Presentation on theme: "Dealing with Uncertainty in Cost-Effectiveness Analyses"— Presentation transcript:
1Dealing with Uncertainty in Cost-Effectiveness Analyses Gerald F. Kominski, Ph.D.Professor, Department of Health Services
2CE Ratios: Decision Rules IVC> 0, E<0 => CE<0never adopt“dominated”C> 0, E>0 => CE>0adopt if CE<CE(max)EIIIIIC< 0, E<0 => CE>0adopt only if savingsis worth health costC<0, E>0 => CE<0always adopt“cost-saving” = dominates
3Consider a New Therapy That Produces a 3-Fold Decrease in Mortality
4ICER Comparing Treatments A and B Treatment Cost Effectiveness A $6, B $10, B-A $4, ICER = $4,200 / = $35,156
5Problems with Point Estimate of ICER Does not express variability in the dataMay lead to the adoption of options that are <WTP threshold, even if the 95% CI exceeds the WTP thresholdSo, what are the options for dealing with the uncertainty of ICERs?
6Dealing With Uncertainty Confidence Intervals95% CIs for ICERs95% confidence ellipsoidNet Monetary BenefitsAcceptability CurvesMonte Carlo Simulations
8How to Calculate 95% CIsFieller’s Theorem and bootstrap methods have been used by various researchersThese methods work when ICERs are confined to one quadrant of the cost-effective plane (as in the previous slide)Bootstrap methods work even when ICERs are found in 2 or 3 of the 4 quadrantsFieller’s Theorem leads to results that are difficult to interpret when the effect difference is insignificant, or when there is no significant difference in costs or effect
9So, Are CIs Worth Calculating? Probably not, unless your data are limited to one quadrant of the cost-effectiveness planeSince the early 2000s, researchers have generally abandoned calculation of 95% CIs in favor of alternative methods:Acceptability Curves and 95% EllipsoidsNet Monetary Benefits (NMB)
10ICER Scatterplot and 95% Ellipsoid for Sample Data
11ICER Scatterplot Data, by Component Comp. Quadrant Incr. Eff. Incr. Cost ICER # Points PercentC1 IV IE>0 IC<0 Superior %C2 I IE>0 IC>0 < %C3 III IE<0 IC<0 > %C4 I IE>0 IC>0 > %C5 III IE<0 IC<0 < %C6 II IE<0 IC>0 Inferior 0 0%
12Interpretation of ICER Scatterplot What does the above table mean?Quadrants begin at "I" in the upper right, and increment counter-clockwise to "IV" in the lower right.To identify cost-effective points, a different component labeling system is used.Cost-effective points for "B" lie below the WTP line, in components 1-3.Component 1 (C1) is where the comparator is dominant ('Superior').Component 2 (C2) is where the comparator is more costly, but lies below the WTP.Component 3 (C3) is where the comparator is less costly, but lies below the WTP.Component 4 (C4) is where the comparator is more costly, and lies above the WTP.Component 5 (C5) is where the comparator is less costly, and lies above the WTP.Component 6 (C6) is where the comparator is dominated ('Inferior').
16How Do We Generate These Measures of Variability? If you have individual level data, you can calculate variability directly from the study dataIf you are conducting a cost-effectiveness analysis using published or aggregate data, you need to either:Have data on the variability for each key variable, orEstimate the variability in each key variable
17Monte Carlo Simulations If you don’t have individual level data, you can simulate the variability in key variables using Monte Carlo techniquesMonte Carlo simulation is a parametric technique, so it requires that you either know or guess the type of distribution each key variable comes fromNonparametric bootstrapping can be used if you have individual level dataTreeAge can produce Monte Carlo simulations, once you specify variable distribution, means, and standard deviationsMore about how to do this in the Lab session