Urology

Systematic review: AI prediction models for endourological stone procedures lack external validation

This systematic review identified 25 studies of AI-labelled prediction models for outcomes after ureteroscopy, RIRS, or PCNL. Most studies reported discrimination but rarely reported calibration or formal clinical-utility assessment. Validation was predominantly internal; external or temporal validation and standardised outcomes were limited. The authors conclude these models show promise but are not ready for routine clinical decision support without further validation. 1

Randomized trial: single-use anal insert reduced FI episodes but missed primary St Mark's endpoint

In this multicenter RCT (n=72), the insert did not significantly increase the proportion achieving ≥3-point St Mark's reduction versus usual care. The device reduced weekly FI episodes by about 3.09 episodes and increased the proportion with >50% episode reduction. Selected FI-specific quality-of-life domains (coping and depression) improved with the insert, while others did not. No serious adverse events occurred; device-related events were common but mild. 2

Randomized trial: audiovisual augmentation of virtual consent did not increase consent or comprehension

In a four-arm randomized substudy (n=1535), adding physician, patient, or animated videos to text did not meaningfully change consent rates. Overall consent rate was 57.9%, and consent comprehension was high (about 86%) across all arms. Relative risks for consent versus text-only were close to 1 for all video modalities, indicating no clear benefit. Findings apply to virtual research consent in one academic health system and may not generalize to other settings. 3