Laboratory correlates of observation and hospitalization decisions in children with acute gastroenteritis: an exploratory association study

Abstract

To examine, in children with acute gastroenteritis (AGE), the association between laboratory (and available clinical) findings and the real-world clinician decision to place a child under observation or to admit. The study is explicitly exploratory and association-based, not a prediction model. Single-centre retrospective analysis of 226 children aged 0–18 years (April 2024–March 2026). Disposition was classified hierarchically (outpatient discharge / observation only / hospitalization). Laboratory tests were obtained at clinician discretion at presentation, at intravenous cannulation and before fluids. Group comparisons used Mann–Whitney U with Benjamini–Hochberg FDR across 11 parameters; three-group comparisons used Kruskal–Wallis with Dunn post-hoc. NLR and PLR were assessed as exploratory composites. Logistic models reported adjusted ORs with VIF; discrimination was summarised by ROC AUC with bootstrap optimism correction. Analyses restricted to the tested subgroup (Python 3; SciPy/statsmodels). Laboratory testing was obtained in 43% of outpatients versus 98% of observation/hospitalization patients, indicating highly differential, clinician-directed (informative) missingness. Compared with outpatients, children escalated to observation or hospitalization had lower lymphocyte counts and higher neutrophil counts and urea (all FDR p < 0.05). Across the three groups, lower lymphocyte count separated outpatients from escalated patients (H = 20.1, p < 0.001) but did not distinguish observation from admission. In the composite multivariable model, lower lymphocyte count was the strongest correlate (adjOR 0.575, 95% CI 0.419–0.789; p = 0.001). Adding lymphocyte count to available clinical variables (fever, vomiting, measured temperature) was associated with improved model fit within the tested subgroup (likelihood-ratio p < 0.001; apparent AUC 0.75→0.81). As a secondary exploratory observation, the neutrophil-to-lymphocyte ratio showed the numerically highest single-marker discrimination (AUC 0.78, 95% CI 0.68–0.86), though its bootstrap confidence interval overlapped substantially with those of lymphocyte and neutrophil counts. The multivariable escalation model had an optimism-corrected c-statistic of 0.77 (apparent 0.82). No parameter survived FDR correction for the hospitalization-specific outcome. Rotavirus positivity was associated with admission (OR 2.74; p = 0.015). Among children who underwent laboratory testing for AGE, lower lymphocyte count was the marker most consistently associated with the clinician’s decision to escalate care, with higher NLR a secondary exploratory signal. Because these same laboratory values informed the disposition decision (incorporation bias) and testing was selective, findings describe associations with real-world management within the tested subgroup and are hypothesis-generating rather than predictive or generalisable to all children with AGE.