For professional bodies

The evidence behind your guidance, with every decision on the record.

Royal colleges, chartered societies and professional associations publish guidance, standards and position statements that members act on. Axelium runs the systematic review behind them, with your librarian, subject experts and committee deciding at defined points, and every extracted value traced to the passage it came from.

Qualitative and quantitative · PRISMA 2020 flow diagram · named sign-off
Sign-off statuson the record
Stage and who decidedState
  • Search strategyLibrarianLocked
  • Title and abstract screeningTwo reviewers, blindedApproved
  • Full-text screeningSubject expertApproved
  • ExtractionReviewer, against each quoteValidated
  • Certainty of evidenceGuideline groupFinalised
  • Manuscript lockCommittee chairAwaiting second approver
Who and when, for every approvalEnds every Word report
The brief

What a professional body needs from its evidence software.

Six needs, and how Axelium meets each one today.

Qualitative and quantitative evidence

Effects are pooled where the studies allow it and summarised with a recorded reason where they do not. Qualitative studies get CASP appraisal, coding, themes and CERQual confidence ratings.

One record behind every output

Every study, decision, value, quote, judgement and reason sits in one record. Reports for a clinical, a patient and public or an internal audience are drafted from the same pinned evidence.

Authority at defined points

Seventeen approval points and five milestones, each milestone requested by one person and approved by another. When your staff disagree with the AI, their decision stands and the reason is kept.

Freedom to leave

Reports leave as Word and Excel, the review as RevMan 5, the included studies as RIS or BibTeX and the approval record as CSV. An organisation can also download an archive of everything it holds.

Your sources and your access

8 databases searched directly, with source profiles by field: Physiotherapy and rehabilitation lists PEDro and CINAHL to import. Subscription papers come in through your staff’s own access.

Methods in public view

The user manual, the versioned method statement, product tours and validation results are public. A release that changes what a person checks, or a decision rule, is logged in the method’s change log.

How a review runs

The AI proposes. Your people decide.

Every stage ends at a named approval point. A draft binds nothing until a person approves it or a fixed rule confirms it, and you choose where those rules apply.

  1. Protocol

    Review lead

    Frame the question with PICO, PEO or PECO, or with SPIDER for a qualitative question. Each element becomes a screening question, and once the protocol is locked nothing changes it.

  2. Search

    Librarian

    Check the queries, edit them and lock the strategy. Automated runs refuse to start until it is locked, and unlocking asks for a reason.

  3. Screening

    Reviewers

    The AI records a verdict, its confidence and the passage behind it for each criterion. In AI-plus-human or two-human mode a person decides every record, blind to the other verdict until their own is in.

  4. Full text and extraction

    Subject experts

    An exclusion needs two supporting passages by default, and missing evidence never excludes a study. Each extracted value sits beside its quote and page, and by default only values a person has validated are pooled.

  5. Appraisal and certainty

    Subject experts

    Risk of bias with RoB 2, ROBINS-I or CASP is proposed with quotes and confirmed by a reviewer. GRADE and CERQual ratings follow from the confirmed appraisals, and a person finalises each one.

  6. Synthesis and sign-off

    Committee

    An outcome with fewer than two usable results is summarised rather than pooled, with the reason recorded. Five milestones, each approved by a second person, close the review.

Why trust it

Every number traces back to its source.

Validation

We reproduced a published meta-analysis: 4 of 4 endpoints within 3%.

One analyst rebuilt a peer-reviewed meta-analysis end to end in 2 days. All four pooled estimates landed within 3% of the published values, with every extracted value checked against its source. The statistics are deterministic: the same inputs always give the same result.

Read the case study

What every review keeps

  • Numbers are calculated, never generated: the statistics run in deterministic code and are reproducible from their recorded inputs.
  • Every extracted value keeps its quote, page and position, and opens the source at the quoted page.
  • A quote that cannot be found in its document is marked unconfirmed, never shown as verified.
  • Reasons are kept in words: for each screening criterion, for every decision against the AI and for every outcome not pooled.
  • The AI’s steps are filed as the AI’s, naming the model and the run; a person’s are filed under their name and role.
How the method is validated →
Sources and access

Search wide. Respect every licence.

For evidence that is not all randomised trials, and for libraries that pay for their access.

Search

8 databases, plus the ones you license

  • PubMed and ClinicalTrials.gov in every review, with Europe PMC, OpenAlex, Semantic Scholar and ERIC when chosen.
  • Exports from CINAHL, Embase or PEDro imported with the database, platform, search date and strategy recorded.
  • Source profiles by field: Physiotherapy and rehabilitation lists PEDro and CINAHL to import.
  • No language limit unless your protocol sets one, and every record checked against the rest for duplicates.
Full text

Only what you may read

  • Axelium fetches only what is free to read.
  • Subscription papers arrive by upload or through the browser companion, using your own access.
  • Every attempt is recorded: the route, the site, the outcome and who made it.
  • A paper that cannot be found is listed as not retrieved, with the reason, and never excluded for it.
Methods

The review your protocol describes

  • Quantitative, qualitative and mixed-methods reviews, end to end.
  • Risk of bias with RoB 2, ROBINS-I, Downs & Black or Jadad, and CASP for qualitative studies.
  • Agreement between raters as percent agreement and Cohen’s kappa.
  • Certainty with GRADE for pairwise outcomes, CINeMA-style domains for networks and CERQual for qualitative findings.
Members and library

Bring member experts in. Publish what they sign off.

An organisation holds your staff, your volunteer experts and your reviews in one place, and publishes signed-off reviews to the members you invite.

Roles
Admin, Lead Reviewer, Reviewer, Commenter and Viewer in each workspace, and roles the organisation defines itself.
Member experts
Invited as guests by an owner or admin; each guest needs a sponsor, and their access is time-limited.
Audit
The organisation’s audit trail downloads as a spreadsheet, filtered by date, kind of actor, action and workspace.
Archive
One archive of everything the organisation holds, with a checksum for every file.

The evidence library

  1. A review is signed off.
  2. A curator attests to it and previews it.
  3. It is published to the readers you invite.
  4. Readers ask the library and get answers with references, drawn only from published reviews.
How we work with you

Proven on your own topics before you rely on it.

We set Axelium up on your review questions, measure it against your own team’s work and hand over once your staff can run it unaided.

  1. Set up on your topics

    Workshops with your librarian and subject experts agree the review questions, the sources and which decisions each person owns.

  2. Measured against your own review

    Your team runs one review the usual way. You set the thresholds; we report search recall, screening and extraction agreement and citation accuracy against them, shortfalls included.

  3. Handed over to your staff

    Training by role on your own topics, recorded for reuse, each session ending with a competence checklist. The manual, product tours and release notes are public.

See a review run on a question from your own field.

Questions

The details.

Only where you allow it. In single-reviewer mode the AI’s title and abstract decision stands unless a person changes it; in AI-plus-human and two-human modes a person decides every record. Automatic confirmation at full text can be switched off, and every automatic decision is labelled as such and never counted as approved by a person.