Predicting prolonged hospital stay in children hospitalised with respiratory syncytial virus infection: model development, internal validation and a calendar-year transportability assessment

Background. Respiratory syncytial virus (RSV) is the commonest cause of lower respiratory tract infection in young children, and length of stay dominates cost. Published RSV models mostly predict intensive-care outcomes from variables unavailable on a ward. We developed a model for prolonged stay from routine admission data and examined its transport across years. Methods. Retrospective cohort of children under 5 years admitted to general paediatric wards in Chengdu, China, with RSV confirmed by multiplex polymerase chain reaction, 2023–2025. Prolonged stay was 7 days or more. Fifteen admission variables were ranked by least absolute shrinkage and selection operator regression with 300 bootstrap resamples; retention required 95% selection frequency and stability across four specifications. Discrimination was assessed with 500-resample optimism correction, calibration by slope, intercept and Brier score against a constant-risk benchmark, and utility by decision curves. The primary model was refitted in 2023–2024 and applied unchanged to 2025. Results. Of 494 admissions, 476 were analysed (median age 20.0 months; 57.6% boys); 106 (22.3%) stayed ≥ 7 days. Age (odds ratio 0.49 per 16.7 months, 95% CI 0.38–0.64), co-detection (1.73, 1.07–2.79) and lactate dehydrogenase (0.66 per 66.6 U/L, 0.51–0.86) were retained; AUC 0.708 (0.653–0.760), optimism-corrected 0.687, slope 1.00, Brier 0.158. Prolonged stay fell from 32.1% (62/193) to 15.5% (44/283) (p < 0.001). On transport discrimination was preserved (AUC 0.701, 0.623–0.772; slope 1.07) but risk was overestimated (intercept −0.82; predicted 28.3% vs observed 15.5%), and the Brier score (0.142) exceeded the constant-risk benchmark (0.131) until recalibrated (0.124). At a 20% threshold, 33.9% of 2025 admissions were low risk (negative predictive value 95.8%; 14 events missed per 1000). Conclusions. Three routine admission variables predicted prolonged RSV stay with moderate discrimination, preserved across calendar years, whereas absolute risk was not and needs recalibration before use. The model suits ruling out a long stay, not escalation.