Economic models contain several parameters ordinarily subject to uncertainty. Unlike most other model parameters, costs can constitute numerous distinct components. For example, a surgical intervention can require a variety of disposables and reusable equipment. A micro-costing output may be disaggregated or presented as a total cost. Uncertainty could be applied to individual cost components or to total cost. We aimed to explore how disaggregation of cost data may impact on uncertainty using a case study.

A set of simulations using hypothetical scenarios were developed with uncertainty set at ± 20 percent. The simulations investigated the impact of number of cost components, balance between components, and correlation between them. A cost-utility model from an assessment of robot-assisted radical prostatectomy was analyzed; procedure cost was divided into 32 individual cost components or treated as a total cost.

Based on five equal cost components, uncertainty reduces from ± 20 percent for correlated variables to ± 9 percent for uncorrelated variables. With increasing numbers of uncorrelated cost components, the uncertainty in the total cost decreases markedly. The uncertainty around total robot-assisted surgery procedure equipment costs was ± 19.7 percent when components were correlated and ± 9.4 percent when uncorrelated. The impact on uncertainty in the incremental cost effectiveness ratio (ICER) was negligible but the ranking of parameters in the univariate sensitivity analysis changed.

Analyzing uncertainty by aggregated or disaggregated costs can have implications for presenting uncertainty in costs to decision makers. Applying uncertainty to aggregated costs essentially implies that variation in the cost of individual components is perfectly correlated. By disaggregating cost components they are being treated as uncorrelated, which can substantially reduce uncertainty in the total cost. In this case study we found that although the reduction in uncertainty could be clearly seen in the cost parameter, it had a negligible impact on uncertainty in the ICER.