OBJECTIVE: McCabe (2020) proposed probabilistic one-way sensitivity analysis (POSA) to replace traditional one-way sensitivity analysis (OWSA). The POSA approach is also considered for the NICE guidance update as it incorporates the whole distribution of uncertain parameters, not only , and better reflects the uncertainty of cost-effectiveness outcomes than deterministic sensitivity analysis. POSA is implemented by fixing each uncertain parameter to selected quantiles one-by-one, and running a traditional probabilistic sensitivity analysis (PSA) on all other parameters. For each quantile, the conditional mean incremental net monetary benefit, E(iNMB), and Prob(iNMB>0) are presented. With p parameters, q quantiles, and N PSA trials for each, this method requires p·q·N deterministic model runs. Although McCabe (2020) claims POSA is a method with modest computational burden, with a typical setup of 10 quantiles, 1000 PSA trials, and 100 parameters included in OWSA, it requires a million deterministic model runs that is often not practically feasible. Our objective is to propose an alternative, simple, computationally more efficient implementation of POSA.

METHODS: Instead of running a PSA for each parameter of interest and each quantile, we propose to run just one PSA with sufficiently large number of trials and analyse its results to obtain the data required by POSA. In such an analysis, for each parameter and each interval defined by the quantiles we select the PSA trials where the parameter falls in the interval and report E(iNMB) and Prob(iNMB>0) for the selected trials. With this method only q·N deterministic runs are required to reach similar precision as POSA with p·q·N runs.

CONCLUSIONS: With our proposed method, in a cost-effectiveness model of typical complexity with 100 uncertain parameters only 1% of the runtime is required to generate the POSA results. Moreover, our method incorporates the distribution within the intervals defined by quantiles, not only their endpoints.