The judge at the finish line.

Knowledge and influence run opposite ways.

Here is the principle at the heart of looking forward, and it's genuinely profound: it's known in technology studies as the Collingridge dilemma, and it governs far more than HTA. Two things you care about move in opposite directions over time.

Knowledge (how much you actually know about a technology: whether it works, how well, at what cost, with what side effects) increases over time. Early on, you know almost nothing; by the time trials are done and the drug is in practice, you know a great deal.

Influence (your ability to change the technology, and how cheaply you can do it) decreases over time. Early on, everything is soft: the trial isn't designed, the comparator isn't chosen, the price isn't set, you could shape any of them. Late on, everything is hard: the pivotal trial is finished and can't be re-run, the price is anchored, the technology is frozen. Changing anything now is enormously expensive or simply impossible.

Plot both against time and the two curves cross. Early, influence is high and knowledge is low; late, knowledge is high and influence is low. And here's the uncomfortable implication: standard reimbursement HTA deliberately sits at the far-right, late end: precisely because it wants knowledge, hard evidence, certainty before it commits public money. That's a defensible choice. But it means standard HTA operates at exactly the point of maximum knowledge and minimum influence: it knows everything and can change nothing. It's the judge at the finish line, by design. Looking forward means moving left along these curves, trading away certainty to buy back the power to shape.

Slide along the technology's life.

Below is the life of a technology, from first concept to everyday practice, with two curves running across it: what you know (knowledge), and how much you can change cheaply (influence). Slide to any moment and see the trade you'd be making if you did your HTA there.

There's the dilemma you can't design away. Drag to the left and influence soars while knowledge collapses: you could shape the whole technology, but you're half-blind. Drag to the right and knowledge fills in while influence drains away: you can see everything and touch nothing. Standard reimbursement HTA plants itself at the far right on purpose: it wants certainty before spending public money, and that's legitimate. But notice the cost of that choice: it surrenders the entire left half of the curve, every moment where the technology could still have been shaped toward value. Early HTA is simply the decision to move left: to accept knowing less, in exchange for the power to change more. The crossover is where that bargain is most balanced, and it's nowhere near where standard HTA chooses to stand.

Two moves, two actors: horizon scanning and early HTA.

Moving left on the curve isn't one activity: it's two quite different ones, done by different people for different reasons. Keeping them apart matters.

Horizon scanning is the system looking ahead. A payer, a national body, a health service asks: what's coming down the line, and are we ready for it? It builds a watchlist of technologies likely to arrive in the next few years (a wave of expensive cell and gene therapies, say) so it can prepare budgets, design care pathways, and set expectations before the wave hits rather than being blindsided by it. The goal isn't to shape the technology; it's to shape the system's readiness for it. Horizon scanning is defensive: don't get surprised.

Early HTA (also called early modelling, or early scientific advice) is the developer looking ahead, often with an HTA body's help. A manufacturer, while the technology is still in development, asks: what will it take for this to be reimbursed, and how do I build it that way from the start? An early model helps choose which comparator the pivotal trial should use (the one the payer will demand, not the one that's easiest to beat), which population to target, which endpoint matters to decision-makers, and what price could plausibly be defended. The goal is to shape the technology and its evidence toward reimbursability, before the expensive, irreversible choices are locked in. Early HTA is constructive: build it right the first time.

Same curve, same leftward move, but one actor is preparing the system to receive a technology, and the other is shaping the technology to fit the system. Confuse them and you'll misread who's doing what, and why.

What an early model is for.

Now the subtle part, the one that trips up almost everyone who tries early HTA and does it wrong. When you build a model early (on thin, immature evidence) what is it actually for?

The naïve answer is: it computes an early ICER, an early verdict. And the naïve criticism follows instantly: garbage in, garbage out: the data are too weak, so the verdict is worthless. Both the answer and the criticism miss the point, because an early model's job is not to deliver a verdict at all. Its job is to deliver direction: to show where the uncertainty lies that will most affect the eventual decision, and to point the evidence-gathering there. An early model that says "we can't tell if this is cost-effective yet" hasn't failed: that's not a null result, it's the result. It means: here is the specific thing you don't know that matters most, so go design the study that will resolve it: with this comparator, this endpoint, this population.

This is exactly the expected value of perfect information from Module 9, put to work in practice. Back then, EVPI answered "how much would it be worth to eliminate uncertainty before deciding?" Early HTA uses that same logic prospectively: it computes what's worth knowing before you go and learn it, so that the trial you're about to spend years and millions on is aimed at the uncertainty that actually moves the decision, not at whatever was easiest to measure. So the whole purpose of the analysis inverts. Standard HTA analyses to judge a finished technology. Early HTA analyses to steer an unfinished one: to direct its development and its evidence. A model that produces a confident early verdict is lying about what early evidence can support. A model that produces a map of uncertainty and a plan for what to learn next is doing precisely the job, because it refuses to pretend it knows the answer.

Now you.

Patients belong early too.

There's one more thing that belongs on the left of the curve, and it's easy to file under "soft" when it's actually about influence over what gets measured: patient involvement.

Patients and carers know things the evidence doesn't capture, and often can't capture unless someone builds it in from the start. What outcomes actually matter to the people living with the condition? (Not always the ones clinicians or trialists assume.) What does quality of life really look like day to day? Which side effects are tolerable and which are dealbreakers? Which endpoints would make a treatment feel genuinely worth having? These aren't decorations on an assessment: they determine what the trial should measure in the first place, and therefore what the eventual HTA will even have data on.

And here's why it lands in this lesson: patient input is most powerful early, for exactly the Collingridge reason. Bring patients in at the reimbursement stage, the far right, and you can record their views, but the trial already measured whatever it measured; the endpoints are fixed, the die is cast. Bring them in early (while the trial is being designed) and their knowledge can shape which endpoints are chosen, which outcomes are captured, what "benefit" is defined to mean. Patient involvement is another form of influence-while-you-still-can: a way of getting the things that matter to real people into the evidence base, at the one moment when the evidence base is still soft enough to change. Late, it's a courtesy. Early, it changes what the whole assessment is built on.

The double-edged sword.

None of this is free of danger, and an expert has to see both edges of the blade, because the very thing that makes early HTA powerful is what makes it risky.

The power is influence before the technology sets. But that same influence cuts the other way. An early model built by a developer isn't a neutral peek at the future: it's often an investment and marketing tool, and it can anchor expectations on numbers that no later evidence will fully dislodge. Float an early, optimistic model (big effect, modest price, huge unmet need) and you plant a stake in everyone's mind: investors, clinicians, even future appraisers. Human judgement clings to the first number it hears, so when the hard evidence finally arrives and it's more sober, the early anchor still exerts a pull it doesn't deserve. The leverage that lets early HTA shape a technology toward value is the same leverage that lets it shape perceptions of a technology before anyone knows if it works. Influence arrives before knowledge, so the anchor drops before there's anything to weigh it against.

And the other cautions are real too. Early models are genuinely speculative: feed them guesses and they return guesses dressed as analysis. Horizon scanning is part forecasting, and forecasting is part lottery: you'll miss things and cry wolf about others. So the mature stance is two-handed: use early analysis to steer your own development and evidence honestly and rigorously, and distrust someone else's early analysis that happens to anchor conveniently favourable expectations. It's the same lesson about transparency and manipulation you met with MCDA and equity, but sharper here for one reason: because influence runs ahead of knowledge, the manipulation can land before the facts exist to catch it.

What's the best reading?

Why this matters for HTA

Early HTA reframes what an HTA practitioner is for, not only a judge of finished technologies, but a shaper of unfinished ones:

• Know where you are on the curve, and why. Every HTA sits somewhere on the knowledge–influence trade-off, and standard appraisal sits deliberately at the low-influence end for the sake of certainty. That's often right, but recognise it as a choice, and recognise that real leverage over a technology's value lives earlier, where you know less. Deciding when to engage is itself a strategic act, not a given.
• Early, produce direction, never a verdict. When you work early, your deliverable is a map of decision-critical uncertainty and a plan for the evidence that would resolve it, not a confident ICER. The most useful early output is often "here is exactly what you don't yet know that matters most": the EVPI question asked before the evidence, so the trial is aimed at what moves the decision.
• Use the leverage honestly, and watch others' use of it. Influence before knowledge is double-edged: it lets you shape a technology and its evidence toward genuine value, and it lets a favourable early model anchor expectations before facts exist to check them. Bring patients in early, where they can shape what's even measured; and treat a slick early model that conveniently pre-commits a decision with the scepticism it deserves. The earlier the influence, the greater both the good it can do and the manipulation it can hide.

The judge at the finish line sees everything and can change nothing. Step back up the course (to where the runners are still training, the distance still unset) and you trade the comfort of certainty for the power to shape the race. HTA's deepest leverage isn't at the end, where it knows the most. It's earlier, in the fog, where it can still make a difference.

Early HTA, in one breath.

• HTA doesn't have to be a judge at the finish line. Knowledge and influence run opposite ways over time (the Collingridge dilemma): early on you can shape everything but know almost nothing; late on you know everything but can change almost nothing. The two curves cross, and standard reimbursement HTA deliberately sits at the far-right, high-knowledge, low-influence end.
• Moving left on that curve is two different acts: horizon scanning (the system looks ahead to ready its budgets and pathways: don't get surprised) and early HTA / early modelling (the developer shapes the technology and its evidence, comparator, endpoint, population, price, toward reimbursability).
• An early model's job is direction, not a verdict: it maps where the decision-critical uncertainty lies and steers what evidence to collect next, the EVPI logic of Module 9, used prospectively. "We can't tell yet" is the result, not a failure. A confident early ICER is a misuse.
• The leverage is double-edged: the same early influence that shapes a technology toward value can anchor expectations on optimistic numbers before any evidence exists to check them. Bring patients in early, where they can shape what's even measured; use your own early analysis honestly, and distrust a convenient early model that pre-commits a decision.

Influence runs ahead of knowledge, which is exactly why the earliest moments are both the most powerful and the most dangerous. Look forward in HTA and you can shape what a technology becomes; just remember you're shaping it in the fog, before anyone can see whether you were right.

That closes Module 13, the special topics where standard drug-HTA had to bend: diagnostics valued through the treatments they trigger, devices whose effect lives in the hands that use them, decisions no single QALY could hold, fairness the ICER was blind to, and now the whole timeline of when to act at all. You've stretched HTA well past the textbook appraisal. What's left is the craft itself: taking everything you've built and doing the job: reading a manufacturer's submission with a critical eye, writing a recommendation, facing a committee, catching the red flags in a dossier. That's Module 14: the practitioner's workshop, and it's where the course comes home.