The problem it solves
Data extraction combines close reading, interpretation and normalisation. Errors can appear when figures are copied incorrectly, denominators are lost, populations are mixed, results are duplicated or the location of the data in the source text is not preserved.
What Vetra does
- Performs data extraction in a structured way based on the review design and outcomes.
- Identifies candidate fields such as population, intervention, comparator, outcomes, sample size, follow-up and effect estimates.
- References each data point in the original paper with the exact section, table, figure or paragraph so human review is easier.
- Flags inconsistencies or incomplete fields before moving to synthesis or meta-analysis.
Human oversight
The AI can speed up the location and pre-structuring of data, but the researcher must verify relevant fields, confirm the clinical interpretation and decide which data are appropriate for each analysis.
What is recorded
- Extracted field, value, unit, source and location.
- Exact reference to the section, table, figure or paragraph in the original paper where each data point appears.
- Review status: pending, verified, corrected or discarded.
- Human corrections and the reason for each change.
- Notes about assumptions, conversions, derived data or imputation decisions.