LAUSR

Introduction: Idiopathic pulmonary arterial hypertension (iPAH) is a rare, life-shortening disease. Symptoms are non-specific with diagnosis usually made at a late stage. Aims and objectives: The study aim was to develop a predictive algorithm (Sheffield Pulmonary Hypertension IndeX, SPHInX) to identify iPAH, with a diagnostic performance that would allow large population screening. Methods: Data collected at Sheffield for iPAH patients (n=852) were linked with pre-diagnosis National Health Service (HES) records capturing in-, outpatient, accident and emergency attendances. Non-iPAH patients (n=11,209,000) were selected from the HES population, enriched by excluding those without selected ICD-10 codes felt to make a diagnosis of iPAH unlikely. Patient history was limited to ≤5 yrs pre-diagnosis. Demographics, timing/frequency of diagnoses, medical specialities visits and procedures undertaken were captured. For predictive modelling a gradient boosting tree with bootstrap aggregation was used to discriminate between iPAH and non-iPAH. Results: To identify 100 iPAH patients, 969 patients would need to be screened: specificity 99.99%, sensitivity 14.1%, positive, negative predictive values 10.3%; 99.99%, respectively. The timing/frequency of hospital visits, the speciality seen, age were key drivers of model performance. Conclusion: Predictive algorithms using existing and accessible real-world data have the potential to determine patients at high risk of having iPAH. Studies to validate this approach to screen for iPAH in the general population are now warranted.