Subtyping intractable functional constipation in children using clinical and laboratory data in a classification model
Yi-Hsuan Huang,Chenjia Xie,Chih-Yi Chou,Yu Jin,Wei Li,Meng Wang,Yan Lu,Zhifeng Liu
DOI: https://doi.org/10.3389/fped.2023.1148753
IF: 3.569
2023-04-25
Frontiers in Pediatrics
Abstract:Background: Children with intractable functional constipation (IFC) who are refractory to traditional pharmacological intervention develop severe symptoms that can persist even in adulthood, resulting in a substantial deterioration in their quality of life. In order to better manage IFC patients, efficient subtyping of IFC into its three subtypes, normal transit constipation (NTC), outlet obstruction constipation (OOC), and slow transit constipation (STC), at early stages is crucial. With advancements in technology, machine learning can classify IFC early through the use of validated questionnaires and the different serum concentrations of gastrointestinal motility-related hormones. Method: A hundred and one children with IFC and 50 controls were enrolled in this study. Three supervised machine-learning methods, support vector machine, random forest, and light gradient boosting machine (LGBM), were used to classify children with IFC into the three subtypes based on their symptom severity, self-efficacy, and quality of life which were quantified using certified questionnaires and their serum concentrations of the gastrointestinal hormones evaluated with enzyme-linked immunosorbent assay. The accuracy of machine learning subtyping was evaluated with respect to radiopaque markers. Results: Of 101 IFC patients, 37 had NTC, 49 had OOC, and 15 had STC. The variables significant for IFC subtype classification, according to SelectKBest, were stool frequency, the satisfaction domain of the Patient Assessment of Constipation Quality of Life questionnaire (PAC-QOL), the emotional self-efficacy for Functional Constipation questionnaire (SEFCQ), motilin serum concentration, and vasoactive intestinal peptide serum concentration. Among the three models, the LGBM model demonstrated an accuracy of 83.8%, a precision of 84.5%, a recall of 83.6%, a f1-score of 83.4%, and an area under the receiver operating characteristic curve (AUROC) of 0.89 in discriminating IFC subtypes. Conclusion: Using clinical characteristics measured by certified questionnaires and serum concentrations of the gastrointestinal hormones, machine learning can efficiently classify pediatric IFC into its three subtypes. Of the three models tested, the LGBM model is the most accurate model for the classification of IFC, with an accuracy of 83.8%, demonstrating that machine learning is an efficient tool for the management of IFC in children.
pediatrics