Novartis continues to try to replicate the crappy paradigm HF trial reviewed brilliantly by VP. Again, they use a submaximal dose of enalapril in the control arm. You can’t fool us.
This again reminds me of why we went into medicine: to help people live longer, and/or feel better. Don’t recall “having a lower BNP” being one of those objectives.
I have lost count of all the studies I have seen comparing Entresto to the standard of ACE inhibitors with or without diuretics. Novartis must have a tremendous profit margin in order to justify endless trials in order to sell this turkey. The pharmaceutical companies and their favorite investigators in the research industry have developed ingenious tricks in order to mislead doctors and others not well versed in the field of statistics. It all started with the definition of "statistical significance" as any "p" value less than 0.05 about 100 years ago---a somewhat arbitrary designation by Ronald Fisher, a British mathematician and statistician who is considered the father of modern statistical analysis. As it has fallen a bit out of favor, it is now often replaced by confidence intervals, number needed to treat, non-inferiority, and other measures not as readily understood by the average person who would never accept a difference of a couple of percentage points as significant. Composite endpoints that often include factors that are subjective and difficult to measure is another proven way to mislead. The use of surrogates in place of actual diseases is also gaining favor. It is time for someone to publish a modern update to the classic text "How to Lie with Statistics".
The assumption that a published paper is marketing>evidence is as relevant that assuming a patient with chest pain is having a heart attack until proven otherwise.
I don't think this is a pessimistic approach. It's what the times call for.
Ugh yes, thank you! I really just want actionable information, like does this thing kill people? Does it make them more sick? How’s their quality of life? Not some BS biomarker with dubious implications.
One thing I don't see in this article or the free preamble of the paper is why we think low EF Chagas heart failure is different from garden variety low EF heart failure and needs to be studied separately in the first place?
And also, even assuming they are different: wouldn't PARADIGM HF provide a relatively strong prior that sacubitril/valsartan works in one type of HFrEF, requiring only a lower burden of evidence that it works in another type?
I feel like the hypothesis “S/V improves mortality and biomarkers compared to enalapril in generic HFrEF, but it only improves biomarkers in Chagas HFrEF and there is actually no mortality benefit” is very non-parsimonious and would require a significant difference between generic and Chagas HFrEF to be plausible.
You may have missed the point. The main problem is treating all endpoint components as equally bad to the patient. Once this bad practice is stopped, you'll likely find that using BNP (not change but absolute) as an endpoint on the milder end of the endpoint severity spectrum will not allow this milder endpoint to carry the day is it does with the faulty pre-specified primary analysis. For resources about ordinal and composite endpoints see https://hbiostat.org/endpoint and https://hbiostat.org/ordinal
The problem in this trial is not the use of a more informative end-point per se - it's the use of end-points of dubious value (change in BNP). There are a whole multitude of things patients with heart failure care about beyond just being dead or hospitalized - if the trialists had measured some of those things and showed a benefit, that would be a good use of a hierarchical end-point.
We shouldn't conflate showing a benefit in a composite with showing a benefit with the only the hard-endpoints (and, if we really think that only the hard-endpoints would justify the use of the drug then we should insist on just the hard endpoints)... but there's an alternative error to be made in dismissing measuring all the other things patients care about and detecting differences that are useful.
This is much more clear in my world (critical care trials) - where trialists historically have only assessed whether someone died or not - collecting no information on potential beneficial effects of treatments like improving functional status, symptoms, speed of recovery etc. - but it applies to cardiology, too. The lesson of "don't use informative end-points" is the wrong one - we should maximize what we learn from trials, but insist that the components of the hierarchy have meaning to patients (or justified data as a surrogate), we should pre-specify what changes will drive practice change and what won't, and we should insist that trials are correctly interpreted for the end point they actually had.
Despite college, med school, and MS statistics courses and careful reading of several series of statistical analysis of studies like this, the statistics still elude me. Is there a source of 'in plain English' statistics that might help?
As I was reading this, I was thinking, "This smells like a pharmaceutical company trying to find a new market for one of its products." And then I read, "The makers of sacubitril/valsartan (Entresto) Novartis funded the trial."
Novartis continues to try to replicate the crappy paradigm HF trial reviewed brilliantly by VP. Again, they use a submaximal dose of enalapril in the control arm. You can’t fool us.
This again reminds me of why we went into medicine: to help people live longer, and/or feel better. Don’t recall “having a lower BNP” being one of those objectives.
I have lost count of all the studies I have seen comparing Entresto to the standard of ACE inhibitors with or without diuretics. Novartis must have a tremendous profit margin in order to justify endless trials in order to sell this turkey. The pharmaceutical companies and their favorite investigators in the research industry have developed ingenious tricks in order to mislead doctors and others not well versed in the field of statistics. It all started with the definition of "statistical significance" as any "p" value less than 0.05 about 100 years ago---a somewhat arbitrary designation by Ronald Fisher, a British mathematician and statistician who is considered the father of modern statistical analysis. As it has fallen a bit out of favor, it is now often replaced by confidence intervals, number needed to treat, non-inferiority, and other measures not as readily understood by the average person who would never accept a difference of a couple of percentage points as significant. Composite endpoints that often include factors that are subjective and difficult to measure is another proven way to mislead. The use of surrogates in place of actual diseases is also gaining favor. It is time for someone to publish a modern update to the classic text "How to Lie with Statistics".
The assumption that a published paper is marketing>evidence is as relevant that assuming a patient with chest pain is having a heart attack until proven otherwise.
I don't think this is a pessimistic approach. It's what the times call for.
Ugh yes, thank you! I really just want actionable information, like does this thing kill people? Does it make them more sick? How’s their quality of life? Not some BS biomarker with dubious implications.
One thing I don't see in this article or the free preamble of the paper is why we think low EF Chagas heart failure is different from garden variety low EF heart failure and needs to be studied separately in the first place?
And also, even assuming they are different: wouldn't PARADIGM HF provide a relatively strong prior that sacubitril/valsartan works in one type of HFrEF, requiring only a lower burden of evidence that it works in another type?
I feel like the hypothesis “S/V improves mortality and biomarkers compared to enalapril in generic HFrEF, but it only improves biomarkers in Chagas HFrEF and there is actually no mortality benefit” is very non-parsimonious and would require a significant difference between generic and Chagas HFrEF to be plausible.
JAMA continues to embarrass itself with what it chooses to publish. smh
You may have missed the point. The main problem is treating all endpoint components as equally bad to the patient. Once this bad practice is stopped, you'll likely find that using BNP (not change but absolute) as an endpoint on the milder end of the endpoint severity spectrum will not allow this milder endpoint to carry the day is it does with the faulty pre-specified primary analysis. For resources about ordinal and composite endpoints see https://hbiostat.org/endpoint and https://hbiostat.org/ordinal
The problem in this trial is not the use of a more informative end-point per se - it's the use of end-points of dubious value (change in BNP). There are a whole multitude of things patients with heart failure care about beyond just being dead or hospitalized - if the trialists had measured some of those things and showed a benefit, that would be a good use of a hierarchical end-point.
We shouldn't conflate showing a benefit in a composite with showing a benefit with the only the hard-endpoints (and, if we really think that only the hard-endpoints would justify the use of the drug then we should insist on just the hard endpoints)... but there's an alternative error to be made in dismissing measuring all the other things patients care about and detecting differences that are useful.
This is much more clear in my world (critical care trials) - where trialists historically have only assessed whether someone died or not - collecting no information on potential beneficial effects of treatments like improving functional status, symptoms, speed of recovery etc. - but it applies to cardiology, too. The lesson of "don't use informative end-points" is the wrong one - we should maximize what we learn from trials, but insist that the components of the hierarchy have meaning to patients (or justified data as a surrogate), we should pre-specify what changes will drive practice change and what won't, and we should insist that trials are correctly interpreted for the end point they actually had.
Despite college, med school, and MS statistics courses and careful reading of several series of statistical analysis of studies like this, the statistics still elude me. Is there a source of 'in plain English' statistics that might help?
So interesting. Thanks for sharing.
As I was reading this, I was thinking, "This smells like a pharmaceutical company trying to find a new market for one of its products." And then I read, "The makers of sacubitril/valsartan (Entresto) Novartis funded the trial."
That pretty much says it all.
Well, once again 95% of scientists agree with whoever is funding them.
Are there any circumstances in heart failure treatment when a lab value rather than a clinical outcome is an appropriate endpoint ?
Fair to say: any trial that requires so many patients to demonstrate an endpoint probably has absolute value of minimal proportion??