An exploratory machine learning study on paediatric abdominal pain phenotyping and prediction

Area of Science:

• Pediatric Gastroenterology
• Computational Biology
• Epidemiology

Background:

• Paediatric abdominal pain (AP) mechanisms are unclear due to patient heterogeneity.
• This study aimed to identify AP phenotypes and associated factors.
• The goal was to guide future research into AP causes.

Purpose of the Study:

• To identify distinct paediatric abdominal pain (AP) phenotypes using machine learning (ML).
• To develop a predictive model for AP risk stratification.
• To explore demographic and clinical factors associated with AP.

Main Methods:

• Utilized ML clustering on data from 13,790 children in a large birth cohort.
• Extracted demographic and comorbidity data from general practitioner records.
• Developed an ML-based predictive model for AP using identified features.

Main Results:

• Identified three AP phenotypes: allergic predisposition (n=137), maternal comorbidities (n=676), and minimal comorbidities (n=340).
• AP frequency increased with allergic diseases and maternal comorbidities (e.g., 25.6% with ≥3 maternal comorbidities).
• ML model showed moderate predictive performance (AUC 0.67), highlighting ethnicity, allergies, and maternal comorbidities as key factors.

Conclusions:

• Distinct AP phenotypes and key risk factors were identified using ML.
• A predictive ML model enabled effective risk stratification for paediatric AP.
• Findings offer insights for future AP mechanism research and targeted interventions.