Side effect Use adjuvant No side effect Chance node Side effect ICER Don't use adjuvant No side effect Decision node
SIMPLE DECISION TREE ICER Side effect QALY 1 Cost 1 Use adjuvant No side effect QALY 2 Cost 1 QALYs adjuvant Cost adjuvant Side effect ICER QALY 1 Cost 2 QALYs no adjuvant Cost no adjuvant Don't use adjuvant No side effect QALY 2 Cost 2
Probability Probability: a number between 0 and 1 expressing likelihood of an event over a specific period of time Can reflect observed frequencies Can reflect strength of belief Sum of probabilities of mutually exclusive Events = 1 Joint probability: P(A and B) Conditional probability: P(A/B) P(A and B) = P(A/B) x P(B)
DECISION TREES: PREVENTION OF VERTICAL TRANSMISSION OF HIV COSTS PROBABILITY Acceptance of interventions p=0.07 £800 0.0665 p=0.95 No vertical transmission Policy of intervening C=£800 p=0.93 £800 0.8835 Vertical transmission No acceptance of interventions £0 p=0.26 0.013 p=0.05 No vertical transmission C=£0 £0 0.037 p=0.74 Vertical transmission £0 0.26 p=0.26 Policy of not intervening No vertical transmission £0 0.74 p=0.74 Adapted from Ratcliffe et al. AIDS 1998;12:1381-1388
Uncertainty Population Parameter Structural Sub-group analysis Sensitivity analysis Structural
Sensitivity analysis Deterministic One-way Multi-way Probabilistic
Model validation What are we validating? What do we validate against? inputs outputs structure mechanics/relationships What do we validate against? RCT results Observational studies all models are wrong, but some are useful