A machine learning-based transcriptomic signature for predicting tumor recurrence after curative resection in T1 colorectal cancer: a retrospective multicenter cohort study (The Tw1CE trial).
코호트
1/5 보강
PICO 자동 추출 (휴리스틱, conf 2/4)
유사 논문P · Population 대상 환자/모집단
138 patients with T1 CRC (2023-2025; ClinicalTrials.
I · Intervention 중재 / 시술
추출되지 않음
C · Comparison 대조 / 비교
추출되지 않음
O · Outcome 결과 / 결론
[CONCLUSION] This study established a validated machine learning-based transcriptomic classifier derived from endoscopic resection specimens that accurately predicts tumor recurrence in patients with T1 CRC. Our findings highlight the potential of this biomarker panel to enable risk-adapted surveillance strategies and guide decisions regarding additional therapy after curative resection.
[BACKGROUND] T1 colorectal cancer (T1 CRC) is increasingly treated with curative-intent endoscopic resection, but tumor recurrence remains a critical factor influencing patient prognosis.
- p-value P < 0.001
- 연구 설계 cohort study
APA
Noma T, Saez de Gordoa K, et al. (2026). A machine learning-based transcriptomic signature for predicting tumor recurrence after curative resection in T1 colorectal cancer: a retrospective multicenter cohort study (The Tw1CE trial).. International journal of surgery (London, England), 112(4), 9039-51. https://doi.org/10.1097/JS9.0000000000004690
MLA
Noma T, et al.. "A machine learning-based transcriptomic signature for predicting tumor recurrence after curative resection in T1 colorectal cancer: a retrospective multicenter cohort study (The Tw1CE trial).." International journal of surgery (London, England), vol. 112, no. 4, 2026, pp. 9039-51.
PMID
41604539
Abstract
[BACKGROUND] T1 colorectal cancer (T1 CRC) is increasingly treated with curative-intent endoscopic resection, but tumor recurrence remains a critical factor influencing patient prognosis. However there is no validated biomarker exists to reliably predict post-resection recurrence, limiting risk-adapted follow-up and adjuvant therapy decisions.
[MATERIALS AND METHODS] In this multicenter retrospective cohort study across academic centers in Spain, 138 patients with T1 CRC (2023-2025; ClinicalTrials.gov NCT06314971) were enrolled. From FFPE endoscopic specimens, expression of five mRNAs and two miRNAs was quantified by RT-qPCR, and an XGBoost-based transcriptomic panel was developed. Patients were assigned to training and independent testing cohorts by treatment type. The primary outcome was 3-year recurrence-free survival (RFS); secondary outcomes included 5-year RFS and overall survival (OS).
[RESULTS] The transcriptomic panel demonstrated high predictive performance in both the training (AUROC = 91.7%) and testing (AUROC = 88.2%) cohorts. Patients classified as high-risk by the panel exhibited significantly worse RFS and OS compared with those classified as low-risk (log-rank P < 0.001). Furthermore, integrating lymphatic invasion with the transcriptomic panel into a combined risk stratification model further improved predictive accuracy (AUROC = 94.6%), and decision curve analysis confirmed its superior clinical utility compared to conventional criteria.
[CONCLUSION] This study established a validated machine learning-based transcriptomic classifier derived from endoscopic resection specimens that accurately predicts tumor recurrence in patients with T1 CRC. Our findings highlight the potential of this biomarker panel to enable risk-adapted surveillance strategies and guide decisions regarding additional therapy after curative resection.
[MATERIALS AND METHODS] In this multicenter retrospective cohort study across academic centers in Spain, 138 patients with T1 CRC (2023-2025; ClinicalTrials.gov NCT06314971) were enrolled. From FFPE endoscopic specimens, expression of five mRNAs and two miRNAs was quantified by RT-qPCR, and an XGBoost-based transcriptomic panel was developed. Patients were assigned to training and independent testing cohorts by treatment type. The primary outcome was 3-year recurrence-free survival (RFS); secondary outcomes included 5-year RFS and overall survival (OS).
[RESULTS] The transcriptomic panel demonstrated high predictive performance in both the training (AUROC = 91.7%) and testing (AUROC = 88.2%) cohorts. Patients classified as high-risk by the panel exhibited significantly worse RFS and OS compared with those classified as low-risk (log-rank P < 0.001). Furthermore, integrating lymphatic invasion with the transcriptomic panel into a combined risk stratification model further improved predictive accuracy (AUROC = 94.6%), and decision curve analysis confirmed its superior clinical utility compared to conventional criteria.
[CONCLUSION] This study established a validated machine learning-based transcriptomic classifier derived from endoscopic resection specimens that accurately predicts tumor recurrence in patients with T1 CRC. Our findings highlight the potential of this biomarker panel to enable risk-adapted surveillance strategies and guide decisions regarding additional therapy after curative resection.