Distinguishing common respiratory pathogens using machine learning of symptom profiles to prioritize diagnostic testing.

Clinical symptoms are critical for diagnosing and managing upper respiratory tract infections, yet systematic comparisons across common pathogens remain limited. We aimed to characterize symptom signatures across common respiratory pathogens and to develop approaches that support the prioritization of diagnostic testing across pathogens. Using a large-scale, home-based multiplex PCR testing dataset, we characterized symptom profiles across common respiratory pathogens stratified by age and sex. Hierarchical clustering was applied to group pathogens into symptom-based categories, and a multilayer perceptron (MLP) model was trained to predict these cluster-level categories. A post-hoc refinement combined MLP outputs with contemporaneous epidemiological data to further prioritize likely pathogens within predicted categories. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC), classification accuracy, and related metrics. Symptom profiles overlapped substantially across pathogens, with differences being more categorical than pathogen-specific: RSV, HCoV, HPIV, and HRV were predominantly characterized by respiratory symptoms, whereas influenza A/B and SARS-CoV-2 showed more pronounced systemic manifestations. The MLP model achieved high AUCs (0.80-0.92, depending on age group) with specificities exceeding 75% but lower sensitivities, reflecting a conservative prediction pattern. Integrating cluster-level predictions with contemporaneous epidemiology improved the prioritization of prevalent pathogens, though performance remained limited for less common ones. When expanding predictions to the top three likely pathogens, the model enabled diagnostic test prioritization with an overall accuracy of 0.83 (95% CI, 0.82-0.84). Our study supports the use of readily available symptom data to inform the prioritization of diagnostic testing, particularly during outbreaks or in resource-limited settings where timely laboratory testing is constrained. In addition, the systematic symptom signatures identified across common respiratory pathogens may inform the development and optimization of symptom-based patient-reported outcomes for future antiviral trials.