Simulation of clinical trials of oral treprostinil in pulmonary arterial hypertension using a virtual population

Abstract

Challenges in drug development for rare diseases such as pulmonary arterial hypertension can be addressed through the use of mathematical modeling. In this study, a quantitative systems pharmacology model of pulmonary arterial hypertension pathophysiology and pharmacology was used to predict changes in pulmonary vascular resistance and six-minute walk distance in the context of oral treprostinil clinical studies. We generated a virtual population that spanned the range of clinical observations and then calibrated virtual patient-specific weights to match clinical trials. We then used this virtual population to predict the results of clinical trials on the basis of disease severity, dosing regimen, time since diagnosis, and co-administered background therapies. The virtual population captured the effect of changes in trial design and patient subpopulation on clinical response. We also demonstrated the virtual trial workflow's potential for enriching populations based on clinical biomarkers to increase likelihood of trial success.

Fig. 1. QSP model overview.

The model consists of endothelial and smooth muscle cell compartments. Communication between cells occurs through endothelial cell production of PGI₂, NO, and ET-1. Within smooth muscle cells, these mediators modulate actin-myosin coupling and the cell proliferation rate. These two factors control PVR. ITGβ1 integrin β1, MSCC mechanosensitive calcium channels, PKC protein kinase C, eNOS endothelial nitric oxide synthase, NO nitric oxide, TRE treprostinil, PGI₂ prostacyclin, ET-1 endothelin-1, ERA endothelin receptor antagonist, PDE5 phosphodiesterase-5, PDE5i phosphodiesterase-5 inhibitor, sGC soluble guanylate cyclase, IP₃ inositol trisphosphate, ROCC receptor operated calcium current, VDCC voltage-dependent calcium current, RyR ryanodine receptor, IP₃R IP₃ receptor, MLCK myosin light chain kinase, MLCP myosin light chain phosphatase, M myosin, M_p phosphorylated myosin, A actin, SMC smooth muscle cell, F force, PVR pulmonary vascular resistance.

Fig. 2. Virtual trial workflow.

The 4 steps of workflow are listed in light orange boxes, with the outputs from each step in dark orange boxes. Sample data is given to illustrate each step’s output for 4 virtual patients (VPs). Trial IDs indicate the trials (if any) in which each VP is a member.

Fig. 3. Effect of population variability on pulmonary vascular resistance.

a Pearson correlation coefficients between population variables and change in PVR across the 4909 patients in the virtual population are shown, with red indicating positive correlation and blue indicating negative correlation. Correlation values are shown for those whose absolute values exceed 0.2. Scatter plots show the correlation of ET-1 production rate with placebo- (b) and ERA-induced (c) effects on PVR. Colors indicate local densities of points.

Fig. 4. Population variability prior to calibration.

Pulmonary vascular resistance (PVR; left column, green), six-minute walk distance (6MWD) calculated from PVR only (middle column, gray), and 6MWD calculated from PVR and corrected for sampling noise (right column, blue). White solid lines indicate virtual population median values, while white upper and lower dashed lines indicate virtual population 75^th and 25^th percentiles, respectively. Box plots for placebo- and treprostinil-treated adjusted 6MWD indicate the median, 75^th, and 25^th percentiles from clinical data gathered in the FREEDOM-M study at weeks 0, 4, 8, and 12.

Fig. 5. Calibration of virtual population with clinical data.

a Differences of ET-1 production rate, and number of SMCs between WHO functional class categories for 100 sampled FREEDOM-M virtual trials. Positive differences indicate larger values in FC III/IV than I/II. Bolded colors represent significant differences (Wilcoxon rank-sum test p < 0.05). Values for ET-1 production rate are shown as log10 differences. b Hodges–Lehmann median difference in 6MWD between treatment and placebo arms for 100 sampled FREEDOM-M trials at week 12 (median difference shown with gray dots, horizontal bars correspond to 95% confidence intervals). Trials are sorted by the lower bound of the 95% CI, and the median sampled trial is highlighted in blue. Results from the FREEDOM-M trial are shown in red. c Hodges–Lehmann median difference in 6MWD between treatment and placebo arms for all fitted trials. Gray regions correspond to 95%CI across 100 trials, with the median trial in blue. Trial data is shown in red. d Probability of a significant 6MWD treatment effect in our sampled virtual trials for each fitted arm. Significance of the real trial is indicated by orange (significant) or blue (not significant).

Fig. 6. Prediction of FREEDOM-EV clinical data.

6MWD treatment effects for 100 sampled virtual trials corresponding to the combination therapy trial FREEDOM-EV. Treatment effects are estimated with mixed model for repeated measures (MMRM). Gray regions correspond to 95% CI across 100 trials, with the median trial in blue. Clinical trial data is shown in red.

Fig. 7. Patient response by trial inputs.

a Pearson correlation coefficients between population variables and change in PVR relative to placebo across the 4909 patients in the FREEDOM-C, -C2, and -EV plausible populations are shown, with orange indicating positive correlation and blue indicating negative correlation. Correlation values are shown for those whose absolute values exceed 0.2. Population enrichment strategies in which virtual patients have been filtered according to ET-1 and PGI₂ production rates (enriched population), with removal of dual (ERA and PDE5i) background therapy in resimulations of the FREEDOM-C (b) and -C2 (c) trials.