Construction Of Postpartum Depression Prediction Model Based On Machine Learning Algorithm

ObjectivesTo investigate the prevalence of postpartum depression at 4-6 weeks after delivery in Guangzhou;to analyze the effects of different demographic characteristics,different social and psychology data and obstetric data on postpartum depression in women;and to combine multiple influencing factors to construct postnatal depression prediction model through machine learning algorithm,so as to provide a simple,sensitive,convenient and feasible tools for the early prediction of the risk of postpartum depression,which also provides a reference for the prevention and intervention of postpartum depression.MethodsIn this study,2396 maternal women who delivered in a third-grade maternity and child hospital in Guangzhou from August 2017 to July2018 and met the inclusion criteria were followed up with questionnaires at 4-6 weeks after delivery.The social demographic data and perinatal period of the study subjects were collected by self-made general information questionnaire;the Edinburgh Postpartum Depression Scale was used to understand the incidence of depression in this population,and obstetric data were collected through medical records system such as hospital integration platform.EpiData 3.1 tool was used for recording the data;SPSS 22.0 was used to explore and analyze the possible factors affecting postpartum depression,and the difference was statistically significant when P<0.05.The machine learning algorithm of Weka software were used to construct the postpartum depression prediction model.Results1.Occurrence and influencing factors of postpartum depressionA total of 2396 postpartum women were enrolled in this study.981cases?40.9%?exposured in depression.21 characteristic indicators,such as age,education level,income level of parturient and spouse,whether the parturient was the only one child,place of motherhood after discharge,family relationship,parity and mode of delivery,were related to postpartum depression?p<0.05?.Logistic regression analysis showed that the maternal spouse had a high income?>10000 yuan??OR=0.471,95%CI:0.227-0.974?,and the maternal relationship with her husband's parents was good?OR=0.359,95%CI:0.144-0.892?are factor to reduce the occurrence of depression;puerpera is the only child?OR=1.429,95%CI:1.043-1.957?,the past year has witnessed life pressure events?OR=2.531,95%CI:1.500-4.270?,mixed infant feeding?OR=1.979,95%CI:1.095-3.578?and artificial feeding?OR=1.598,95%CI:1.210-2.112?do harm for depression?P<0.05?.2.Comparisons of Several Algorithms for Establishing Predictive Model of Postpartum DepressionBased on the 21 statistical characteristics mentioned above,a data set is constructed.Seven classical machine learning algorithms?Bayesian network,Naive Bayesian,Decision Tree,Random Forest,Artificial Neural Network,Support Vector Machine and Logical Regression?in Weka software tool are used to construct prediction models and compare their performances.The results show that the Bayesian network algorithm achieves good prediction performance on the two performance indicators of F₁ measurement and ROC curve,71.4%and 76.3%namely.In terms of running speed,the maximum efficiency of bayesian network is 0.01seconds.Therefore,Bayesian network is the best algorithm for predicting postpartum depression.Conclusion1.The incidence of PPD was higher in Guangzhou 4-6 weeks postpartum,which was related to screening time,screening tools and population characteristics.2.The influencing factors of postpartum depression are various.High income of maternal spouse and good relationship between maternal and husband's parents are conducive to the reduction of PPD risk.Only child,life stress events and non-breastfeeding mothers are all high risk groups of postpartum depression.3.Bayesian network algorithm is the best algorithm of postpartum depression prediction model.In the future,the model can be further put into use to help clinical decision-making.