The Construction Of The Primary Screening Model And The Discriminant Model For Chronic Obstructive Pulmonary Disease In Shenyang·China

Objective:The gold standard for diagnosis and grading of chronic obstructive pulmonary disease?COPD?is pulmonary function test,but the high application cost and technical requirements make it unsuitable for widely usage in primary medical institutions.The establishment of the discriminant model can reduce the rate of missed diagnosis caused by the reasons above and increase the diagnostic rate of COPD.The purpose of this study was to compare the effectiveness of different models about primary screening and discriminant of COPD,and to establish a reasonable and efficient method to screening and discriminate COPD.Methods:The normal controls and COPD patients from north-east part of China were enrolled from December 2017 to April 2019.Pulmonary function and the questionnaire were carried out for all subjects in the First Hospital of China Medical University.Verify the effectiveness of Guangdong model in Shenyang.Taking screening illness or not as a goal for screening model.Taking the disease severity as a goal for the discriminant model.Using R language software,the multivariate linear regression,logical regression,linear discriminant analysis,and K-nearest neighbor,decision tree,conditional decision tree and support vector were respectively applied to construct the models and compare the effectiveness among them.The most suitable primary screening model and discriminant model were established for COPD.Results:232 COPD patients?128 male,aged 40-80 years,124 GOLD I-II patients,and 108 GOLD III-IV patients?and 115 normal controls?42 male,aged 40-80 years?were enrolled.Guangzhou model was used to verified the accuracy in Shenyang,the two-class accuracy were:normal:0.9826,patient:0.4483,overall accuracy:0.6254.The three classification accuracy rates were:normal:0.9826,mild COPD:0.0403,severe COPD:0.5556,and overall accuracy:0.5130.Nine primary screening models were established.The optimal one was:Y=-3.6522-1.1012X₄+3.0346X₅+0.9241X₆+1.1462X₇+1.7173X₉-0.3096X₁₁+2.2282X₁₆+1.2225X₁₇-0.5788X₁₈+0.9945X₁₉+0.7663X₂₀.Seven discriminant screening models were established.The optimal one was decision tree.Based on the information value of decision tree,we got that shortness of breath?x5?,cooking fuel level?x6?,smoking index?x7?,BMI?x11?,cough?x12?,farmer or not?x16?,kitchen ventilation?x18?,childhood heating?x19?are the best variables for the discriminant model.The evaluation for the primary screening model showed that 0.084 was the best critical point and the AUC value was0.941.The regression accuracy of the discriminant model was 0.9884 for normal subjects,0.6522 for mild and modest COPD patients,0.7901 for severe COPD patients and overall accuracy was 0.8038.The prediction accuracy was 0.8571 for normal subjects,0.5934 for mild and modest COPD,0.7901 for severe COPD and overall accuracy was 0.7778.The corresponding program code was established to combine the discriminant model with computer technology?Conclusion:Birth season,education level,BMI,resident type,winter location AQI,smoking/passive smoking,mother's smoking history during pregnancy,fuel explosure level,heating history during childhood and current heating were be found have statistical differences between COPD patients and normal people in Shenyang·China.Stepwise logistic regression is the optimal screening model for COPD,and decision tree is the optimal discriminant model for COPD,which can be applied to the screening and discrimination of COPD.