Different Statistical Models Applied To The Researchs In Public Health

Obiective:Used different statistical models to fit the prevalence data of Tuberculosis(TB), to analysis the factors which affect the prevalence of TB, in order to provide clues and evidences for preventing TB; Analyzed and compared the fitting of different regression models and their effects. Through this study, providing methods and clues to analyze the similar characteristics of clustering data in relevant researchs in the field of public health.Methods:This study used findings and results of the prevention and treatment of AIDS, viral hepatitis and TB and other major infectious diseases large-scale field epidemiology and intervention research projects. Collected and collated the specific data of13prevention demonstration areas of TB in Zhejiang province during2009-2012. The data contented factors like region, gender, age and the prevalence of TB. Depending on the conditions applicable to different types of statistical models, the data were collated and recoded before model fitting. Using different statistical models to fit the data. Regression models were mainly involved:Logistic regression model, Probit regression model, Poisson regression model, negative binomial regression model, multilevel Poisson regression model, overdispersion adjusted multilevel Poisson regression model and multilevel negative binomial regression model, by fitting different models, analyzed and understood the influences of region, gender and age factors to the prevalence of TB. Compared the fittings of different statistical models at the same time. In this study, the collating and recoding of data were processed by Excel2007and SAS9.2, models fitting and parameters estimation were mainly dealt by SAS9.2statistical analysis software.Results:The results of all seven statistical models showed that, region, gender and age factors had impacts on the prevalence of TB. Logistic regression, Probit regression, Poisson regression and negative binomial regression models suggested that, compared with district A, region factors C and B were risk factors, OR value was more than2. The differences of TB prevalence between population of male and female was statistically significant, OR value was more than2, then the relative risk of suffering from TB of the male population was more than twice of the female population. The rate of TB prevalence between population aged more than50and less than50was statistically significant, OR value was over2, the relative risk that people aged more than50suffering from TB was more than twice of the people aged less than50.The former6models dealt with the discrete case of data partly, but the results showed that the minimum Pearson chi-square/degrees of freedom value of the6models was4.44, which was much larger than1, and indicated that there existing discrete phenomenon in the data, we must deal with the problem of discrete when analyze the data. The value of Pearson chi-square/degrees of freedom was1.02when fitted the data with multilevel negative binomial model, which closed to the ideal value of1, solved the problem of over-discreted residuals. By comparing between different models, multilevel models, especially the multilevel negative binomial regression model was better than single-level models when fitting present research data. Conclusions:Statistical analysis show that region, gender and age factors affecting the prevalence of TB, then we could develop appropriate measures and policies to improve the efficiency of TB prevention and controling; By compare the results of different statistical models, suggesting that multilevel models are better tools when deal the data with characteristic of aggregation or clustering. Which have better model fitting, and could deal with data which has discrete problems properly.