Methodology: sex-representation evidence signals
FemHealth adds a sex-representation evidence layer over the biomedical literature, with search as the free surface. Every result shows who was studied, what was reported and where each signal came from.
FemHealth searches PubMed, EuropePMC, OpenAlex, ClinicalTrials.gov by default, with bioRxiv, medRxiv, SciELO, WHO, CORE, OpenAIRE, Semantic Scholar, DHARA, AYUSH Portal, TCM, AJOL, IndMED, CAM-Quest behind explicit opt-in controls. Results are deduplicated and ranked by relevance.
For each result, FemHealth uses deterministic pattern matching over abstracts and structured source fields to identify participant counts by sex, whether sex data were reported or analysed, and stated pregnancy or breastfeeding exclusions. Unclear text is not guessed.
Sample sizes are extracted from abstracts using pattern matching on common reporting formats. When participant sex breakdown is explicitly reported (e.g. "n=120 women, n=80 men"), FemHealth surfaces this in the result card - before the user opens the paper.
Research quality is assessed using a multi-factor model considering: study type (systematic review > RCT > cohort > case study), peer review status, citation count, publication recency, and open access availability.
FemHealth scans abstracts and methods sections for common exclusion patterns: pregnancy exclusion, breastfeeding exclusion, age-range restrictions, and hormonal contraceptive exclusion - flagging them in every result card.
Published validation reports 92.2% precision and 78.1% recall for the sex-reported signal on a 149-abstract, single-annotator test set. The sex-stratified signal is also published, including its 7.7% recall limitation.
The methodology DOI resolves to version 4 (24 July 2026), which states 92.3% precision and 78.3% recall. A reference-set label was corrected on 15 August 2026, so the figures above supersede the deposit until version 5 is published.
Sex-data extraction is only as good as the available source text. Many studies do not report a participant sex breakdown in the abstract, so FemHealth labels the signal as not found instead of treating absence as proof that analysis did not happen.