Discounting Costs and Health Benefits in Cost-Effectiveness Analysis

Cost-effectiveness analysis (CEA) routinely compares interventions whose costs and health effects unfold over years or decades — a vaccination program, a screening pathway, a chronic-disease therapy. Before those future costs and future QALYs can be combined into a single present-day incremental cost-effectiveness ratio, they have to be converted onto a common time footing. That conversion is discounting, and it is one of the few modeling choices where a single parameter — the discount rate applied to costs, and the discount rate applied to health — can shift an ICER enough to flip a reimbursement decision for anything with a long time horizon.

Why discount future costs and health benefits at all?

Discounting rests on two distinct justifications, and conflating them is a common source of confusion.

The first is time preference: individuals and societies place less weight on a benefit or cost that arrives later, purely because of when it occurs. A dollar, or a year of full health, received ten years from now is conventionally treated as worth less than the same dollar or year received today — not because it is smaller, but because it is deferred.

The second is opportunity cost: resources spent today could instead have been invested and grown, so a future cost avoided is worth less than a present cost avoided, and a health gain delayed forecloses other uses those resources (or that time in good health) could have had in the interim. In practice, most health technology assessment (HTA) bodies set their discount rate with explicit reference to a real, risk-free social rate of time preference — in the UK, NICE’s rate is set to match HM Treasury’s Green Book social discount rate rather than derived independently by NICE itself.

Mechanically, discounting is simple: a cost or a QALY occurring in year t is multiplied by 1/(1+r)^t, where r is the annual discount rate, before it is summed into the analysis’s total present value. A 3.5% rate roughly halves the weight given to a benefit 20 years out; a 1.5% rate barely touches it. That difference in curve steepness is exactly why the choice of rate is not a rounding detail for interventions with long-tail effects — a childhood vaccine, a one-time gene therapy, a primary-prevention program — where most of the health benefit and much of the averted cost sit decades in the future.

Discount rates used by major HTA bodies

There is no single global standard rate. Each HTA body sets its own reference-case rate, usually tied to that country’s own long-run government borrowing or social discount rate, and the rates genuinely differ:

• NICE (England and Wales): 3.5% per year, applied equally to costs and health effects, in its current reference case — see the NICE technology appraisal process guide for how that reference case fits into the wider appraisal.
• United States — Second Panel on Cost-Effectiveness in Health and Medicine (2016): 3% per year, applied equally to costs and QALYs, as the reference-case recommendation for U.S. analyses (there is no single binding U.S. HTA body, but the Second Panel’s recommendations function as the de facto U.S. methodological standard).
• CDA-AMC, Canada (formerly CADTH): 1.5% per year, applied equally to costs and outcomes, in the fourth edition of its national economic evaluation guidelines — see the CDA-AMC reimbursement review guide.
• Netherlands (Zorginstituut Nederland): a differential rate — 4% for costs, 1.5% for health effects — discussed in more detail below, since it is the standard real-world example of differential discounting in practice.

A rate under 2% and a rate over 3% are both live, currently-used reference-case choices at major national HTA bodies — not a historic-versus-current split. An analyst preparing a submission for more than one jurisdiction (or a reviewer checking a manufacturer’s model) needs the rate that specific body actually specifies, not a remembered “standard” figure, and needs to re-confirm it against the current version of that body’s methods guide, since these rates are periodically revisited.

Differential discounting: a genuinely unsettled question

Almost every major HTA body’s reference case discounts costs and health effects at the same rate. A minority of health economists have argued, on both theoretical and equity grounds, that this is the wrong default.

The core argument, developed most prominently by Brouwer, Niessen, Postma and Rutten in a widely cited 2005 paper and by Claxton and colleagues in later work, runs roughly as follows: if the relative price of buying health (through healthcare spending) is expected to rise over time relative to consumption generally, then discounting health benefits at the same rate as costs systematically undervalues interventions whose benefits accrue mostly in the future — prevention, vaccination, and public health programs in particular — relative to interventions with more immediate benefits, even when the two are otherwise equally efficient uses of the same budget. On this view, a lower discount rate for health effects than for costs is not a subsidy for future health; it is a correction for a bias that equal discounting otherwise builds in.

The Netherlands has applied exactly this logic since its 2006 guideline: a 4% rate for costs against 1.5% for health effects, on the reasoning that the relative price of healthcare interventions is expected to rise faster than costs generally. The United Kingdom took the opposite path at almost the same moment: NICE moved to equal discounting (currently 3.5%/3.5%) in its 2004 methods guide, reversing an earlier practice of differential rates, and has been criticized in the health-economics literature for that reversal since — O’Mahony’s 2014 analysis of NICE’s practice is titled, pointedly, “NICE’s Selective Application of Differential Discounting,” because NICE does still permit a lower rate for health effects, but only as a narrow, non-reference-case exception for interventions producing substantial health gains sustained over a very long duration, presented as a sensitivity analysis alongside the standard equal-rate base case rather than as the default.

The counter-position is also real, not just institutional inertia: equal discounting is simpler, avoids having to defend a second, separate rate with its own justification, and sidesteps the risk that a lower health-effects rate becomes a way to make a preferred intervention look more favorable by construction. Most major HTA bodies — NICE, the U.S. Second Panel, CDA-AMC among them — currently side with that position in their reference case. Treat differential discounting as a live methodological debate with real advocates and real institutional adoption in at least one major jurisdiction, not as a fringe position and not as settled practice either.

What this means for building or reviewing a model

A few practical consequences follow directly from the above:

• State the rate and the source explicitly. A CEA or budget-impact model should name the exact discount rate used for costs and for health effects, and cite the HTA body or methods guide it was drawn from — not just “standard discounting.” The CHEERS 2022 reporting checklist requires this as a distinct reporting item precisely because it is so easy to leave implicit.
• Run the discount rate as a one-way sensitivity analysis alongside the reference-case rate, especially for interventions with a long time horizon (vaccines, one-time curative therapies, primary prevention) where the ICER is genuinely sensitive to this choice. A tornado diagram is the standard way to show how much the result moves relative to other structural assumptions.
• Match the rate to the submission’s destination. A model built for a NICE submission and then reused for a CDA-AMC submission without re-running it at 1.5% instead of 3.5% will materially misstate the ICER for anything with a multi-decade horizon — this is a genuinely different number, not a rounding adjustment.
• Distinguish discounting from other time-horizon choices. Discounting is separate from the analytic time horizon itself, from how a budget impact analysis handles near-term affordability (which typically uses little or no discounting over its shorter horizon), and from parameter uncertainty, which is what tools like expected value of perfect information analysis address instead.

Frequently asked questions

What discount rate should I use?

Whatever rate the HTA body you are submitting to specifies in its current reference case — there is no universal default. If the analysis is not tied to a specific jurisdiction’s submission, state clearly which convention you followed (e.g., the U.S. Second Panel’s 3%) and report the alternative as a sensitivity analysis.

Why not just discount everything at the same rate and avoid the debate?

Most major HTA bodies do exactly that in their reference case, for the simplicity and transparency reasons described above. The debate persists because equal discounting has a real, published theoretical objection behind it (the relative-price argument), not because the question is unresolved through neglect.

Does discounting apply to a one-year or short-horizon analysis?

Its effect is minimal over a short horizon and becomes material only as the time horizon lengthens, since the discount factor compounds. Most guidelines, including CDA-AMC’s, still specify discounting any costs or effects occurring beyond one year, even if the practical impact on a short-horizon model is small.

Is the Netherlands the only country using differential discounting?

It is the most frequently cited real-world example in the literature, and the one worth knowing by name, but analysts should check the current methods guide for any jurisdiction directly rather than assuming the Netherlands’ 4%/1.5% split generalizes elsewhere — NICE’s narrow exception aside, equal discounting remains the dominant reference-case default internationally.

How does the discount rate interact with the QALY itself?

They are separate steps. The QALY calculation (time in a health state multiplied by its utility weight) happens first, year by year; discounting is then applied to each year’s QALY total before those yearly figures are summed into the lifetime total used in the cost-utility analysis.