A multicenter retrospective case-control study across 4 centers (China, Japan, South Korea) developed and externally validated a 7-exosomal miRNA (exo-miRNA) machine learning model to preoperatively detect occult early liver metastasis (early-LiM, defined as liver recurrence within 6 months of curative resection) in 372 PDAC patients enrolled from 2011–2024.
The 7-exo-miRNA XGBoost model achieved AUC 0.899 (95% CI, 0.822–0.976) in training and AUCs of 0.876 and 0.862 in two independent external cohorts. Early-LiM was associated with dramatically worse median OS (9.1 months vs. 26.6–31.8 months for other recurrence patterns; log-rank P < .001), and the panel score was an independent predictor of early-LiM (OR 26.49; 95% CI, 18.45–55.28; P < .001).
- Retrospective case-control design with inherent selection bias; prospective validation is still needed. - All centers are in East Asia (China, Japan, South Korea), limiting generalizability to Western populations. - Early-LiM definition (recurrence within 6 months) is threshold-based and may misclassify borderline cases.
A preoperative blood-based exo-miRNA panel may flag PDAC patients at high risk for occult liver micrometastasis before pancreatectomy, potentially guiding decisions about neoadjuvant therapy versus upfront surgery. Prospective trials are needed before clinical adoption.
Explore related topics
