Health Economics Metrics

Topic 10 of 76 · Economic reasoning foundations

Dominance and the Efficiency Frontier

An option is dominated if another option costs less and delivers more. The efficiency frontier is what remains after eliminating dominated options: the set of choices where getting more requires paying more.

Why it matters

Before any debate about thresholds or budgets, health technology assessment first eliminates options nobody should ever pick. Plotting every option on a cost-vs-effect plane and drawing the frontier is a five-minute exercise that routinely kills half a shortlist. Incremental comparisons (ICERs) are then computed only along the frontier, each option against the next-cheapest non-dominated one — never against "do nothing" when better intermediate options exist.

The math

Strict dominance:   A dominates B if Cost_A ≤ Cost_B and Effect_A ≥ Effect_B
                    (with at least one strict inequality)

Extended dominance: B is ruled out if a mix of A and C achieves more effect
                    per pound — detected when ICERs decrease as you move up
                    the frontier. Valid frontier ICERs must be increasing.

Procedure: sort options by effect; remove strictly dominated ones; compute pairwise ICERs between neighbors; remove any option whose ICER exceeds that of the next more-effective option (extended dominance); repeat until ICERs increase monotonically.

Worked example

Four options for reducing missed appointments (effect = appointments recovered/year):

Option          Cost/yr    Recovered
Do nothing      £0         0
SMS reminders   £20,000    2,000
Phone calls     £120,000   2,200
SMS + AI triage £90,000    3,500

Phone calls are strictly dominated by SMS + AI triage (costs more, recovers fewer). Frontier: nothing → SMS → SMS + AI.

ICER(SMS vs nothing)   = 20,000 / 2,000  = £10 per appointment recovered
ICER(SMS+AI vs SMS)    = (90,000 − 20,000) / (3,500 − 2,000) = £46.67 per appointment

Increasing ICERs → valid frontier. At ~£160 saved per recovered hospital appointment (see did-not-attend rate), both frontier steps are worth taking; the phone-bank proposal should never reach the committee.

Software engineering connection

Build the same chart for any tooling decision: cost per year on one axis, measured outcome (hours saved, incidents avoided, deploys enabled) on the other. Points up-and-left of the frontier are eliminated before anyone argues about budget. This reframes vendor selection from feature-checklist debates to "you are dominated; the meeting is over." It also exposes the common enterprise pattern of buying the most expensive option for a marginal gain — legitimate only if the incremental price per incremental unit is one the org would knowingly pay.

Pitfalls

  • Comparing everything to baseline instead of to the next option on the frontier — this flatters expensive options by hiding cheaper near-equivalents.
  • Single-dimension effect scores that hide what matters; if two outcomes count, either combine them defensibly (see cost-utility analysis) or show two frontiers.
  • Forgetting uncertainty: options near the frontier may swap places under sensitivity analysis.

Sources