By appropriate multiplication of the probability differences, the uptake percentage and the parameters in the consequences table, it will be possible to obtain an estimate of the impact of the HMR in each of the terms described against each of the consequences. As multiple assessors will be involved, multiple estimates will be made, providing an upper and lower range for the estimate. Extrapolation to the remainder of the sample will be based on the types and categories assigned in step 2.

A cost-consequence analysis will then be undertaken to calculate the range of potential economic benefits (or cost) of the intervention compared to the counterfactual (or no intervention). Where data permits, a cost-effectiveness analysis will also be undertaken using intermediate outcomes of change in adverse events or change in hospitalisations as a consequence of the intervention. The economic model will include extensive sensitivity and uncertainty analyses. Sensitivity analysis will be used to test the results to variations model parameters including uptake and attrition rate. Uncertainty analysis will be conducted around the epidemiological and costing estimates on the final results. In the primary analyses, point estimates are calculated to measure benefits and costs.

The commercial package @RISK will be used to allow estimates and assumptions to be entered as probability distributions in a spreadsheet. The program then recalculates the spreadsheet many times over—each time picking a value out of all defined probability distributions—and provides summary statistics across all iterations (usually 2000 or more) for selected outcome variables. From the values generated by the iterations of the simulation, a 95 per cent uncertainty interval can be calculated by taking the 2.5 and 97.5 percentiles to mark the lower and upper bounds. This uncertainty interval can be interpreted as the range within which the true result lies with 95 per cent certainty. An uncertainty interval differs from a confidence interval in that it includes both type I and type II errors.