Risk factors for invasive mechanical ventilation, extracorporeal membrane oxygenation, and mortality in children with severe adenovirus infection in the pediatric intensive care unit: a retrospective study.

Abstract

Background: Adenovirus infection causes considerable morbidity and mortality in pediatric patients, primarily those affected by severe respiratory system involvement. Although prevalent, it often presents vague indications, making accurate diagnosis and management challenging. This study aims to set some risk factors for invasive mechanical ventilation, ECMO, and mortality in children with severe adenovirus infection admitted to PICU.

Methods: We evaluated 66 children with severe adenovirus infection admitted to the PICU of Children's Hospital, Zhejiang University School of Medicine, from 2018 to 2019. Data on general conditions, clinical manifestations, laboratory findings, pathogenetic and radiological discoveries, treatments, therapeutic efficacy, and outcomes were collected. Machine learning models were used to predict the need for invasive mechanical ventilation, ECMO, and mortality.

Results: Of the 66 patients, 5 died, and 61 survived. Significant factors related to mortality included heart failure (p = 0.005), pericardial effusion (p = 0.032), septic shock (p = 0.009), hemoglobin levels (p = 0.013), lactate dehydrogenase (p = 0.022), albumin (p = 0.035), normal creatinine levels (p = 0.037), and pneumothorax (p = 0.002). Additional risk factors for invasive mechanical ventilation included acute respiratory distress syndrome and encephalopathy. Low breath sounds were identified as a risk factor for ECMO. For predicting poor outcomes, including invasive mechanical ventilation, ECMO, or mortality, the random forest model using these factors demonstrated high accuracy, with an area under the curve of 0.968.

Conclusions: The study indicates poor prognosis in children with severe adenovirus infection is significantly related to comorbidities and clinical symptoms. Machine learning models can accurately predict adverse outcomes, providing valuable insights for management and treatment. Identifying high-risk patients using these models can improve clinical outcomes by guiding timely and appropriate interventions.

Trial registration: The article is a retrospective study without a clinical trial number, so it is not applicable.