A programme that "isn't cost-effective." Look closer at why.

A technology isn't a moment. It's a stream of costs and benefits spread across time.

Every intervention plays out over a timeline, and when things happen turns out to matter as much as whether they happen.

Some technologies are time-compact: a cancer drug for advanced disease costs money now and delivers its survival benefit now-ish — cost and effect sit close together on the timeline. Others are radically time-spread: a prevention programme pays its cost over the early years and delivers health years or decades later. A cure that costs £1m today but grants forty years of health is the extreme case — one big cost now, a long thin stream of benefit stretching out for a lifetime.

You cannot simply add up a pound spent today and a pound spent in twenty years as if they were the same thing — and, more controversially, you cannot obviously add a QALY gained today to a QALY gained in twenty years either. Before you can combine them into a single cost-effectiveness ratio, you have to convert everything to a common point in time. That conversion is discounting, and the rest of this lesson is about how it works and what it costs you.

Start with money, where the logic is airtight. £100 next year is worth less than £100 now.

Why? Not inflation — assume prices are perfectly stable. It's still true, for two reasons: you'd rather have money now (you can use it, enjoy it, or need it), and money now can be invested to grow. If £100 invested becomes £103.50 in a year, then £103.50 next year is equivalent to £100 today — so £100 next year is worth only about £96.60 today.

That equivalence is present value (PV): what a future amount is worth in today's terms. The formula runs the growth backwards:

where r is the discount rate and t is the number of years away. The ^t — raising to the power of time — is the crucial part: discounting compounds, just like interest. A pound ten years out is divided by (1+r) ten times over. So the further into the future a value sits, the more dramatically it shrinks — not linearly, but exponentially. Twenty, thirty years out, and even a modest rate cuts a value to a fraction of its face amount. Hold that exponential shape; it's the engine of everything that follows.

For money, discounting is just correct. Almost nobody argues.

Apply present value to a technology's costs and there's no real controversy, because the justification is solid: money has a genuine opportunity cost of capital. A pound the health system spends in ten years is a pound it could otherwise have invested, or spent on something else for a decade first. Committing to spend it later is genuinely less costly, in real terms, than spending it now — so future costs are discounted to a smaller present value.

Note what this is not: it isn't an inflation adjustment. Health-economic models work in constant (real) prices — inflation is already stripped out — and they still discount, because the time value of money survives even in a world with zero inflation. Inflation is about prices rising; discounting is about a pound now being worth more than a pound later regardless of prices.

The standard rate is set by the methods guideline — NICE uses 3.5% for both costs and effects; some jurisdictions have used higher (Poland has historically used 5% for costs). Applied to costs, this is technical housekeeping everyone accepts. The controversy begins the moment you take this same rate and apply it to health.

The shrinking future.

Watch the discount rate decide whether prevention is worth funding.

Here's a prevention programme: it costs £2,500 a year for ten years — £25,000 nominal — and it delivers 2 QALYs of health — but that health arrives late, mostly fifteen to twenty-five years out. The timeline below shows the cost stream (early) and the health stream (late). Costs and health each get their own discount rate below — drag either one and watch the ICER — cost divided by discounted health — cross the £30,000 cost-effectiveness threshold.

Each slider only reshapes its own row — health bars respond to the health rate, cost bars to the cost rate.

Compute a present value.

Put numbers on the shrink. First health, then money.

A single QALY, gained 20 years from now, discounted at NICE's 3.5%. Build the present-value expression — pick each piece:

Discounting money has one clean justification. Discounting health needs a different one — and it's shakier.

The case for discounting costs was the opportunity cost of capital: a pound can be invested. But a QALY cannot be invested. You can't put a year of health in a bank and earn interest on it. So "we discount costs, therefore we discount health" is not an argument — it's a category error. Discounting health needs its own justification, and three are usually offered, each contestable:

• Pure time preference — people simply prefer health now over health later. Real, but ethically loaded: should society's choices treat a future person's health as inherently worth less, purely for being later?
• Uncertainty — the future is unsure; that distant QALY might not materialise. True, but this is an argument for modelling uncertainty directly (sensitivity analysis), not for baking it into the discount rate, where it double-counts.
• Growing wealth/health — future people will likely be healthier and richer, so an extra QALY means comparatively less to them. Plausible for money; much weaker for health, since a year of health arguably means the same to anyone, any era.

There's also a powerful consistency argument for discounting both at the same rate (if you discount health less than costs, you can always improve the ratio by delaying a programme forever). So it's not a simple matter — but none of it makes health-discounting the obvious, settled move that routine practice implies.

Once you see the exponential shape, you can predict exactly which technologies discounting punishes.

A fixed positive discount rate isn't neutral across technologies — it has a systematic bias, and it always points the same way: against anything whose benefit is far in the future. The losers are predictable:

• Prevention and public health — vaccination, screening, lifestyle programmes. Cost now, health decades later. Discounting shrinks the payoff while leaving the bill nearly intact. This is the structural reason prevention chronically struggles in cost-effectiveness terms.
• Paediatrics — treat a child and the health benefits stretch across an entire lifetime, much of it heavily discounted.
• One-shot durable therapies — a gene therapy costing millions now, granting decades of health, has almost its entire benefit stream discounted while the whole cost hits at full value up front. This is a real and current headache in appraising curative therapies.

Because the effect is so consequential, some argue health should be discounted at a lower rate than costs — differential discounting. NICE's base case uses 3.5% for both, but allows a 1.5% health rate in specific cases (notably where benefits are long-term and largely irreversible). Others defend equal rates on the consistency grounds from the last screen. There's no global consensus: rates and rules vary by country and shift over time. The one thing you can't do is treat the rate as a neutral technicality — it is, in effect, a policy lever that decides how much the far future counts.

A related choice quietly shapes the same result: how far out do you look at all?

Discounting decides how much future costs and effects shrink. The time horizon decides whether you count them at all. And getting it wrong is one of the easiest ways to distort an analysis.

The rule is that the horizon must be long enough to capture all costs and effects that differ between the options. For a short, self-limiting condition, a few years suffices. But for a chronic disease — or any technology whose consequences ripple across a patient's remaining life — the correct horizon is often a lifetime horizon, running until death. A common error (sometimes innocent, sometimes not) is to set the horizon to the length of the clinical trial, because that's where the data stops. That systematically truncates the story: it cuts off late costs (which flatters an expensive technology) and cuts off late benefits (which buries prevention and durable cures all over again).

So horizon and discounting interact. A lifetime horizon dutifully captures a prevention programme's distant benefits — and then discounting shrinks them anyway. Choose a short horizon and you don't even let them in the door. Both are decisions about how much the future counts, made before the arithmetic starts. (Modelling long horizons in yearly cycles brings a small timing refinement called the half-cycle correction — a detail you'll meet when we build Markov models in Module 8.)

The other chair

Same skill from both chairs — seeing the discount rate and horizon not as settings to fill in, but as decisions about how much the future is allowed to count, decisions that can matter more than the effect size itself.

Why this matters for HTA

When it lands on your desk: discounting and the time horizon look like the driest cells in the model. They are, in fact, where a great deal of the real decision quietly happens — because they govern how the future is weighed, and most of a technology's value often lives in the future.

• You treat the discount rate as a value lever, not a technicality. A shift from 3.5% to 1.5% can rescue a prevention programme or a gene therapy that failed the base case — not by changing a single fact about the technology, but by revaluing its future. Knowing that, you interrogate the rate as carefully as you'd interrogate an effect estimate.
• You police the time horizon against strategic truncation. A horizon set to trial length rather than the technology's real reach is a classic way to drop inconvenient late costs or late benefits. Insisting on a horizon that captures the full profile is often where an assessor recovers a distorted ICER.
• You recognise the equity built into the arithmetic. Standard discounting systematically disadvantages prevention, children, and durable cures — populations and technologies whose benefits are simply distant. That's not a neutral feature; it's a distributional choice embedded in a parameter, and part of your job is to make it visible rather than let it pass as maths.

The discount rate never announces itself as an ethical decision. It just sits there, a small number in a methods table, deciding how much tomorrow's health is allowed to weigh against today's.

Discounting, in one breath.

• Costs and health effects arrive at different times; before you can combine them, you convert everything to present value — today's-terms worth.
• Discounting compounds: PV = value ÷ (1+r)^t, so distant values shrink exponentially. A QALY 20 years out at 3.5% is worth about half a QALY today.
• Discounting costs is uncontroversial — money has a time value (opportunity cost of capital), and this holds even with zero inflation. It is not an inflation adjustment.
• Discounting health is contested — a QALY can't be invested, so it needs its own shakier justifications (time preference, uncertainty, growing wealth), and whether its rate should match the cost rate is a live argument.
• The rate is a value lever, not a technicality. A fixed rate systematically punishes prevention, paediatrics, and durable cures — which is why differential discounting (a lower health rate) exists and is debated.
• The time horizon decides what you count at all: it must be long enough (often lifetime) to capture every cost and effect — and shortening it to trial length quietly distorts the result.

The future doesn't vanish. We shrink it, on purpose, at a rate almost no one voted on.

That closes Module 5. You've now taken the cost-effectiveness ratio completely apart: what "cost" really means (opportunity cost), how you measure the effect (the four analysis types), whose costs count (perspective), where the numbers come from (costing), and how time reshapes them all (discounting). Every one turned out to be a choice, not a fact — and each choice can move the verdict. But we've kept saying one word as if it were solid: QALY. We've discounted them, divided by them, built entire decisions on them. Where do they actually come from? How do you put a number on a year of life in imperfect health? That question opens Module 6 — measuring health outcomes — and it's where we go next.