Health Economics Metrics

Topic 14 of 76 · Outcome measures

Incremental Cost-Effectiveness Ratio (ICER)

The ICER is the extra cost per extra unit of health effect when you choose one option over the next-best alternative. It is the headline number of health technology assessment. (When the effect unit is QALYs, it is also called the incremental cost-utility ratio, ICUR.)

Why it matters

Health systems never evaluate a technology in isolation — always incrementally, against what would otherwise be done. NICE compares a technology's ICER to its £20,000–£30,000 per QALY threshold; the US ICER institute reports across $50,000–$200,000/QALY; Canada works to roughly CAD$50,000/QALY. Whether your product is "worth it" to a national health service is, formally, whether its ICER clears the local threshold. See willingness-to-pay thresholds.

The math

ICER = (Cost_new − Cost_comparator) / (Effect_new − Effect_comparator)
     = ΔC / ΔE

Rules of interpretation:

  • ΔC < 0, ΔE > 0: new option dominates — cheaper and better; no ratio needed.
  • ΔC > 0, ΔE > 0: compute ICER, compare to threshold λ; adopt if ICER < λ.
  • ΔC > 0, ΔE < 0: new option is dominated — reject.
  • Ratios behave badly near ΔE = 0 — prefer net monetary benefit for ranking.

The comparator must be the next-best non-dominated option, not "do nothing" — see dominance and the efficiency frontier.

Worked example

A remote-monitoring service for heart-failure patients, per 1,000 patients/year, versus usual care:

Costs:   service £900,000; admissions avoided save £600,000
         ΔC = 900,000 − 600,000 = £300,000
Effects: earlier intervention gains 25 QALYs
         ΔE = 25

ICER = 300,000 / 25 = £12,000 per QALY

£12,000/QALY is comfortably below NICE's £20,000 threshold — a strong case. Note how the net cost matters: without the £600,000 offset the ICER would be £36,000/QALY and the case would likely fail. Cost offsets and their evidence quality are where these analyses are won and lost (see avoided downstream costs).

Software engineering connection

The ICER discipline transfers wholesale to engineering decisions:

(cost of option B − cost of option A) / (outcome B − outcome A)

— incremental cost per additional deploy, per engineer-hour saved, per incident avoided — always against the next-best alternative, not against doing nothing. The two habits worth stealing: (1) name the comparator explicitly; most tool ROI claims secretly compare against a strawman; (2) net the costs first — a tool that costs £100k but displaces £80k of existing spend has ΔC = £20k.

Pitfalls

  • Comparator gaming: comparing against an obsolete or artificially bad baseline inflates ΔE and flatters the ICER.
  • Averages instead of increments: cost per QALY of a whole program is not the ICER of expanding or adopting it.
  • Point-estimate worship: ICERs are ratios of two uncertain differences; report uncertainty via PSA and CEACs.
  • Negative ICERs are ambiguous (cheaper-and-better vs costlier-and-worse give the same sign) — never report a negative ICER without saying which quadrant it is.

Sources