Article Text

Download PDFPDF

Original research
Survival analysis from the INCREASE study in PH-ILD: evaluating the impact of treatment crossover on overall mortality
  1. Steven D Nathan1,
  2. Shilpa Johri2,
  3. Joanna M Joly3,
  4. Christopher S King1,
  5. Amresh Raina4,
  6. Colleen A McEvoy5,
  7. Dasom Lee6,
  8. Eric Shen6,
  9. Peter Smith6,
  10. Chunqin Deng6,
  11. Aaron B Waxman7
  1. 1Advanced Lung Disease and Lung Transplant Program, Inova Heart and Vascular Institute, Inova Fairfax Hospital, Falls Church, Virginia, USA
  2. 2Pulmonary and Critical Care Medicine, Pulmonary Associates of Richmond Inc, Richmond, Virginia, USA
  3. 3Division of Cardiovascular Disease, University of Alabama at Birmingham, Birmingham, Alabama, USA
  4. 4Advanced Heart Failure and Transplant, Allegheny General Hospital, Pittsburgh, Pennsylvania, USA
  5. 5Division of Pulmonary and Critical Care Medicine, Washington University in St Louis, St Louis, Missouri, USA
  6. 6United Therapeutics, Research Triangle Park, North Carolina, USA
  7. 7Pulmonary Vascular Disease Program, Brigham and Women's Hospital/Harvard Medical School, Boston, Massachusetts, USA
  1. Correspondence to Dr Steven D Nathan, Heart and Lung Transplant Center, Inova Fairfax Hospital, Falls Church, VA 22042, USA; steven.nathan{at}


Objective A post-hoc analysis of the INCREASE trial and its open-label extension (OLE) was performed to evaluate whether inhaled treprostinil has a long-term survival benefit in patients with pulmonary hypertension associated with interstitial lung disease (PH-ILD).

Methods Two different models of survival were employed; the inverse probability of censoring weighting (IPCW) and the rank-preserving structural failure time (RPSFT) models both allow construction of a pseudo-placebo group, thereby allowing for long-term survival evaluation of patients with PH-ILD receiving inhaled treprostinil. Time-varying stabilised weights were calculated by fitting Cox proportional hazards models based on the baseline and time-varying prognostic factors to generate weighted Cox regression models with associated adjusted HRs.

Results In the INCREASE trial, there were 10 and 12 deaths in the inhaled treprostinil and placebo arms, respectively, during the 16-week randomised trial. During the OLE, all patients received inhaled treprostinil and there were 29 and 33 deaths in the prior inhaled treprostinil arm and prior placebo arm, respectively. With a conventional analysis, the HR for death was 0.71 (95% CI 0.46 to 1.10; p=0.1227). Both models demonstrated significant reductions in death associated with inhaled treprostinil treatment with HRs of 0.62 (95% CI 0.39 to 0.99; p=0.0483) and 0.26 (95% CI 0.07 to 0.98; p=0.0473) for the IPCW and RPSFT methods, respectively.

Conclusion Two independent modelling techniques that have been employed in the oncology literature both suggest a long-term survival benefit associated with inhaled treprostinil treatment in patients with PH-ILD.

  • Rare lung diseases
  • Primary Pulmonary Hypertension
  • Idiopathic pulmonary fibrosis

Data availability statement

Data are available upon reasonable request.

This is an open access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited, appropriate credit is given, any changes made indicated, and the use is non-commercial. See:

Statistics from

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.


  • In the INCREASE trial, inhaled treprostinil was evaluated in patients with pulmonary hypertension associated with lung disease (PH-ILD). This study demonstrated that treatment with inhaled treprostinil improved the 6-min walking distance and time to clinical worsening. Secondary analyses of this clinical trial have demonstrated other benefits of inhaled treprostinil such as improvement in forced vital capacity and a decrease in the total number of disease progression events. However, the long-term effects of inhaled treprostinil on patient survival are unknown.


  • In this post-hoc analysis of the INCREASE trial, we used statistical methodology drawn from oncology clinical research to demonstrate a long-term reduction in the risk of death associated with inhaled treprostinil.


  • This data provides further support for inhaled treprostinil as a treatment option in patients with PH-ILD and additional rationale for wider adoption, as well as for physicians to actively screen and recommend right heart catheterisations in patients with ILD at risk of PH.


Patients with interstitial lung disease (ILD) who develop pulmonary hypertension (PH-ILD) have a very poor prognosis with a reported survival ranging from 23% to 40% at 3 years.1–3 Until recently, it was uncertain whether treating patients with PH-ILD with pulmonary vasoactive therapy improved outcomes. While some studies and reports have suggested benefit,4 there have been others that have shown harm with increased mortality observed in one randomised controlled trial (RCT) that resulted in its early termination.5 6 The first successful study in PH-ILD was the INCREASE study, which was a double blind, placebo-controlled trial that met its primary endpoint of change in placebo-adjusted 6-min walk distance (6MWD) at 16 weeks.7 In addition, multiple secondary endpoints were positive including time to clinical worsening and change in the N-terminal pro–B-type natriuretic peptide (NT-proBNP). This led to US Food and Drug Administration (FDA) approval of inhaled treprostinil for PH-ILD in 2021. Patients who completed the RCT were also given the opportunity to enrol in the open-label extension (OLE) study and results on the long-term safety and efficacy of inhaled treprostinil were recently published.8

The design of clinical trials in both ILD and PH-ILD usually employs a surrogate endpoint, typically forced vital capacity, 6MWD, or a composite endpoint of clinical worsening. Mortality as an endpoint in ILD clinical trials is generally regarded as impractical due to multiple barriers including the number of patients required, as well as the duration and ethical issue in the context of available effective therapies.9 Patients with PH-ILD have distinctly worse outcomes and hence mortality studies might be more feasible as enough events may accrue over the study timeframe. However, similar constraints exist with regard to recruitment, retention and ethics of such studies. The same issues have vexed oncology studies where patients with similarly poor outcomes have been enrolled in clinical trials of novel therapies, resulting in the emergence of novel statistical methodologies to estimate mortality benefits as accurately as possible.10–12 Two such methods are the inverse probability of censoring weighting (IPCW) and the rank-preserving structural failure time (RPSFT), which can be used to simulate a placebo group’s survival probability in an OLE assuming the group remained on placebo treatment rather than crossing over to active drug.13–17 These two methodologies may therefore help us better understand clinical trial results in PH-ILD when there are no placebo group data available due to ethical concerns or open-label study designs. Herein, we present a post-hoc analysis of the overall survival data from the combined INCREASE RCT and OLE populations to assess the impact of active treatment on patient survival.


The methods and results of the INCREASE study have previously been reported.7 Briefly, it was a 16-week, randomised, double-blinded, placebo-controlled study evaluating inhaled treprostinil in patients with PH-ILD. Confirmation of PH-ILD within 1 year of randomisation was required and was defined by a pulmonary vascular resistance greater than 3 Wood units, pulmonary capillary wedge pressure of 15 mm Hg or lower and mean pulmonary arterial pressure of 25 mm Hg or higher. Study participants who completed the 16-week study were offered the option to enrol in the OLE study.8 Once in the OLE, all patients, even those randomised to the active treatment arm of the RCT, initiated inhaled treprostinil at three breaths four times daily (QID) to preserve blinding, and titrated to a maximum dose of 15 breaths QID. Time to all-cause death was calculated from the randomisation date to the date of death in the RCT or OLE. Participants who survived were censored at the last contact date in the RCT or OLE, and the number of deaths in the RCT and OLE was tabulated. Time to all-cause death was analysed with the Cox-regression model using intention-to-treatment (ITT) analysis without adjusting for the treatment crossover.


Treatment switching methods

The IPCW and the RPSFT are two methods that account for switching and have been widely used and validated in oncology research. In this study, these methodologies were used to adjust for the treatment crossover where the treatment crossover was defined as participants in the placebo group switching to open-label inhaled treprostinil.13–17

The IPCW method constructs a pseudo-placebo arm that has the same prognostic characteristics as the original placebo arm but assumed to have stayed on placebo during the OLE.13 14 This is achieved by artificially censoring control group subjects at the time of crossover and estimating statistical weights based on baseline and time-varying demographic as well as disease-related characteristics. This mitigates the potential bias due to censoring and allows for the estimation of the treatment effect without the impact of crossover to active treatment. The IPCW method relies on ‘no unmeasured confounding’, assuming that all prognostic covariates that predict crossover and outcomes are adequately included in the model. In this study, time-varying stabilised weights were calculated by fitting Cox proportional hazards models based on the baseline and time-varying prognostic factors shown in table 1.14 These weights were then used in a weighted Cox regression model to estimate an adjusted HR with a robust SE.

Table 1

Baseline and time-varying covariates for the observational-based inverse probability of censoring weighting methods

The RPSFT method assumes that each subject proceeds through the disease process at their individual pace towards death, and that the experimental treatment slows this rate of progression down by the same factor regardless of whether treatment is administered from randomisation or from the time of crossover; this is known as the ‘common treatment effect’ assumption.15 16 The slow-down factor was estimated and then applied to placebo crossover subjects to emulate the hypothetical survival times that would have been observed had they not crossed over. A Cox proportional hazards model was fitted using the observed survival time for inhaled treprostinil and placebo non-crossover subjects, along with the adjusted survival times for the placebo crossover subjects. The HR adjusting for the treatment crossover was then estimated.17 The bootstrap method was used to estimate the CIs and p values to account for the uncertainty of the RPSFT method. As a sensitivity analysis, the two methodologies were rerun using week 4 as time zero, therefore excluding deaths occurring in the first 4 weeks, considering that patients were in the uptitration phase during this time period and had not yet achieved an optimal dose.


There were 326 patients randomised to inhaled treprostinil (n=163) or placebo (n=163) in the INCREASE RCT. Details regarding patient disposition are included in the INCREASE primary publication and OLE papers.8 There were 10/163 (6.1%) deaths in the inhaled treprostinil group and 12/163 (7.4%) deaths in the placebo group during the RCT (figure 1). At the end of the RCT, 119 participants from the previous inhaled treprostinil group and 121 participants from the previous placebo group enrolled in the OLE, during which all patients received inhaled treprostinil. Among the patients who died, the mean time to death was 58.5 weeks for inhaled treprostinil (n=39) and 43.5 weeks for placebo (n=45). Time to death and estimated survival rates for the conventional survival analysis are shown in table 2.

Figure 1

Patient disposition CONSORT diagram. OLE, open-label extension; RCT, randomised controlled trial. *Two patients entered the OLE who were excluded from the INCREASE RCT due to a study drug labeling issue.

Table 2

Time to death using the conventional survival analysis of the ITT population for the RCT and OLE

The HRs from the analyses adjusted for treatment crossover are provided in figure 2. With a conventional ITT analysis, the HR for death was 0.71 (95% CI 0.46 to 1.10; p=0.1227; log-rank p=0.1209) (figure 3A Kaplan-Meier plot). When using the IPCW and RPSFT methods to calculate overall survival, there are significant reductions in death associated with inhaled treprostinil treatment with the IPCW method resulting in an HR of 0.62 (95% CI 0.39 to 0.99; p=0.0483) and the RPSFT method yielding an HR of 0.26 (95% CI 0.07 to 0.98; p=0.0473) (figure 3B Kaplan-Meier plot). A Kaplan-Meier plot for the IPCW method was not generated since this entails a time-varying scale versus the other two methodologies which are time-fixed. Using the RPSFT method to adjust for treatment crossover, the estimated 52 week survival of the constructed placebo group was 59.2% (95% CI 25.1 to 64.9). By comparison, the Kaplan-Meier estimate of 52-week survival for the inhaled treprostinil group is 89.0% (95% CI 82.3 to 93.3).

Figure 2

Estimates of overall survival HR. IPCW, the observational-based inverse probability of censoring weighting; ITT, intention to treat; RPSFT, randomised-based rank-preserving structural failure time.

Figure 3

KM plots of overall survival. (A) Conventional ITT analysis. (B) RPSFT method adjusting for treatment crossover. ITT, intention to treat; KM, Kaplan-Meier; RPSFT, randomised-based rank-preserving structural failure time.

In the sensitivity analysis, five deaths during the first 4 weeks of the uptitration phase in the RCT were excluded from the analysis. Four of these deaths were in the active treatment arm and one was in the placebo arm. This sensitivity analyses showed numerical improvements in survival estimates; specifically, for the unadjusted ITT, the HR was 0.65 (95% CI 0.41 to 1.01; p=0.0546), while for the IPCW the HR was 0.56 (95% CI 0.34 to 0.93; p=0.0237) and for the RPSFT method the HR was 0.17 (95% CI 0.04 to 0.69; p=0.0136).


In this current analysis of the INCREASE study of inhaled treprostinil, we demonstrate that the conventional ITT analysis was confounded by treatment crossover at the end of the double-blind phase resulting in an underestimation of the benefit of inhaled treprostinil on overall survival. By adjusting for treatment crossover through the IPCW and RPSFT models, we demonstrate an estimated 38% and 74% reduction in risk of mortality, respectively, with inhaled treprostinil vs placebo. The differences in the estimated mortality benefits using the two models are due to their differing techniques and sets of information employed. Both models demonstrating improved survival with inhaled treprostinil support a true mortality benefit which is further underscored by our sensitivity analyses.

The INCREASE study, the largest study to date in PH-ILD, met its primary endpoint of 6MWD improvement at week 16, resulting in the US FDA approval of inhaled treprostinil in the USA for this indication.7 In the primary and subsequent post-hoc analyses, inhaled treprostinil was shown to have multiple benefits in patients with PH-ILD, including a reduction in time to clinical worsening, ameliorating multiple progression events and more clinical improvement with less worsening at higher doses.7 18 19 However, the issue of whether these benefits translate to an overall mortality benefit remains unanswered. Indeed, there are examples in the literature of drugs with short-term benefits but long-term harm.20 21 In this analysis, we provide evidence of a mortality benefit to inhaled treprostinil in patients with PH-ILD. We adopted two statistical methodologies, that were originally popularised and accepted in oncology trials, but which have now also been employed in cardiology studies and more recently in a pulmonary arterial hypertension trial of macitentan.22 This mortality benefit is further supported through the sensitivity analysis which confirmed the reduced risk of death for patients on inhaled treprostinil. Both the IPCW and RPSFT methods demonstrated statistical significance when adjusting for treatment crossover, potentially indicating a significant mortality benefit associated with inhaled treprostinil over 124 weeks. It is well-established that right ventricular dysfunction and decompensation drive outcomes in pulmonary arterial hypertension (PAH).23 This same phenomenon is established in heart failure with evidence to suggest that the same holds true in PH-ILD.24 25 Over the course of the 16-week double-blind INCREASE RCT, there was a reduction in the NT-proBNP in the treatment arm as opposed to an increase in this biomarker in the placebo arm.7 This suggests reduction of right ventricular (RV) strain in those patients on active therapy versus increasing strain over time in the placebo arm. It is also well established that once PH supervenes in patients with ILD, the natural history is one of relentless progression.25 26 The survival benefit associated with inhaled treprostinil that we observed in this post-hoc analysis can perhaps be attributed to the downstream effects of PH on the RV and the ability of inhaled treprostinil to ameliorate this. We cannot rule out that other factors played a role in this apparent survival benefit including but not limited to the possible antifibrotic properties of inhaled treprostinil.27

The only randomised controlled survival study in PAH is the original intravenous epoprostenol study.28 For similar practical and ethical reasons, mortality as an endpoint in PAH clinical trials is unrealistic. Nonetheless, it is well established through registries and other reports that the survival of patients with PAH has improved dramatically since the 1990s with the advent and use of PAH medications.29 30 Similarly, in idiopathic pulmonary fibrosis (IPF), survival has improved despite no RCT demonstrating a survival benefit or being powered to do so. In this regard, post-hoc analyses, registry reports, and meta-analyses have provided evidence of a survival benefit associated with antifibrotic therapy.31–37 For similar reasons, it is somewhat unlikely that there will be future double-blind RCTs in PH-ILD with mortality as an endpoint; although such studies may have merit, they are impractical to implement. Similar to IPF, ongoing registries will evaluate whether there is a possible mortality benefit associated with inhaled treprostinil treatment in patients with PH-ILD. In the interim, reliance will need to be placed on novel statistical methodologies such as those employed in this report. A cautionary note, however, is that the data from the INCREASE trial should not be extrapolated to any other medications approved for the treatment of PAH, since harm and increased mortality have previously been demonstrated with the use of riociguat in the same population in what was the largest RCT in PH-ILD that that time.6

There are certain limitations to this study. For one, both the RPSFT and IPCW methods rely on strong assumptions and the violation of these assumptions could affect the estimation of the treatment effect. The IPCW method relies on the ‘no unmeasured confounding’, meaning that all information related to the treatment crossover and the outcome is captured in the dataset. This assumption is plausible as several baseline and time-varying prognostic factors that predict treatment crossover and outcomes collected during the trial were included in the model. Another potential source of error with the IPCW method is when there is a small number of subjects with a very large crossover proportion. However, given the large size of the INCREASE study and the proportion of crossover, the IPCW can be considered an appropriate method. The RPSFT method assumes that the survival of both arms would have been similar in the absence of inhaled treprostinil. Given that the INCREASE trial was a well-designed study with well-matched cohorts, this assumption is also likely true. The RPSFT method also relies on the ‘common treatment effect’ assumption, meaning that the treatment effect is the same no matter when initially administered. It is apparent from the OLE of the INCREASE RCT that the former placebo group has a more muted 6MWD response compared with the former inhaled treprostinil group,12 but whether this translates to a diminished survival benefit is uncertain. It is noteworthy though, that the former placebo arm had about a 50% decrease in their NT-proBNP once on open-label drug suggesting clinical benefit that might not have been reflected in the 6MWT results.12 On the other hand, the true mortality benefit might be underestimated since deaths in the first month during the uptitration phase of treatment were treated equivalently to later deaths in the RPSFT and IPCW models. Indeed, our sensitivity analysis which includes deaths occurring when most patients had attained the targeted dose demonstrated an even greater mortality benefit, due to four deaths happening during the first 4 weeks of the RCT in the active treatment group versus one in the placebo arm. This is not a surprising finding given that higher doses have previously been shown to result in less clinical worsening and more clinical improvement.19

In conclusion, these two well-established statistical methodologies demonstrate a survival benefit for inhaled treprostinil in patients with PH-ILD. This data provides further support for this treatment option in patients with PH-ILD and additional rationale for physicians to actively screen and recommend right heart catheterisations in patients with ILD at risk of PH. It is also our hope that this analysis provides further evidence for this therapy thus enabling its broader availability, especially given the dismal prognosis and dearth of other treatment options for patients with PH-ILD.

Data availability statement

Data are available upon reasonable request.



  • Contributors DL and CD analysed the data and contributed to data interpretation. SDN, ES and CD wrote the draft, revised the paper, and act as guarantors for the work. All authors, including SJ, JMJ, CSK, AR, CAM, ABW and PS contributed to data interpretation and revised the draft for intellectual content. SDN, SJ, JMJ, CSK, CAM, PS, CD and ABW contributed to data acquisition for the original clinical trial.

  • Funding The INCREASE study was sponsored by United Therapeutics.

  • Competing interests SDN has received consulting fees or payment or honoraria, from United Therapeutics, Boehringer-Ingleheim, Roche, Bellerophon Therapeutics, Altavant Sciences and Merck. SJ has received consulting fees or payment or honoraria from United Therapeutics, Bellerophon Therapeutics, Bayer and Acceleron Pharma, and has participated in advisory boards for Janssen Pharmaceuticals and Merck. CSK has received consulting fees or payment or honoraria, from Merck, Janssen Pharmaceuticals, United Therapeutics and Altavant Sciences. JMJ has nothing to disclose. AR has received payment or honoraria from Merck. CAM has received payment or honoraria from United Therapeutics and has participated on a data safety monitoring board or advisory board, for United Therapeutics, Merck and Janssen Pharmaceuticals. CD, DL, ES and PS are employees of and may have received stock or stock options from United Therapeutics. ABW has received grant support or consulting fees or payment or honoraria, from AI Therapeutics, ARIA-CV, Acceleron/Merck, Janssen Pharmaceuticals, and has participated in advisory boards for Insmed.

  • Provenance and peer review Not commissioned; externally peer reviewed.