Decision analytic models, sensitivity analysis and value of information in economic evaluations in health care

Economic evaluations of health care technologies are now commonly carried out to
assess the economic value of new pharmaceuticals, medical devices and procedures. The
aim of health economic evaluations is to measure, value and compare the costs and
benefits of different health care interventions. To date, cost-effectiveness analysis (CEA)
and cost-utility analysis (CUA) are the two types of economic evaluations that are
applied in the vast majority of economic evaluation studies. Cost-effectiveness estimates
in CEAs and CUAs can either be derived from data collected alongside a randomized
controlled trial or by means of decision analytic modelling.
In situations where evidence from a trial is (e.g. short time horizon of the trial; small
sample size), decision analytic modelling provides a structure within which evidence
from a range of sources can be directed at a specific decision problem for a defined
population and context. Decision analytic models use mathematical relationships to
define a series of possible consequences. Based on the inputs of the model, the
likelihood of each consequence is expressed in terms of probabilities. It is thus possible
to calculate the expected costs and expected outcome for different interventions
analyzed in the model.
In the first study the cost-effectiveness of extended prophylaxis with fondaparinux of
one month versus one week in patients undergoing hip fracture surgery and total hip
replacement was analysed. The analysis was based on a decision tree, using a time
horizon of 30 days and 5 years. In this CEA the health effect was measured in life-years
gained. Depending on the patient population and the time horizon, the extended
prophylaxis with fondaparinux was found to be cost-effective or cost-saving.
In the second study the cost-effectiveness of risedronate was examined for Swiss
osteoporotic women. In this study we developed a time-dependent Markov model to
examine the cost-effectiveness of risedronate for women who start a 5 year risedronate
therapy between 60 and 90 years of age. This CUA was carried out from a Swiss health
care perspective. For osteoporotic women or women with severe osteoporosis we found
that risedronate treatment is cost-effective. As expected, the cost-effectiveness estimate
is influenced by the patients’ age and disease severity.
Two chapters of this thesis are based on a cost-utility analysis of 2 drug-eluting stents
(the sirolimus- and the paclitaxel-eluting stent; DES) compared to bare metal stents
(BMS). Although DES are now used for several years, concerns remain about their long
term safety. Given the threefold higher acquisition costs, it was unclear whether DES are
cost-effective when compared to BMS. Based on clinical data with 3-year follow-up we
developed a Markov CUA model to shed light on this question. Both DES under analysis
were found to not be cost-effective from a US Medicare payer’s perspective.
In a further study the decision uncertainty was examined in full depth. With expected
value of perfect information (EVPI) analysis total decision uncertainty was assessed. EVPI provides the value a rational decision maker should be willing to spend in order to
acquire perfect information. Through expected value of partial perfect information
analysis the contribution of groups of parameters towards total decision uncertainty was
examined. The uncertainty in the cost-effectiveness estimate is largely driven by the
uncertainty in the clinical model input parameters. More precise clinical parameter
estimates could be derived from a future clinical trial. To assess the value of such a trial,
analysis of expected value of sample information was performed. Although the value of a
future trial would be enormous, we show diminishing marginal returns and a linear
increase in the costs of the future trial per additional patient enrolled into the trial for
sample sizes larger than 2000 patients. The optimal sample size was estimated to be
4700 patients for a 3 year time horizon.
To conclude, decision analytic models have a range of uses and are thus an important
and powerful tool for economic evaluations in health care. Decision analytic models that
incorporate probabilistic sensitivity analysis and closely related expected value of
perfect information analysis are best suited to provide decision makers not only with a
point estimate for the cost-effectiveness estimate but to quantify in addition decision
uncertainty and the value of future research.