Phenotyping and prediction of paediatric abdominal pain: A machine learning approach

Abstract

Due to patient heterogeneity, the exact mechanisms of paediatric abdominal pain (AP) remain unknown. We sought to resolve this by identifying paediatric AP phenotypes and developing predictive models to determine their associated factors. In 13,790 children from a large birth cohort, the frequencies of paediatric and maternal demographics and comorbidities were catalogued from general practitioner records. Unsupervised machine learning clustering was used to identify phenotypes of paediatric AP with shared characteristics. Predictive paediatric AP models were constructed using paediatric and maternal demographics and comorbidities.

1,274 children experienced AP (9.2 %) (average age: 8.4 ± 1.1 years old, male/female: 615/659), who clustered into 3 distinct phenotypes: phenotype 1 with an allergic predisposition (n = 137), phenotype 2 with maternal comorbidities (n = 676), and phenotype 3 with minimal other comorbidities (n = 340). As the number of allergic diseases or maternal comorbidities increased, so did the frequency of AP, with 17.6% of children with ≥ 3 allergic diseases and 25.6% of children with ≥ 3 maternal comorbidities. Furthermore, in high-risk children who met both ≥ 3 allergic diseases and ≥ 3 maternal comorbidities, 30.8% had AP. Predictive models demonstrated modest fidelity in predicting paediatric AP (AUC 0.66), showing that a child’s ethnicity and paediatric/maternal comorbidities were strongly predictive factors. Our findings reveal distinct phenotypes and associated factors of paediatric AP, suggesting targets for future research to elucidate the exact mechanisms of paediatric AP related to allergic diseases, ethnicity, and maternal comorbidities.