Direct comparison
NMB vs ICER: Ranking Health Interventions
Net Monetary Benefit is a linear, bootstrap-friendly value; ICER is a ratio HTA bodies still report. Compare formulas, ranking, and statistics.
Written and maintained by CASRAI Editorial Board
Last updated
Ask CASRAI · included with Regulatory Radar
Ask about NMB vs ICER: Ranking Health Interventions
Ask CASRAI answers research-administration questions and cites the passages behind every claim — and says so when the corpus does not cover something, instead of guessing. It comes with a Regulatory Radar subscription at $29 a month, alongside the daily digest of regulatory changes and the dashboard of what changed.
150 questions a day, on this site, over the API, or inside your own tools through the CASRAI MCP server.
Everything CASRAI publishes — this page, the dictionary, the guides and the news — stays free to read, with no account and no card.
How do Net Monetary Benefit (NMB), ICER (Incremental Cost-Effectiveness Ratio) compare side by side?
The table below compares Net Monetary Benefit (NMB), ICER (Incremental Cost-Effectiveness Ratio) across 11 procurement-relevant dimensions, from formula through origin of the modern framework.
Side-by-side comparison
| Dimension | Net Monetary Benefit (NMB) | ICER (Incremental Cost-Effectiveness Ratio) |
|---|---|---|
| Formula | (λ × ΔQALY) − ΔCost, where λ is the willingness-to-pay threshold per QALY | ΔCost ÷ ΔQALY (incremental cost divided by incremental effect) |
| Output type | A single monetary value (e.g. +£4,200) — positive means cost-effective at that threshold | A ratio expressed as cost per unit of effect (e.g. £18,500/QALY) |
| Decision rule | Cost-effective if NMB > 0 at the chosen threshold; among several options, the one with the highest NMB is preferred | Cost-effective if the ICER falls below the threshold — but only unambiguous when the new option costs more and is more effective than the comparator |
| Statistical behaviour | Linear function of cost and effect — approximately normally distributed, so parametric and non-parametric (bootstrap) confidence intervals are straightforward to compute and interpret | A ratio of two random variables — its sampling distribution is skewed and can be bimodal; confidence intervals become unstable or uninterpretable when ΔQALY is close to zero, since the ratio can approach infinity or flip sign |
| Ranking more than two options | Direct — rank all options by NMB at the chosen threshold and select the highest; extends cleanly to any number of comparators | Awkward — requires ordering options by cost, eliminating any that are dominated (more costly, less effective than another option) or extendedly dominated, then computing ICERs only between adjacent, non-dominated options on the resulting efficiency frontier |
| Handling of negative or small denominators | No special case — the formula is defined and continuous for any combination of cost and effect differences | Loses meaning — a negative ICER can mean either "dominant" (cheaper and more effective) or "dominated" (costlier and less effective), and the ratio alone cannot distinguish the two without checking the sign of ΔQALY and ΔCost separately |
| Dependence on a threshold value | Threshold-dependent by construction — λ must be specified before NMB can be calculated, and the result (and any ranking) can change if λ changes | Threshold-independent to compute — the ratio itself does not require λ; a threshold is only needed afterward, to judge whether the reported ICER counts as cost-effective |
| Intuitive legibility to non-specialists | Lower — a signed monetary figure that is only meaningful once the reader also knows the assumed threshold behind it | Higher — "cost per QALY gained" reads directly as value for money, which is why it is the figure most clinicians, payers and journalists actually quote |
| How it is typically visualised | Cost-effectiveness acceptability curves (CEACs) plot the probability NMB > 0 across a range of threshold values | Cost-effectiveness planes plot each simulation's incremental cost against incremental effect, with the ICER as the slope of a line through the origin |
| Typical use in HTA reporting | Used mainly in the statistical/uncertainty analysis behind an appraisal (bootstrapping, probabilistic sensitivity analysis, EVPI) rather than as the headline reported figure | The headline figure most HTA bodies (e.g. NICE) report and compare against a cost-effectiveness threshold |
| Origin of the modern framework | Formalised by Stinnett and Mullahy (1998) as "net health benefit," recasting cost-effectiveness analysis as a linear function to solve exactly the statistical problems ICER has | Long-standing convention in cost-effectiveness analysis, predating the net-benefit reformulation |
Common questions
Common questions about Net Monetary Benefit (NMB) vs ICER (Incremental Cost-Effectiveness Ratio)
Do NMB and ICER ever disagree about which intervention is cost-effective?
+
For a single new option against a single comparator, no — at a given willingness-to-pay threshold, NMB is positive exactly when the ICER falls below that same threshold (given ΔQALY > 0). They are mathematically consistent restatements of the same comparison. They diverge in practice once you move past a single pairwise comparison: ranking three or more mutually exclusive options by ICER requires eliminating dominated and extendedly dominated alternatives before the remaining ICERs are even comparable, while ranking the same options by NMB is a direct sort.
What is the "net health benefit" and how does it relate to NMB?
+
Net health benefit (NHB) is the same idea expressed in QALYs instead of currency: NHB = ΔQALY − (ΔCost ÷ λ). Multiplying NHB by λ gives NMB. Stinnett and Mullahy's 1998 paper introduced both forms together as a single linear framework; which one a study reports is a units choice, not a different method.
Why does the NICE DSU and similar guidance recommend bootstrapping NMB rather than the ICER directly?
+
Because ICER is a ratio, its sampling distribution from resampled trial or model data is frequently skewed, multimodal, or undefined when the denominator (incremental QALYs) straddles zero across bootstrap replicates. NMB, as a linear combination of cost and effect, has none of these pathologies, so its bootstrap distribution behaves the way standard confidence-interval theory expects — which is why probabilistic sensitivity analysis and cost-effectiveness acceptability curves are typically built on NMB even when the headline result is still reported as an ICER.
If NMB is statistically better-behaved, why do NICE and most HTA bodies still report the ICER?
+
Convention, communication, and history. Reviewers, clinicians and journalists reading "£20,000 per QALY gained" understand it as a rate of value for money without needing to know the underlying threshold — a skill most audiences already have from unit pricing generally. NMB requires the reader to already know (or be told) the threshold λ before the number means anything, and a raw monetary figure like "+£4,200" does not self-explain the way a per-QALY rate does. The ICER also predates the net-benefit reformulation by decades and is embedded in how cost-effectiveness thresholds themselves are defined and communicated in guidance documents.
Does choosing a different willingness-to-pay threshold change the ICER?
+
No — the ICER itself (ΔCost ÷ ΔQALY) does not depend on any threshold; it is computed purely from the cost and effect data. The threshold is applied afterward, as the yardstick against which the reported ICER is judged. NMB works the opposite way: the threshold λ is a required input to the calculation itself, so NMB (and any ranking based on it) is only defined once λ is specified, and can change if a different threshold is used.
Going deeper
