Daisuke Shigemi1,2, Satoru Yamaguchi1, Shotaro Aso2, Hideo Yasunaga2. 1. a Yamaguchi Women's Hospital , Chiba , Japan. 2. b Department of Clinical Epidemiology and Health Economics , School of Public Health, The University of Tokyo , Tokyo , Japan.
Abstract
Objective: We aimed to develop new predictive models for excluding macrosomia using only maternal physical parameters, without sonographic examination. Methods: The present study retrospectively analyzed the medical records of pregnant women who delivered singleton infants at term at one obstetric hospital in an urban area in Japan from May 2005 to April 2017. We performed logistic regression analysis to predict macrosomia and created an integer risk scoring system based on the significant predictors. We also developed an alternative predictive regression model using machine learning with the random forest algorithm. Results: There were 203 cases of macrosomia among 15,263 eligible women. Although our scoring system had low specificity and positive predictive value, the negative predictive value for screening macrosomia was very high (0.996-1.000). The other model, using machine learning with the random forest algorithm to predict macrosomia, showed a negative predictive value of 0.99, which was similar to the results of our scoring system. Conclusions: Our integer scoring system is an easy and useful method for excluding macrosomia among pregnant women without sonographic examination.
Objective: We aimed to develop new predictive models for excluding macrosomia using only maternal physical parameters, without sonographic examination. Methods: The present study retrospectively analyzed the medical records of pregnant women who delivered singleton infants at term at one obstetric hospital in an urban area in Japan from May 2005 to April 2017. We performed logistic regression analysis to predict macrosomia and created an integer risk scoring system based on the significant predictors. We also developed an alternative predictive regression model using machine learning with the random forest algorithm. Results: There were 203 cases of macrosomia among 15,263 eligible women. Although our scoring system had low specificity and positive predictive value, the negative predictive value for screening macrosomia was very high (0.996-1.000). The other model, using machine learning with the random forest algorithm to predict macrosomia, showed a negative predictive value of 0.99, which was similar to the results of our scoring system. Conclusions: Our integer scoring system is an easy and useful method for excluding macrosomia among pregnant women without sonographic examination.
Entities:
Keywords:
Machine learning; macrosomic infant; random forest; scoring system; screening