Novel Liver Injury Phenotypes and Outcomes in Clinical Trial Participants with Pulmonary Hypertension

Abstract

Rationale: Pulmonary arterial hypertension (PAH) and chronic thromboembolic pulmonary hypertension (CTEPH) cause right ventricular dysfunction, which can impact other solid organs. However, the profiles and consequences of hepatic injury resulting from PAH and CTEPH have not been well studied. Objectives: We aimed to identify underlying patterns of liver injury in a cohort of patients with PAH and CTEPH enrolled in 15 randomized clinical trials conducted between 1998 and 2014. Methods: We used unsupervised machine learning to identify liver injury clusters in 13 trials and validated the findings in two additional trials. We then determined whether these liver injury clusters were associated with clinical outcomes or treatment effect heterogeneity. Measurements and Main Results: Our training dataset included 4,219 patients and our validation dataset included 1,756 patients with serum total bilirubin, alkaline phosphatase, aspartate aminotransferase, alanine aminotransferase, and albumin data. Using k-means clustering, we identified phenotypes with no liver injury, hepatocellular injury, cholestatic injury, and combined injury patterns. Patients in the cholestatic injury liver cluster had the shortest time to clinical worsening and the highest risk of mortality. The cholestatic injury group also experienced the greatest placebo-corrected treatment effect on 6-minute-walk distance. Randomization to the experimental arm transitioned patients to a healthier liver status. Conclusions: Liver injury was associated with adverse outcomes in patients with PAH and CTEPH. Randomization to active treatment had beneficial effects on liver health compared with placebo. The role of liver disease (often subclinical) in determining outcomes warrants prospective studies.

Keywords: liver injury phenotypes; machine learning; pulmonary hypertension.

Figure 1.

Participant flowchart. FDA = U.S. Food and Drug Administration; PH = pulmonary hypertension.

Figure 2.

k-Means clustering results in factor dimensional space. AP = alkaline phosphatase.

Figure 3.

Changes in liver injury phenotype from baseline to Week 12 or 16, by assignment to active versus control arms. Liver injury phenotype modeled as an ordinal variable (from better to worse): no liver injury > hepatocellular injury pattern > cholestatic injury pattern > combined injury pattern. The overall effect of pulmonary arterial hypertension treatment on the change in liver injury phenotype from baseline to 12 or 16 weeks was assessed using ordinal regression (odds ratio for change to more severe liver injury phenotype for active drug vs. control, 0.67; 95% confidence interval, 0.56, 0.81; P < 0.001). The model was adjusted for baseline phenotype, age, sex, baseline 6-minute-walk distance, pulmonary hypertension etiology, and body mass index as fixed effects and trial as a random effect.

Figure 4.

Kaplan-Meier curves for (A) time to clinical worsening and (B) overall survival.

Figure 5.

Treatment assignment by liver injury phenotype interaction in terms of (A) change in 6-minute-walk distance at 12 or 16 weeks, (B) time to clinical worsening, (C) time to death, and (D) worsened World Health Organization functional class at 12 or 16 weeks. All models were adjusted for age, sex, body mass index, pulmonary arterial hypertension etiology, and baseline 6-minute-walk distance (except for baseline World Health Organization functional class in that model). aHR = adjusted hazard ratio; aOR = adjusted odds ratio; CI = confidence interval.