Classification Tree Model For Risk Factors Of Preterm Birth And Interaction Analysis

Objective:(1).To learn the current incidence of preterm birth.(2).To analyze the risk factors of preterm birth.Using classification tree to investigate factors which influence the incidence of preterm birth and discuss the interaction between those risk factors.To provide theoretical evidence for early prediction and preventing from preterm birth.Methods:Questionnaires of pregnant women were collected from the First Affiliated Hospital of Shanxi Medical University between Mar.2012 and Jul. 2014. We collected the data of the general demographic characteristics of pregnant women, pregnancy and health status.Epidata 3.0 is used to data record and SPSS 20.0 to data analysis.By using χ2test,logistic regression model and classification trees to explore risk factors of preterm birth and its interaction between these factors.The risk factors were determined by logistic regression model and exhaustive CHAID decision tree methods.Select interaction factors by using classification trees model and calculate the addictive and multiplication interaction.Then verifying the interaction by logistic regression model to investigate risk factors and interaction of preterm birth.Results:(1).We collected valid questionnaires of 4804 pregnant women who gave birth to singleton in First Affiliated Hospital of Shanxi Medical University between Mar.2012 andJul. 2014. The age of those pregnant women ranges from 14 to 59.The medium age is28.The average pregnancy week is( 38.5±1.9).The average weight of newborn babies is(3199.8±596.6)g.(2).Of 4804 newborn singletons,there were 494 preterm birth babies.The incidence of preterm birth is 10.28%.There were 347 late preterm births which account for 70.24% of all.77 of moderate preterm account for 15.59% of preterm birth.70 of very preterm account for 14.17%.190 cases of spontaneous preterm delivery account for 38.46% of preterm birth.While the cases of preterm birth for medical and obstetrical indications were 304 which account for 61.54%.(3).Chi-square test showed that differences were statistically significant(P < 0.05) in age of pregnant women,living places,income per month,education level,parity,pre-pregnant BMI,weight gain during pregnancy,vomit during pregnancy,vaginal bleeding occurance,gestational hypertension,folic acid in early pregnancy,parental care,pregnancy health education,passive smoking,delivery season,mode of delivery,birth weight,placental abruption,premature rupture of membranes,abnormal position of fetus, distressing, amniotic fluid and placenta previa between preterm birth as cases and full-term infants as control group.(4).Unconditional regression model showed that the risk factor of preterm birth include:gestational hypertension disease(OR=10.658,95%CI:8.329~13.638),premature rupture of membranes(OR=2.353,95%CI:1.760~3.146),abnormal position of fetus(OR=2.763,95%CI:1.601~4.770), placental abruption(OR=3.883,95%CI:2.060~7.322), residence in rural area(OR=1.738,95%CI:1.372~2.201),the number of parental care less than10( OR=1.688, 95%CI:1.326~2.149), weight gain below IOM recommendation(OR=1.713, 95%CI:1.267~2.314),pregnant age older than 35 years old(OR=1.432,95%CI:1.094~1.874), slight(OR=1.890,95%CI:1.387~2.576) or severe vaginal bleeding(OR=7.218,95%CI:3.204~16.263).If pregnant season was summer or fall,the incidence of preterm birth is lower than in spring.The odds ratio were 0.755(95%CI:0.578~0.986) 、0.689(95%CI:0.510~0.930).primiparas have lower preterm birth rate than multiparas(OR=0.759,95%CI:0.584~0.988).(5).The CART model of preterm birth showed that the classification tree model consisted of three stratifications,13 nodes and 7 terminal nodes.6 explanatory variables were selected from the model,namely gestational hypertension, the number of premature rupture of membranes, maternal history,vaginal bleeding during pregnancy, and residence of pregnant women.The primary risk factor of preterm birth is gestational hypertension disease.The proportion of preterm in gestational hypertension disease pregnant women(42.3%)is significantly higher than those who were not(6.3%).In pregnant women who experience gestational hypertension disease,residence became the main factor which affect the incidence of preterm birth.Of those women,parity is the most important factor for residence in rural area.However,those who had not any gestational hypertension disease,their prior factor was parental care.Of these women,if their parental care was more than 10 times,premature rupture of membranes mainly influence the preterm birth rate.Having more than 10 times parental care make the proportion of preterm birth(10.7%)lower than those who had not(2.5%).Vaginal bleeding during pregnancy is the main factor for insufficient parental care and their preterm birth rate is higher.(6).Use classification tree model to filter the possible factors between which existed interactions.The result showed that there existed addictive interaction between gestational hypertension disease and parental care RERI=10.296,(95%CI:4.907~15.686),AP=0.520,(95%CI:0.384~0.657),S=2.213,(95%CI:1.628~3.008).Addictive interaction was found between gestational hypertension disease and residence of pregnant women RERI=6.761,(95%CI:0.643~12.878), AP= 0.341,(95%CI:0.106~0.575),S=1.559,(95%CI:1.067~2.280).Parental care and premature rupture of membranes have multiplication interaction to the preterm birth rate.OR is 2.026(95%CI:1.401~2.932),P<0.05.Conclusions:(1).The most important risk factor of preterm birth is gestational hypertension disease.In pregnant women who experience gestational hypertension disease,residence became the main factor which influence the incidence of preterm birth.While those who had not any gestational hypertension disease,their prior factor was parental care.(2).Risk factors for preterm birth are premature rupture of membranes,abnormal position of fetus,placental abruption,residence in rural area,the number of parental care less than 10,weight gain below IOM recommendation age older than 35 years old,vaginal bleeding during pregnancy.Delivery in summer or autumn and primiparity.(3).There exists addictive interaction between gastational hypertension disease and parental care or residence in rural area.There are multiplication interaction between parental care and premature rupture of membranes.Classification tree model is able to detect interaction factors quickly.(4).Those who had gastational hypertension disease and lived in rural area should be paid most attention to in order to prevent preterm from happening.Parental care in pregnancy is essential for all of pregnancy women.Women with gestational hypertension disease are supposed to treat properly.Also PROM should be avoided in pregnancy to make preterm birth rate decrease.We are supposed to be aware of these factors to make the preterm birth rate lower and promote health of mother and baby for further.(5).Combing this model with logistic regression model,more information could be found in factors which may influence preterm birth rate and between which interaction exist.