Clinically applicable machine learning prediction model for pulmonary hypertension due to left heart disease

Current non-invasive prediction tools for pulmonary hypertension due to left heart disease (PH-LHD) in suspected pulmonary arterial hypertension (PAH) patients lack sensitivity. We hypothesized that machine learning (ML) can improve the prediction of PH-LHD in a mixed population of PAH and PH-LHD patients.To build the ML model, potential non-invasive PH-LHD predictors, including demographics, medical history, echocardiographic, lung function test, lab and ECG variables, were recorded from medical files of 213 PAH and 174 PH-LHD patients from the University Hospitals of Leuven PH centre database. The dataset was randomly split (70:30) in a training and a test dataset, to evaluate the performance of the model on unseen data. The Jacobs score was used as a benchmark to compare the performance of the ML model (Jacobs W. et al, ERJ 2015). Predictive accuracy was assessed by area under the receiver operating curve (AUROC); and sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) were calculated.Ten-fold cross-validation showed that Extreme Gradient Boosting had the best accuracy among all explored ML techniques. In the independent test dataset (N=117), the model correctly diagnosed 63% (sensitivity: n= 37/59) of PH-LHD patients, with a PPV of 100%, a NPV of 64% and 100% specificity. The model outperformed the Jacobs score that identified 19% (n= 11/59) of the patients with PH-LHD without false positives. The AUROC of the ML model was 0.98.ML...
Source: European Respiratory Journal - Category: Respiratory Medicine Authors: Tags: Pulmonary hypertension Source Type: research