{
  "abstract": "Background Pulmonary hypertension is a rare, progressive and life-threatening condition, which is subject to delayed diagnosis due to its non-specific symptoms. Machine learning (ML) when applied to electronic health records (EHR) may aid earlier detection of PH. This study evaluated whether ML models could accurately classify PH using OCD-10 codes and identify comorbidities with strong predictive contributions using the UK Biobank dataset.Methods The UK Biobank dataset was pre-processed to extract participants with 2 or more of 20 predefined conditions associated with PH, creating an unmatched cohort (n=152,853). PH cases were identified by the ICD-10 codes (I27.0, I27.2, I27.9) and matched 1:1 with controls based on age and sex (n=5,142). Three supervised ML models: K-Nearest Neighbour (KNN), Support Vector Machine (SVM) and Random Forest (RF), were trained on an 80% of the matched cohort. Shapley Additive exPlanations (SHAP) was applied for both feature selection and final model interpretation. A temporal analysis of comorbidity diagnosis timing was conducted for PH cases, prior to diagnosis.Results The SVM model using 12 features achieved the best performance with (AUC 0.88, sensitivity 0.82 and accuracy 0.82). Key predictive comorbidities included left heart failure, restrictive airway disease, atrial fibrillation, valvular disease, pulmonary embolism and chronic obstructive pulmonary disease. Temporal analysis showed that many predictive comorbidities emerge up to 3 years prior to PH diagnosis, with increasing prevalence closer to diagnosis date.Conclusions ML models can accurately classify PH using fewer, clinically relevant features. SHAP improves model transparency and identifies high-impact predictors. The findings support the feasibility of developing early screening tools for PH based on comorbidity profiles. Future work should include more diverse EHR features and externally validate models with alternate datasets.",
  "authors": [
    {
      "affiliations": [
        "Faculty of Medicine, National Heart and Lung Institute, Imperial College London, London, UK"
      ],
      "name": "NB Bhamji"
    }
  ],
  "title": "S80 Exploring machine learning techniques for classifying pulmonary hypertension using UK Biobank electronic health records",
  "uid": "a4a2daff-fdb4-5beb-a930-62e8c41574ee"
}