Development of a machine-learning model for therapeutic efficacy prediction of preoperative treatment for esophageal cancer using single nucleotide variants of autophagy-related genes

Abstract

AbstractNeoadjuvant chemotherapy with cisplatin + 5-fluorouracil followed by radical surgery is the standard treatment for stage II and III esophageal cancers. Although, a more potent regimen comprising cisplatin + 5-fluorouracil with docetaxel, has shown superiority in overall survival compared to the cisplatin + 5-fluorouracil regimen, it involves worsening of Grade 3 or higher adverse events due to docetaxel. Based on these reports, this study aimed to construct a prognostic system for cisplatin + 5-fluorouracil regimens, particularly for locally advanced cancers, to guide selection of neoadjuvant chemotherapy. Biopsy specimens from 82 patients who underwent a cisplatin + 5-fluorouracil regimen plus radical surgery at Saitama Medical University International Medical Center between May 2012 and June 2020 were analyzed. Variants in 56 autophagy- and esophageal cancer-related genes were identified using targeted enrichment sequencing. Overall, 13 single nucleotide variants, including eight non-synonymous group single nucleotide variants predicting recurrence were identified using Fisher’s exact test with recurrence as a two-group event, which showed a significant difference (p < 0.05). Additionally, machine learning was used to predict recurrence using 21 features, including eight patient backgrounds. The results showed that the Naive Bayes was highly reliable with an accuracy of 0.88 and Area Under the Curve of 0.9. Thus, we constructed a machine learning model to predict recurrence in patients with esophageal cancer treated with a cisplatin + 5-fluorouracil regimen. We believe that our results will provide useful guidance for the selection of neoadjuvant adjuvant chemotherapy, including the avoidance of docetaxel.