Research On The Related Problems Of Sample Size Determination In Logistic Regression Model

The determination of sample size is the basic work for the implementation of statistical experiment such as sampling survey or medical clinical practice.Reasonable sample size can not only ensure the reliability and accuracy of the survey or test results,but also control the cost and cost of the survey or test,which plays a particularly important role in ensuring the credibility of the test and resource saving.Logistic regression model is widely used in clinical medicine and drug research and development.In this paper,for the problem of sample size calculation of the model,combined with a variety of different information,the corresponding theory of sample size calculation is given,and the corresponding formula is derived.Based on the brief introduction of empirical method,the applicable scenarios of formula method are discussed in detail.First of all,when the independent variable in the logistic regression model is binary type variable,the sample size of three different logistic regression models are calculated: single risk factor model,one exposure factor and one hybrid factor model,and two risk factor model with interaction.When the independent variable is a continuous variable,this paper briefly introduces the theory of Wald test,likelihood ratio test and score test in sample size calculation,and focuses on the specific derivation process of Wald test method of single factor model.We also discuss the sample size calculation method under the Bayesian framework,and compare the advantages and disadvantages of the frequency method and the Bayesian method.Moreover,the paper introduces the threshold test method,which includes the logical regression method,cross validation method,distance method,etc.through the construction of classifiers,and through a large number of data simulation to compare and analyze several methods.In particular,using the characteristics of logistic regression model,considering the calculation of sample size of logistic regression model based on logistic normal distribution,according to the characteristics of logistic normal distribution,using delta method and optimization method to convert it into normal distribution,and through simulation data analysis,comparing the sample size results calculated before and after the conversion.We get the following conclusion: under the same level of efficacy and confidence,the sample size of transformed logistic regression model is significantly better than that of untransform-ed model.