economic evaluation of health programmes
DESCRIPTION
Economic evaluation of health programmes. Department of Epidemiology, Biostatistics and Occupational Health Class no. 22: Applying the net benefit framework, assessing the value of information, reporting economic evaluations Nov 17, 2008. Plan of class. Net benefit framework - PowerPoint PPT PresentationTRANSCRIPT
Economic evaluation of health programmes
Department of Epidemiology, Biostatistics and Occupational Health
Class no. 22: Applying the net benefit framework, assessing the value of information, reporting
economic evaluations
Nov 17, 2008
Plan of class
Net benefit frameworkCost-effectiveness acceptability curves‘Marrying econometrics and cost-
effectiveness analysis’
Why are p-values for the NMB regression coefficient estimates equal to those
obtained with effect as the dependent variable?
When λ goes to infinity in
the variation in Ci becomes less and less important compared to that in Ei, so that the regression:
is mostly explaining variation in Ei: the Ci part represents a smaller and smaller share of the variation in NMBi. Hence, while the coefficients are different (NMBi is becoming larger and larger), the p-values are the same.
Additional points concerning Hoch et al. 06
Nature of outcome (1)
If a fainting episode (syncope) occurs, and the device that measures heart rhythm is activated on time, it is possible to see whether the syncope is or is not associated with heart arrhythmia.
Success occurs if : a) a syncope occurs b) the device is activated.
Nature of outcome (2)
Loop recorder: 63.27% of patients have symptom reccurrence and successful activation, vs. 23.53% of Holter group
The loop recorder is more expensive but has a higher chance of detecting whether arrhythmia is the cause of syncope
How the dependent variable (net benefit) is constructed
Review: why the p-value on the intervention dummy from the net benefit regression can be
used to construct the CEAC (1)
The CEAC shows, for a given value of λ, the probability that the intervention is cost-effective.
In the regression:
statistical software computes a t-statistic for the estimate of δ to test the null hypothesis that δ=0; if the p-value is small enough (usually less than 0.05), we reject the null.
(Under what condition does this become the probability density function of a t-distribution?)
Review: why the p-value on the intervention dummy from the net benefit regresion can be
used to construct the CEAC (2)
But we want to test the null hypothesis that δ <= (negative or equal to) 0.
If the estimate of δ is positive, then the associated t-statistic is also positive – so we locate it on the right-hand side of the t-distribution. Half of the corresponding p-value gives us the area to the right of that value. This is the significance level at which we can reject the null that δ <= (negative or equal to) 0. In other words, the probability that δ>0 is 1- half of the p-value.
Review: why the p-value on the intervention dummy from the net benefit regresion can be
used to construct the CEAC (3)
If the estimate of δ is negative, then the associated t-statistic is also negative – so we locate it on the left-hand side of the distribution. Half of the corresponding p-value gives us the area to the left of that value. This is the significance level at which we can reject the null that δ >= (positive or equal to) 0. In other words, the probability that δ>0 is half of the p-value.
Regression estimates for different values of λ
Calculating probabilities of cost-effectiveness
Note how similar probabilities of cost-effectiveness derived from regression and from bootstrapping are
CEAC from probabilities of cost-effectiveness derived from one-sided p-values in previous table
Comments
Here, no variation in cost across individuals with same Tx
When there is variation in cost, skewness or heteroskedasticity may make p-values less valid – then use bootstrapping
NBRF enables estimates of mean net benefit of : Usual care (β0) New treatment (β0+ β1) as well as incremental net
benefit (β1)
Using models to assess value of additional research
Concept of decision uncertainty
Methodological uncertainty Sampling variation/
parameter uncertainty Modelling uncertainty Generalizability
Decision uncertainty
Is additional research necessary?
Expected value of perfect information (EVPI)
Probabilistic sensitivity analysis (PSA) to yield expected costs and effects of alternative options: identify preferred option
Determine probability of making wrong decision = 1 - probability that this is indeed best option (use CEAC)
Use PSA to determine cost of making wrong decision: Foregone health Wasted resources
Calculate expected cost of uncertainty by multiplication (in terms of health and dollars)
Multiply by number of patients to get population EVPI
Example of EVPI analysis
Implications of EVPI analysis
Additional research must cost less than EVPI
Can also use EVPI analysis to assess value of information to be yielded by alternative research designs
Frontier area of investigation EVPI absolutely not taken into account by
CIHR
Presentation and use of economic evaluation results
Economic evaluation is widely used…
Oregon Medicaid experiment Combined with public deliberative process
Requirement for formal economic evaluation for drugs to be approved for reimbursement Australia Several Canadian provinces U.K.
Wider use in England (NICE)
…but has many limitations
Validity can be hard to assess => Standardize reporting
Generalizability may be issueVery method of funding interventions
meeting $/QALY threshold has been criticized
Common reporting format Alleged benefits:
Transparency Comparability across studies
• Limits of league tables
Improve quality of evaluations?• Stifle innovation?
US Public Health Services Panel on Cost-effectiveness in Health and Medicine Include reference case in report
British Medical Journal Working Party on Economic Evaluation (1996) Too many methodological controversies, so just focus on common
reporting standards
Some common recommendations
To include in reporting of economic evaluation
Background: Question(s) to be addressed and its(their) importance
Viewpoint/perspective(s) of analysis
Justification for type of analysis
For whom is this relevant?
Comparators being assessed
Source and quality of medical evidence
Range of costs and how measured
Measure of benefits
Methods for dealing with uncertainty
Incremental analysis of costs and benefits
Overall study results and limitations
Concept of league table
Rank interventions in order of $/QALYWhy?
Place findings in broader context Inform decisions about which interventions to
fund• Use of league tables for this purpose has been
criticized
Examples from published studies (1998 US$)
Source: http://www.hsph.harvard.edu/cearegistry/comprehensive-revised.pdf