Colorectal Surgery

Systematic review: LLMs have varied utility across colorectal cancer care but evidence quality is limited

This systematic review of 37 studies (2023–2026) found LLMs useful for data extraction, patient education, diagnosis support, and decision aid in CRC care. Domain-specific and multimodal models outperformed general models in certain tasks, and prompt design strongly influenced performance. Only 27% of studies had low risk of bias; common concerns included outcome measurement, patient selection, and lack of blinded assessment. Authors conclude LLMs are promising but require robust privacy safeguards, prospective validation, and generalizability testing before clinical deployment. 1

Nationwide multimodal prehabilitation associated with fewer complications and shorter hospital stay

In 18 Dutch hospitals, 1192 prehabilitated patients were propensity-matched to 1192 historical controls undergoing elective CRC resection. Overall complications were lower with prehabilitation (30.1% vs 37.8%; OR 0.71; 95% CI 0.60–0.84). Medical complications and median length of stay were reduced (median 4 vs 5 days), with readmission and ICU admission rates also lower. Benefit was observed across age and ASA subgroups, supporting broad applicability in unselected elective CRC patients. 2

Biomarker Toolkit scores distinguish successful versus stalled diagnostic CRC biomarkers

The Biomarker Toolkit was validated using diagnostic CRC biomarker datasets and differentiated successful from stalled biomarkers. Successful biomarkers had higher total scores (44.7% vs 25.1%, p < 0.001) and higher clinical utility scores (42.1% vs 20.2%, p < 0.001). Higher clinical utility signal appeared even before regulatory approval, suggesting earlier identification of promising biomarkers is possible. The Toolkit may guide prioritization of biomarkers for further clinical development and resource allocation. 3