Marginal Cost-Effectiveness

The additional cost required to produce one more unit of health effect at the margin of an intervention's current scale, distinct from its average.

Source: Drummond MF, Sculpher MJ, Claxton K, Stoddart GL, Torrance GW. Methods for the Economic Evaluation of Health Care Programmes. 4th ed. Oxford University Press; 2015.

Average cost-effectiveness divides the total cost of a programme by the total health it produces, which describes the programme as a whole and includes the most productive activity within it. Marginal cost-effectiveness describes the cost of the next unit of health at the current scale, once the easiest gains have already been taken. Because programmes typically reach the most responsive patients first, the marginal figure is worse than the average and diverges further as the programme expands, which is why decisions about whether to extend a service cannot be taken on the average. Reporting both figures makes the divergence visible, and the gap between them indicates how far the programme has already moved along its declining return.

Source: Drummond et al. 2015

Because the patients easiest to reach and most likely to benefit are treated first. A screening programme extending from high-risk groups to the general population detects progressively fewer cases per person screened while the cost per person screened stays similar or rises. The same pattern applies to case finding, vaccination and outreach. The consequence is that a programme can be highly cost-effective at its current scale and poor value at the margin of expansion, and both statements are true simultaneously. The same logic applies in reverse to contraction, since withdrawing provision removes the least productive activity first and the health lost per unit of resource released is therefore small at the margin.

Source: Phelps 2017

Any decision about changing the scale of an existing service, including expanding eligibility, extending screening intervals, adding sites or withdrawing provision. The relevant question is what the next increment costs and produces, not what the programme as a whole has achieved. Using the average for these decisions systematically overstates the value of expansion and overstates the loss from contraction, and the error grows with how far the programme has already moved along its declining return. The distinction is also what makes threshold comparison meaningful, since a threshold expresses what the system will pay for the next unit of health rather than for the average unit it already buys.

Source: Drummond et al. 2015

By modelling or observing the cost and effect of a defined increment rather than of the whole programme, which means specifying exactly what the extension would be and to whom. Where a programme is being extended to a new group, the baseline risk in that group determines the absolute benefit and is usually the parameter the result turns on. Where evidence exists from services that have already expanded, observed marginal figures are preferable to modelled ones. Where no such evidence exists, the analysis should present results at several possible scales rather than a single point, since the reader can then see where the marginal figure crosses the threshold.

Source: Gold, Siegel, Russell & Weinstein 1996

That the question is rarely whether to run a programme but how far to run it, and that the answer is the scale at which the marginal figure reaches the threshold. Programmes designed without this in mind tend to be either too narrow, leaving worthwhile gains unrealised, or extended past the point where the additional cost exceeds what the displaced resources would have produced. Reporting the marginal figure at several possible scales is more useful to a decision maker than a single result for one configuration. Presenting a curve of marginal cost-effectiveness against scale is the most informative format, since it identifies the optimum directly rather than requiring the reader to interpolate between scenarios.

Source: Drummond et al. 2015