Tongue image information for prediction esophageal precancerous lesion: A multi-centric logistic regression analysis of esophageal cancer high risk population in China.

Abstract

e15579 Background: More than 50% esophageal cancer cases of the world are in China. The main causative factors that contribute to extremely high incidence of esophageal cancer (9.15/ million) in certain rural areas of China has not yet been elucidated. Screening and diagnosis of precancerous lesions and early stage cancer are main neasures of decreasing incidence rate and mortality rate. Methods: High-risk population (40-69 years old) in high risk areas of esophageal cancer, Cixian and Yangcheng were screened, and 4742 cases were included. They were divided into 4 groups: normal group, esophagitis/gastritis, esophageal neoplasia low-level group and esophageal neoplasia high-level group, according to pathology and electronic gastroscope diagnosis. Diagnostic testing lingual information collection system (DS01-B) was used and tongue image were collected, including tongue color, coating color, fur character, tongue shape, local ecchymosis, et al. Difference of tongue image information was analyzed by multi-factor logistic regression using SPSS and R statistical software. Results: Logistic regression analysis showed that dark purplish tongue colour, and tongue local ecchymosis were correlated to precancerous lesions and early stage cancer (P < 0.01 or P < 0.05). Red and crimson tongue, yellow and white coated tongue, spotted tongue and tongue local ecchymosis were correlated to esophagitis and gastritis(P < 0.01 or P < 0.05). Logistic regression model had diagnostic total coincidence rate (69.4%) for early esophageal cancer screening. Conclusions: Multi-factor logistic regression model based on tongue image information has clinical value of prediction esophageal precancerous lesion.