Treprostinil for Pulmonary Hypertension

As you have seen, the Cardio-Renal Division reviewers and supervisors have consistently reached the conclusion that treprostinil should not be approved.

Does that mean we all think what to do is clear? No. That's why we’re here.

1. All reviewers have found that the decision is a very close call, although they lean toward no

2. The disease is bad and available treatment (Flolan) is onerous and difficult to use (central venous line), although it may have a survival effect. Potential alternatives (tomorrow) are not serious risk-free. In that situation, how much weight should be given to a noxious, but not dangerous, symptom (pain) that the patient can presumably assess herself and, of course, consider the benefit? What is the implication of the 95% entry into long-term maintenance? You will probably guess that I think it does tell you something

3. The effect on the primary endpoint plainly failed to meet the pre-specified statistical criteria, but is very close. Is very close good enough? Could other findings buttress the result? And is an effect of the magnitude seen a meaningful benefit at all?

4. There is a complex question of whether the 6 minute walk results, as analyzed, in light of dropouts, are really valid or are falsely favorable so that the small effect is even smaller, or non-existent.

5. The other effectiveness findings (symptomatic benefits) are generally favorable but, with one exception (the Borg score), are potentially contaminated by unblinding because of injection site pain. And even the Borg and exercise test could be unblinded for the patient. Could some of them be persuasive anyway (e.g., rate of new appearance of a problem not reported at baseline?) and how does one deal with the high level of multiplicity?

6. What does the increasing dose mean: disturbing evidence of tolerance to whatever effect the drug has, or alternatively, an inevitable consequence of a design that allowed dose increases if patients were still symptomatic, which, of course, all were and would be expected to be, despite treatment. The increase in dose on placebo was about twice that on treprostinil (? evidence of effect)

A Few DetailsI. The primary endpoint - 6 minute walk

Planned2 studies

p < 0.049 both studies, orp < 0.049 one study and > 0.01 pooled

Observedp < 0.061p < 0.055p < 0.006 combined

We like to say we’re not the slave of p-values. We have two studies quite close to goal and, analyzed together, moderately extreme. Is this result convincing statistically? And does the effect seem clinically meaningful by itself, or together with the other data?

II. Possible “informative” censoring in Log-Rank analyses (non-parametric)

The MOR notes many patients on treprostinil dropped out of the study and that this was an “inherent bias” because those patients could never get a low value in the analysis, even if they did badly (e.g., died)

In the planned analysis:

DC death, deterioration, transplant = lowest rank

DC ADR = LOCF

It is not true, I think, that this procedure must be biased. It is only biased if the procedure removes patients who are worse than the remaining patients and worse in a way not reflected in the exercise test used to score them. Could this be the case? Yes. Suppose discontinuation for ADR selected patients not doing well (that is, the decision to drop out for pain might represent both the ADR and deteriorating exercise ability leading to less willingness to accept the pain) and suppose this was not reflected in the last observation used in the analysis, either because it was obtained earlier (a contemporary exercise test might have revealed deterioration) or because doing badly is not reflected in the exercise test, then censoring at dropouts is informative, i.e., leads to a bias. Obviously, given the result, even a little of this would be a problem

The review shows that assigning a low rank to selected WD’s [e.g., (1) those who died or were terminated in < 100 days, (2) those who received Flolan within one month of DC, (3) those who received Flolan up to 100 days] eliminated statistical significance

All these analyses, as expected, reduce statistical significance. But they are appropriate only if, in fact, censoring was informative, and it is hard to know that. That is, the withdrawn patients, even the ones who died, may have been a random sample of the group, or may have been adequately located by their LOCF.

III. Use of other (non primary) measures: combined ET/Borg; symptoms

A. ET

The possibility that ET isn’t all one hoped is not unreasonable, but it was not anticipated. The possibilities include:

1. People with relatively preserved exercise (say > 300 M) don’t improve much, perhaps because major limit is not dyspnea (Implication, look more at the low baseline group)

2. People don’t exercise to real exhaustion, but to some other endpoint, e.g., a maximum walking speed (Implication, look at both distance and comfort, i.e., the Borg score)

All this implies that the true measure of benefit is some combination of how you walk and how you feel when walking

You’ve seen the combined result and it is statistically robust (including to various ways of handling the drop outs)

It is, however, inevitably after the fact

Is it, then, a finding or a hypothesis?

Obviously, we and the Committee have a long history of skepticism regarding such analyses, but even skepticism needs to consider plausibility and likelihood of chance, and, in this case, the near significant finding on the primary endpoint may help

Similarly, we are skeptical about subsets (low baseline) as a way to conclude there is an effect, but if the overall result is positive (a debatable question) there may be some legitimacy to seeing whether the effect seems more meaningful in a subset. Consistency of such a finding in the two studies could support such an approach

IV. Symptoms

Note that the review has tended to discount the symptom results even though many show very extreme p-values (Borg score, composite symptoms, dyspnea/fatigue score)

As you have heard, it can be argued that some symptom findings might be more plausible, even if there was unblinding

That is obviously a matter of judgement. In addition, however, it is important to consider the findings of all important symptoms, even those not thought resistant to blinding. You would not, for example, want the “plausible” ones to be only the ones that went in the right direction