Extraction matrix: systematic review of explainable AI validation in clinical decision support

Supplementary material for a systematic review of the validation of explainable AI (XAI) in clinical contexts, conducted according to PRISMA 2020 (30 included studies). The review examines four dimensions jointly across the same set of studies: transparency and explainability, cybersecurity and robustness, fairness, privacy and ethics, and regulatory alignment. The workbook (Excel, 14 sheets) contains:– a per-study traceability matrix (Overview and one sheet for each of the four dimensions), with the item-level coding behind the risk-of-bias and certainty-of-evidence ratings and the justification recorded for each rating;– the data dictionary (definition and coding rule of each of the 55 extraction variables);– the controlled vocabularies (XAI techniques, model architectures, data modalities, clinical domains and others) with the studies assigned to each category;– the full extraction matrix (55 variables for each of the 30 studies) and the frequency synthesis (sheet 13A);– the change log of all corrections made to the dataset;– a consistency sheet that recomputes the figures reported in the paper.