Hai-Feng Liu ¹ Xue-Zu Zhang ² Cong-Yun Liu ³ Wang Li ⁴ Wen-Hong Li ⁵ Ya-Yu Wang ⁶ He- Yun Li ⁷ Mei Xiang ⁸ Rui Lu ⁹ Ting-Yun Yuan ¹ Hong-Min Fu ^1*

• ¹ Kunming Medical University, Kunming, Yunnan Province, China
• ² The People’s Hospital of Lincang, Lincang, China
• ³ The People’s Hospital of Baoshan, Baoshan, China
• ⁴ The Fifth People’s Hospital of Kunming, Kunming, China
• ⁵ The People’s Hospital of Lufeng, Lufeng, China
• ⁶ The Third Affiliated Hospital of Dali University, Dali, China
• ⁷ Zhaotong Hospital Affiliated to Kunming Medical University, Zhaotong, Yunnan, China
• ⁸ The People’s Hospital of Honghe, Honghe, China
• ⁹ The People’s Hospital of Wenshan Zhuang & Miao Autonomous Prefecture, Wenshan, Yunnan, China

Off-season upsurge of respiratory syncytial virus (RSV) infection with changed characteristics and heightened clinical severity during the post-COVID-19 era are raising serious concerns. This study aimed to develop and validate a nomogram for predicting the risk of severe acute lower respiratory tract infection (SALRTI) in children hospitalized for RSV infection during the post-COVID-19 era using machine learning techniques. Methods: A multicenter retrospective study was performed in nine tertiary hospitals in Yunnan, China, enrolling children hospitalized for RSV infection at seven of the nine participating hospitals during January-December 2023 into the development dataset. Thirty-nine variables covering demographic, clinical, and laboratory characteristics were collected. Primary screening and dimension reduction of data were performed using Least Absolute Shrinkage and Selection Operator (LASSO) regression, followed by identification of independent risk factors for RSV-associated SALRTI using Logistic regression, thus finally establishing a predictive nomogram model. Performance of the nomogram was internally evaluated by receiver operating characteristic (ROC) curve, calibration curve, and decision curve analysis (DCA) based on the development dataset. External validation of our model was conducted using same methods based on two independent RSV cohorts comprising pediatric RSV inpatients from another two participating hospitals between January-March 2024. Results: The development dataset included 1102 patients, 239 (21.7%) of whom developed SALRTI; while the external validation dataset included 249 patients (142 in Lincang subset and 107 in Dali subset), 58 (23.3%) of whom were diagnosed as SALRTI. Nine variables, including age, preterm birth, underlying condition, seizures, neutrophil-lymphocyte ratio (NLR), interleukin-6 (IL-6), lactate dehydrogenase (LDH), D-dimer, and co-infection, were eventually confirmed as the independent risk factors of RSV-associated SALRTI. A predictive nomogram was established via integrating these nine predictors. In both internal and external validations, ROC curves indicated that the nomogram had satisfactory discrimination ability, calibration curves demonstrated good agreement between the nomogram-predicted and observed probabilities of outcome, and DCA showed that the nomogram possessed favorable clinical application potential. Conclusion: A novel nomogram combining several common clinical and inflammatory indicators was successfully developed to predict RSV-associated SALRTI. Good performance and clinical effectiveness of this model were confirmed by internal and external validations.