Pathway: Researchers

This pathway is for academic researchers who study, develop or validate AI for clinical use. It concentrates on the parts of the lifecycle where study design decides how much a result can be trusted: defining the problem, finding sources of bias, documenting data and models, and testing a model on data from other hospitals (external validation).

Reporting guidelines¶

Journals increasingly expect AI studies to follow a reporting guideline. Which one applies depends on the study design:

Study	Guideline
Developing or validating a clinical prediction model, including machine learning models	TRIPOD+AI Collins et al. (2024)
AI in medical imaging	CLAIM Mongan et al. (2020), updated in 2024
Protocol for a randomised trial of an AI intervention	SPIRIT-AI Cruz Rivera et al. (2020)
Randomised trial of an AI intervention	CONSORT-AI Liu et al. (2020)
First small-scale use of an AI decision support tool in live clinical care	DECIDE-AI Vasey et al. (2022)

The Evaluation chapters refer to these guidelines where they apply.

If your study uses patient records from Dutch hospitals, check early whether it falls under the WMO (the Dutch Medical Research Involving Human Subjects Act). Research that only reuses existing records is usually outside the WMO and goes through the hospital’s own review. Design: ethical review and approval explains the routes.

Tools¶

These tools from the Toolbox suit your role.

Tools for you

AI Fairness 360: An open-source Python and R toolkit for measuring bias in datasets and models and for trying out algorithms that reduce it before, during or after training.
ARX: Free desktop software and Java library for anonymising structured personal data, with built-in checks of re-identification risk and of how much useful information remains.
CodeCarbon: An open-source Python package that estimates the electricity use and carbon emissions of your code while it runs, for measuring and reporting the footprint of training and running models.
CONSORT-AI: An extension of the CONSORT checklist that tells you what to report when a randomised trial tests an intervention with an AI component.
Data Ethics Canvas: A one-page canvas of questions that helps you spot and manage the ethical issues of a project that collects, shares or uses data, at the start and as the project evolves.
Datasheets for Datasets: A set of questions you answer about a dataset, from why it was made to how it is maintained, so others can judge whether it suits their purpose.
DECIDE-AI: A reporting checklist for small, early clinical studies of AI decision support, where you first test a tool with real users before a large trial.
Ethical Matrix: A simple grid that makes you check, for every group affected by a technology, how it touches their wellbeing, autonomy and fairness, so you can weigh the concerns openly.
Fairlearn: An open-source Python library that breaks model performance down by subgroup and offers algorithms to reduce the gaps you find.
Flower: An open-source Python framework for federated learning: you train one model across several sites while patient data stays where it is.
GRIPP2: A reporting checklist that helps you describe how patients and the public were involved in your research, what they did and what difference it made.
Model Cards: A short, structured document that travels with a trained model and tells readers what it is for, how it was tested and where it performs worse.
PROBAST+AI: A structured tool for judging the quality, risk of bias and applicability of studies that develop or test a clinical prediction model, including AI models.
SHAP: A Python library that explains individual predictions by estimating how much each input feature pushed the model’s output up or down, based on Shapley values from game theory.
SPIRIT-AI: An extension of the SPIRIT checklist that tells you what to put in a clinical trial protocol when the intervention includes AI.
System Usability Scale: A ten-item questionnaire that produces a single score from 0 to 100 for how usable people find a system.
TRIPOD+AI: A 27-item checklist for writing up studies that develop or validate a clinical prediction model, whether it uses regression or machine learning.
Value Sensitive Design: A design approach, with ready-made methods and card decks, for identifying the people and human values a technology affects and building those values into the design from the start.

See the Toolbox for all tools.

References¶

Collins, G. S., Dhiman, P., & others. (2024). TRIPOD+AI statement: updated guidance for reporting clinical prediction models that use regression or machine learning methods. BMJ. 10.1136/bmj-2023-078378
Mongan, J., Moy, L., & Kahn, C. E. (2020). Checklist for Artificial Intelligence in Medical Imaging (CLAIM): a guide for authors and reviewers. Radiology: Artificial Intelligence, 2(2), e200029. 10.1148/ryai.2020200029
Cruz Rivera, S., Liu, X., Chan, A.-W., Denniston, A. K., & Calvert, M. J. (2020). Guidelines for clinical trial protocols for interventions involving artificial intelligence: the SPIRIT-AI extension. BMJ, 370, m3210. 10.1136/bmj.m3210
Liu, X., Cruz Rivera, S., Moher, D., Calvert, M. J., & Denniston, A. K. (2020). Reporting guidelines for clinical trial reports for interventions involving artificial intelligence: the CONSORT-AI extension. BMJ, 370, m3164. 10.1136/bmj.m3164
Vasey, B., Nagendran, M., Campbell, B., & others. (2022). Reporting guideline for the early-stage clinical evaluation of decision support systems driven by artificial intelligence: DECIDE-AI. Nature Medicine, 28, 924–933. 10.1038/s41591-022-01772-9

Suggested reading order¶

Reporting guidelines¶

Tools¶