Clinical characteristics, pathological comorbidities, BMI-for-age Z-score, and machine learning-based exploration of postoperative enterocolitis in iIntestinal neuronal dysplasia: a 20-year retrospective study

Abstract

Intestinal neuronal dysplasia (IND) is a congenital enteric neuropathy characterized by abnormalities of the enteric nervous system, most commonly involving hyperplasia of the submucosal plexus and the presence of giant ganglia. This study aimed to characterize the clinical features of IND, evaluate postoperative enterocolitis (POEC), and explore the utility of machine learning (ML) approaches for POEC risk prediction. Patients diagnosed with IND who underwent pull-through surgery at our institution between 2005 and 2025 were retrospectively reviewed. Clinical characteristics and postoperative outcomes were analyzed. POEC was retrospectively assessed with reference to established HAEC diagnostic criteria. Multiple feature-selection strategies and five machine learning algorithms were combined to construct and evaluate 101 predictive models using out-of-fold area under the receiver operating characteristic curve (AUC). A total of 84 patients with isolated IND who underwent pull-through surgery were included. The mean age at symptom onset was 2.57 ± 3.59 years and the mean age at surgery was 5.11 ± 4.22 years. Of these, 77.4% underwent laparoscopic pull-through surgery, 57.1% were male, and constipation was the most common presenting symptom (67.9%). Mixed IND with IMG and/or HG accounted for 64.3% of cases. POEC occurred in 10 patients (11.9%) during follow-up. Female patients and those with mixed IND had significantly lower BMI-for-age Z-scores (BMIZ). Conventional statistical analyses identified no significant individual factors associated with POEC, whereas exploratory machine learning analyses consistently ranked BMIZ among the potentially important predictive features. In this study, we characterized the clinical and pathological features of children with isolated intestinal neuronal dysplasia over a 20-year period. Patients were predominantly male, frequently presented with constipation, and most exhibited mixed IND rather than pure IND. Nutritional impairment was common, with female patients and those with mixed IND showing lower BMIZ values. Machine learning analyses consistently identified BMIZ as a potential predictive feature associated with postoperative enterocolitis, highlighting the potential role of nutritional status in postoperative risk stratification.