Untangling uncertainty with common random numbers: a simulation study

Cost-effectiveness analysis microsimulation for low- and middle-income countries can guide decision makers in allocating limited health budgets. But in order to compare alternative options, estimates of cost and effect must account for a high level of uncertainty in input parameters (parameter uncertainty) and also include an appropriate amount of stochastic uncertainty. It can be a challenge to incorporate the stochastic uncertainty appropriately, particularly when there is a lot of parameter uncertainty, due to large variance in the quantities of interest (such as incremental cost and health of an intervention scenario as compared to a baseline scenario). We investigated the utility of the common-random-numbers approach to variance reduction in simulation and found it very useful. We found that without variance reduction our intervention erroneously appeared unacceptable at any threshold, but with variance reduced using common random numbers it was cost effective.