Risk Factors For Isolated Maternal Hypothyroxine-Mia And The Relationship With Pregnancy Outcomes

Objective:To learn the incidence of isolated maternal hypothyroidemia(IMH)during pregnancy in Hangzhou Obstetrics and Gynecology Hospital;to identify the risk factors for IMH and their diagnostic value;to explore the relationship between IMH and maternal/neonatal outcomes,so as to provide theoretical basis for the prevention and clinical treatment of IMH.Methods:1.Research object and data source:The study population consisted of pregnant women who were admitted to the Hangzhou Obstetrics and Gynecology Hospital from March to December 2018 and participated the“Hangzhou Maternal and Child Health Cohort”(registration number:Chi CTR1900026149).Related data were collected by questionnaires and electronic medical record system retrospectively.In our study,pregnant women diagnosed with IMH and had complete medical records were enrolled as the IMH group,and the control group(with normal thyroid function,NTF)was 1:4matched by age(±1 year).The following data were collected:1)general demographic data;2)maternal lifestyle data during pregnancy;3)maternal examination data;4)ultrasound indexes of the newborn.By comparing the data between the two groups,possible risk factors for IMH were identified,and the effects of maternal IMH combined with independent risk factors on adverse maternal/neonatal outcomes were analyzed in the context.2.Statistical analysis:SPSS25.0,Graphpad prism8.0,and anaconda were used to analyze the data.The t-test and Mann-Whitney U test were used to analyze continuous numerical variables and chi-square test was used to analyze classification variables.After adjusting confounders to identify independent risk factors,a logistic regression model combined with receiver operating characteristic(ROC)curve was used to evaluate the diagnostic value of single risk factors or multiple risk factors combined to predict IMH.Logistic regression analysis was conducted,using the variables with statistically significant difference as independent variables,IMH and adverse maternal/neonatal outcomes as dependent variables to obtain the odds ratio(OR)and95%confidence interval(CI)before and after adjusting for confounding factors.All results were weighed statistically significant with P<0.05.3.A machine learning-based diagnostic prediction model for IMH using XGBoost(extreme gradient boosting)algorithm was constructed,which was interpreted using a Shapley value inspired SHAP(shapley additive ex Planation).In this study,the proportion of training set and test set of machine learning was 7:3.All sample data were divided into IMH group and NTF group,with IMH group marked as 1 and NTF group marked as 0.This study took IMH as the prediction label,and the baseline characteristics,biochemical indicators,lifestyle and other indicators of pregnant women as the input characteristics to build the model.Results:1.The incidence of IMH:A total of 2477pregnant women with complete medical records were enrolled in“Hangzhou Maternal and Child Health Cohort”,among them127 pregnant women were diagnosed with IMH(5.13%).2.Analysis of risk factors related to IMH disease:In this study,127 pregnant women were chosen as the IMH group(case group),508 pregnant women with NTF were randomly matched with a 1:4 ratio as the control group,the matching factor was age(±1 year).1)The mean±standard deviation of pre-pregnancy body mass index(p BMI)in the IMH group was higher than that in the NTF group(21.90±2.79 kg/m~2 vs 20.94±2.74kg/m~2,P<0.001).Pregnancy weight gain(GWG)in the IMH group was also significantly higher than that in the NTF group(13.89±4.73 vs.12.75±3.81,P=0.03).2)The proportion of pregnant women who were overweight or obese was significantly higher in the IMH group than in the NTF group(36.2%vs.21.1%,P=0.002).3)The proportion of people with high triglyceride(TG)levels in the IMH group was significantly higher than that in the NTF group(23.8%vs.18.5%).The proportion of patients with high TG levels+high serum total cholesterol(TC)was higher than that of NTF group(37.3%vs.20.1%).4)p BMI[OR(95%CI):1.04(1.01～1.06),P=0.014],GWG[OR(95%CI):1.1(1.04～1.16),P<0.0001],neutrophil proportion(NP)[OR(95%CI):1.07(1.01～1.13),P=0.015],FBG[OR(95%CI):1.77(1.06～2.93),P=0.036],and TG[OR(95%CI):1.58(1.24～2),P<0.001]were independent risk factors for IMH.Creatinine[OR(95%CI):0.93(0.89～0.97),P=0.001]was a protective factor for IMH.3.Analysis of the diagnostic values of individual risk factors for prediction of IMH revealed that TG was superior to p BMI,GWG,NR,and FBG.The area under the ROC curve of TG was 0.643(AUC:0.643;95%CI:0.588-0.699);and when the value was set at 2.435 mmol/L,the sensitivity and specificity of TG for IMH diagnosis were 56%and 70.4%,resepectively.When multiple risk factors were combined,the combination of p BMI+TG+GWG had the highest diagnostic efficacy for IMH,with a sensitivity of 44.4%and specificity of 86%,respectively,at the maximum Jordans index.4.A machine learning-based diagnosis prediction model for IMH patients was constructed and verified.The corresponding AUC value under the XGBoost model in this study was 0.99.The AUC value for the prediction on the test set was 0.73.SHAP’s interpretation of XGBoost model in a single sample showed that TG,total bile acid(TBA),GWG,and Homocysteine may increase the possibility of IMH in pregnant women.Meanwhile,high density liptein cholesterol(HDL-C)had a negative effect on the occurrence of IMH.The summary diagram of feature importance showed that TG was the feature that contributed most to the prediction model.Direct Bilirubin was the second largest contributor,and the third to fifth largest contributors to prediction were GWG,LR,and p BMI,respectively.In the summary graph of the results of a particular classification,the red dots of GWG and p BMI were concentrated in the region of SHAP>0,indicating that increased GWG and p BMI can increase the risk of IMH.All blue dots of DB,LR,and Total protein(TP)were mainly in areas of SHAP>0,indicating that reduced DB,LR,and TP can increase the risk of IMH.In addition,dependence scatter plots showed that with FBG<4.5 mmol/L and TG<2.5 mmol/L,pregnant women had a lower risk of being diagnosed with IMH and vice versa.5.Effects of IMH on adverse maternal/neonatal outcomes:Pregnant women with IMH were 1.779,2.309,2.748,and 1.737 times more likely to have hypertensive disorders of pregnancy(HDP),gestational diabetes mellitus(GDM),hyperlipidemia(HLD),and amniotic fluid abnormalities,respectively,compared to those with NTF.After adjusting for confoundings,IMH women were 3.60 times and 2.91 times more likely to have GDM and HLD than NTF women.Conclusion:1.The incidence of IMH in this study was 5.13%,which is within the range from previous studies.2.p BMI,GWG,NP,FBG and TG were independent risk factors for IMH,while creatinine is a protective factor for IMH.3.In the logistic regression model combined with ROC curve,TG had the best diagnostic value to predict IMH.In the combined prediction of multiple indicators,p BMI+TG+GWG had the highest diagnostic efficacy for IMH.The efficacy for the prediction of IMH diagnosis model constructed by XGBoost is average.However,it can identify important factors which normally might be easily ignored.In addition,the prediction effect is better than the traditional logistic regression model.4.IMH is closely associated with adverse outcomes in pregnant women,the risk for GDM and HLD in pregnant women with IMH is 3.60 times and 2.91 times of that in pregnant women with NTF.