Development and validation of a nomogram to predict occult cervical metastasis in early oral squamous cell carcinoma.

Background: Lack of adequate objectivity and universality, available models are still difficult to be applied to clinical practice in predicting occult cervical metastasis of early oral squamous cell carcinoma (OSCC). Taking abnormal metabolic state into consideration, the current model is helpful to distinguish those patients with or without occult cervical metastasis.

Methods: This study retrospectively analyzed 330 OSCC patients initially diagnosed cT1-2N0M0 stage and received neck dissection from January 2020 to July 2022. The occult cervical metastasis was identified by pathological examination.. After screening independent risk factors using logistic regression, patients were divided into training and validation cohorts at the ratio of 2:1 randomly, and a novel diagnostic model was constructed. Performances of this model were evaluated by the area under the curve (AUC), calibrating curve, decision curve analysis (DCA) and clinical impact curve (CIC).

Results: Of the 330 included patients {age mean [standard deviation (SD)], 61.24 (12.99) years; 202 (61.2%) males}, 49 (14.8%) had occult nodal metastasis. Five variables, including body mass index (BMI) [high odds ratio (OR): 1.132; 95% confidence interval (CI): 1.019-1.258, P=0.021], primary tumor site (tongue & floor of mouth (TF) OR: 3.756; 95% CI: 1.295-10.898, P=0.015), depth of invasion (DOI) (5-10 mm OR: 2.973; 95% CI: 1.266-6.981; P=0.012), pathological differentiation (Poor differentiation OR: 2.65; 95% CI: 1.341-5.239; P=0.005), and diabetes (OR: 3.123; 95% CI: 1.23-7.929; P=0.017) were screened to establish the predictive model. In training cohort (n=220), this model achieved an AUC of 0.814 and had a sensitivity of 78.1% and specificity of 70.2%. Calibration plots showed favorable consistency between the prediction of the model and actual observations (Hosmer-Lemeshow value >0.05). Decision curve analysis (DCA) and clinical impact curve (CIC) showed the model was clinically useful and had better discriminative ability under the threshold probability of 0.5. Above evaluations were verified in the validation cohort (n=110). Compared to previous reported models, the concordance index (C-index), net reclassification index (NRI), and integrated discrimination improvement (IDI) values were superior in both training and validation cohorts (P<0.05).

Conclusions: This constructed model might have reference value for clinicians in making neck management decisions of early OSCC patients.