A Study On Prediction For Alzheimer’s Disease In Urban Community Based On Artificial Neural Network Model

Objective:Through the follow-up investigation and related biomarkers detection of old people over 60 years old in the Nanchang Hong du community of Jiang Xi province,we aim to establish an artificial neural network model for the prediction of early stage of Alzheimer’s disease within city community.Methods:10 communitys（Hong Dong,Hong Gan,Hong Ying,Hong Xiao,Hong Fang,Hong Yuan,Hong Xiang,Hong Ke,Hong Xi,Hong Zhao）of Nanchang Hongdu street were selected as the study fields.We used stratified cluster sampling method to sample 9733 healthy old people who over 60 years old as our research object,after three years（June 2013 to June 2015）follow-up screening（two stages of investigation international standards method was used to the diagnosis of AD）,364new alzheimer’s disease patients were confirmed,then we selected 200 cases AD patients and 200 cases normal controls who had good adherence,willing to detect blood and urine sample as ANN modeling object.We used questionnaire investigation to obtain the general situation of objects and the influence factors of AD,and ELISA method to detect the urine and blood related biomarkers.We built ANN model by SPSS Modeler software and features participants work curve（ROC curve）analysis by Med Calc v.14.0 software.Results:（1）The incidence rate of AD was 13.13/1000 person years.AD incidence rate increases with age(χ²_trend=45.500,p<0.01),people aged 60～69 AD incidence rate was 10.71/1000 person year,70～79 was 13.32/1000 person year,over80 was 19.34/1000 person year;The incidence rate of women was higher than men,15.25/1000 and 10.94/1000 person year respectively;（2）The multi-factor analysis showed that the influence factors of Alzheimer disease（AD）were age,smoking,educational level,occupation,family per capita income,diabetes.70～79（OR=4.651,95%CI:2.699-8.014）or over than 80 years old（OR=14.585,95%CI:6.567,32.394）group were risk factors for AD.in addition,smoking（OR=4.214,95%CI:2.112,8.409）,and history of diabetes（OR=1.874,95%CI:1.057-3.323）original professional was farmer（OR=3.391,95%CI:2.573-4.327）were also risk factors for AD;Protect factors of AD includes high degree education（OR=0.669,95%CI:0.477-0.939）,the original professional were administrative cadres（OR=0.158,95%CI:0.056-0.445）,scientific or technical personnel（OR=0.091,95%CI:0.009-0.958）,per capita monthly income of family more than 3000 yuan（OR=0.221,95%CI:0.115-0.426）,often study or read newspaper（OR=0.428,95%CI:0.239-0.766）（p<0.05）;（3）The sensitivity and specific degrees of early diagnosis of AD by Plasma Aβ42 were 88.0%,92.5%,and the optimal cutoff value is 33.7pg/ml,AUC is 0.938,95%CI（0.910,0.960）;The sensitivity and specific degrees of early diagnosis of AD by Urine AD7c-NTP were 81.5%,83.5%,and the optimal cutoff value is 0.725ng/ml,AUC is0.852,95%CI（0.813,0.885）;The sensitivity and specific degrees of early diagnosis of AD by plasma Aβ40 were 47.5%,67.5%respectively,the optimal cutoff value is160.7pg/ml,AUC is 0.578,95%CI（0.528,0.627）;The sensitivity and specific degrees of early diagnosis of AD by Aβ42:Aβ40 ratio were 84.0%,91.0%respectively,and the optimal cutoff value is 0.197,AUC is 0.91,95%CI（0.878,0.937）;The comparison AUC of early diagnosis of AD among different variables showed that plasma Aβ42>plasma Aβ42:Aβ40 ratio>urine AD7c-NTP>plasma Aβ40（p<0.001）;（4）original occupation,whether take part in physical exercise or physical activity,family per capita income,cultural degree,whether learning or read the newspaper,with major adverse life history,marital status,history of diabetes,urine AD7c-NTP,smoking,ADL,Aβ42:Aβ40,plasma Aβ42,MMSE,age,A total of15 variables were selected as the input variables to build ANN model,output results show that the model’s accuracy is 82.1%.The prediction of Model AUC is 0.987,the sensitivity and specific degrees was 96.5%,94.0%.Conclusion:（1）The incidence rate of AD in Jiangxi province city community is high,the incidence rate of AD increases with age,the incidence rate of women was high than men;（2）age,smoking,diabetes were risk factors of AD;High degree of culture,the original occupation were officer or individual freedom occupation,high family per capita monthly income were protection factors of AD;（3）In AD group,urine AD7c-NTP content is higher than normal control group,the plasma Aβ42,Aβ40protein and Aβ42:Aβ40 ratio were lower than normal control group.The comparison of early diagnosis effectiveness of AD among different variables showed that the descending order was plasma Aβ42,plasma Aβ42:Aβ40 ratio,urine AD7c-NTP,plasma Aβ40;（4）original occupation,whether take part in physical exercise or physical activity,family per capita income,cultural degree,whether learning or read the newspaper,with major adverse life history,marital status,history of diabetes,urine AD7c-NTP,smoking,ADL,Aβ42:Aβ40,plasma Aβ42,MMSE,age,A total of15 variables were selected as the input variables to build ANN model,the results show that the accuracy of the model is higher,authenticity is good.The model is suitable for early stage of AD screening or prediction in large scale urban community.