Development And Validation Of A Risk Prediction Model For Spontaneous Preterm Birth

ObjectiveWe aimed to perform systematic review and meta-analysis to identify risk factors and their risk values of spontaneous preterm birth for singleton pregnant women in China.Based on the risk values of risk factors,the parameters of prediction model were calculated.Logistic regression model and Rothman-Keller model were used to establish the risk prediction model of spontaneous preterm birth,and the model’s performance was further verified by the clinical real-world data,which aimed to provide reference for clinical evaluation for obstetric medical workers.MethodsThis study was consisted of three parts.Part I:Systematic review and meta-analysis of risk factors of spontaneous preterm birth.CNKI,Wanfang,Pub Med,Embase databases were systematically searched to collect cohort studies and case-control studies that investigated risk factors of spontaneous preterm birth in singleton pregnant women from inception to October 29,2022.Two researchers independently screened literature and extracted data.NOS（Newcastle-Ottawa Scale）tool was used to assess the risk bias of included studies.Meta-analysis was performed by Stata15.1 software,and using OR（Odds Ratio）and 95%CI（95%Confidence Interval）as the pooled effect size of risk factors.PartⅡ:Construction of risk prediction model for spontaneous preterm birth.We obtained the pooled OR values of risk factors via meta-analysis,and obtained the minimum incidence of spontaneous preterm birth as well as the rate of population exposure of each risk factor through literature review.The constant termβ₀was calculated based on the calculation formula of parameter in Logistic regression model,and the pooled OR values of each risk factor were converted intoβto construct Logistic regression prediction model for spontaneous preterm birth.Additionally,the Rothman-Keller model was developed by calculating by risk scores of each risk factor using the pooled OR value and the population exposure rate.By using the binomial distribution function method in R4.2.2 software to generate a group of random data set containing 20000 individuals,we caculeted the risk probability of these data through the constructed models,and sorted the probabilities from smallest to largest.With serial number as the X-axis and risk probability as the Y-axis,we drew a risk trend chart.According to the trend of risk probability,we divided the risk of spontaneous preterm birth into low,moderate and high risk levels.PartⅢ:To verify the prediction performance of risk prediction model of spontaneous preterm birth.From March to December 2022,clinical data were collected from singleton pregnant women at tertially-A hospitals in Lanzhou city,including basic information about the pregnant woman,maternity history,mode of conception,pregnancy complications,mode of delivery and gestational week of delivery.The established prediction model was used to predict the risk probability of spontaneous preterm birth,and the results were compared with the actual occurrence of spontaneous preterm birth,to verify the prediction performance of Logistic regression model and Rothman-Keller model.We used SPSS27.0 software and R4.2.2software for data analysis.The discrimination of the model was evaluated by the area under the ROC curve,and the calibration of the model was evaluated using the Hosmer-Lemeshow goodness of fit test.Results1 Screening for risk factors of spontaneous preterm birth:A total of 5115literatures were obtained through systematic search.By a step-by-step screening process,26 articles including 23 case-control studies and 3 cohort studies,with a total of 149,846 singleton pregnant women were reruited.The NOS scores of included studies ranged from 5 to 8 points,with 15 moderate quality studies（5 to 6 points）,and11 high quality studies（>6 points）.The methodological flaw of included studies mainly included no description of ascertainment of exposure,comparability between groups and assessment of outcome.The results of meta-analysis showed that there were 10 statiscally significant risk factors for spontaneous preterm birth,including the history of preterm birth（OR=7.194,95%CI 3.813～13.572）,premature rupture of membranes（OR=3.103,95%CI 2.061～4.672）,assisted reproductive technology（OR=2.382,95%CI 1.232～4.605）,gestational diabetes（OR=2.704,95%CI1.752～4.172）,gestational hypertension（OR=2.809,95%CI 1.142～6.909）,maternal age≥35 years old（OR=1.342,95%CI 1.069～1.685）,lower reproductive tract infection（OR=2.869,95%CI 2.036～4.043）,polyhydramnios（OR=3.277,95%CI2.093～5.131）,preeclampsia（OR=2.271,95%CI 1.455～3.543）and work during pregnancy（OR=2.178,95%CI 1.720～2.758）.2 Construction of risk prediction model:The Logistic regression model’s calculation formula is:Logit（P）=-4.219+1.973×the history of preterm birth+1.131×premature rupture of membranes+0.868×reproductive technology conception+0.995×gestational diabetes+1.003×gestational hypertension+0.294×age+1.054×lower reproductive tract infection+1.187×polyhydramnios+0.820×preeclampsia+0.778×work during pregnancy.According to the trend of risk probability chart,the probabilities of 17097^th（with P value of 0.046）and 19495^th（with P value of 0.117）were respectively selected as the cut-off value to predict low,moderate,and high risk of spontaneous preterm birth in singleton pregnant women.The risk scores of each factor in the Rothman-Keller model were 7.020（the history of preterm birth）,2.346（premature rupture of membranes）,2.211（assisted reproductive technology）,2.334（gestational diabetes）,2.660（gestational hypertension）,1.295（maternal age≥35 years）,2.702（lower reproductive tract infection）,3.162（polyhydramnios）,2.225（preeclampsia）,and 1.712（work during pregnancy）.According to the trend of risk probability chart,the probabilities of the16333^th（with P value of 0.081）and 19551^th（with P value of 0.159）were respectively selected as the cut-off value to predict low,moderate,and high-risk of spontaneous preterm birth in singleton pregnant women.（3）Validation of risk prediction model:A total of 523 singleton pregnant women were recruited,among whom there were 33 cases of spontaneous preterm birth,with an incidence rate of about 6.310%.The AUC of Logistic regression model was 0.830,and the sensitivity and specificity were 0.667 and 0.878,respectively.The Hosmer-Lemeshow goodness of fit test showed that P>0.05.The AUC of Rothman-Keller model was 0.807,and the sensitivity and specificity were 0.667 and0.863,respectively.The Hosmer-Lemeshow goodness of fit test showed that P>0.05.ConclusionWe constructed the Logistic regression prediction model and Rothman-Keller prediction model for spontaneous preterm birth that included 10 risk fators.External validation of clinical real-world data showed that the prediction performance for both models was good,and the performance was similar,which could provide reference for obstetric medical workers to evaluate risk of spontaneous preterm birth.