We are conducting a systematic review comparing intrinsically interpretable statistical models, such as logistic regression and Cox proportional hazards models, with explainable artificial intelligence (XAI) approaches, including machine learning and deep learning models supported by SHAP, LIME, attention mechanisms, or other explanation methods, for predicting breast cancer recurrence after primary treatment.
The review is registered in PROSPERO: CRD420251145602; https://www.crd.york.ac.uk/PROSPERO/view/CRD420251145602
Our goal is to critically appraise published models across the black-box, grey-box, and white-box spectrum, identify current methodological and reporting gaps, and outline future directions for trustworthy, clinically usable prognostic models in oncology.
We are currently completing data extraction and would welcome feedback from colleagues working in breast cancer prognosis, prediction modelling, oncology, or explainable/interpretable AI before we move to risk-of-bias assessment and evidence synthesis.
The full protocol, including the eligibility criteria, search strategy, and data extraction/appraisal plan, is attached to this post.
We would be grateful for comments on:
• the review question and eligibility criteria (PICO); • the search strategy across PubMed, Scopus, IEEE Xplore, Google Scholar, and Research4Life; • the planned data extraction and risk-of-bias/quality appraisal tools, including CHARMS, PROBAST+AI, and TRIPOD+AI.
Please share any comments under this post or contact the corresponding author directly.
Corresponding author: Eiman Sahly, University of Benghazi, Libya eiman.sahly@uob.edu.ly ORCID: 0000-0002-5888-6305
Review team: Eiman Sahly, Sophie Pilleron, Aiman Gannous, and Abdelfattah Elbarsha