Overview

A Markov state transition model was developed to replicate the disease course of PAH and compare long-term health outcomes associated with immediate treatment initiation with sotatercept plus BGT vs delayed treatment initiation with sotatercept plus BGT in adult patients with symptomatic PAH. This model was based on a previously published population model in PAH and simulated patients’ clinical journeys in line with the definition of the intent-to-treat (ITT) population of STELLAR [12]. STELLAR data were analyzed to determine patients’ risk strata distribution at baseline and their transition probabilities across risk strata at weeks 3, 12, and 24. Furthermore, data from the COMPERA PAH registry were used to inform patients’ mortality and transplantation rates. This study relied on previously conducted studies and did not involve any new studies requiring the direct participation of human subjects; thus, ethics board approval was not required. The authors sought and received permission from the registry owners to access and use data from the COMPERA PAH registry.

The model adopted a lifetime horizon, which was assumed to be 30 years, and utilized a cycle length of 3 weeks for the first cycle, 9 weeks for the second cycle, and 12 weeks for the third and subsequent cycles, corresponding to the visit schedule in STELLAR. All model outcomes were discounted at 3% annually beyond the first model cycle [16].

Model Framework

A Markov-type model was developed that consisted of six mutually exclusive health states: low risk, intermediate-low risk, intermediate-high risk, high risk, lung/heart-lung transplant, and death (Fig. 1). These health states were established to align with the refined four-strata risk assessment, which has been validated across multiple PAH registries [17,18,19] and is recommended in the 2022 European Society of Cardiology (ESC) and European Respiratory Society (ERS) treatment guideline for PAH [20]. Patients in higher-risk strata have a heightened risk of mortality, PAH hospitalization, and lung transplant [20].

Fig. 1

At the model’s start, all patients were assigned to one of the four risk strata health states (Tables 1 and 2). Over each model cycle, patients could either remain in their current health state, move between risk strata health states, or progress to the lung/heart-lung transplant or death states, governed by a set of transition probabilities obtained from STELLAR (Table 1). The model tracked the patient cohort as they moved between these health states. Patients accrued LYs over time, determined by their health state occupancy during each model cycle.

Table 1 Distribution of risk as measured in STELLARTable 2 Model input parametersModel ParametersPopulation Characteristics

The baseline patient characteristics of the simulated patient population are detailed in Table 2. Additional patient characteristics that were not directly used as model inputs but used for validation purposes are shown in Supplementary Table 2.

Risk Strata Transition Probabilities

STELLAR data were used to calculate transition probabilities across risk strata (either worsening, improvement, or maintenance). Short-term transition probabilities between health states used in the model (i.e., model baseline to week 3, week 3 to 12, and week 12–24) were derived from observed patient transition counts across weeks 0, 3, 12, and 24 in STELLAR. For long-term transition probabilities beyond week 24, probabilities were based on observed transition counts from weeks 12 to 24 in STELLAR, assuming that the short-term outcomes were generalizable to long-term outcomes. Transition probabilities for patients receiving immediate treatment with sotatercept were calculated using the observed patient transition counts for the sotatercept plus BGT arm in STELLAR. Transition probabilities for patients with delayed sotatercept treatment were calculated using a two-step approach: (1) during year 1 and year 2, observed patient transition counts for the placebo plus BGT arm in STELLAR were applied; (2) from year 3 onwards, observed patient transition counts for the sotatercept plus BGT arm in STELLAR were used. Since patients initiated treatment with sotatercept in year 3, short-term transition probabilities were applied from week 1 to week 24 in year 3. After week 24, long-term transition probabilities were used. Transition probabilities for patients receiving BGT alone were derived using the observed patient transition counts for the placebo plus BGT arm in STELLAR.

Mortality, by Risk Stratum

Mortality risk for patients with PAH escalates as the disease progresses, driven by progressive vascular remodeling [19, 21]. Therefore, to inform long-term all-cause mortality probabilities by risk stratum, parametric regression models were fitted to long-term 5-year survival curves from the COMPERA registry and the French PAH registry (Supplementary Figs. 1–3), both of which are large, representative, prospective, multicenter PAH registries.

Sotatercept demonstrated a numerical reduction in mortality versus placebo in STELLAR (HR for all-cause mortality = 0.25; 95% CI: 0.05 to 1.19 per Merck & Co., Inc., Rahway, NJ, USA, data on file 2023; including a mixed population with any of the four risk strata) was applied to the four risk health states but not the lung/heart-lung transplant state. This assumption was based on the clinical understanding that after transplantation, patients would discontinue PAH therapy [22, 23]. For patients with delayed sotatercept, the mortality rates were determined through a two-step approach. In year 1 and year 2, the mortality rates of patients on BGT alone were applied, meaning there was no effect on mortality. After year 2, the survival benefit of sotatercept was factored in, based on the treatment effect (HR = 0.25) observed in the overall STELLAR population.

Lung/Heart-lung Transplant by Risk Stratum and Post-transplant Mortality

Lung or heart-lung transplantation remains a final treatment option for patients who reach end-stage PAH [20]. Consistent with current clinical practice, only patients with intermediate-high or high risk were considered eligible for lung/heart-lung transplantation (Table 2) [20]. Furthermore, it was assumed that patients transitioning to the lung/heart-lung transplant health state would remain in this state until death. COMPERA registry data informed transplant probabilities, while post-transplant mortality rates were sourced from published literature [24].

PAH Hospitalization by Risk Stratum

Given the frequent need for hospitalization in PAH, the model incorporated PAH hospitalizations by applying risk stratum–adjusted probabilities of PAH hospitalization to each health state, with higher-risk patients associated with increased hospitalization rates. These probabilities were based on observed data from the COMPERA registry (Table 2).

Infused Prostacyclin Escalation by Risk Stratum

As the disease evolves, patients with PAH often experience treatment escalation, including the introduction of infused prostacyclin, particularly for those at intermediate-high and high risk [24]. The model applied risk stratum-specific probabilities for prostacyclin use (Table 2) and assumed treatment changes would typically occur 1 year from baseline, with sensitivity analyses exploring alternative time cut-off points. In the base case, sotatercept’s impact on reducing prostacyclin use was modeled both directly, via observed treatment effects, and indirectly, through its influence on risk strata distribution.

Model Outcomes

Primary model outcomes included: (1) per patient total LYs (accumulated in each risk stratum), (2) per patient infused prostacyclin-free LYs, (3) the number of PAH hospitalizations (accumulated in each risk stratum) and lung/heart-lung transplants per 1000 patients, and (4) the number needed to treat (NNT) to prevent one PAH hospitalization and one transplant event.

Sensitivity Analyses

A deterministic sensitivity analysis (DSA) and probabilistic sensitivity analysis (PSA) were conducted to test the impact of individual model parameters and the uncertainty of the deterministic base case results. The DSA investigates the impact of the individual model parameters on model results by varying these parameters one at a time and compared immediate treatment with sotatercept to delayed treatment with sotatercept. The PSA varies all input parameters simultaneously around their uncertainty distribution to assess the uncertainty of the model results and was conducted by simultaneously drawing values for each parameter based on their point estimates and uncertainty information (Table 3). The PSA was run 1000 times to increase the stability of the results.

Table 3 Deterministic and probabilistic base case resultsa