1427P - Predicting overall survival and prognostic indicator genes in esophagogastric cancer patients using machine learning and bioinformatics analysis

Background

Esophagogastric cancer presents challenges in prognosis and treatment. We aimed to utilize machine learning (ML) and bioinformatics analysis to predict overall survival and identify prognostic indicator genes in patients diagnosed with esophagogastric cancer.

Methods

This study utilized a database of 902 patients from the Memorial Sloan Kettering Cancer Center (egc_msk_2023). We focused on analyzing the dataset related to copy number alterations (CNAs), which refer to changes in the number of copies of specific DNA segments in cancer cells, and the clinical dataset included patients' diagnoses clinical symptoms, and other relevant information. ML models were developed to predict overall survival and identify the top 20 genes with the most significant prognostic value. Moreover, bioinformatics analyses were employed to shed light on the role of these genes in esophagogastric cancer prognosis.

Results

The ML models achieved the following accuracies in predicting patient survival status using CNAs data: Random Forest (0.60), Support Vector Machine (0.58), Logistic Regression (0.53), and Ensemble (0.53). To enhance the prognostic effectiveness of the ML models, we integrated the CNAs dataset with clinical data. The improved results were as follows: Random Forest (0.75), Support Vector Machine (0.68), Logistic Regression (0.71), and Ensemble (0.70). Additionally, we identified important clinical factors for overall survival prediction, including age, BMI, metastasis status, and stage. The top 20 genes found to be most influential in the diagnostic prognosis were MET, CCNE1, PTEN, RTEL1, PIK3CA, CDKN2A, CDKN2Ap16INK4A, CDK4, CCND3, MLH1, RB1, CDK12, CREBBP, FGF3, IGF1R, CDKN2Ap14ARF, CDC42, FGF19, and MDM2. Through bioinformatics analysis, we discovered the significant involvement of these genes in various cancer pathways, as well as the P53 signaling pathway, PI3K-Akt signaling pathway, and cellular senescence.

Conclusions

Our research holds the potential to contribute to improved patient prognosis and treatment approaches and enhanced clinical decision-making in the management of esophagogastric cancer.

Clinical trial identification

Editorial acknowledgement

Legal entity responsible for the study

S-H. Hung and N-K. Viet-Nhi.

Funding

Has not received any funding.

Disclosure

All authors have declared no conflicts of interest.