A Bayesian network for identifying causes of breathlessness using a national electronic medical records (EMR) database.

Breathlessness is a common symptom linked to various respiratory and cardiac conditions, often causing delays in diagnosis and management in primary care. This study aimed to develop an algorithm using Bayesian Networks (BN) to predict ten conditions associated with presentations of breathlessness. The study included individuals aged 16 years and older with breathlessness recorded as the reason for visiting a general practitioner (GP) between 2002 and 2024 in 50 practices in the United Kingdom (UK). Thirty-four characteristics including demographic, symptoms, comorbidities and medication histories were considered as predictors. A total of 384,994 breathlessness episodes from 136,215 patients were analysed, with 55% being female and 42% aged over 60 years. The BN achieved good performance with area under the curve (ROC-AUC) values ranging from 0.68 for non-pneumonia lower respiratory tract infections (LRTI) to 0.94 for heart failure. The algorithm demonstrated strong predictive performance and could help GPs prioritise early and targeted diagnostic tests for the most likely causes of breathlessness, potentially reducing both diagnostic delays and costs.