M5 · HEALTH ECONOMICS
The same operation. Two costings. Two different answers.
A hip replacement. Ask the finance office what it costs and you get a clean number: £6,000 — the national tariff, the same for every patient.
Ask the ward that actually did the operations, patient by patient, and the clean number falls apart. The straightforward cases cost around £4,200. The ones that developed a wound infection or a clot cost £11,000. Nobody actually cost £6,000. The £6,000 is an average that describes no real patient.
So which number goes in your model? It sounds like a technical footnote. It isn't. Because here's what's hiding underneath: a new drug that halves post-op complications is worth a great deal — but only if your costing can see the complications in the first place. Feed the model the £6,000 tariff and that drug looks worthless, because the tariff already blended the complications away. Feed it the patient-level costs and the drug's value appears.
The number you pick doesn't just change a line in a spreadsheet. It decides whether a technology's entire benefit is visible or invisible. This lesson is about where cost figures come from — and why the method that produces them is never neutral.
Price is not cost.
The first trap: the number on the invoice is not the resources consumed.
Two words get used interchangeably and mean completely different things.
Price is what someone pays — the tariff, the reimbursement rate, the sticker figure. It's set by negotiation, policy, and accounting convention. Cost is what's actually used up — the surgeon's ninety minutes, the titanium implant, the three nights on the ward, the physiotherapist's time. Costing is the discipline of measuring the second thing, not the first.
Why does the gap matter? Because prices are often blunt instruments that bear only a loose relationship to real resource use. A tariff might pay the same £6,000 for a two-hour uncomplicated case and a two-week disaster — the price is flat, the cost is anything but. If you build a health-economic model on prices when you needed costs, you inherit every distortion baked into how those prices were set. Cost versus price is the distinction the rest of this lesson turns on: a good costing asks "what resources did this consume?", never just "what did we pay?"
Four steps to cost anything.
Every cost figure in every model — however it's dressed up — came from these four steps.
To put a defensible number on any resource, you do four things in order:
- Identify. What resources does this actually consume? Staff time, drugs, devices, bed-days, tests, the building itself. Miss a resource here and it's invisible forever downstream.
- Measure. How much of each? Ninety minutes of surgeon, three bed-days, one implant. This is the quantity, in natural units, before any money is attached.
- Value. What's the unit cost — the price of one unit of each resource? £X per surgeon-minute, £Y per bed-day. Now money enters.
- Aggregate. Multiply quantity by unit cost for each resource, and sum. That total is the cost of the episode.
The whole of costing is variations on these four steps. What separates the two great methods you're about to meet is simply where you start — at the top with a big total you break down, or at the bottom with individual resources you build up.
Two philosophies: top-down and bottom-up.
Same four steps, opposite directions. And they can disagree violently.
Top-down costing starts with a big number and divides. Take a hospital's total annual spend on hip surgery — everything, all in — and divide by the number of operations. Out comes an average cost per patient. Fast, cheap, uses data that already exists. This is where most tariffs come from: someone's top-down average, published for everyone to reuse.
Bottom-up costing — also called micro-costing — starts at the individual resource and builds up. Follow one patient's actual pathway: this many surgeon-minutes, this implant, these bed-days, this much physio, and cost each component separately, then sum. Slow, expensive, data-hungry — and far more precise about what actually drives cost for whom.
The critical thing is that these are not two roads to the same number. Top-down gives you the system average; bottom-up gives you the cost of a specific pathway. When patients are similar, the two roughly agree. When patients vary wildly — some cheap, some catastrophic — they can diverge by a factor of two or three, and the gap between them is exactly the information top-down threw away. The next screen makes that visible.
The distribution a tariff hides.
Here are 100 hip-replacement patients, costed bottom-up. Watch what the tariff does to them.
Each bar below is a group of patients at a given cost. Notice the shape: a big cluster of straightforward cases around £4,500–5,000, and a long right tail of complicated ones running out past £10,000. That tail is where infections, clots, and re-operations live. Tap the two highlighted bars to see what's actually inside them.
Now switch the costing method — and then add the drug — and watch.
£6,000 — national tariff, every patient
You can see the whole distribution — including the expensive tail.
Mean ≈ £6,000
Sit with what the drug toggle just did. The drug is identical in both modes — it genuinely stops complications. In bottom-up, you watch the expensive tail melt and £900 per patient appear. In top-down, the drug is invisible: the tariff blended the complications into its average years ago, so there's nothing left for the drug to reduce. The costing method didn't just change the precision of the answer. It decided whether the technology had any value at all.
Build a cost from the bottom up.
Do one patient by hand — and see how far the tariff is from the truth.
Take one straightforward hip replacement and micro-cost it, component by component: Theatre time £1,500, Implant £2,000, Ward stay 3 days × £400 = £1,200, Physiotherapy £300.
Theatre £1,500 + Implant £2,000 + Ward (3 days × £400) £1,200 + Physiotherapy £300 = £?
National tariff £6,000 − this patient's micro-costed £5,000 = £?
£5,000. Now compare it to the national tariff of £6,000: this uncomplicated patient actually cost £1,000 less than the tariff claims.
The tariff isn't wrong as an average — it's wrong about this patient, and about every individual patient, because averages describe no one. Bottom-up recovers what the average erased: who is cheap, who is expensive, and therefore where a new technology could actually save money.
Who benefits from the blur.
If the method decides what's visible, then choosing the method is choosing who wins.
You've now seen the core asymmetry: top-down hides the tail, bottom-up reveals it. A manufacturer who understands this doesn't pick a costing method for accuracy — they pick the one that flatters their product. This is the same move as cherry-picking a perspective last lesson: select the lens that shows your best face.
So think it through. A technology whose whole value is preventing expensive complications needs the tail to be visible — its benefit lives in that right-hand tail, and a tariff that averaged the tail away makes the benefit vanish.
A manufacturer's new device prevents costly surgical complications. Which costing method should they push for — and why?
A terminology trap: "indirect" means two different things.
One word, two unrelated meanings — and mixing them up is a classic submission error.
You met indirect costs last lesson, meaning lost productivity — the societal-perspective bucket. In costing, "indirect" means something else entirely, and you must keep them apart.
In costing, costs split into:
- Direct costs — resources you can trace straight to a specific patient or episode: this patient's implant, this patient's surgeon-time. You can point at them.
- Indirect costs (overheads) — the shared running costs that keep the whole hospital going but can't be pinned to one patient: management, cleaning, heating, IT, the building. Overhead costs.
Every real episode has to carry a slice of the overheads — a hip replacement uses the heated, cleaned, administered building — so costers allocate a fair share to each episode (often by "step-down" rules that spread overhead across departments). Get the allocation wrong and you either under- or over-state the true cost.
The trap: "indirect cost" in a perspective discussion (productivity) and "indirect cost" in a costing discussion (overhead) are unrelated. A submission that blurs them — or double-counts by putting productivity under overheads — is making a real error. Same two words, two different worlds.
Accuracy has a price, and a limit.
Bottom-up isn't automatically "better." Sometimes the tariff is the right call.
Micro-costing is more precise, but precision isn't free, and it isn't always worth buying:
- It's expensive and slow. Following individual patients, timing every resource, pricing every component — it can cost more than the study it feeds.
- It's local. Your unit's micro-costing reflects your unit — your staff mix, your prices, your efficiencies. Lift it into a national model and you've imported one hospital's quirks. This is the transferability problem: precise here doesn't mean valid there.
- It can over-claim precision. A number built from 40 components looks authoritative, but every component carries its own error, and they compound. A confident-looking micro-cost can be less reliable than an honest average.
So the real skill isn't "always micro-cost." It's matching the resolution to the question. If your technology changes the care pathway — shifts who ends up in the expensive tail — you need bottom-up to see it. If it doesn't touch the pathway and you just need a reasonable total, a top-down tariff is faster, cheaper, and good enough. Resolution is a choice you calibrate, not a virtue you maximise.
The other chair
The other chair. Reading a submission: costs arrive looking like facts — they're measurement decisions in disguise. Ask how each major cost was derived: tariff or micro-costed, and from where? If the technology's value is in avoiding an expensive complication, a top-down tariff will bury that benefit — so a manufacturer understating their own product's value with a crude tariff is as revealing as one inflating it. Check that "indirect" costs weren't double-counted across the costing and perspective senses. And probe transferability: is this a foreign unit's micro-costing wearing a local badge? Building one: choose the resolution your value needs, and justify it. If your benefit lives in the tail, micro-cost the tail and show the distribution, not just a mean — a well-argued bottom-up costing that reveals avoided complications is your strongest card. But don't over-reach: a micro-costing lifted from one specialist centre invites a transferability challenge that can sink the whole model. Match the method to where your costs and savings actually sit, and say so out loud.
Same skill from both chairs — reading a cost not as a fact but as the output of a method, and knowing that the method chosen already decided what the number could show.
Why this matters for HTA
When it lands on your desk: cost inputs look like the dull, settled part of a submission — which is exactly why they're worth scrutiny. A model's ICER is only as trustworthy as the costs feeding its numerator, and those costs were produced by methods that made choices about what to reveal and what to blur.
- You audit how costs were derived, not just what they are. A tariff and a micro-costing of "the same" episode can differ two- or three-fold. Knowing which method produced a figure — and what that method conceals — tells you whether the numerator can be trusted, before you ever look at the ICER.
- You watch for method-shopping. The costing method that flatters a technology depends on where its value sits in the cost distribution. A manufacturer reaching for a tariff where micro-costing is warranted (or vice versa) is often steering the result — the same cherry-picking reflex you learned to spot with perspective.
- You match resolution to the decision. When a technology's benefit is preventing rare, expensive events, an averaged tariff will hide it and the technology will look worse than it is. Insisting on the right resolution isn't pedantry — it's the difference between seeing a real benefit and missing it entirely.
The costs in a model aren't found, they're made. Ask who made them, how, and what their method was built to keep out of view.
Costing, in one breath.
- Price is not cost. Price is what you pay; cost is the resources consumed. Costing measures resources, and a tariff is already someone's averaged costing.
- Every cost figure comes from four steps: identify, measure, value, aggregate — the only question is where you start.
- Top-down divides a big total into an average; bottom-up (micro-costing) builds up from individual resources. They can differ two- or three-fold, and the gap is the information the average threw away.
- A single cost figure hides a distribution. A tariff blends the expensive tail into a mean — and goes blind to any technology whose whole value is shrinking that tail.
- So the method is strategic, not neutral. Whoever's value sits in the tail wants bottom-up; whoever's cost sits above the average wants the blur of top-down.
- More resolution isn't always better: micro-costing is costly, local, and can over-claim precision. Match the resolution to where the value lives.
Every cost in the model is a measurement decision wearing the disguise of a fact.
You've now opened up both halves of the cost-effectiveness ratio — the effect and the cost — and seen that each is shaped by choices, not handed down. One dimension is still missing, and it's the sneakiest: time. The costs and health effects of a technology don't all arrive at once — some land this year, others in twenty years. Are a pound spent today and a pound spent in 2045 really worth the same? Is a QALY gained now equal to one gained decades away? The answer — discounting — is where the module ends.