Research On Statistical Inference Method And Its Application In Disease Data Analysis

With the arrival of big data era,more and more subjects are being combined with statistics.In recent years,public health has become a hot research field of statistics.Under the background of advocating"smart medical",it is a hotspot that using statistical inference method to solve the problem of disease data analysis in public health.Thyroid nodule is one of the most common diseases in public health.To discriminate benign from malignant ones,a risk-score model was proposed.We used information value to screen variables,recoded variables with weight of evidence,and obtained risk score based on logistic regression model.This model was applied to a real ultrasound diagnostic dataset of thyroid nodule,and the accuracy rate was 96.5%.Our model could not only improve the accuracy of ultrasound diagnosis,but also realize the risk assessment of patients'condition.Foodborne disease caused by food safety is one of the maj or public health issues.Zhejiang Province is one of the areas seriously affected by foodborne disease.In the paper,spatial and temporal inference methods,such as correlation analysis,spatial autocorrelation analysis and time-space scanning analysis,were carried out to mine multi-dimensional features of foodbome disease.The results could provide data support for the prevention and control of foodborne disease.Aiming at dealing with the outbreak warning problem of foodborne disease,we put forward a series of BHM-ARxy models based on B ayesian hierarchical model theory.In our model,we first used mixed Gaussian distributions nested to Poisson distribution to fit data distribution;then we used the combination of 0,1,2 order autoregressive models to model the mean parameters under two different states;Markov chain was used to simulate the generation mechanism of state parameters;finally we set prior distributions for each submodel and used MCMC method to calculate the posterior predictive distribution of the outbreak probability.Through the evaluation of DIC criteria and AUC indicator,the results showed that BHM-ARO1 model had the best comprehensive performance.