Abstract: This paper discusses the risk factors related to gallbladder disease in Shanghai, improves the accuracy of risk prediction, and provides a theoretical basis for scientific diagnosis and universality of gallbladder disease.We selected 3 462 data of middle-aged and elderly health check-up patients in a general hospital in Shanghai,and divided into gallbladder disease group according to color doppler ultrasound diagnosis results. Single-factor analysis screened out 8 important risk factors, which were used as an analysis variable of multi-layer perceptron neural network and binary logistic regression to construct the prediction model of gallbladder disease. The prediction accuracy of the multi-layer perceptron neural network risk prediction model is 76%. The area under the receiver operating characteristic curve (AUC) is 0.82, the maximum Youden index is 0.44, the sensitivity is 79.51, and the specificity is 64.23. The prediction accuracy of the multi-layer perceptron neural network model is better than that of the binary logistic regression prediction model. The overall prediction accuracy of the binary logistic regression prediction model is 75.60%, the AUC is 0.81, the maximum Youden index is 0.42, the sensitivity is 74.48, and the specificity is 57.60. In the objective risk prediction of gallbladder disease in middle-aged and elderly people in Shanghai, the risk prediction model based on the multi-layer perceptron neural network has a better prediction performance than the binary logistic regression model, which provides a theoretical basis for preventive treatment and intervention.

Key words: neural networks, gallbladder disease, risk prediction model