J.V. Scott, L.C. Lohmueller , C.G. Garnett , M.K. Kanwar , R.L. Benza. A105 GLORY DAYS: THE LATEST CLINICAL RESEARCH IN PAH / Poster Discussion Session / American Thoracic Society 2019;199:A7346
Rationale: The Registry to Evaluate Early and Long-Term PAH Disease Management recently updated their linear multivariable risk-stratification tool (REVEAL 2.0). Our ongoing study aims to develop and validate an improved prognostic model (PHORA) that determines a patient’s probability of one-year (1-yr) survival by applying Bayesian machine-learning to REVEAL data. For this analysis, we compared the performance of PHORA to REVEAL 2.0 on AMBITION clinical trial data. Methods: Both risk models were applied retrospectively to AMBITION and performance was assessed by: 1) prediction of 1-yr survival using Area Under the Curve (AUC) 2) detection of increased risk of death using a log-rank test, 3) detection of a significant treatment effect with combination therapy versus monotherapy using Cochran-Mantel-Haenszel. Risk assessments were performed both on trial baseline values and on 16-week reassessment.
Results: Both models perform similarly at baseline assessment for 1-yr survival, but PHORA significantly outperforms REVEAL 2.0 at 16-week reassessment (AUC 0.80 versus 0.73, respectively). Change in either REVEAL or PHORA risk achieved a significant hazard ratio for time to death at last follow-up. However, an increase of at least 1-point in REVEAL did not translate to a higher hazard ratio for time to death (p = 0.10), whereas a decrease in PHORA risk of at least 0.8% did translate (p = 0.04). Unlike REVEAL 2.0, PHORA detected a significant treatment effect: a higher proportion of patients on combination therapy (versus monotherapy) were able to substantially decrease 16-week PHORA risk relative to baseline (OR = 1.55, p = 0.045) and the effect was confounded by baseline PHORA risk (low risk OR = 1.18, average OR = 1.69, high OR = 1.80). Using the original trial endpoint, time to clinical failure, average and high-risk groups saw significantly lower hazard ratios on combination therapy compared to monotherapy, but low risk patients did not experience a significantly lower hazard ratio on combination therapy versus monotherapy (Figure 1).
Conclusion: A Bayesian model for PAH prognosis demonstrated an improvement in accuracy, sensitivity for change, and ability to detect a treatment effect in a clinical trial setting over the existing REVEAL 2.0 model. PHORA’s ability to identify a low-risk group that did not benefit from combination therapy versus monotherapy (per original trial endpoint) shows promise that Bayesian models are viable for PAH clinical trial enrichment. Further, such a risk assessment tool could assist clinical decision-making regarding upfront treatment benefit versus cost and/or potential for adverse effects.
This abstract is funded by: Food and Drug Administration
https://www.atsjournals.org/doi/abs/10.1164/ajrccm-conference.2019.199.1_MeetingAbstracts.A7346