Scholarlink

Background <p>Early identification of pediatric pre-diabetes remains challenging due to limited availability of scalable and non-invasive screening approaches. Recent advances in machine learning enable integration of clinical and image-derived indicators to improve early risk prediction and population-level screening reliability.&#xa0;</p> Objective <p>To develop and validate machine learning–based predictive models for pediatric pre-diabetes risk assessment using a harmonized multi-cohort dataset and to evaluate model robustness across demographic strata and validation settings.</p> Methods <p>A multi-cohort observational dataset comprising pooled and harmonized participant populations was analyzed for predictive modeling. Statistical modeling and machine learning algorithms, including logistic regression and ensemble-based classifiers, were implemented for risk prediction. Model performance was evaluated using internal validation procedures and internal–external validation strategies to assess transportability. Stratified demographic performance analysis was conducted across age, sex, and population subgroups.</p> Results <p>The proposed modeling framework demonstrated strong discriminative performance, achieving high predictive accuracy and area under the receiver operating characteristic curve (AUC). Validation analyses indicated consistent performance across demographic strata, supporting model stability and generalizability within heterogeneous cohorts.</p> Conclusions <p>The findings demonstrate that multi-cohort machine learning–based modeling can provide reliable and transportable prediction of pediatric pre-diabetes risk. The integration of demographic stratification and validation across cohorts strengthens clinical applicability and supports future large-scale deployment and prospective validation.</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

Advanced statistical modeling of image-augmented markers for early detection of pediatric pre-diabetes: an internally validated exploratory study

  • 期刊论文

  • 发表日期：

    2026年3月21日

查看全文
  • Ravinder Kaur,
  • Swetta Kukreja

摘要

Background

Early identification of pediatric pre-diabetes remains challenging due to limited availability of scalable and non-invasive screening approaches. Recent advances in machine learning enable integration of clinical and image-derived indicators to improve early risk prediction and population-level screening reliability. 

Objective

To develop and validate machine learning–based predictive models for pediatric pre-diabetes risk assessment using a harmonized multi-cohort dataset and to evaluate model robustness across demographic strata and validation settings.

Methods

A multi-cohort observational dataset comprising pooled and harmonized participant populations was analyzed for predictive modeling. Statistical modeling and machine learning algorithms, including logistic regression and ensemble-based classifiers, were implemented for risk prediction. Model performance was evaluated using internal validation procedures and internal–external validation strategies to assess transportability. Stratified demographic performance analysis was conducted across age, sex, and population subgroups.

Results

The proposed modeling framework demonstrated strong discriminative performance, achieving high predictive accuracy and area under the receiver operating characteristic curve (AUC). Validation analyses indicated consistent performance across demographic strata, supporting model stability and generalizability within heterogeneous cohorts.

Conclusions

The findings demonstrate that multi-cohort machine learning–based modeling can provide reliable and transportable prediction of pediatric pre-diabetes risk. The integration of demographic stratification and validation across cohorts strengthens clinical applicability and supports future large-scale deployment and prospective validation.