Scholarlink

Background <p>Compared with normoglycaemic pregnancies, gestational diabetes mellitus (GDM) confers a markedly elevated risk of dysglycaemia at 6-12 weeks postpartum.</p> Objective <p>This study aimed to develop and externally validate a prediction model for the risk of dysglycaemia at 6–12 weeks postpartum in a Chinese population of women with gestational diabetes mellitus and to provide a clinically usable risk-assessment tool.</p> Methods <p>The derivation cohort comprised 500 Chinese women diagnosed with GDM at 24–28 weeks’gestation in the obstetric outpatient clinic of Beijing Friendship Hospital affiliated to Capital Medical University (a single tertiary center in China) between January 2016 and December 2022, who completed a postpartum oral glucose tolerance test. Predictors were selected using LASSO regression, and four machine learning algorithms (logistic regression, decision tree, random forest, and support vector machine) were trained to construct the prediction models, followed by internal validation. For external validation, a prospective cohort of 170 Chinese women with GDM was recruited from Beijing Chaoyang Hospital(another single center in China) from May to November 2023 and followed until 6 weeks postpartum. Finally, an R-Shiny web calculator was developed to facilitate real-time risk estimation.</p> Results <p>Among 500 women in the development cohort, 209 (41.8 %) with GDM developed dysglycaemia at 6-12 weeks postpartum. LASSO regression identified prior GDM, family history of diabetes, HbA1c, 2-hour plasma glucose from the diagnostic 75-g OGTT, and total bilirubin as independent predictors. The four machine-learning models achieved AUCs of 0.606-0.769, accuracies of 0.600-0.729, sensitivities of 0.526-0.877, and specificities of 0.581-0.828; the logistic model was superior. In the external validation cohort of 170 women, 54 (31.8%) developed dysglycaemia. The validated logistic model yielded an AUC of 0.808 (95 % CI 0.740-0.875), sensitivity 0.810, specificity 0.644, and Youden index 0.455, with excellent calibration.</p> Conclusions <p>The validated logistic model offers health care personnel a ready-to-use web tool for early postpartum dysglycaemia identification and targeted intervention to curb progression to type 2 diabetes.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Development and validation of a risk prediction model for dysglycaemia at 6–12 weeks postpartum in Chinese women with gestational diabetes: a retrospective cohort study

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月6日

查看全文
  • Zhe Liu,
  • Jie Liu,
  • Weiwei Liu

摘要

Background

Compared with normoglycaemic pregnancies, gestational diabetes mellitus (GDM) confers a markedly elevated risk of dysglycaemia at 6-12 weeks postpartum.

Objective

This study aimed to develop and externally validate a prediction model for the risk of dysglycaemia at 6–12 weeks postpartum in a Chinese population of women with gestational diabetes mellitus and to provide a clinically usable risk-assessment tool.

Methods

The derivation cohort comprised 500 Chinese women diagnosed with GDM at 24–28 weeks’gestation in the obstetric outpatient clinic of Beijing Friendship Hospital affiliated to Capital Medical University (a single tertiary center in China) between January 2016 and December 2022, who completed a postpartum oral glucose tolerance test. Predictors were selected using LASSO regression, and four machine learning algorithms (logistic regression, decision tree, random forest, and support vector machine) were trained to construct the prediction models, followed by internal validation. For external validation, a prospective cohort of 170 Chinese women with GDM was recruited from Beijing Chaoyang Hospital(another single center in China) from May to November 2023 and followed until 6 weeks postpartum. Finally, an R-Shiny web calculator was developed to facilitate real-time risk estimation.

Results

Among 500 women in the development cohort, 209 (41.8 %) with GDM developed dysglycaemia at 6-12 weeks postpartum. LASSO regression identified prior GDM, family history of diabetes, HbA1c, 2-hour plasma glucose from the diagnostic 75-g OGTT, and total bilirubin as independent predictors. The four machine-learning models achieved AUCs of 0.606-0.769, accuracies of 0.600-0.729, sensitivities of 0.526-0.877, and specificities of 0.581-0.828; the logistic model was superior. In the external validation cohort of 170 women, 54 (31.8%) developed dysglycaemia. The validated logistic model yielded an AUC of 0.808 (95 % CI 0.740-0.875), sensitivity 0.810, specificity 0.644, and Youden index 0.455, with excellent calibration.

Conclusions

The validated logistic model offers health care personnel a ready-to-use web tool for early postpartum dysglycaemia identification and targeted intervention to curb progression to type 2 diabetes.