The report nobody reads.

Documentation versus a decision tool.

Here's the reframe that fixes the page-38 disaster, and it's more fundamental than any writing tip. There are two completely different documents you could write from the same analysis, and confusing them is the root of almost every bad HTA report.

The first is documentation: a faithful record of what you did. It's chronological (background, then methods, then data, then results, then conclusion) and exhaustive, because its job is completeness. It answers the question "what did I do?" This is the document your instincts, fresh from the analysis, want to write. You did painstaking work; you want to show it, in the order you did it.

The second is a decision tool: an instrument built to help someone else make a defensible decision, fast. It answers a completely different question ("what does the decision-maker need to know to decide?") and it's organised not by what you did, but by what they need, in the order they need it. Most bad HTA reports are good documentation masquerading as a recommendation: complete, orderly, thorough, and useless to the overloaded person who has to vote. Because that person doesn't want to retrace your analysis: they can't, and they won't. They need your judgement, and enough basis to trust it, delivered before their attention runs out. The moment you internalise that you're writing a decision tool for a busy, sceptical reader (not a record for the file) everything about how you structure the report changes.

The inverted pyramid.

So how should a decision tool be ordered? The answer comes from journalism, and it's exactly right for HTA: the inverted pyramid. Put the most important thing first, the conclusion, then what supports it, then the details, in strict descending order of importance. The recommendation and its basis go at the very top. The reasoning goes underneath. The methodological detail goes at the bottom, or in an annex.

This feels wrong to an analyst. You built to a conclusion; you want to lead the reader through the logic and arrive at the recommendation as a climax, the way you arrived at it. Resist that completely. Here's why it matters, and it's not aesthetic: the decision-maker reads from the top and stops when they have enough to decide. Their attention is finite and it runs out early. So whatever sits at the top is what reaches everyone; whatever sits at the bottom reaches almost no one. Which means the order determines which information actually exists in the decision-maker's mind when they vote. Put the recommendation and its key conditions first, and they land with everyone. Bury them at the end, and you've hidden them from the very people who most needed them, exactly the page-38 catastrophe. In HTA, structure isn't presentation. Structure is which of your findings survive the reader's attention, and that makes ordering one of the most consequential decisions in the whole report.

Build the recommendation, see what reaches the room.

Here are the pieces of an HTA report. Put them in the order you'd write them. A busy committee member reads the top 3 before they've formed their decision, so watch what actually reaches the room, and what gets left behind.

That's why order is content, not decoration. The facts didn't change as you reordered them, but which facts reached the decision-maker changed completely. Bury the recommendation and it dies unread; bury the restriction and the committee makes the wrong call while believing your report supported it. The inverted pyramid guarantees that the things a decision genuinely depends on arrive first, regardless of how far anyone reads. Everything else (the methods, the background, the careful justification) still belongs in the report, just lower down, where it serves the reader who wants to verify without blocking the reader who only needs to decide.

Calibrate the uncertainty.

Ordering gets your recommendation read. The next principle gets it trusted, and it's the one analysts most often get backwards. A report that sounds equally confident about everything is useless, because the decision-maker can't tell which parts to lean on and which to treat gingerly. If your rock-solid trial result and your speculative long-term extrapolation are written in the same certain tone, you've flattened exactly the information the decision-maker needs most: what can I rely on?

So calibrate. Say the solid things plainly. Flag the shaky things explicitly as shaky. Give the decision-maker a map of the risk (this finding is firm, that one hinges on a fragile assumption) so they know precisely where your recommendation is load-bearing and where it's provisional. A calibrated report doesn't hedge everything into mush; it draws sharp lines between what's certain and what isn't.

And here's the counterintuitive heart of it: admitting your own uncertainty is the single strongest thing you can do for your credibility. The instinct is to sound confident everywhere, as if flagging a weakness would undermine you. The opposite is true. The decision-maker knows perfectly well that no analysis is certain: a report confident about everything reads as naïve or evasive. But a report that volunteers its own weak points earns something precious: when you say "this extrapolation is fragile," the reader infers that where you don't flag a weakness, you're genuinely confident, because you've proven you'll tell them when you're not. Your admitted uncertainties become the collateral that makes your stated certainties believable. Conceal a weakness instead, and the logic runs the other way, catastrophically: when the manufacturer or another committee member exposes the uncertainty you hid, and they will, trust collapses not just in that point, but in the entire report. If you hid that, what else did you hide? Calibration isn't modesty. It's how you build credibility with a reader who starts out not trusting you.

Now you.

Separate fact from judgement.

There's a third discipline that separates a trustworthy recommendation from a presumptuous one, and it ties together much of this course: keep fact and judgement distinct, and mark where one ends and the other begins.

A recommendation is never pure fact. It fuses two different things: what the evidence shows (fact: "the reconstructed ICER is £34,000 in the full population"), and how that evidence should be weighed against values (judgement: "given the severity of this disease and the lack of alternatives, we consider that acceptable in the subgroup but not more widely"). These are genuinely different kinds of claim. The fact is, within its uncertainty, not up for debate. The judgement is a value call that reasonable people (including committee members who outrank you) might make differently.

Blur them together and you do the decision-maker a disservice in both directions. State a judgement as if it were a fact ("the drug is not worth funding") and you've smuggled your values in as though they were arithmetic, usurping a decision that isn't yours to make. Bury the fact under the judgement and the decision-maker can't see what would change if they weighed things differently. So mark the seam explicitly: here is what the evidence establishes; here is the value judgement I've applied to reach a recommendation; and here is what would change under a different judgement. This is the same honesty you met throughout the course: MCDA insisting weights be visible, equity insisting the theory of fairness be named, critical appraisal separating what the manufacturer did from how you judged it. A recommendation that shows its seam lets the decision-maker accept your facts while substituting their own values, which is exactly their right, and exactly what a decision tool should permit. You inform the decision; you don't quietly make it for them under cover of analysis.

The reader who doesn't trust you.

Step back and see the single figure all three principles serve: the person you're actually writing for. Not yourself, not the file, not a methodological reviewer, a specific, difficult reader with three defining features. They are overloaded (limited attention, many decisions, little time). They cannot verify you (they won't re-run your model; they have to take your analysis largely on trust). And they don't trust you by default (you're one input, possibly from an interested party, into a high-stakes decision they're accountable for).

Once you picture that reader clearly, the three principles stop being writing tips and become the obvious response to their situation. Structure (the inverted pyramid) answers their overload: put what matters first, because their attention will run out. Calibration answers their inability to verify: tell them exactly how much to trust each part, and prove you're honest by volunteering your weaknesses, so they can rely on you without checking. Fact-judgement separation answers their scepticism and their authority: show them where the evidence ends and your values begin, so they can trust the facts while keeping the value call that is rightfully theirs. Every one of these is a bridge across the gap between your analysis and their decision, a gap made of finite attention and default distrust.

That's the real craft of the HTA report. It isn't dressing up your analysis; it's engineering the transfer of judgement through a narrow channel of attention and trust. All the work of the previous fourteen modules (every effect measure, every model, every reconstructed base case) reaches the world only through this last, thin conduit: a few pages read by a tired person deciding for millions. The report is the bandwidth of that conduit. Write it as documentation and your judgement stays trapped in your own head. Write it as a decision tool (ordered, calibrated, honest about its seams) and your judgement becomes their decision. Which was the entire point of doing the analysis at all.

Which approach best serves the decision?

Why this matters for HTA

The report is where an assessment either becomes a decision or dies unread, and writing it well is a distinct professional skill, not an afterthought to the analysis:

• Write for the reader, not the record. Before structuring anything, picture the actual decision-maker: overloaded, unable to verify you, not trusting you by default. Your report is a decision tool for them, not documentation of your work for the file. That single reframe ("what do they need to decide?" instead of "what did I do?") reorganises everything, and it's the difference between a report that changes a decision and one that merely records an analysis.
• Put the load-bearing findings where attention actually is. The recommendation, its key restrictions, and the headline result belong at the very top, because readers stop when they have enough, and whatever sits below the attention line effectively didn't get written. Calibrate as you go: state the solid plainly, flag the fragile loudly, and volunteer your uncertainties, because admitted weakness is what makes your stated confidence believable to a sceptical reader.
• Inform the decision; don't quietly make it. Keep fact and judgement visibly separate, so the decision-maker can accept what the evidence shows while applying their own values to the call that is rightfully theirs. A recommendation that hides its value judgements inside apparent arithmetic overreaches; one that shows its seams respects both the evidence and the decision-maker, and, not incidentally, is far harder to attack.

An assessment is only as good as the decision it enables, and a decision-maker can only act on what reaches them, in time, and earns their trust. The report is not the paperwork after the analysis: it is the analysis completing its journey into the world. Write it as the tool it is: ordered so the essential arrives first, calibrated so the reader knows what to lean on, and honest about where your facts end and your judgement begins.

Writing the recommendation, in one breath.

• An HTA report is not documentation of what you did: it's a decision tool for an overloaded, sceptical reader who can't verify you and must trust your judgement to act. Most bad reports are good documentation ("what did I do?") posing as a recommendation ("what do you need to decide?").
• Structure by the inverted pyramid: recommendation and key restrictions first, support beneath, detail at the bottom. The reader stops when they've read enough, so order determines which findings actually reach them: bury the crucial restriction and the committee decides without it. Structure is content, not decoration.
• Calibrate the uncertainty: state the solid plainly, flag the fragile explicitly, give the reader a map of what to trust. Volunteering your own weaknesses is the strongest credibility move there is: it makes your stated confidence believable, while a concealed weakness, once exposed, collapses trust in the whole report.
• Separate fact from judgement: mark where the evidence ends ("the ICER is £34,000") and your value call begins ("acceptable given severity, in the subgroup"), so the decision-maker can accept your facts while applying their own values, the decision that's rightfully theirs. Structure, calibration, and this seam are the three bridges across a gap made of finite attention and default distrust.

All the analysis in the world reaches the decision only through a few pages read by a tired person deciding for millions. Order them so the essentials arrive first, calibrate them so the reader knows what to trust, and show where your facts stop and your judgement starts, and your analysis finally becomes someone's decision, which was the point all along.

You can now write a recommendation that gets read, trusted, and acted on. But the biggest decisions are rarely settled by the document alone. They're settled in the room: where you present the report live, field hostile questions from a manufacturer, and have to defend your judgement under pressure, in real time. Turning a written recommendation into a case you can hold in a live committee is the next lesson.