Internal Medicine

Most medical AI studies are preclinical; RCT evidence is scarce and concentrated geographically.

A scoping review of 218 systematic reviews identified 4667 primary AI studies, of which 88.2% were preclinical and 2.4% were RCTs. RCTs were geographically concentrated, with the top 10 countries contributing 75.5% and the US and China contributing 47.5%. Among RCTs, 67.3% were single-center and 71.7% did not report adherence to reporting guidelines. Authors report 82.3% of RCTs had favorable outcomes but note frequent methodological concerns in allocation concealment and blinding. 1

EMR-integrated risk assessment identified additional patients meeting guidelines for genetic counseling.

In a single-arm EMR-integrated trial, 1685 consented patients were offered a 24-syndrome family health history assessment and 636 completed it. Of completers, 155 (28.5%) met guideline criteria for genetic counseling, and only 31 attended counseling. About half (47.7%) of those meeting criteria had previously been identified by billing codes, showing added case-finding value. The intervention increased identification of at-risk adults in routine primary care but produced modest downstream counseling uptake. 2

Cancer drug trials in China commonly use restrictive eligibility beyond NCI guidance.

Reviewing 2448 registered cancer drug trials (2013–2021) found 97.0% applied at least one eligibility criterion considered restrictive versus NCI guidance. Median restrictive criteria per trial rose from 2 (IQR 1–3) in 2013–2015 to 3 (IQR 2–4) by 2019–2021. Use of restrictive rules increased over time for performance status, cardiac function, and prior/concurrent malignancies. Authors conclude narrower eligibility likely reduces generalizability and call for more inclusive, modernized criteria. 3