Decoding Informed Consent: Evaluating AI Disclosures in Clinical Research

Artificial intelligence is rapidly shifting from experimental code to point-of-care medical decision-making. As machine learning models gain autonomy in diagnosing diseases and tailoring treatments, ensuring true patient autonomy through informed consent has become a vital ethical challenge.

In a cross-sectional study published in the Journal of Medical Internet Research, researcher Hankun Su and colleagues at Xiangya Hospital Central South University evaluated the transparency, readability, and data governance of informed consent documentation across 114 AI-involved clinical trials registered on ClinicalTrials.gov. Their findings reveal a significant gap between ethical principles and real-world consent practices.

Critical Gaps in Transparency and Readability

While public surveys show that over 80% of individuals believe they should be explicitly notified whenever AI is used in their care, current consent documentation often keeps participants in the dark:

• Omission of AI Mechanisms: Over half (58%, 66/114) of informed consent documents failed to disclose the type or intended role of the AI system, leaving participants unaware of the technology driving their care.
• Incomplete Risk Disclosure: While 88.6% of forms outlined potential benefits, 18.4% (21/114) omitted risks entirely, creating a risk of therapeutic misconception.
• Dense and Complex Language: Only 14% (16/114) of documents met basic criteria for brevity (fewer than 15,000 characters) and readability (Simple Measure of Gobbledygook score under 13). Forms for higher-risk trials were no easier to comprehend than those for low-risk studies.
• Inconsistent Post-Withdrawal Data Handling: When participants withdraw from a trial, protocols for handling their data varied widely: 44.7% provided no clear policy, 26.3% specified data destruction, 25.4% allowed continued data use, and only 3.5% offered participants a choice.

A Framework for Minimum Requirements in AI Consent

To address these deficits, Su and colleagues proposed the Minimum Requirements for Informed Consent in AI-Related Clinical Trials (MRIC-AI) checklist to help researchers and institutional review boards upgrade consent quality:

1. AI-Specific Disclosure: Clearly state the AI type (e.g., deep learning vs. machine learning), its specific clinical role, and its degree of autonomy.
2. Balanced Risk-Benefit Communication: Present algorithmic risks (e.g., bias, re-identification, misdiagnosis) with equal prominence to potential clinical benefits.
3. Participant-Centered Readability: Target a reading level of 10th grade or lower for high-risk trials, keep length under 15,000 characters, and incorporate tested visual aids.
4. Transparent Data Governance: Clearly outline data storage, retention, applicable privacy laws, and explicit participant choices regarding post-withdrawal data use.
5. Dynamic Consent: Establish protocols to provide accessible updates if an AI model evolves during the trial.

The authors selected the Journal of Medical Internet Research to present this work due to its focus on digital health ethics, AI governance, and clinical trial transparency. As AI systems become deeply embedded in human research, this study provides a crucial roadmap for ensuring that patient consent remains meaningful, transparent, and ethically robust.

Curious about how AI transparency frameworks are reshaping patient rights in clinical research? Watch the video featuring Hankun Su and read the full study to explore the MRIC-AI checklist, expert validation results, and regulatory analysis.