目的:在一项前瞻性队列研究中,比较通过不同方法确定的体重指数(BMI)轨迹与儿童早期超重风险之间的关联,并确定在儿童早期关键生长窗口期肥胖风险较高的儿童。
方法:本研究纳入了来自通州北京大学出生队列(PKUBC-T)的1330名儿童。孩子出生时被跟踪,1、3、6、9、12、18和24个月和3岁,以获取其身高/身长和体重数据,并计算BMIZ评分。采用潜在类生长混合模型(GMM)和基于纵向数据的k-means聚类算法(KML)确定从出生到24口的幼儿BMI轨迹分组。使用线性回归比较通过不同方法确定的幼儿BMI轨迹与3岁时BMIZ评分之间的关联。采用Logistic回归模型中5倍交叉验证的平均曲线下面积(AUC)比较不同方法确定的幼儿BMI轨迹在预测3年超重风险(BMIZ评分>1)中的预测性能。
结果:在本研究中包括的研究人群中,使用GMM确定的三类轨迹被分类为低,中等,高,占39.7%,54.1%,6.2%的参与者,分别。使用KML方法确定的两类轨迹被分类为低和高,代表503%和49。7%的参与者分别。使用KML方法确定的三类轨迹被分类为低,中等,高,占31.1%,47.4%,21.5%的参与者,分别。使用不同方法确定的幼儿BMI轨迹反映的生长模式存在一定差异。线性回归分析发现,在调整母亲种族后,教育水平,交货方式,奇偶校验,产妇分娩年龄,分娩时的孕周,儿童的性别,在1个月大时母乳喂养,通过KML方法确定的三类轨迹中的高轨迹组之间的关联(表现为出生时BMI略高,其次是婴儿期的快速增长和直到24个月的稳定高BMI),3年的BMIZ评分最强。Logistic回归分析显示,KML方法确定的三类轨迹分组对3年超重风险的预测性能最好。对儿童饮食平衡指数的高界限评分进行额外调整后,结果基本一致。平均每日体力活动时间,屏幕时间。
结论:这项研究使用不同的方法来识别具有不同特征的幼儿BMI轨迹,并发现由KML方法确定的高轨迹组能够更好地识别儿童早期超重风险较高的儿童。这为选择合适的方法来定义幼儿BMI轨迹提供了科学依据。
OBJECTIVE: To compare the association between body mass index (BMI) trajectories determined by different methods and the risk of overweight in early childhood in a prospective cohort study, and to identify children with higher risk of obesity during critical growth windows of early childhood.
METHODS: A total of 1 330 children from Peking University Birth Cohort in Tongzhou (PKUBC-T) were included in this study. The children were followed up at birth, 1, 3, 6, 9, 12, 18, and 24 months and 3 years of age to obtain their height/length and weight data, and calculate BMI Z-score. Latent class growth mixture modeling (GMM) and longitudinal data-based k-means clustering algorithm (KML) were used to determine the grouping of early childhood BMI trajectories from birth to 24 mouths. Linear regression was used to compare the association between early childhood BMI trajectories determined by different methods and BMI Z-score at 3 years of age. The predictive performance of early childhood BMI trajectories determined by different methods in predicting the risk of overweight (BMI Z-score > 1) at 3 years was compared using the average area under the curve (AUC) of 5-fold cross-validation in Logistic regression models.
RESULTS: In the study population included in this research, the three-category trajectories determined using GMM were classified as low, medium, and high, accounting for 39.7%, 54.1%, and 6.2% of the participants, respectively. The two-category trajectories determined using the KML method were classified as low and high, representing 50. 3% and 49. 7% of the participants, respectively. The three-category trajectories determined using the KML method were classified as low, medium, and high, accounting for 31.1%, 47.4%, and 21.5% of the participants, respectively. There were certain differences in the growth patterns reflected by the early childhood BMI trajectories determined using different methods. Linear regression analysis found that after adjusting for maternal ethnicity, educational level, delivery mode, parity, maternal age at delivery, gestational week at delivery, children\' s gender, and breastfeeding at 1 month of age, the association between the high trajectory group in the three-category trajectories determined by the KML method (manifested by a slightly higher BMI at birth, followed by rapid growth during infancy and a stable-high BMI until 24 months) and BMI Z-scores at 3 years was the strongest. Logistic regression analysis revealed that the three-category trajectory grouping determined by the KML method had the best predictive performance for the risk of overweight at 3 years. The results were basically consistent after additional adjustment for the high bound score of the child\' s diet balanced index, average daily physical activity time, and screen time.
CONCLUSIONS: This study used different methods to identify early childhood BMI trajectories with varying characteristics, and found that the high trajectory group determined by the KML method was better able to identify children with a higher risk of overweight in early childhood. This provides scientific evidence for selecting appropriate methods to define early childhood BMI trajectories.