One drug. Two departments. Two opposite verdicts.

Last lesson you learned the denominator is a choice. Now the numerator turns out to be one too.

In the previous lesson you saw that the effect (the bottom of the cost-effectiveness fraction) is a choice of ruler. Now look at the top: the cost. It feels like the solid, objective part. A drug costs what it costs; add up the invoices.

But add up whose invoices? When a patient takes this drug, costs and savings ripple outward in rings. The drug itself and the hospital stays it prevents: that's the health budget. But also: the social-care package that shrinks or grows. The carer who no longer has to quit their job. The patient's own return to work. The taxis to the clinic. Every one of those is a real financial consequence of the treatment, but they land in different pockets, belonging to different people and different budgets.

The perspective of an analysis is simply the decision about which of those pockets you're going to look in. Draw the boundary tight (only the health budget) and you get one cost. Draw it wide (every pocket in society) and you get another. Same drug. Same ripples. Different boundary, different number in the numerator.

The numerator was never handed to you. You drew it.

There isn't one "correct" boundary: there are three standard places to draw it, nested inside each other.

From tightest to widest:

• Payer perspective. Count only costs and savings falling on the health budget: drugs, procedures, hospital stays, tests. This is the narrowest boundary and, in many systems, the default: it answers the payer's literal question, "what does my budget gain or lose?"
• Health and social care perspective. Widen the ring to include the social-care budget too: home help, residential care, the support a condition demands beyond the clinic. This is NICE's official reference case in England: health and social care, because the two budgets are so entangled that ignoring one distorts the other.
• Societal perspective. Widen the ring all the way: every cost, wherever it falls. Add lost productivity when illness stops people working, the value of unpaid care from family members, patients' travel and out-of-pocket costs, effects on other public budgets entirely. It asks not "what does the health system pay?" but "what does the nation pay?"

Each ring contains the one inside it. And here's the crucial property: moving outward can only ever add costs and savings, never remove them. So the ICER doesn't just drift as you widen the boundary; it can move decisively, because you're pouring whole new categories of money into the numerator. Which is exactly what makes the choice consequential enough to fight over.

To draw any boundary, you need to know what's on the table. Costs come in four buckets.

• Direct medical costs. The obvious ones: the drug, the doctor, the ward, the scan. These sit in the innermost ring: they're in every perspective, because they always land on the health budget.
• Direct non-medical costs. Real, treatment-caused spending that isn't clinical: the patient's travel to appointments, home adaptations, special diets. Small per person, occasionally large in aggregate.
• Indirect costs (productivity). The economic value of work not done because someone is too ill to do it: sick leave, reduced output, early retirement, in the starkest cases premature death removing a worker from the economy for decades. These are usually the largest number in the whole analysis when a working-age population is involved, which is precisely why they're the most contested.
• Informal care costs. The unpaid labour of family and friends: the spouse who cuts to part-time, the daughter who moves back home to provide care. It has real economic value even though no invoice is ever raised.

Now overlay this on the boundaries: the payer perspective sees essentially bucket one. The health-and-social-care perspective adds social-care spending. The societal perspective is the only one that reaches buckets three and four: productivity and informal care. Those two buckets are where the action is, because they're huge, and because they fall on people who aren't the payer.

Watch a verdict flip, driven by nothing but where you draw the line.

Here's our drug: +0.2 QALYs over the comparator, for a working-age patient. The rings below show the costs. Move the perspective boundary outward and watch the ICER (and the verdict against a £30,000 threshold) recompute live. Watch which direction it moves at each step.

Put the numbers on the flip.

The ICER is net cost divided by the QALY gain (0.2 throughout).

First, the payer. The drug costs £10,000 and prevents £1,000 of hospital care → net health-budget cost £9,000. Now go all the way to the societal view. Two things change at once: extending the patient's functional life adds £1,200 of home-care cost, and their return to work produces £5,400 of value: net societal cost = £9,000 + £1,200 − £5,400 = £4,800.

The societal perspective looks more complete, and more generous. Now watch it turn on someone.

It's tempting to conclude the societal perspective is simply better: fuller, more honest, and kinder to good drugs. But look again at why it flipped our drug to fund: because the patient went back to work. The £5,400 of productivity did the heavy lifting. Which raises an uncomfortable question: what about a patient who doesn't, and won't, work?

Take the identical drug (same 0.2 QALYs, same direct medical costs) given to an 80-year-old retiree. They generate no productivity gain, because they weren't in the labour market to begin with. So the societal numerator gets no productivity offset, but it still carries the £1,200 of extended-life care cost. Net societal cost = £10,200. Societal ICER = £51,000: worse than the payer's £45,000. The wider boundary that rescued the 40-year-old doesn't just fail the 80-year-old; it actively condemns them.

That's not a quirk. It's structural. A perspective that counts productivity systematically values the health of working-age, employed, higher-earning people above the retired, the unemployed, the carers, the disabled. The "fuller" picture has a bias baked into it: the same ability-to-pay problem you met with cost-benefit analysis, wearing a different coat.

If you open the boundary for savings, you must open it for costs too.

Here's how perspective gets weaponised. A manufacturer wants a societal perspective because their drug sends patients back to work: a big productivity saving that improves the ICER. Fine. But a societal boundary doesn't only let the good news in. Widen the ring for productivity gains, and you must also count societal costs the drug creates: like the £1,200 of home care you already watched push the ICER the wrong way, or the years of informal care a longer-surviving patient may need from a family member.

The rule is symmetry: a perspective is a boundary, and a boundary lets everything at its distance cross: costs and savings alike. You don't get to draw a line that admits productivity gains but excludes informal-care costs. Yet that asymmetry (societal benefits, payer-only costs) is one of the most common manipulations in submitted models. A submission that kept our £5,400 productivity gain but quietly dropped the £1,200 of care would be manufacturing a better ICER out of a rigged boundary. It is structurally the same move as the false cost-minimisation from last lesson: include only the half of reality that flatters your product.

So when you see a societal perspective in a submission, the question isn't just "did they count productivity?" It's "did they count everything at that boundary, or only the parts that helped?" A perspective applied to the savings but not the costs isn't a societal analysis. It's a payer analysis with a productivity bonus stapled on.

Even when you play fair, one of these buckets is genuinely hard to measure.

Suppose you commit to a symmetric societal perspective. You still have to put a number on lost productivity, and there is no agreed way to do it.

• The human-capital approach values lost work at the person's full expected earnings for the whole time they're off: potentially decades if illness ends a career. It treats every hour of lost work as permanently lost to the economy.
• The friction-cost approach says that's far too much: in a real labour market, a sick worker is eventually replaced, so society only loses output during the "friction period" until someone fills the gap. It typically yields a number many times smaller than human capital.

Pick one and the productivity figure (the very number that flipped our verdict) can swing by an order of magnitude. Worse, there's a double-counting trap: the QALY may already capture some of the wellbeing of being well enough to work, so adding full productivity on top can count the same benefit twice.

This is why reference cases are cautious. It isn't that productivity doesn't matter: it's that the number is soft, method-dependent, and easy to inflate. When you see a large productivity offset carrying a submission, the method used to compute it is the first thing to interrogate.

The other chair

Same skill from both chairs: seeing the perspective as a boundary that must let everything at its distance across, and knowing that where you draw it is a claim about whose costs society is willing to count.

Why this matters for HTA

When it lands on your desk: perspective is rarely left to the manufacturer's discretion: the agency's reference case fixes it, precisely because it's powerful enough to flip a decision. Knowing the perspective rules is knowing which number is the real one and which is a scenario dressed up as the answer.

• You hold the mandated perspective as non-negotiable. NICE works to a health-and-social-care perspective; AOTMiT in Poland centres the public-payer perspective. A submission whose base case quietly widens the boundary to improve the ICER hasn't done a more thorough analysis: it has answered a different question than the one the agency asked.
• You treat productivity and informal care as scenarios, not base-case facts. Where a societal view is presented, it belongs in sensitivity analysis, applied symmetrically, with its method exposed. A base case leaning on productivity is a base case leaning on the softest, most manipulable number in the model.
• You recognise the equity stakes in the boundary itself. A productivity-inclusive perspective structurally favours the young and employed. That's not a neutral technical enrichment: it's a distributional choice, and keeping it out of the base case is itself an equity decision the reference case is making on society's behalf.

The perspective is chosen before the first cost is counted, and by then, half the verdict is already decided. Find out where the line was drawn before you trust the number it produced.

Perspective, in one breath.

• The cost in a cost-effectiveness ratio isn't objective; it depends on the perspective: the boundary around whose costs and savings you count.
• Three standard boundaries nest inside each other: payer (health budget) → health and social care (NICE's reference case) → societal (every cost, wherever it falls).
• Widening the boundary can add costs and savings, and can move the ICER either way: our drug went £45,000 (payer, reject) → £51,000 (health & social care, worse) → £24,000 (societal, fund).
• "Wider" is not "fairer." A productivity-inclusive perspective systematically values the working-age and employed over the retired, unemployed, and carers: the ability-to-pay bias in a new coat.
• Symmetry is the rule. A boundary that admits societal savings must admit societal costs too. Counting productivity gains but not informal-care costs is the commonest manipulation in the book.
• Even done honestly, the productivity number is soft (human-capital versus friction-cost methods differ severalfold) which is why reference cases keep it out of the base case.

Draw the boundary before you run the numbers, because the boundary already contains the answer.

You've now seen that both halves of the cost-effectiveness fraction (the effect and the cost) are shaped by choices made before any arithmetic. But we've still been treating each individual cost as a known figure. Where do those figures actually come from? How do you attach a real, defensible number to "a day in hospital" or "a course of treatment"? That's costing (from broad tariffs down to micro-costing) and it's where we go next.