Virtual PAH Trials Reproduce Treprostinil Response Trends but Predict Significance Too Early
A United Therapeutics systems-pharmacology model linked pulmonary vascular biology to simulated six-minute walk responses. It reproduced several oral treprostinil trial trends, but underestimated later variability and predicted a significant response earlier than the FREEDOM-EV trial observed.
Turning Pulmonary Vascular Pathways Into a Virtual Population
United Therapeutics researchers developed a quantitative systems pharmacology model to examine how trial design and patient composition affect oral treprostinil responses in pulmonary arterial hypertension, or PAH. The model represents endothelial and smooth muscle signaling rather than a digital reconstruction of individual patients. Its principal trial outcome was change in six-minute walk distance.
The biological model links prostacyclin, nitric oxide and endothelin-1 pathways to smooth muscle contraction and proliferation. These processes change pulmonary vascular resistance. A separate regression then translates vascular resistance into six-minute walk distance, with added measurement variability and a placebo-response term. That regression used baseline patient data from FREEDOM-EV, a distinction relevant to the later use of that trial as a test case.
The investigators generated 5,000 sets of patient-specific parameters. After excluding sets that produced walking distances outside 100–600 m, 4,909 remained as a plausible population. The varying features included mediator production, receptor expression, smooth muscle cell biology and time since diagnosis. These are simulated physiological differences, not measurements collected from 4,909 trial participants.
Prevalence weights were fitted to five published trials covering treprostinil, a phosphodiesterase-5 inhibitor and an endothelin receptor antagonist, used alone or in combination. Weighting was performed separately for functional classes I/II and III/IV. Trial inputs then specified patient numbers, background therapy, disease-severity composition and dose-titration patterns.
Fitted Trials and a Later Test Case Answered Different Questions
For each fitted trial, the researchers sampled 100 virtual trials. In the FREEDOM-M monotherapy comparison, simulated 12-week treatment effects ranged from +6 to +39 m, with a median of +25 m. The published clinical estimate was +23 m, with a 95% confidence interval of 4–41 m. The simulated confidence intervals were somewhat narrower than the clinical intervals.
The model also reproduced the relatively small effects in the fitted FREEDOM-C and FREEDOM-C2 combination trials. Simulated probabilities of a statistically significant effect were 31% for both, consistent with the actual trials’ failure to show significance on the walking-distance endpoint. Agreement in these comparisons describes calibration performance because these trials contributed to prevalence-weight fitting.
FREEDOM-EV was evaluated after fitting those weights. It differed from the earlier combination trials through three-times-daily treprostinil, a single background therapy, a larger proportion of less severe functional-class patients and longer follow-up. It was a later test of the weighted trial workflow, although its baseline data had already informed the resistance-to-walking-distance relationship.
The simulated probabilities of significance at weeks 4, 8, 12, 24, 36 and 48 were 6%, 43%, 66%, 99%, 99% and close to 100%, respectively. The actual trial showed significance at weeks 36 and 48, but not at weeks 4–24. The model captured the increasing treatment-effect trend while predicting success too early. Its virtual population had less noise at later observations, inflating the estimated probability of significance.
Enrichment Scenarios Remained Computational Proposals
The researchers also examined which virtual physiological features were associated with greater treprostinil benefit. Higher endothelin-1 production and lower prostacyclin production identified a more responsive simulated population. They resampled 1,000 trial realizations per scenario to explore these features together with reduced use of dual background therapy.
At the real FREEDOM-C and FREEDOM-C2 trial sizes of 350 and 310, the modeled proportions of successful trials increased from 23% and 22% to 81% and 75% when enrichment and reduced dual background therapy were combined. These scenario results used a different resampling analysis from the earlier 100-trial calibration comparisons.
The authors explicitly state that the proposed mediator-based enrichment variables are not clinically actionable biomarkers. Circulating endothelin-1 may not reflect local pulmonary concentrations, and prostacyclin degrades rapidly. The scenarios therefore illustrate a trial-design analysis rather than a validated patient-selection rule.
Other constraints include a single-compartment pulmonary circulation, incomplete disease pathways and limited patient-specific pharmacokinetics. In the current model, a higher treprostinil dose always produces higher exposure; actual absorption and metabolism can vary among patients. Code, virtual-population simulations and weight vectors were available only on reasonable request at publication. The reported work supports conditional exploration of trial design, with uncertainty in both physiological translation and outcome variability.