Evidence Based Evaluation Of Prediabetes Screening Tools

Research Purpose:This article evaluates the accuracy of various prediabetes screening tools and the relative accuracy of the tools through a systematic review,in order to find the best screening tool among many tools,and provide a basis for early intervention and treatment of prediabetes.Research Methods:First,search Cochrane Library,Pub Med,EMBASE,CNKI,Wanfang,VIP,Baidu scholar,Google scholar and other databases;Retrieval time: from January 1972 to January 2022.QUADAS-2 was applied to bias risk assessment in the study of the accuracy of diabetes and glucose metabolism prediction models used in prediabetes screening;PROBAST was used to conduct bias risk assessment on the accuracy of prediabetes prediction models;QUADAS-C was used to conduct a bias risk assessment on the comparative accuracy of prediabetes screening tools.Finally,Meta Disc 1.4 and Stata 14.0 were used to combine the effect sizes of the literature.If the results could not be combined,they were described in words.Research Results:24 studies based on the prediction of diabetes or glucose metabolism abnormality in predicting the accuracy of prediabetes were included in the study.There were 16 models.The study included 33872 prediabetes.The study came from 12 countries.There are 19 articles in English and 5 in Chinese.There are 8 FINDRISC models.The combined values of FINDRISC are Sen=0.60,Spe=0.64,LR+=1.69,LR-=0.63,DOR=2.85,AUC=0.675;There are 8 PRT models.The combined value of PRT is Sen=0.76,Spe=0.57,LR+=1.94,LR-=0.47,DOR=4.18,AUC=0.750;There were 4 studies on the Chinese Diabetes Risk Score,with combined values of Sen=0.70,Spe=0.61,LR+=2.06,LR-=0.47,DOR=4.35 and AUC=0.724.There are four PST models,with combined values of Sen=0.73,Spe=0.52,LR+=1.36,LR-=0.51,DOR=2.77,AUC=0.667;The US diabetes calculator research has 2,with combined values of Sen=0.43,Spe=0.66,LR+=1.55,LR-=0.78,DOR=2.01,the rest of the models are 1,Sen range is 0.24-0.99,Spe range is 0.15-0.78,LR+ range is 0.84-2.39,LR-range is 0.02-1.10,DOR range is 0.76-98.37.A total of 13 studies of prediction prediabetes were included,including 14 models.There are two studies on TAG-IT,with combined values of Sen=0.89,Spe=0.35,LR+=1.47,LR-=0.28,DOR=5.24.The other 12 risk prediction models are single studies.According to the purpose of the study,the literature was divided into 6 diagnostic models and 7 prognostic models.The Sen range of diagnostic model is 0.55-0.92,the Spe range is 0.26-0.72,and the AUC range is0.65-0.79;The Sen range of the prognostic model was 0.45-0.86,the Spe range was 0.75-0.91,and the AUC range was 0.58-0.98;Only two studies reported calibration values: PERSEUS model calibration Hosmer lemeshow χ2=5.50,Guangdong（2015）model Hosmer lemeshow χ 2 the range is 0.56-0.96.A comparative accuracy study of 7 prediabetes screening tools was included.Among them,16 models is included.Two tools were verified twice in one study,which were counted as two studies: The data of three studies directly compared the PRT model Sen,SPE and DOR,which were better than the PST model;The direct comparison between the two research data shows that the Sen（0.92）of ADA 2016 model is better than that of PRT（0.76）;In indirect comparison,Sen of ADA 2016 is better than PRT and PST.There are 1 studies comparing the accuracy of China Diabetes Risk Score with the Danish Diabetes Risk Score.There are 1 studies on the accuracy of China Diabetes Risk Score compared with Jiangsu Diabetes Risk Score and Qingdao Diabetes Risk Score.Jiangsu Diabetes Risk Score（AUC）> China Diabetes Risk Score（AUC）> Danish Diabetes Risk Score（AUC）was indirectly detected;There are 1 studies comparing FINDRISC with the Australian Diabetes Risk Score,the German Diabetes Risk Score,the Canadian Diabetes Risk Score,the British Diabetes Risk Score and the US Diabetes Calculator accuracy.There is one study on the accuracy comparison between FINDRISC and JAMRISC.Among the seven models,the specificity of FINDRISC was the highest.Research Conclusion:The accuracy of Jiangsu Diabetes Risk Score and PRT model is better than other models,and China’s Diabetes Risk Score is moderate.FINDRISC has high specificity.However,the external validation of PRT and FINDRISC is mostly for the overseas population,which is rarely verified in the prediabetes population in China.Secondly,the bias risk of most prediabetes prediction models is high or unclear,because most developed models rarely go through external validation or lack sufficient report on the prediction performance of the model,so the validity of the model is still unknown.Therefore,the focus of future research should not focus on developing new prediction models,but on increasing the accuracy of prediction models.