Health Economics Metrics

Topic 23 of 76 · Economic analysis types

Cost-Minimization Analysis (CMA)

CMA compares only costs, and picks the cheapest option — legitimate only when the outcomes of the alternatives have been demonstrated to be equivalent.

Why it matters

CMA is the simplest analysis and the most abused. The equivalence claim is doing all the work: if outcomes genuinely don't differ (a biosimilar vs its originator; two suppliers of the same service meeting the same specification), then cost is the only question and CMA is correct. The rigor lives in proving equivalence first — typically via a non-inferiority study with a pre-specified margin — which is exactly the step buyers usually skip.

The math

Given evidence that Effect_A ≈ Effect_B (within a pre-specified margin δ):
Choose min(Cost_A, Cost_B)

Costs measured from the same perspective, over the same horizon,
including switching/transition costs.

If equivalence cannot be evidenced, CMA is invalid — use CEA/CUA instead.

Worked example

A trust chooses between two video-consultation platforms. A 3-month parallel pilot shows completion rates 94.1% vs 93.8%, patient satisfaction 4.4 vs 4.4 — differences inside the pre-agreed δ of 2 percentage points. Outcomes: equivalent. Costs over 3 years:

                     Platform A     Platform B
Licences             £360,000       £210,000
Integration          £80,000        £150,000
Training/support     £60,000        £90,000
Total                £500,000       £450,000

Platform B wins by £50,000 — including its higher integration cost. Without the pilot, the equivalence claim would rest on vendor brochures, and a 1-point completion-rate difference (≈ thousands of failed consultations/year) would dwarf £50,000.

Software engineering connection

CMA is the formal shape of commodity procurement: two CI providers meeting identical SLOs, two object stores with the same durability spec. The health-economics lesson is the order of operations: first evidence equivalence (benchmark against your workload, pilot against your SLOs, with the margin agreed in advance), then compare total costs including migration. "They're basically the same, B is cheaper" without the first step is how orgs buy the tool that's 10% cheaper and 40% worse. Corollary: when a vendor argues price, make them stipulate equivalence — it's binding in the other direction too.

Pitfalls

  • Assumed equivalence — the defining sin; absence of evidence of difference is not evidence of equivalence (underpowered pilots "show" equivalence for free).
  • Omitting switching costs — migration, retraining, and parallel running belong in the cost side.
  • Equivalence on the wrong outcomes: equivalent on the measured metric, different on one that matters (accessibility, tail latency, data egress).

Sources