Background: Risk assessment is crucial to predict survival and then guide therapeutic management for pulmonary arterial hypertension (PAH) patients. Several risk assessment methods have been developed, but the comparison of their performances using the same dataset of PAH patients remains to be investigated.
Aim: In the same patient’s dataset, to compare the performance of the main PAH risk assessment methods used nowadays.
Methods: We used a harmonized dataset of patients from several PAH trials. First, we built a new version of the Pulmonary Hypertension Outcome and Risk Assessment (PHORA) model, (PHORA 2.1) using random forest and Bayesian network analysis, as previously published. In the same dataset, the Registry to Evaluate Early and Long-term PAH Disease Management (REVEAL) 2.0 and REVEAL Lite 2.0, 3- and 4-strata Comparative Prospective Registry of Newly Initiated Therapies for Pulmonary Hypertension (COMPERA) 2.0 and non-invasive French Pulmonary Hypertension Registry (FPHR) risk assessment methods were applied. The discrimination performance to predict 1-year survival was compared between all these methods using the area under the curve (AUC) after fivefold cross-validation and the Wilcoxon signed rank test.
Results: A dataset of 494 PAH patients was used for this analysis (mean age 47 ± 18 years old, 77% female, 62% idiopathic or heritable PAH, mean pulmonary artery pressure 53 mmHg). The PHORA 2.1 method included 13 variables, with an AUC of 0.84 compared with 0.71, 0.78, 0.72, 0.81 and 0.80 for non-invasive FPHR, 4-stata and 3-strata COMPERA, REVEAL 2.0 and REVEAL Lite 2.0, respectively, after the fivefold cross-validation. In general, the 95% confidence interval AUC of PHORA 2.0 covers higher values in general, with a Wilcoxon signed rank test significantly higher compared with all the other methods (P < 0.001).
Conclusion: The PHORA 2.1 model, based on a Bayesian network analysis, depicted the highest AUC to predict the outcome, compared with other PAH risk assessment methods widely used.
KEY CONTRIBUTORS
Charles Fauvel, MD; Zilu Liu, MS; Shili Lin, PhD; Priscilla Correa-Jaque; Amy Webb, PhD; Rebecca Vanderpool, PhD; Manreet Kanwar, MD; Jidapa Kraisangka, PhD; Puneet Mathur, MD MBA; Adam Perer, PhD; Allen D.Everett, MD; Raymond Benza, MD Charles Fauvel*, Internal Medicine Department, Division of Cardiovascular Medicine, Wexner Medical Center, The Ohio State University, Columbus, OH, USA Zilu Liu*, Shili Lin, Department of Statistics, The Ohio State University, Columbus, OH, USA Priscilla Correa-Jaque, Internal Medicine Department, Division of Cardiovascular Medicine, Wexner Medical Center, The Ohio State University, Columbus, OH, USA Amy Webb, Department of Biomedical Informatics, The Ohio State University, Columbus, OH, USA Rebecca Vanderpool, Internal Medicine Department, Division of Cardiovascular Medicine, Wexner Medical Center, The Ohio State University, Columbus, OH, USA Manreet Kanwar, Temple University School of Medicine, Philadelphia PA, USA Jidapa Kraisangka, Mahidol University, Thailand Puneet Mathur, Research Information Technology, The Ohio State University, Columbus, OH, USA Adam Perer, Carnegie Mellon University, Pittsburgh, PA, USA Allen D. Everett, The John Hopkins University, Baltimore, MD, USA Raymond Benza, Internal Medicine Department, Division of Cardiovascular Medicine, Wexner Medical Center, The Ohio State University, Columbus, OH, USA
