Fully Incremental Analysis

A method comparing more than two mutually exclusive interventions by ranking them by effectiveness and calculating each option's ICER against the next best alternative.

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.

All mutually exclusive options are listed in order of increasing effect with their costs. Options that cost more and achieve less than another are removed, then those beaten by a combination of two others. Incremental ratios are calculated between each surviving option and the next less effective one, producing a sequence of ratios rather than a set of comparisons against a common baseline. The most effective option whose incremental ratio remains below the threshold is selected. Presenting the full table, including the options removed and the reason for each, is what allows a reader to verify the sequence rather than accepting the final ratio on trust.

Source: Drummond et al. 2015

Because the decision at each step is whether to move up from the option below, and the relevant question is what that step costs per unit of additional health. Comparing every option against a common baseline answers a different question and produces ratios that cannot be used sequentially. An option can appear acceptable against the baseline while the step from the option immediately beneath it is poor value, and only the incremental comparison reveals that. The distinction becomes visible only when the options are laid out in order, which is why the tabular presentation is part of the method rather than a formatting preference.

Source: Drummond et al. 2015

That the options are mutually exclusive, so choosing one precludes the others, and that the set is complete, since an option omitted at the start cannot enter later and its absence may change which of the others sits on the frontier. It also requires costs and effects estimated on a consistent basis across all options, since a difference in method between two of them will appear as a difference in performance. Where an option was excluded before analysis on clinical or practical grounds, that should be stated, since the exclusion affects the frontier as much as any calculation.

Source: Gold, Siegel, Russell & Weinstein 1996

Calculating each option's ratio against a fixed comparator such as usual care and then reading them as a ranking. This omits both the ordering and the extended dominance check, so options are compared against something other than the alternative they would actually displace. The resulting figures look like incremental ratios and cannot be used with a threshold, and the error is not visible from the output alone. A ranking produced this way frequently recommends a different option from the correct procedure, and the discrepancy is largest where the options are closely spaced.

Source: Drummond et al. 2015

Sequential removal based on point estimates can be reversed once distributions are considered, so options discarded as dominated may be optimal in some simulations. Probabilistic analysis therefore compares expected net benefit across the whole option set rather than working through the sequence, and reports acceptability for each option across a range of thresholds. The deterministic sequence retains its value for structuring and presenting the comparison. Reporting both the deterministic sequence and the probabilistic comparison is now common practice, since each communicates something the other does not. The two also serve different audiences, since the deterministic table shows what was compared and the probabilistic result shows how confidently the recommendation can be made.

Source: Briggs, Claxton & Sculpher 2006