Hybrid machine learning models for fetal growth restriction assessment:a 5-year follow-up study
Objective To explore the predictive value of three hybrid machine learning models(nomogram,decision tree and random forest)for fetal growth restriction(FGR).Methods A total of 111 pregnant women with FGR who underwent regular obstetric examination and delivered in Nanning Red Cross Hospital from January 2018 to December 2022 were retrospectively selected,and another 87 pregnant women without FGR in the same period of time were also selected.Clinical data,color Doppler ultrasound indexes,laboratory test indexes,and fetal conditions of all pregnant women were collected.Multifactorial Logistic regression was used to analysis the influencing factors of FGR in pregnant women.Three prediction models,nomogram,decision tree and random forest,were constructed using independent influencing factors as predictor variables.The predictive performance of the models was evaluated using accuracy,sensitivity,specificity,precision,recall,F1 value,and receiver operating characteristic(ROC)curves.Results Multifactorial logistic regression analysis showed that irregular folic acid supplementation,hypertensive disorders of pregnancy(HDP),middle cerebral artery middle cerebral artery RI(cut-off value=0.86),low fasting venous blood VEGF(cut-off value=49.85ng/mL),and low umbilical cord venous blood VEGF(cut-off value=67.10ng/mL)were independent factors for the occurrence of FGR in pregnant women(OR=0.013-57.563,respectively,all P＜0.05).Based on the independent risk factors,the area under the curve(AUC),sensitivity,and specificity of the three models were constructed respectively,and were 0.903(0.823-0.982),0.925,0.750 for the nomogram model;were 0.894(0.809-0.979),0.875,0.850 for the decision tree model;were 0.968(0.931-1.000),0.875,0.950 for the random forest model.The AUC value of random forest model was higher than that of decision tree model(Z=-2.582,P＜0.05).Conclusion The prediction results of nomogram,decision tree and random forest prediction models on FGR all have a high degree of differentiation,among which the random forest model is better than the decision tree model in predicting.There is no significant difference in the prediction efficacy between the random forest model and the nomogram model,and the two can be complementary to each other in the application of predicting FGR.
万方数据
维普
