The early detection of second primary esophageal squamous cell carcinoma (SPESC) is crucial to preventing death in head and neck cancer (HNC) survivors. The study aims to establish a reliable prediction model for second primary esophageal squamous cancer after HNC, which has not been well established. The retrospective study included 2798 patients who primary head and neck cancer without second primary malignancies (Non-SPESC) and 65 patients with second primary malignancies in the esophagus (SPESC). In addition, clinical risk characteristics were individually tested and compared with seven machine learning models. As a result, CART (classification and regression trees) provided the highest AUC (the area under the curve) of 0.9240, as well as the best of accuracy (0.8678), recall rate (0.8573), F1 score (0.8871), Matthews correlation coefficient (0.7305) and lowest False Positive Rate (0.1990). The clinically significant risk factors for SPESC were Age < 65 years (OR=2.96, 95%, CI=1.07- 8.19, p), Grade/Differentiation > 2 (OR=2.54, 95%, CI=1.44-4.46), Smoking Behavior (OR=3.20, 95%, CI=1.16-8.84), Drinking Behavior (OR=5.92, 95%, CI=2.37-14.78), and Measured Depth in Pathology Report (OR=1.64, 95%, CI=1.00-2.70). Furthermore, certain risk factors can be incorporated into prediction models for the development of clinical evidence-based decision rules. The results of this study can be used as a theoretical model for physicians to detect HNC survivors early, who have these clinical risk factors. Finally, these findings offer useful information for avoiding unnecessary screening during cancer follow-up.

Download