Health Economics Metrics

Topic 20 of 76 · Economic analysis types

Cost-Effectiveness Analysis (CEA)

CEA compares the costs of alternative interventions against a single outcome measured in natural units — life-years, cases detected, admissions avoided, mmHg of blood pressure reduced. Its output is a cost per unit of outcome.

Why it matters

CEA is the workhorse comparison when all options target the same outcome. It answers "which of these ways of achieving X is the best use of money?" — but not "is X worth achieving at all?" (that needs cost-benefit analysis) and not "how does X compare with unrelated priorities?" (that needs cost-utility analysis and a generic outcome like the QALY).

The math

The comparison statistic is the ICER in natural units:

ICER = (Cost_A − Cost_B) / (Effect_A − Effect_B)
     = £ per additional case detected / admission avoided / etc.

Procedure: define the outcome unit; cost every option from the same perspective over the same time horizon; eliminate dominated options (efficiency frontier); compute incremental ratios along the frontier.

Worked example

Three ways to find undiagnosed atrial fibrillation in a population of 100,000:

Option                    Cost        Cases found
Opportunistic pulse checks £150,000       300
Pharmacy screening events  £400,000       520
Wearable-based screening   £900,000       610

ICER pharmacy vs pulse:  (400k−150k)/(520−300) = £1,136 per additional case
ICER wearable vs pharmacy:(900k−400k)/(610−520) = £5,556 per additional case

Whether £5,556 per additional case is "worth it" depends on the value of a found case (stroke prevention downstream) — CEA ranks the options but the adoption decision needs that external valuation. Note the wearable option's average cost per case (900k/610 = £1,475) looks fine; the incremental £5,556 is the honest number for the expansion decision.

Software engineering connection

CEA is the right template whenever options share one outcome: cost per flaky test eliminated across three remediation approaches; cost per incident avoided across observability vendors; cost per successful deployment across CI architectures. The discipline it enforces — one declared outcome unit, incremental (not average) ratios, dominated options eliminated first — kills most bad vendor comparisons before the pricing discussion starts.

Pitfalls

  • Comparing options with different outcomes ("cases found" vs "satisfaction") in one CEA — that needs cost-consequence analysis or a generic outcome.
  • Average cost-effectiveness ratios presented where incremental ones are needed (the wearable example above).
  • Outcome units chosen for flattery: "alerts generated" is an output, not an outcome; insist on units that carry value.

Sources