Multiple Test Procedures Of Disease Prevalence Based On Partially Validated Series In The Presence Of A Gold Standard

Estimation of disease prevalence is an important research issue in biomedical statistical research.In order to obtain the prevalence of a disease,an individual needs to be diagnosed.Screening methods are inexpensive but often misclassified,while gold standards do not misclassify but are often costly,time-consuming to diagnose or have side effects on individuals.Therefore,the double sampling method,in which some individuals are randomly selected from the population of interest and diagnosed by screening(or misjudgment test),and then some individuals are randomly selected to receive the gold standard test,can compromise the shortcomings of the screening method and the gold standard.Data obtained by the double sampling method is also known as partial verification data.Since the prevalence of disease may be affected by factors such as gender,age and living environment,statistical inference of disease prevalence can avoid the confounding effects of these factors by considering partial verification data in a hierarchical design.From the perspective of multiple test,this paper will study the statistical inference of whether there are significant differences in disease prevalence in stratified designs.Based on partial verification data in a hierarchical design,from the perspective of multiple test,Bonferroni test procedure,Single-step adjusted Max T procedure and single-step adjusted Min P procedure based on two types of Wald type test statistics,inverse hyperbolic tangent transformation test statistic,score test statistic and likelihood ratio test statistic are proposed.Monte Carlo simulation studies are conducted to evaluate the performance of these test procedure in various empirical type I error rates and empirical power between different layers.The simulation results show that the single-step adjusted Max T procedure performs best when the sample is large,and the test procedure of all test statistics between different layers can be controlled within a given significance level in both large and small samples.The second is the single-step adjusted Min P procedure,and the test results are well controlled within the significance level.When the sample size is small,Bonferroni procedure based on likelihood ratio and Score also has good statistical properties,so it is recommended to be used.Finally,two real data analyses are carried out to further verify the validity of the proposed method.