A real-time prediction model for post-irradiation malignant cervical lymph nodes.

OBJECTIVE: To establish a real-time predictive scoring model based on sonographic characteristics for identifying malignant cervical lymph nodes (LNs) in cancer patients after neck irradiation.

METHODS: One-hundred forty-four irradiation-treated patients underwent ultrasonography and ultrasound-guided fine-needle aspirations (USgFNAs), and the resultant data were used to construct a real-time and computerised predictive scoring model. This scoring system was further compared with our previously proposed prediction model.

RESULTS: A predictive scoring model, 1.35 × (L axis) + 2.03 × (S axis) + 2.27 × (margin) + 1.48 × (echogenic hilum) + 3.7, was generated by stepwise multivariate logistic regression analysis. Neck LNs were considered to be malignant when the score was ≥ 7, corresponding to a sensitivity of 85.5%, specificity of 79.4%, positive predictive value (PPV) of 82.3%, negative predictive value (NPV) of 83.1%, and overall accuracy of 82.6%. When this new model and the original model were compared, the areas under the receiver operating characteristic curve (c-statistic) were 0.89 and 0.81, respectively (P < .05).

CONCLUSIONS: A real-time sonographic predictive scoring model was constructed to provide prompt and reliable guidance for USgFNA biopsies to manage cervical LNs after neck irradiation.