南宁地区女性献血者HLA抗体与妊娠因素相关性及随机森林模型预测价值研究OA

Objective To explore the association between the HLA antibody positivity rate in female blood donors and pregnancy history,number of pregnancies,interval from the last pregnancy to blood donation,and age,to identify associat-ed variables using a univariate generalized additive model(GAM),and to further analyze the predictive role of characteris-tic variables for HLA antibody positivity using a random forest model.Methods HLA antibody detection was performed on 391 female blood donors using the Luminex immunomagnetic bead method.The correlation between pregnancy-related fac-tors and HLA antibodies was analyzed using the Chi-square test.Based on R software,a univariate GAM was first construc-ted to analyze the association types between characteristic variables and the HLA antibody positivity rate,followed by the construction of a random forest model to evaluate the predictive value of the variables.Results Among the 391 female blood donors without a transfusion history,the overall HLA antibody positivity rate was 26.34%.The positivity rate in do-nors with a pregnancy history was significantly higher than that in those without(30.09%vs 9.72%,P<0.05),and HLA antibody positivity rate increased linearly with the number of pregnancies(P<0.05).In the univariate GAM,age and num-ber of deliveries exhibited a non-linear association with the HLA antibody positivity rate(the positivity rate increased sharply between 25-35 years of age and stabilized after 3 deliveries).Besides,the interval from the last pregnancy to blood dona-tion showed a linear association with the HLA antibody positivity rate,and the positivity rate decreased as the interval pro-longed(P<0.05).In the random forest model,age(mean decrease gini=29.26)and interval from the last pregnancy to blood donation(mean decrease gini=22.02)were core predictive variables:age was more conducive to identifying positive samples,while the interval from the last pregnancy to blood donation was more helpful for excluding negative samples.The number of deliveries(mean decrease accuracy=16.98)made a significant contribution to predicting positive samples,whereas the number of abortions had no impact.The model had an AUC of 0.583(95%CI:0.593 8-0.770 2),indicating a certain predictive value.Conclusion The associated variables identified by the univariate GAM model,including age,interval from the last pregnancy to blood donation,and number of deliveries,provide a basis for key variables in the random forest model.All three variables have predictive value for HLA antibody positivity,which can provide evidence-based sup-port for personalized transfusion management and stratified screening of female blood donors in this region.