Objective To investigate the application value of a joint prediction model based on lymph node imaging features in evaluating lymph node metastasis of locally advanced rectal cancer (LARC) patients after neoadjuvant chemoradiotherapy (nCRT).

Methods The retrospective cohort study was conducted. The clinicopathological data of 215 LARC patients who were admitted to Peking University Cancer Hospital & Institute from July 2010 to June 2015 were collected. There were 131 males and 84 females, aged (56.7±10.1)years. All 215 patients were randomly divided into a training set of 143 cases and a testing set of 72 cases using a 2∶1 ratio of random seed numbers. The training set was used to construct the prediction model, and the testing set was used to validate the performance of prediction model. Observation indicators: (1) lymph node metastasis in LARC patients after nCRT; (2) imaging feature selection and model construction and evaluation. Com-parison of measurement data with normal distribution between groups was conducted using the independent sample t test. Comparison of measurement data with skewed distribution between groups was conducted using the Mann-Whitney U test. Comparison of count data between groups was conducted using the chi‑square test. Univariate and multivariate analyses were conducted using the Logistic regression model. Performance evaluation of prediction model was conducted using the receiver operating characteristic (ROC) curve. The area under the curve (AUC), accuracy, sensitivity, and specificity were calculated. Calibration curves and decision curves were used to evaluate the consistency and clinical application value of the prediction model.

Results (1) Lymph node metastasis in LARC patients after nCRT. Of the 215 LARC patients after nCRT, results of postoperative pathological examination showed that there were 162 cases with negative lymph node metastasis and 53 cases with positive lymph node metastasis, showing significant differences in age and maximum short‑axis diameter of lymph node between them (t=2.178, Z=-5.305, P<0.05). (2) Imaging feature selection and model construction and evaluation. Forty‑one imaging features were extracted from the 215 LARC patients after nCRT, including 9 gray-level first‑order features, 24 gray‑level co‑occurrence matrix features and 8 shape features. The score of lymph node (LNscore) in 162 cases with negative lymph node metastasis and 53 cases with positive lymph node metastasis were 0.18(0.10,0.33) and 0.39(0.23,0.54), respectively, showing a significant difference between them (Z=-5.487, P<0.05). Results of multivariate analysis showed that maximum short‑axis diameter of lymph node and LNscore were independent factors influencing lymph node metastasis of LARC patients after nCRT (odds ratio=1.277, 25.514, 95% confidence interval as 1.010-1.614, 2.003-324.964, P<0.05). A Logistic regression joint prediction model was constructed by incorporating the maximum short-axis diameter of lymph node and LNscore. The ROC curves results showed that the AUC, accuracy, sensitivity, and specificity of the joint prediction model in the training set were 0.779 (95% confidence interval as 0.702-0.844), 72.7%, 71.4%, and 73.2%, respectively. The above indicators in the testing set were 0.805 (95% confidence interval as 0.694-0.889), 80.6%, 66.7%, and 85.2%, respectively. Calibration curves in both training set and test set showed good agreement with the ideal curve, indicating high calibration. Decision curves demonstrated the model′s clinical utility with a high net benefit.

Conclusion The maximum short‑axis diameter of lymph node and LNscore are independent factors influencing lymph node metastasis of LARC patients after nCRT. The joint prediction model constructed based on the above indicators can be used to predict lymph node metastasis in LARC patients after nCRT.