Purpose
This policy is distinct from the AI and LLM Policy, which governs AI as a writing tool. This policy sets out what JAIHL expects when the research itself develops, trains, or evaluates an AI or machine learning system intended for a healthcare or life-sciences application.
Applicable Reporting Guidelines
Where the submission matches one of the following study types, authors are expected to follow the corresponding reporting guideline and submit its completed checklist:
- CONSORT-AI: Randomized controlled trials evaluating an AI/ML intervention.
- SPIRIT-AI: Clinical trial protocols involving an AI/ML intervention.
- TRIPOD-AI: Studies developing or validating a diagnostic or prognostic prediction model using AI/ML methods.
- CLAIM (Checklist for Artificial Intelligence in Medical Imaging): Studies applying AI/ML to medical imaging or digital pathology.
- Where a submission does not clearly fit one of these designs, authors identify the closest applicable standard and explain any material deviation.
Minimum Reporting Expectations
- A clear statement of the clinical or biological problem and the intended use case for the model.
- Description of training, validation, and test datasets, including source, size, demographic composition, and how data splits were constructed.
- Model architecture, key hyperparameters, and training procedure sufficient for independent replication.
- Performance metrics appropriate to the task, reported with uncertainty estimates (e.g., confidence intervals) rather than single point values alone.
- A statement on external validation, or an explanation of why external validation was not performed.
Fairness and Bias Reporting
Where a model is intended for use across a patient population, authors report performance disaggregated by relevant demographic or clinical subgroups where ethically and legally permissible, and discuss known or possible sources of dataset bias. Authors are not required to demonstrate the complete absence of bias, but are expected to transparently report what was and was not assessed.
Explainability
For models informing clinical decisions, authors describe what interpretability or explainability methods, if any, were applied, and discuss the clinical implications of the model’s transparency or lack thereof.
Regulatory Status Disclosure
Where a study evaluates a system that is marketed, intended for marketing, or seeking regulatory clearance as a medical device (for example, under FDA, CE marking, or an equivalent regulatory framework), authors disclose the system’s current regulatory status. This disclosure is informational and does not substitute for the manuscript’s own scientific evaluation.
Editorial Evaluation
Reviewers and editors assess reporting completeness against the applicable guideline above as part of standard peer review. Incomplete reporting is addressed through revision rather than automatic rejection, except where omissions materially prevent evaluation of the study’s validity.