JMIR Publications

ABSTRACT

Background:

Numerous models have been developed to predict the overall survival in postoperative patients with ESCC, but only few have focused on predicting postoperative recurrence.

Objective:

We aimed to develop a support vector machine (SVM) predictive model and assess its significance in evaluating recurrence rates and related factors in patients with ESCC.

Methods:

Clinical data from 311 patients with Esophageal Squamous Cell Carcinoma (ESCC) who underwent surgery at Jinling Hospital between June 2014 and November 2016 were collected, and the patients were followed up until October 2021 (range, zero to 93.5 months). Data were analyzed using SPSS (version 22.0) and R (version 3.6.1).

Results:

In the test, validation 1, and validation 2 groups, the sensitivity of SVM6 in predicting recurrence was lower than that of SVM7 (P<0.001). The sensitivities of SVM6, SVM7, and SVM8 for predicting recurrence in patients with ESCC were 94.00%, 79.59%, and 72.73%, respectively, and the specificities were 98.11%, 69.84%, and 78.43%, respectively. These sensitivities were comparable with that of SVM6+TNM and higher than that of SVM6+TNM (P<0.001). Postoperative survival analysis showed that the disease-free survival (DFS) duration of the predicted low recurrence risk group in SVM6 and SVM6+TNM were much longer than those of the predicted high recurrence risk group, and a notable difference in the recurrence rates was observed(P<0.001).

Conclusions:

We used a nomogram to integrate the indices from the optimal SVM model into an artificial-intelligence model for patients with ESCC who had not yet had an individualized treatment plan developed. This approach allows for accurate prediction and evaluation of postoperative recurrence outcomes with high sensitivity, specificity, and accuracy.