How we grade a pattern

A pattern is the strongest thing this publication says. It is a claim that the same thing is happening across more than one part of an older person’s care, and that it has gone on happening over time. Because it is the strongest claim, it needs the clearest rules. This page sets them out. Each pattern’s own page shows how the rules were applied to it.

Why we wrote this down

We did not always have these rules. The register’s first seven patterns were recorded in June 2026 as established on the day they entered, before any test for promotion existed. When we went back and read their evidence against the papers themselves, some of it did not test the claim it was cited for. Several endorsements had been counted as evidence. One study had been counted more than once. Nothing had been invented, but the confidence had been allowed to grow faster than the evidence.

So we rebuilt the grading. The test we set ourselves was simple: would the method stand up to the thinkers this publication says it reasons with? An invented scale would not. The rules below are borrowed from methods the evidence community already trusts. Where we have added a rule of our own, we say so.

Start from what we already said

No pattern is graded from nothing. Each of the five domains has a written assessment of what the evidence in that field shows, with its own sources and its own limits. A pattern’s grade starts from the assessments of the domains it spans. It then says whether it confirms, qualifies or contradicts them.

Beneath each assessment sits a baseline built to a fixed standard, so its depth does not depend on what the person building it happened to think of. It has nine layers. The first is a one-paragraph reading of the whole. The others are: what happens to these patients in the care setting; what the setting itself does to them; what works; how patients and carers experience it; what national standards ask and measure; which systems abroad do better, judged on outcomes; what practitioners take as self-evident, tested part by part; and what is genuinely unsettled. Each entry says which kind of claim it is and is graded by that kind’s own test. A domain whose baseline is not yet built says so.

The assessment is a starting point, not proof. It is graded by the same method and can be wrong. It is revised only from outside evidence, never from the patterns that cite it, because a register that cites its own earlier view as corroboration is only agreeing with itself. When we re-assessed all five domains in October 2026, the clearest finding was this. Three of them treated how early specialists become involved as a lever, on evidence that tests whether the intervention works, not when it is done. Only palliative care has a trial of timing. That finding now sits at the start of every pattern that leans on the idea.

What kind of claim is it?

Different claims need different tests, and the test is chosen by the question the claim answers, never by how strong its evidence looks. A pattern can make more than one kind of claim. Each is graded separately, and the weakest decides the pattern’s standing.

• An effect — X improves or worsens Y. Graded the way clinical guidelines grade evidence, using GRADE (Guyatt et al., 2008). It asks about bias, consistency, how directly the evidence applies, how precise it is, and whether studies are missing.
• A mechanism or context — X works through Z, for these people, in these circumstances. Graded by realist reasoning (Wong et al., 2013). It asks whether the account of what works, for whom and why is stated, and whether it is supported and coherent.
• An experience — how care is lived or received. Graded with GRADE-CERQual (Lewin et al., 2018). Its four questions ask how well the studies were done, whether the finding hangs together, whether there is enough data, and how relevant it is. Patients’ and carers’ accounts are evidence here, not colour.
• A state of the system — the health system is in condition X: funding, workforce, eligibility, policy. This test is our own. It asks who collected the figure, how and when, whether the source is independent of the body whose interests it touches, and whether the number can be traced back to a primary dataset.
• Natural history, and harm from exposure — how often something happens to people in a setting, and whether being exposed to it causes harm: how many older people leave hospital less able than they arrived, what days in bed do to muscle, what a long wait does. No one randomises a frail person to lie in bed, so judging these claims by the standard for treatments would understate the most solid knowledge in the field. We follow GRADE’s own rules instead (Schünemann et al., 2018). Studies that follow people over time start at high confidence for how often something happens, because association, not causation, is the question. For whether an exposure causes harm, observational studies start low and move up when the harm is large or rises with the dose, and experiments on healthy volunteers count as direct evidence of the mechanism. “How often” and “does it cause” are always graded apart, because sicker people stay longer and lie in bed longer.

All five end on the same four levels of confidence: high, moderate, low and very low. That shared scale is what lets one register hold them all.

What a confidence grade does and does not mean

A grade answers one narrow question: how sure can we be, from the evidence this publication has assessed, that the claim is true as stated? It is not a judgement on whether the idea is sensible, and it is not a ranking of the whole medical literature.

Three things follow.

• Low is not the same as wrong. In the methods we borrow, low certainty means the trials do not yet pin down whether, or by how much, something works. A claim can be highly plausible from first principles and still be low on this scale, because plausibility and proof are different questions. Where a claim’s mechanism is strong but its trial evidence is thin, we show the two separately rather than letting one word stand for both.
• The grade is only as wide as the evidence read. Each pattern is graded against its own cited sources and against the settled evidence for its domains. That is a deliberate, recorded body of evidence, not a full systematic review, and a grade can move when wider evidence is read.
• The grade is of the claim as worded. “Specialist care helps” and “earlier specialist care helps more” are different claims, and they can deserve different grades. When one pattern makes both, each is shown with its own grade.

The checks every pattern faces

Whatever kind of claim it makes, every pattern is put through the same seven questions.

1. Does each source actually test the claim? Each source is marked as testing it, bearing on it or not testing it. Only the first two are graded. A paper cited beside a claim is not the same as a paper that tests it.
2. Is each source independent? A study counts once, however many times it is cited. Mandates, inquiries, guidance, board papers and this publication’s own articles are listed as recognition: evidence that the issue is noticed, never evidence that the claim is true.
3. Does it apply here? That means to NHS practice, and to populations in areas of high deprivation or spread thinly across rural areas. Average findings do not always reach the people this publication is most concerned with.
4. What kind of outcome does it move? Something the patient would value, a clinical measure, or a process measure such as length of stay. A pattern that moves only process measures says so.
5. Is every domain earned? A pattern claims a domain only when a source in that domain supports it.
6. What else could explain it? The most plausible rival explanation is named, with the reason it is or is not ruled out.
7. What would prove it wrong? Every pattern states an observable result that would count against it, and that result is looked for each month.

Different kinds of evidence, and why agreement between them matters

Every source is also sorted by the kind of knowledge it comes from, because each kind has its own blind spot.

• Published research sees what gets funded and studied, not always where the problem is.
• Official data and audit sees what can be counted, measured by the body being measured.
• Practice and grey literature, such as service evaluations and professional reports, is often written by the people running the service.
• Lived experience speaks in small numbers, and some voices are heard more than others.

When kinds of knowledge with different blind spots point the same way, confidence rises more than it would from the same amount of agreement within one kind. Ten studies that share one blind spot count as one line of evidence, not ten.

The reverse guard matters as much. Peer review is not proof, and the absence of a study is not the absence of a problem: sometimes the published record is looking in the wrong place. But “it shows up across several kinds of knowledge” must not become a way round evidence. A pattern that rests on no published or official source says so on its face, and can never rise above low confidence.

What each status means

A pattern climbs one step at a time and only when it passes a stated test. Downgrades are published with the reason, never made quietly. A register in which nothing has ever gone down is one that has stopped looking.

How we keep our own judgement in check

Grading is judgement, and two careful readers can disagree. So every pattern is graded twice, by two independent passes. Where they disagree, the disagreement is shown on the pattern’s page rather than smoothed over, and we publish how often the two agreed overall. Judgement is also noisy from day to day; the second pass is how we find out how noisy ours is.

The point of all of it

The scale is not the point. This publication judges care by value-added time: what an encounter actually does for the person receiving it. A pattern matters if it changes that. So the grade’s last question is always about the outcome. Does the evidence reach something the patient would recognise, such as being alive and at home, or able to do what they want to, or dying where they chose? Or does it stop at what the system finds easy to count? A rigorously graded claim about a process measure is still only a claim about a process.

Where to see it applied

Each pattern’s page starts from the domain assessments behind it. It then shows the kind of claim it makes, its confidence, the sources and what each one does and does not show, the rival explanation, the test that would prove it wrong, and where the two grading passes disagreed. The register shows every pattern’s status and the path it has taken.