Systematic review

Search on strings anyone can rerun, screen and assess bias as one of two independent reviewers, and report to PRISMA 2020. The flow diagram matches the screening log

Customize this Agent

DOCXXLSX20 steps

The problem

One trial in several papers, depression on different scales, villages randomized instead of people

A search on this question returns thousands of records from medical journals, economics journals and working paper series, and one trial can be reported in several papers. Many trials randomize villages rather than people, and they measure depression on different scales. Count papers instead of trials, ignore the clustering or pool raw scale points, and the pooled effect is wrong before anyone reads it

The request

Run the systematic review in our registered protocol, attached: do cash transfers reduce depressive symptoms in adults in low and middle income countries? Individual or cluster randomized trials, conditional or unconditional transfers against none. Write rerunnable searches per database; our information specialist PRESS checks each first. You run PubMed, LILACS, trial registries and working papers; I'll run Embase, PsycInfo, EconLit and CENTRAL. Be one of two independent reviewers for screening, extraction and RoB 2; we're the other. Give every exclusion a reason; show me full text decisions before extracting. Pool only if trials are close enough; we sign off bias, pooling and GRADE before writing. I need the review, PRISMA 2020 flow diagram and extraction workbook, every included study cited.

Review protocol.pdf

How Axe works

ApprovedChanges asked

Specialist check and searchesStrategies checked before any search

Second reviewer screeningA team member screens independently

Confirm the included studiesTeam settles every disagreement

Second reviewer extractionOutcome data and RoB 2, independently

Sign off the synthesisBias, pooling and GRADE approved

Your reviewApprove, or say what to change

Approved or changes

Review and extraction workbookReview, workbook and check report

Final reviewApprove the revised files

Revised review and extraction workbookRevised files and check report

Step

Synthesize the findings

Prompt

Sent to the model at this step

Compare the team member's form with the workbook, value by value and judgement by judgement, and list every disagreement with both entries and the page or table. Recheck each disputed outcome value against the report, use the value it supports and mark it for the team to confirm. Leave each RoB 2 disagreement for the team to settle, and use the higher risk judgement until they do. Then decide whether the trials are similar enough to pool on participants, transfers, comparison and outcome, and say why. Put every depression result on one scale as a standardized mean difference (Hedges' g), signed so that a negative value means fewer symptoms, and do not mix final values with change scores. Economics trials often standardize by the control group standard deviation: record which standard deviation each trial used and convert consistently. Take one result per trial at the follow up the protocol names, and pool later follow ups separately. Where a trial has spillover and pure controls, use the pure control. Where a trial has several transfer arms against one control group, combine the arms so no participant is counted twice. Where depression is reported only inside a wider wellbeing index, use the depression scale if it is given, otherwise describe the trial in the text. For cluster trials analysed as if individuals had been randomized, correct the standard error with the design effect from the trial's intracluster correlation, or a stated external value tested in a sensitivity analysis. Pool by REML with the Hartung-Knapp adjustment, and report tau squared, I squared and the prediction interval. Run only the subgroups the protocol names and a sensitivity analysis without results at high risk of bias. Test funnel plot asymmetry only with ten or more trials. If pooling is not sensible, synthesize the trials without pooling, following SWiM, and never count trials by statistical significance. Rate the certainty of each outcome with GRADE and word every finding to that certainty.

Uses

Eligibility criteria
Protocol
Extraction workbook
Second extraction

Produces

Extraction workbook
Synthesis
Disagreements

What you get back

DOCXSystematic reviewPRISMA 2020 flow diagram, RoB 2 and summary of findings tables, every trial cited

XLSXExtraction workbookSearches, screening by both reviewers, outcome data, RoB 2 by result, disagreements and analysis

Check and recalculate reportEach count, value and citation marked verified, corrected or unresolved

Make it yours

Put in your own question and protocol, add the databases your library holds, or change the extraction form and the subgroups. Every step and every prompt can be edited

Customize this Agent

Opens a private copy of this Agent in your workspace, ready to adapt to your own sources

More agent examples

ResearchersLiterature reviewDOCXXLSX15 steps

Capital markets analystsBroker opinion of valuePPTXXLSX16 steps

Financial planning analystsBudget variance analysisXLSXDOCX13 steps

Credit research analystsCapital structure analysisXLSX12 steps

Life sciences analystsClinical trial landscapeXLSXDOCX14 steps