How it works

One method, shared by all Current platforms.

CInterval Research develops and maintains the common method used across all Current platforms. Each platform has its own journal selection, inclusion criteria and quality reporting.

In short

A Brief is a structured summary of one study, generated using AI from its published title and abstract; the full text is not used. Fixed rules and separate AI readings then check each new Brief against the abstract, and what does not pass is left out or held for the editor. A Brief holds only what the abstract holds and errors can occur, so read the article before relying on a finding.

Journal selection

Each platform follows a defined list of specialty and general medical journals. The list is selected by the editor, Christian Byskov Fur, MD, and reviewed annually. Journals proposed for a new platform must be indexed in MEDLINE.

Written rules define what counts as research in each specialty. Studies published in specialty journals are generally considered within scope. Studies in general medical journals are assessed individually, on whether their main question belongs to the field.

Crossref and PubMed are checked hourly for new articles in these journals.

Study selection

Not all publications from the selected journals are included. Studies are screened against predefined criteria using automated rules and AI-assisted assessment. The criteria are set by the editor.

The common criteria are the same on all platforms: a study must be research in humans, published as a full article with an abstract, in one of the publication types listed below. The specialty criteria differ by platform: its journal list and the written scope rules that decide whether a study's main focus belongs to the specialty. Each platform's What's included page gives these rules.

Included

  • Randomized controlled trials
  • Observational, registry and epidemiological studies
  • Systematic reviews and meta-analyses
  • Diagnostic, prognostic and risk-prediction studies
  • Imaging studies
  • Secondary analyses and substudies
  • Health economics and implementation research
  • Study protocols and design papers
  • Mechanistic or translational studies in humans

Guidelines and consensus and scientific statements are included as their own type, summarized without a results section.

Not included

  • Editorials, viewpoints and comments
  • Letters, brief reports and short communications
  • Case reports and small case series
  • Narrative reviews
  • Conference abstracts and congress presentations without a published article
  • Preprints
  • Basic science, and animal and laboratory research without human data
  • Articles without an abstract
  • Studies outside the platform's scope

Screening can miss an eligible study or include one that should not be there. An abstract published more than a week after the article is first listed may also be missed.

Brief creation and verification

Each Brief is generated using AI from the published title and abstract, then checked against the original abstract. Full-text articles are not used.

How AI is used

AI models read the abstract and fill a fixed structure: study type, population, groups, the main result, further results, safety and the authors' conclusion. Numbers are kept as printed, with nothing rounded, converted or recalculated. Details the abstract leaves out are marked as not reported.

How results are checked

New Briefs go through two kinds of check. Fixed rules check that:

  • every number appears in the abstract;
  • quotations and the authors' conclusion are verbatim;
  • group labels and definitions appear in the abstract or the title;
  • a ratio does not clearly conflict with the groups' figures shown beside it;
  • a difference carries a sign or a direction word, or has both groups' values beside it;
  • a ratio under an outcome such as survival names the event it measures.

For the same Brief, abstract and rule version, these rules always give the same result. They confirm that a number is printed in the abstract, not that it belongs to the group or outcome shown, and they can compare a ratio's direction with the groups' figures only where both figures are shown.

Two AI readings, by models other than the one that wrote the Brief, then compare it with the abstract. One checks that each field is supported and that each value stands with the right group, outcome, time point and direction. The other checks that no main result is missing. These readings are not guaranteed to give the same answer if repeated, and they can miss errors or object to correct text.

When a check fails

A Brief that fails a fixed rule is generated once more from the abstract, with the failure named. A line or field that fails again is left out, and the page notes that parts of the Brief are withheld. If the failure is in the main result, the safety results or a statement, or concerns direction, the whole Brief is held for the editor.

An objection from an AI reading is handled according to its kind. A question about which group a value belongs to, the direction of a result, which groups are compared, or whether a value is correct goes to the editor, and the Brief is not published until the editor has decided. A further AI reading settles other objections: it keeps the line only with a verbatim quote from the abstract, or replaces it with a line whose numbers are printed in the abstract; otherwise the line is left out. A main result found to be missing is not added afterward, and a Brief whose study type is not confirmed is not published.

The editor also audits published Briefs against their abstracts. These are internal audits, not independent validation.

No clinical interpretation

A Brief adds no clinical interpretation of its own. An abstract does not hold the details, limitations and context such a judgment would need, and an interpretation could not be checked against the source the way a number can. The only conclusion shown is the authors' own, in their words.

Weekly updates and archive

Featured studies

Each weekly issue holds up to three Featured studies and a short list of further selected studies. A Featured study is the same Brief as any other, with no text added.

To produce a shortlist, an AI model reads the Briefs dated in that week and proposes, in ranked order, five or six Featured candidates and up to seven further studies. For each Brief it receives the journal, topic, title and study type and, where the Brief has them, the population, participants, comparison, groups, primary endpoint, trial registration, index test or model, or the document and scope of a statement. It also receives the number of Briefs per topic in the last four published issues, to be used only between studies of comparable merit.

The model is instructed to judge clinical relevance, methodological strength, novelty and potential clinical impact from the question, design, population, comparison, endpoints, size and journal. Result sections, conclusions and funding are not among the fields it receives, and any line containing an estimate or a P value is removed. The title is not filtered, so a result stated in a title does reach the model.

Fixed rules then check the answer, including that its rationales use no result wording and no number absent from the fields provided. An answer that fails is requested once more. If it fails again, no shortlist is produced.

The editor chooses the Featured studies, from the shortlist or from other Briefs. An issue not published by the deadline, the end of Sunday Danish time, is published automatically: any Featured place the editor has not filled goes to the highest-ranked candidate the editor has not rejected. If no shortlist has passed its checks, one is requested first; without one, only the editor's Featured choices are used.

Personal updates

Readers who confirm an e-mail address can follow topics, journals and searches. My Week lists new Briefs that match them, using fixed rules. The Monday e-mail can include up to three Featured studies and up to five of these matches, as the reader chooses. My Week changes which Briefs a reader sees, never what a Brief says.

Archive

Briefs can be searched by trial name, drug, condition, outcome or journal on each platform's Explore page, and filtered by topic, study design, journal and date. A DOI or PMID opens the Brief directly.

Limitations

Briefs rest only on published abstracts. Abstracts leave out details, limitations and results that may matter, and those limits are the Brief's limits.

A Brief does not replace reading the original article, and every Brief links to it. Read the article before relying on a finding.

The checks do not make a Brief infallible, and errors can occur. A Brief is not re-checked when a rule changes. Briefs made when the archive was built passed an earlier version of the checks: fixed rules applied to every Brief, while an AI reading covered a random sample of about one in ten, was advisory and removed nothing. Errors can be reported to contact@cinterval.com; corrections are noted on the Brief.

Methods by specialty