Concepts and core ideas in Axelium

Questions and frameworks

A framework turns a broad research idea into criteria that can guide search, screening, extraction, and synthesis. Choose the framework that matches the evidence you expect to review.

www.axelium.ai · Configure

Frameworks, policies and versions

In the limited review-settings preview, a framework defines your question elements, a screening policy defines who reviews each study, and the appraisal and certainty choices name the methods used to assess evidence. Each choice displays its version so that a later change to a method does not silently change your review. A review that appraises with a combination outside the standard sets is shown as using custom instruments, rather than being labelled with a set it does not use.

Start with a blank review, a template, settings from an earlier review, or an exported settings file. Copied settings do not include the original search strategy, locks, schedules or results. Review every question and outcome before starting work. Preview frameworks are not a promise of end-to-end workflow support.

Configure explains where effective settings come from and keeps a history of saves. Restoring a save creates a new version, subject to existing locks. A freshness notice means that retained results used different settings for that stage; it does not delete or automatically re-run them. Results without comparable saved settings are marked unknown. These checks concern settings, not changes to the study data.

Alternative wordings the screener proposes for your population, intervention or comparator are listed under “Suggested by the screener” until a reviewer answers them. They help match studies during screening and map study arms during extraction; they are not part of your search strategy. Keeping one leaves the wording in your settings and clears the note; removing one takes it out. Editing the wording yourself answers the suggestion the same way, and results already produced are not changed either way.

Search sources and study discovery

Search is intentionally additive. PubMed and ClinicalTrials.gov cover the core clinical workflow; optional source coverage can broaden the search for multidisciplinary, preprint, device, education, or trial registry needs. Results are deduplicated before screening and keep enough source information for transparent PRISMA reporting.

www.axelium.ai · Search

Citation chasing and pearls

Citation chasing means using known-relevant studies as seeds, then checking references, citing papers, and related recommendations for studies that keyword search may miss. Pearls are known-relevant papers you provide to improve search vocabulary and recall.

• Use pearls before the main search when you already know must-have papers.
• Use Discover after an initial set of included studies exists.
• Review discovered records through the same screening workflow as every other source.

Screening, confidence, and conflicts

Screening applies the protocol to candidate records. AI-assisted screening can suggest decisions and rationales, but formal reviews should preserve reviewer judgement, exclusion reasons, and conflict resolution. Review modes support single reviewer, AI plus human, or two independent human decisions.

Screening has two phases. Title and abstract screening decides which records are worth obtaining; full-text screening then checks each included or unsure study against the paper itself before extraction. The AI reviewer settles a criterion only when enough independent passages in the paper support it (two unless the review asks for one or three), and a criterion the paper does not settle goes to a reviewer as Needs review — absence of evidence never excludes a study. Studies whose full text could not be obtained are marked Full text unavailable rather than excluded.

Whether conclusive full-text decisions are confirmed automatically is a review setting that follows the review mode. In single-reviewer review it is on by default, and everything the evidence does not settle clearly still waits for a person; you can turn it off to confirm every decision yourself. In dual review it is disabled, and full-text decisions are made, and their conflicts adjudicated, separately from the abstract phase. See Review settings.

For teams, assignments and blinding help separate independent decisions from adjudication. See Collaboration for the team workflow.

Qualitative and mixed-methods synthesis

Qualitative reviews code passages from interview, focus-group, process-evaluation, or lived-experience studies into a codebook, develop themes, and rate confidence in findings with GRADE-CERQual. Mixed-methods reviews place confidence-rated qualitative findings next to certainty-rated quantitative effects.

• CASP appraisal is the current qualitative appraisal path.
• JBI-style appraisal is a planned extension rather than the current default.
• Every final theme, finding, and CERQual judgement remains reviewer-owned.

RevMan and Cochrane handoff

RevMan export supports included and excluded studies, screening reasons, and supported quantitative outcomes for Cochrane-style authoring. Studies excluded at full text are listed with their reason, marked “Full text: ” followed by the reason. Risk-of-bias assessment exists in Axelium, but the current RevMan quality export is limited; authors should check and complete any RevMan-specific quality fields after import.

Endpoints and effect measures

Endpoints are the specific outcomes you plan to analyse. Effect measures describe how each endpoint is represented across studies: for example risk ratio, odds ratio, hazard ratio, mean difference, standardised mean difference, prevalence, response ratio, or correlation. Choose measures before extraction when possible.

Evidence provenance and audit trail

Values, decisions, and report evidence should be traceable back to the source documents and reviewer actions that support them. Axelium keeps source snippets, reviewer decisions, protocol changes, sign-offs, and generated evidence together so teams can explain how a conclusion was reached.

Traceability includes knowing when it is absent. Each quoted excerpt is checked against the document it is attributed to, and one that cannot be found there is marked as unconfirmed rather than shown as a verified quotation — so a value you have not yet checked yourself is never mistaken for one that has been.

www.axelium.ai · Analysis

Custom analysis, the Workbench, and user-defined results

Most analyses use Axelium's curated, deterministic tools. When a method falls outside that toolset, the statistics assistant can run your own — a single custom script inline, or a larger job handed to the Analysis Workbench, which iterates on multi-step work in the background and reports live progress in the chat. Both run against a frozen, versioned copy of your dataset with network access disabled, and both require your explicit approval before any code executes.

Their outputs are user-defined results: they appear alongside curated results and plots with a distinct “User-defined” badge, and are deliberately kept out of the curated pipeline — they never feed certainty-of-evidence assessments or generated reports. A custom analysis can also be frozen as a re-runnable recipe that reproduces its estimates on the same frozen data. The reviewer decides whether a user-defined result belongs in the review.

Living reviews

A living review monitors for new evidence after the initial review is complete. New records should return through screening, full-text retrieval, full-text screening, extraction, analysis, and reviewer approval. Updates should never silently overwrite a reviewed conclusion.

Notifications and settings

Notifications cover assignments, mentions, conflict requests, sign-off requests, living-review updates, and account events. Review your notification preferences before starting a team or living review so the right people receive the right prompts.

Reports and evidence board

The Evidence board collects the plots, tables, notes, decisions, and snapshots that should support a report. Generated drafts are starting points for reviewer editing, not final manuscripts. Always compare the report against the protocol, included studies, and finalised evidence.

Common export formats include DOCX for narrative editing, XLSX for structured tables, RIS/BibTeX for reference managers, and RevMan for supported Cochrane handoff scenarios.