Study objectives: Children with late-onset (2-5 years) or persistent (3 months-5 years) sleep-related breathing disorder (SRBD) have an increased risk of behavior problems compared to children with no or early-onset SRBD. We sought to determine whether a combination of urine metabolites and sleep questionnaires could identify children at risk for SRBD-associated behavior problems.
Methods: Urine and data were analyzed from the Edmonton site of the CHILD birth cohort study. We measured urine metabolites (random, mid-stream) at age three-years among a sub-cohort of participants (n = 165). Random Forest with a Boruta wrapper was used to identify important metabolites (creatinine-corrected, z-scores) for late/persistent SRBD versus no/early SRBD (reference). An algorithm was subsequently generated to predict late/persistent SRBD in children with a history of snoring using a metabolite composite score (z-scores < or ≥ 0) plus the SDBeasy score defined as [age (yrs.) of most recent positive SRBD]2 - [age (yrs.) first reported ever snoring]2.
Results: Of the 165 children with SRBD data, 40 participants had late/persistent SRBD. Seven urinary metabolites in addition to the SDBeasy score were confirmed as important for late/persistent SRBD (AUC = 0.87). Among children with an ever-snoring history and a metabolite composite score ≥0, those with SDBeasy score ≥3 were over 13-fold more likely to have late/persistent SRBD (OR 13.7; 95%CI: 3.0, 62.1; p = 0.001). This algorithm has a Sensitivity of 69.6%, Specificity of 85.7% and a positive likelihood ratio (+LR) of 4.9.
Conclusions: We developed a predictive algorithm using a combination of questionnaires and urine metabolites at age three-years to identify children with late/persistent SRBD by five-years of age.
Keywords: Boruta; Prediction; Random forest; Scoring system; Sleep-disordered breathing; Urine metabolomics.
Copyright © 2022. Published by Elsevier B.V.