Seventeen trials reported non-inferiority or superiority for an AI-based strategy. Across the review, 84% of trials were rated as having a high risk of bias or some concerns.
On October 12, 2026, European Heart Journal – Digital Health published an accepted manuscript reviewing 25 randomised trials of AI in cardiovascular medicine. The trials compared AI tools with standard care. October 12 is the review’s publication date, rather than the date of the underlying trials. The literature search ended on April 10, 2026.
According to the public abstract, 17 trials, or 68%, reported non-inferiority or superiority for the AI-based strategy. Non-inferiority means the strategy was not worse than its comparator by more than a prespecified margin for the measured outcome. This count does not establish that all 17 trials improved patients’ health.
The 84% figure combines trials assessed as having a high risk of bias with those rated as having some concerns. Also, 36% explicitly reported following CONSORT-/SPIRIT-AI guidelines. That measures reported adherence; it does not establish that the remaining trials breached the guidelines.
The authors describe the evidence as limited and call for robust international, multicentre trials that assess patient-relevant outcomes and whether findings apply in other settings. The version checked on October 12 was an accepted manuscript before final copyediting and typesetting; the final version may contain changes.
Questions for a small clinic
Our editorial takeaway for a clinic considering an AI tool is to ask which evidence supports the proposed use. Four questions can make that discussion more specific:
- How closely do the trial’s participants, setting and task match your clinic?
- What was the comparator, and does it reflect your current process?
- Which outcome was measured, and did success mean non-inferiority or superiority?
- How was the risk of bias assessed, and which limitations affect the conclusion?
These questions are an editorial application of the review, rather than a recommendation to buy or reject a particular product. If a supplier cites this review, ask them to identify the relevant trial and explain how it relates to the version of the tool being offered. That makes a headline figure something you can examine against a specific proposal.
This report draws on the public abstract and publisher information. We have not checked the full methods, risk-of-bias tables or individual trial results.
Prepared by the NADALI editorial team using AI
Sources checked on 12 October 2026. The review’s accepted manuscript was published that day; the literature search ended on 10 April 2026.
