Sensitivity analyses
To explore parameter uncertainty of the model inputs, a probabilistic
sensitivity analysis was conducted by randomly sampling from each of the
parameter distributions (beta distribution in the case of relative risk,
and utilities, Dirichlet distribution for multinomial data in the case
of transition probabilities, and gamma distribution in the case of
costs). The expected costs and expected QALYs for each treatment
strategy were calculated using that combination of parameter values in
the model. This process was replicated one thousand times (i.e.,
second-order Monte Carlo simulation) for each treatment option resulting
in the expected cost-utility. Decision uncertainty is represented in the
cost-effectiveness acceptability frontiers, which plot the probability
that the treatment strategy with the maximum expected net monetary
benefit is the most cost-effective over a range of willingness-to-pay
threshold values. Net monetary benefit was calculated by multiplying
effect by societal willingness to pay and subtracting cost, with
willingness to pay set at a ratio of US$ 20,000 per QALY. We estimated
the expected value of perfect information (EVPI). The EVPI is the
maximum value that the health care system would be willing to pay for
additional evidence to inform the reimbursement decision in the future.
The population expected value of perfect information (PEVPI) was
calculated to inform the expected cost of uncertainty (expected
opportunity loss surrounding the decision) (17). Microsoft Exel ®was
used in all analyzes.