Comparison of the efficacy of different models in predicting the progression of Pulmonary Hypertension

Abstract

Abstract Objective To identify the risk factors associated with the progression of pulmonary hypertension (PH), develop two distinct risk prediction models, and provide valuable insights for clinical management. Methods This study employed a retrospective analysis to examine the clinical data of 346 individuals diagnosed with PH by transthoracic echocardiography (TTE). The participants were allocated randomly to either a training set (n = 243) or a validation set (n = 103) at a 7:3 ratio. Subsequently, the individuals were further categorized into the control and case groups according to PH progression. The training set was utilized to perform single- and multifactor logistic regression analysis, as well as random forest feature priority ranking, to determine the most effective predictive variables. Subsequently, logistic regression and random forest models were developed. The performance of both models was evaluated and compared based on the validation set, using the receiver operating characteristic (ROC) curve, accuracy, sensitivity, specificity, positive predictive value, and negative predictive value. Results Logistic regression analysis revealed that World Health Organization Function Class (WHO FC), tricuspid annular plane systolic excursion (TAPSE)/pulmonary artery systolic pressure (PASP), right atrial diameter (RAD)/left atrial diameter (LAD), right ventricular end-diastolic diameter (RVDd)/left ventricular end-diastolic diameter (LVDd), main pulmonary artery (MPA), MPA/ascending aorta (AA), MPA/descending aorta (DA), red blood cell distribution width (RDW)-coefficient of variation (RDW-CV), Neutrophil-to-Lymphocyte Ratio (NLR), N-Terminal Pro-B-Type Natriuretic Peptide (NT-proBNP) and D-dimer were risk factors for PH progression. Among these, WHO FC, TAPSE/PASP, RVDd/LVDd, MPA/AA, and NT-proBNP were independent risk factors for PH progression. The random forest model identified the top five predictors of PH progression as TAPSE/PASP, MPA/AA, RVDd/LVDd, NT-proBNP, and NLR. The area under the ROC curve (AUC), accuracy, sensitivity, specificity, positive predictive value, and negative predictive value of the logistic regression and random forest model were 0.958 (95% CI: 0.919–0.997) and 0.959 (95% CI: 0.921–0.997), 93.2% and 92.23%, 90.91% and 90.91%, 94.92% and 93.22%, 93.02% and 90.91%, 93.33% and 93.22% respectively. Conclusions Both the logistic regression and the random forest model demonstrated significant predictive power for PH progression, providing clinical utility in identifying high-risk patients and implementing effective interventions to prevent PH progression in clinical practice.