About ortHOTopic

Once a week, an automated run:

1. pulls everything the whitelisted journals published in the last week from OpenAlex, and drops anything already in the archive;
2. adds PubMed's publication-type tags (that's where “RCT” and “meta-analysis” come from) and discards anything retracted;
3. runs a quick metadata prescreen to decide which papers are worth reading in full;
4. tries to fetch each one's full text from open-access sources such as Europe PMC and PubMed Central;
5. for every paper it can actually read, has an AI model (Claude) write the summary and fill in a structured critical-appraisal checklist from the full text;
6. turns that appraisal into the rigor and impact scores, ranks the week by heat, and hands the top of the list to editorial review.

The rule the whole pipeline enforces: no paper receives a heat score until its full text has been read and appraised. Papers whose full text isn't available yet simply wait, unscored, and are retried every week; open-access copies often appear weeks after publication.

Rigor asks: can you believe the result? Every paper starts from a prior based on its study design (a randomized trial or meta-analysis can exclude more bias than a retrospective series, so it starts higher), plus a small credit for journal tier, which is halved once the methods have been measured directly so reputation is never counted twice. Sample size counts only until we know whether the study was actually powered.

The appraisal then moves the score on the things a journal club would check: loss to follow-up, an a-priori power calculation, allocation concealment, blinded outcome assessment, intention-to-treat analysis, prospective registration, and whether the reported outcome matches the registered one. For significant dichotomous results we also compute the fragility index (how many patients would have to flip outcomes for significance to vanish) and flag the classic tell where more patients were lost to follow-up than the result can absorb. Meta-analyses and systematic reviews are further judged on what they pool (a meta-analysis of retrospective series is not Level I evidence), heterogeneity, and GRADE certainty. Industry funding costs a little; industry funding plus author conflicts costs more.

When a paper simply doesn't report something, that scores as neutral, never negative: absence of evidence about the methods is not evidence of bad methods.

Impact asks: even if it's true, does it matter? The starting point is the strength of the paper's conclusion (practice-changing, confirmatory, or inconclusive), and the single biggest adjustment is whether the effect exceeds the minimal clinically important difference (MCID) of the outcome instrument. Orthopaedic outcomes are dominated by patient-reported scores with published MCIDs, and a statistically significant three-point change on a scale whose MCID is ten is noise wearing a p-value.

Beyond that: patient-important outcomes score above radiographic surrogates, follow-up has to be long enough for the question, contradicting current practice is worth more than confirming it, first evidence on a question earns a bump, reporting harms earns a small one, and a weak or missing comparator costs points. A paper that answers one of the open questions on our open questions page gets extra credit; that is roughly the definition of hot.

The two axes are composed with two rules borrowed from actual critical appraisal rather than summed as bonuses. First, impact is gated by rigor: rigor sets a ceiling on how much impact can count, because a paper cannot be practice-changing on weak methods, no matter how exciting the claim. Second, a retraction zeroes the score outright; nothing else can. Heat is then an even split (half rigor, half impact) minus a small recency decay so this week's papers outrank last month's leftovers. The score is deterministic, and every card shows the factors that moved it: the same reasoning the ranking used, not a post-hoc rationalisation.

Two things are deliberately absent. Citation counts: everything here is a week or two old, and at that age a citation count measures indexing speed, not importance. And reader-relevance signals like Canadian authorship or subspecialty: they say nothing about how good a paper is, so they inform the editorial triage below instead of the score. Either way, heat is a screening tool, not a verdict: cold papers can still be great papers.

Three gates stand between “published this week” and the front page. The full-text gate: no readable full text means no score and no summary; the paper waits in a queue until an open-access copy appears or one is deposited by hand. The heat cut: only the strongest scored papers of each run go forward. And editorial triage: every remaining paper is approved or rejected before it appears, by hand and with help from a small classifier that learns the editor's taste from every past decision and can handle the confident calls automatically; the uncertain middle is always reviewed by a human. Only approved papers are published to the site.