Integration of Machine-Learning Algorithm to Identify Early Life Risk Factors for Future Overweight or Obesity among Preterm Infants: A Prospective Birth Cohort

Yuanqing Fu,Wanglong Gou,Wensheng Hu,Yingying Mao,Yuhong Guan,Tao Huang,Kelei Li,Xiaofei Guo,Huijuan Liu,Duo Li,Ju-Sheng Zheng

DOI: https://doi.org/10.1093/cdn/nzaa054_060

2020-01-01

Current Developments in Nutrition

Abstract:To identify early life risk factors for childhood overweight/obesity among preterm infants and to determine feeding practices that could modify the identified risk factors. Jiaxing Birth Cohort is a prospective cohort involving 338,413 mother-child pairs who were enrolled in between 1999 and 2013, of whom 2125 singleton preterm born children with adequate information documented were included in the analyses. Infant and maternal variables were summarized into 25 features. The LightGBM model based on a gradient-boosting framework was used to link input features with future overweight/obesity and a novel unified framework, SHAP (Shapley Additive exPlanations), was used to interpret predictions and identify predictive factors from the summarized features. Poisson regression model was used to examine the association between feeding practices and the identified leading predictive factor. Of the eligible 2125 preterm infants, 274 (12.9%) developed overweight/obesity at age 4–7 years. Using an interpretable machine learning-based analytic framework, we identified two most important features as predictors of Childhood overweight/obesity: trajectory of infant BMI Z-score change during the first year of corrected age and maternal BMI at enrollment. The identified features in the model showed similar predictive capacity compared with all features. According to the impacts of different BMI Z-score trajectories on model outputs, we classified this feature into favored and unfavored trajectory. Compared with early introduction of solid foods (≤3 months of corrected age), introducing solid foods after 6 months of corrected age was significantly associated with 11% lower risk (risk ratio, 0.89; 95% CI, 0.82 to 0.97, P < 0.01) of being in the unfavored trajectory. Our results suggest that the trajectory of BMI Z-score change within the first year of life is the most important predictor for childhood overweight/obesity among preterm infants. Introducing solid foods after 6 months of corrected age is a recommended feeding practice for mitigating the risk of unfavored trajectories of BMI Z-score change early in life. This study was funded by the Open Project Program of China-Canada Joint Lab of Food Nutrition and Health, Beijing Technology and Business University (BTBU) (KFKT-ZJ-201,801).