Preventive medicine has an unusual problem: the tests are advancing far faster than the outcome evidence for them. That gap is where most of the marketing lives. Our grades are an attempt to state the size of the gap for each test, in one character, without pretending it is smaller or larger than it is.

What a grade applies to

A grade attaches to a test or intervention, for a stated purpose, in a stated population. Not to a company, not to a clinic, not to a price. Whole-body MRI for cancer screening in asymptomatic adults gets a grade; whole-body MRI for staging a known malignancy is a different question with a different answer, and we would grade it separately.

This matters because grades get quoted out of context. A provider that offers a grade C modality is not thereby a grade C provider. What the grade tells you is what the provider should be claiming about it.

The grades

Randomised trial evidence of benefit on a hard outcome (mortality, or morbidity that matters to patients), replicated.

Consistent evidence of benefit, but from trials with surrogate endpoints, single trials, or strong longitudinal cohorts.

B-

Promising trial evidence that does not yet reach a hard outcome, or a single registrational study without independent replication.

Mechanistically plausible and widely used, but no trial evidence of net benefit in the screened population.

C-

Plausible with documented harms that are not yet quantified against the benefit.

Marketed ahead of its evidence. Measurement is unreliable, or the claimed inference is not supported.

The two rules that do the work

Hard outcomes, not surrogates

Grade A requires replicated randomised evidence of benefit on an outcome a patient would care about: death, or morbidity that affects how a life is lived. Detection rates, stage shift, biomarker change and image findings are surrogates. Surrogates can be excellent, and stage shift in cancer screening is a reasonable one, but the history of screening contains several technologies that produced clean surrogate improvements and no survival benefit, because they were detecting disease that was never going to progress. We do not treat a surrogate as a hard outcome, however good it looks.

Harms are counted, not mentioned

A test's grade reflects net benefit in the population screened, so documented harms sit on the same ledger as documented benefits. Incidental findings, false positives, downstream biopsies and overdiagnosis are not footnotes to a screening test; for many screening tests they are the largest measured effect. A modality with real detection and unquantified harms cannot reach B, which is what grade C- exists to express.

What we do not grade

Reporting carries no grade. When we write about how a market is structured or what a programme costs, we are describing the world rather than appraising a literature, and stamping a letter on that would be false precision. Those pieces are marked ungraded.

How a grade changes

Every graded article ends with a section stating what would move it. This is not a formality. It is the part that makes a grade falsifiable rather than an opinion. If the stated condition is met, we regrade and log the change on the article.

We also regrade downward. A test whose registrational trial fails to replicate independently, or whose reliability turns out to be worse than claimed, loses ground. Grades carry the date they were last reviewed.

Who grades

Grades are set by the editorial staff against the criteria above and checked against the primary sources cited in the article. Every graded article carries its references, the date the figures were verified, and the condition under which the grade would move, so a reader can audit the judgement rather than take it on trust.

Where the evidence genuinely splits and a defensible case exists for two grades, we say so in the article and grade to the more conservative of the two. A grade is a floor on scepticism, not a summary of the best case available.

The Index method

The Baseline Index is a different instrument and deliberately a dumber one: it records what providers offer and what they charge, with a source and a date for each figure. It does not score, rank or recommend. Modality columns record presence, not quality. We separated the two deliberately: a table that mixes verifiable facts with editorial judgement invites readers to treat the judgement as a fact.

We do make one editorial judgement in the Index, and we flag it as such: a provider that does not publish a price is marked as such in its own column, because non-disclosure is a decision about the buyer's position and worth knowing before a consultation.

The grades as data

Every standing grade, with its key findings, references and assessment date, is published at /method/grades.json under CC BY 4.0. Quote it with attribution to Baseline. If you are building something that needs the whole corpus rather than the grades, use llms-full.txt.

Errors

We correct in place, with a dated note on the article. Nothing is quietly edited. Corrections policy.