Novel GHSROC And MGHSROC Model Construction For Meta-analysis Of Diagnostic Test Accuracy Based On The Generalized Logistic Distribution

Background:In recent years,the evidence based medicine(EBM)has developed rapidly,and the result of EBM based on the meta-analysis was regards as the superior evidence to guide clinical practice.Nowadays,in the diagnostic field,the methodology attracted lot of attentions because that the meta-analysis of diagnostic test can not meet the application requirements.The HSROC(hierarchical summary receiver-operator curves)model introduced by Rutter&Gatsonis and the bivariate random-effects model introduced by Reitsma were applied widely.The HSROC model took the influence of different cutpoint parameters(positive criteria)into consideration,which was the primary meta-analysis method when involved to the cutpoint effect.The HSROC model has a latent assumption that the sensitivity and specificity has a liner relationship when they were transformed by logit function.This assumption was established when the diagnostic result of the patients and control cohorts was follow logistic distribution.The logistic distribution is symmetrical,when the data was not symmetrical,it can not meet the assumption of HSROC.However,most of data in the medical field are skew despite transformed by some functions,so it is necessary that develop a model that can be used for symmetrical and asymmetrical data.Moreover,HSROC which is based on a pair of sensitivity and specificity provided by each of study has a disadvantage that it is difficult to examine the linear relationship because concentration about 0.95 of sensitivity and specificity in many studies.It can be found that many studies provided ROC in most of studies which diagnostic value are continuous,so if we can make full use of information about ROC,the SROC estimated by model will be more precise.Objective:?For the data that each of study provided a pair of sensitivity and specificity,construct GHSROC(Genalized HSROC)model which can be used for symmetrical and asymmetrical data base on the generalized logistic distribution.?For the data that each of study provided ROC,construct MGHSROC(Multi-points data based Genalized HSROC)model which can be used for symmetrical and asymmetrical data base on the multi-points data in the ROC.Method:?Assuming diagnostic value fellow generalized logistic distribution,we can develop a linear relationship between sensitivity and specificity after power exponent and logit transformation.Then we can develop GHSROC based on this linear relation and HSROC construction method.Simulation was done in order to examine whether HSROC can only used for symmetricaldata and whether GHSROC are more superior to HSROC when data are asymmetrical.Number of study and the skewness of data were considered when doing simulation.?We extract nine pairs of sensitivity and specificity in the ROC,in which specificity is 0.1-0.9 interval of 0.1.Regarding nine pairs of sensitivity and specificity as the cumulative probability corresponding to nine order and regarding intercept which are in the ordinal logistic regression as cutpoint,we can develop MGHSROC Model base on the generalized logistic distribution.Specifically,the MGHSROC which did not take shape parameter into consideration is called simple model,and which take shape parater into account is called complex model.Simulation was done in order to examine whether simple model can only used for symmetrical data and whether complex model are more superior to simple model when data are asymmetrical.Number of study and the skewness of data are considered when doing simulation.Result:?HSROC are superior to GHSROC when the data are symmetrical and number of study is less than 10.And HSROC are close to GHSROC when the data are skew and number of study is less than 10.In addition that GHSROC are superior to HSROC when the data are skew and number of study is large than 10.?Simple model are qual to complex model when the data are symmetrical.And complex model are superior to simple model when the data are skew.Discussion:?For the data that each of study provided a pair of sensitivity and specificity,our study constructed GHSROC model and we can conclude that HSROC is suggested when number of study is less than 10 and GHSROC is recommend when number of study is large than 10.?For the data that each of study provided ROC,our study constructed MGHSROC model and we can conclude that complex model is recommended whatever the data are asymmetrical or symmetrical.