Machine learning prediction model for lateral lymph node metastasis in rectal cancer.

Abstract

Background: The preoperative diagnosis of lateral lymph node metastasis presents a significant challenge. In this study, we aimed to predict the pathological characteristics of lateral lymph nodes in patients with rectal cancer using preoperative clinical information and to develop a logistic prediction model for lateral lymph node metastasis.

Methods: A retrospective analysis of 143 patients who underwent total mesorectal excision (TME) and lateral lymph node dissection (LLND) at Tianjin Union Medical Center, from January 2017 to June 2024 was conducted. Patients were categorized into lateral lymph node metastasis and non-metastasis groups based on postoperative pathological findings. Basic information, tumor markers, and MRI reports were compared. Patients were segmented into training and validation sets at an 8:2 ratio. The R software was used to create a logistic prediction model and a nomogram.

Results: This study included 66 pathologically positive and 77 pathologically negative lateral lymph node cases. Extramural vascular invasion (EMVI), MRI clinical N stage (MRI cN stage), and the number of enlarged lateral lymph nodes (NoELLN) were used to construct the logistic prediction model. The model achieved an accuracy of 0.62, sensitivity of 0.80, specificity of 0.43, and area under the curve (AUC) of 0.80 in predicting the pathological characteristics of lateral lymph nodes using the test dataset.

Conclusion: EMVI, MRI cN stage, and NoELLN are significant predictive factors for predicting lateral lymph node pathology in patients with rectal cancer. These findings offer guidance for determining patient eligibility for LLND surgery.