Serum levels of chemical elements in esophageal squamous cell carcinoma in Anyang, China: a case-control study based...

Esophageal squamous cell carcinoma (ESCC) is the predominant form of esophageal carcinoma with extremely aggressive nature and low survival rate. The risk factors for ESCC in the high-incidence areas of China remain unclear. We used machine learning methods to investigate whether there was an association between the alterations of serum levels of certain chemical elements and ESCC.

Primary healthcare unit in Anyang city, Henan Province of China.

100 patients with ESCC and 100 healthy controls matched for age, sex and region were included.

Primary outcome was the classification accuracy. Secondary outcome was the p Value of the t-test or rank-sum test.

Both traditional statistical methods of t-test and rank-sum test and fashionable machine learning approaches were employed.

Random Forest achieves the best accuracy of 98.38% on the original feature vectors (without dimensionality reduction), and support vector machine outperforms other classifiers by yielding accuracy of 96.56% on embedding spaces (with dimensionality reduction). All six classifiers can achieve accuracies more than 90% based on the single most important element Sr. The other two elements with distinctive difference are S and P, providing accuracies around 80%. More than half of chemical elements were found to be significantly different between patients with ESCC and the controls.

These results suggest clear differences between patients with ESCC and controls, implying some potential promising applications in diagnosis, prognosis, pharmacy and nutrition of ESCC. However, the results should be interpreted with caution due to the retrospective design nature, limited sample size and the lack of several potential confounding factors (including obesity, nutritional status, and fruit and vegetable consumption and potential regional carcinogen contacts).