Research On Prediction Model Of Adherence To Prep In Msm

Objective:To predict the adherence to PrEP in MSM through establishing a good predictive model,and then conduct precise interventions for people with low adherence to improve their adherence and reduce the risk of HIV infection.At the same time,it can also provide some reference significance for the "13th Five-Year" national science and technology major project "Research on reducing HIV infection in MSM based on intelligent reminder system to improve the adherence to PrEP ",and improve the strategy of improving adherence.Methods: From April 2013 to March 2015,in the four provinces of Guangxi,Sichuan,Chongqing,and Xinjiang,MSM were recruited through non-probability sampling methods such as publishing information on comrades' websites,cooperating with NGOs,and using core personnel to snowball.Questionnaires and HIV testing were conducted.Firstly Univariate analysis and multi-factor Logisitic regression analysis were performed on the factors affecting the adherence to PrEP in MSM.The data were then randomly divided into training set sampled 756 and test set sampled 100.Variables that is meaningful in multivariate logistic regression analysis are used as input and the adherence as output.Severalprediction models are constructed through the training set,and the adherence is predicted by the test set.Finally,the prediction effects of several prediction models are compared by the prediction agreement rate and the area under the ROC curve(AUC).Results:(1)Among the 856 subjects included in the analysis,those is morely 30 years old and below,urban registered residence,junior college or bachelor degree and above,and unmarried;AIDS knowledge scores less than 8 points accounted for 78.15%;And those who are anxiety or depressed respectively account for 48.60% and 20.56%.(2)Multivariate logistic regression analysis showed that the factors influencing the adherence to PrEP in MSM were research center,trial group,frequency of using condom for anal sex with male partner,and frequency of searching for sexual partner through the Internet.Those whose research center is Chongqing Medical University,Guangxi Medical University,and West China Medical College of Sichuan University is higher adherence than those whose research center is Xinjiang Medical University,the OR values ??are respectively OR=1.552(95%CI:1.086~2.219)?OR=32.32.878(95%CI:9.825~110.028)and OR=3.367(95%CI:2.147~5.280);Daily medication group have low adherence with medication before and after sexual behavior(OR=0.585(95%CI:0.426~0.804));Those who always use condom for anal sex with male partner is higher adherence than never use condom(OR=1.782(95%CI:1.001~3.173));Those who ofen find sexual partners through the Internet is higher adherence than never find sexual partners(OR=1.910(95%CI:1.020~3.573)).(3)The prediction agreement rate and the area under the ROC curve(AUC)of several prediction models are as follows: the prediction agreement rate of the logistic regression prediction model is 64.00%,the AUC is 0.608,the sensitivity is 47.37%,and the specificity is 74.19%;The prediction agreement rate of GA-BP neural network model is 69.00%,AUC is 0.699,sensitivity is 73.68%,and specificity is 66.13%;The prediction agreement rate of LVQ neural network model is 73.00%,AUC is 0.736,sensitivity is 76.32%,and specificity is 70.97%;The prediction agreement rate of the SVM model is 69.00%,AUC is 0.704,sensitivity is 76.32%,and the specificity is 64.52%;The prediction agreement rate of the decision tree model is 69.00%,AUC is 0.684,sensitivity is 73.68 %,and specificity is62.90%.The AUC of the logistic regression prediction model compared with 0.5 were not statistically significant(P>0.05).The AUC of the other prediction models compared with 0.5 were statistically significant(P<0.05).Conclusion: The adherence to PrEP in MSM is affected by multiple factors together.It is closely related to research center,trial group,frequency of using condom for anal sex with male partner,and frequency of searching for sexual partner through the Internet.In the prediction model ofadhenence to PrEP in MSM,the logistic regression prediction model has poor prediction effect,while the GA-BP neural network model,LVQ neural network model,SVM model,and decision tree model have good prediction effects.The latter can be applied to future prediction studies for adherence.