73 Leveraging data-driven machine learning for enhanced paediatric case definitions in Severe Acute Respiratory Infections (SARI) surveillance

Abstract Background Acute respiratory infections are one of the leading causes of death globally in both children and adults. The World Health Organization (WHO) has developed a standardized case definition for Severe Acute Respiratory Infection (SARI) to detect and monitor respiratory virus outbreaks. However, previous studies have demonstrated limited accuracy of the WHO SARI case definition to detect viral respiratory infections in children. Objectives To develop a machine learning (ML)-based clinical prediction model/case definition to detect SARI in children and youth hospitalized with an acute respiratory infection using machine learning. Design/Methods Children and youth under 18 years hospitalized with a suspected or confirmed respiratory infections at two large Canadian children's hospitals were included. Demographics, clinical symptoms, and microbiologic testing were extracted and preprocessed to create model inputs. Acute respiratory infections were defined through microbiological viral tests. SARI case definition was developed using an L1-regularized logistic regression (LASSO) approach to assess positive respiratory virus results for hospitalized children. The sample was split randomly into 70% training and 30% testing sub-samples through set.seed() function in R. We considered 18 socio-demographic and symptom indicators for the model. The ML-based case definition was trained using 10-fold cross-validation on the training set, and performance metrics, including model discrimination, diagnostic accuracy, sensitivity and specificity, were assessed on the test set. Results Overall, 1887 participants (1135 male, 752 female) were included in the analysis, with a median age of 2.5 years. Most (86%) of participants had a positive viral test, of which 38% were positive for RSV. The most common reported symptoms included cough (83%), increased work of breathing (68%), reported fever (70%), and nasal congestion (61%). The final case definition included following variables: reported fever, cough, nasal congestion, dehydration, increased work of breathing, tachypnea, irritability, poor feeding, wheezing and vomiting. Cough and sore throat were the most important positive and negative predictors, respectively. The model achieved an accuracy of 73% (95% CI: 70-75%), sensitivity of 77% (95% CI 74-79%) and specificity of 47% (95% CI 39-55%) maximizing Youden index, and area under the receiver operating characteristics curve (AUC) of 64% (95% CI 59-69%). Conclusion The novel paediatric ML-based SARI case definition had superior performance compared to the WHO SARI case definition. This study highlights a promising opportunity to use machine learning to generate case definitions which could be embedded into electronic medical records to enhance surveillance for novel and emerging infections globally in paediatrics.

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.