Efficiency Frontier

The boundary formed by non-dominated interventions plotted by cost and effect, showing the best attainable outcome at each level of spending.

Source: Black 1990

Options are plotted with effect on one axis and cost on the other. Any option costing more and producing less than another is removed as dominated. Any option lying above the line joining two others is removed as extendedly dominated, since a combination of those two would deliver more health for the same outlay. The remaining options are joined in order of increasing effect, and the resulting boundary shows the best attainable outcome at each level of spending. Options lying on the frontier are described as efficient in the narrow sense that no other available option or combination delivers more health for the same outlay, which says nothing about whether any of them is worth funding.

Source: Black 1990

Each segment's slope is the incremental cost-effectiveness ratio between the two options it connects, so the frontier displays the whole sequence of ratios geometrically. Because dominated and extendedly dominated options have been removed, the slope rises at every step, meaning each further increment of health costs more than the last. Where the slope exceeds the threshold, moving further along the frontier is no longer worthwhile, which identifies the option to choose. This is why the frontier is read from the bottom upward, moving to the next option only while the additional cost per unit of health remains acceptable.

Source: Drummond et al. 2015

Because the construction assumes only one will be chosen, so that selecting a point means forgoing the others. Where options are independent and could all be funded, dominance in this sense does not apply, and the correct procedure ranks them by their individual ratios and funds downward until the budget is exhausted. Applying frontier logic to independent options discards alternatives that should have been assessed on their own merits. Independent options are also assessed differently in another respect, since adopting several of them together is possible and the frontier construction has no way to represent that.

Source: Drummond et al. 2015

It identifies which options remain candidates and what each successive improvement costs, which turns a set of pairwise comparisons into a single ordered picture. Some appraisal frameworks use it to establish the price at which a new treatment would sit on the frontier for its indication, working backwards from the ratios of existing options rather than applying an external threshold. That approach makes the comparison internal to the therapeutic area rather than across the whole system. Using the frontier this way makes the standard internal to the therapeutic area, which avoids relying on a threshold and imports whatever inefficiency the existing options embody.

Source: Black 1990

It is drawn from point estimates and shows nothing of the uncertainty around them, so an option removed as dominated may not be dominated once distributions are considered. Extended dominance assumes options can be mixed in any proportion, which real treatments rarely permit. And it collapses all health into a single measure, so options differing in the kind of benefit they produce appear directly comparable when they may not be. Presenting the scatter from probabilistic analysis alongside the frontier is the usual response, since it shows how firmly each option's position is established.

Source: Briggs, Claxton & Sculpher 2006