Nomogram vs. Depth of invasion for predicting occult lymph node metastasis in cT1-2N0 buccal squamous cell carcinoma.

OBJECTIVE: To develop a nomogram prediction model for occult lymph node metastasis (LNM) in patients with cT1-2N0 buccal squamous cell carcinoma (BSCC), then to compare its predictive efficacy against depth of invasion (DOI). METHODS: Clinical data were retrieved for patients undergoing primary tumor resection and neck dissection from June 2020 to August 2024. Based on the risk factors screened by Lasso regression, we established four candidate models: logistic regression, random forest, support vector machine, and XGboost. The optimal model was determined by comparing the values of areas under the receiver-operating characteristic curve (AUC), then the nomogram was ultimately plotted accordingly to visualize the results. RESULTS: Two hundred and fifty patients were enrolled. The screened variables include Ki-67, tumor differentiation grade, surgical margin status, perineural invasion, DOI, and smoking. With similar good performance from both the training and test cohorts (AUC, 0.726 vs. 0.782) and good calibration, the logistic regression model performed the best overall, and was thus selected for creating a nomogram. The nomogram was superior to DOI cut-off values of 3 mm and 4 mm in predicting occult LNM, with a higher AUC (0.741 vs. 0.543 and 0.595) and more net benefits. Compared with DOI < 4 mm, at a 9.51 % risk of LNM, the nomogram identified an equivalent number of cases (n = 64) for not undergoing elective neck dissection (END), while successfully reducing 2 false-negative cases (2 vs. 4) with insufficient treatment. CONCLUSIONS: The nomogram described here prevails over DOI in predicting occult LNM in early-stage BSCC, and provide effective guidance for END.