Medical studies are often hyped as "positive." Consumers of medical evidence must always be on alert for the endpoint measured. Not all positive results are the same.
The following fixes less than 1/2 of the problem but may still be worth pondering. Part of the problem comes from treating all hospitalizations alike, independent of (1) duration and (2) intensity of care needed. Restricting attention to duration, one can think of all-cause mortality and all-cause hospitalization this way: As a function of time since randomization, treatment, and baseline covariates, what is the probability that a patient will either be in the hospital or be dead? This can readily be analyzed with a state transition model that considers death as an absorbing state and (at home, in hospital alive, dead) as 3 ordinal outcome levels. Long or multiple hospitalizations will elevate the probability of being in the hospital on a given day post randomization. More info is at https://www.fharrell.com/talk/cmstat/
From your description John it seems that recurrent hospitalizations were not counted and only the first hospitalization was used. This is problematic.
Thanks Dr. Mandrola for another great article. Your TWIC articles on heart.org are also weekly required reading.
Trials have traditionally be done with “time-to-first-event”. There appears to be a small trend to doing “win-ratio” analyses to address this, but they all tend to be post hoc efforts. I wonder if future RCT will adopt more of this type of statistical analysis on a prospective basis.
When dealing with old people, asking a new drug to reduce total mortality or total hospitalizations is asking the wrong question. We all have to die of something and preventing one death or hospitalization among the many that take the life or cause a hospitalization in old people makes it impossible for one drug to effect improvement in all those events. The only rationale for another heart failure drug to be considered useful is if it improves the life of the patient who uses the drug.
Does the drug make you feel better , live better, be more active, useful or enjoy their life more.
These metrics are hard to quantify but at the end of their day they are often the only things that justify the use and the cost of the additional drug.
Right on with endpoints! Look at the Premarin/Provera rabbit hole. For decades we looked at cholesterol levels rather than death rates. Unfortunately we are no closer than we were 80 years ago to understanding the secrets of estradiol (E2) and real progesterone. Much damage has been done to patients both before and after the WHI study 22 years ago.
Very interesting, thank you. Of course the benefit does depend hugely on the cost. I understand these are very expensive in the US, although interestingly in the UK they are not particularly so. £30/month here, so they probably are cost-effective in terms of improving quality of life and symptoms. I appreciate that the latter were secondary endpoints.
Super article, thank you.
The following fixes less than 1/2 of the problem but may still be worth pondering. Part of the problem comes from treating all hospitalizations alike, independent of (1) duration and (2) intensity of care needed. Restricting attention to duration, one can think of all-cause mortality and all-cause hospitalization this way: As a function of time since randomization, treatment, and baseline covariates, what is the probability that a patient will either be in the hospital or be dead? This can readily be analyzed with a state transition model that considers death as an absorbing state and (at home, in hospital alive, dead) as 3 ordinal outcome levels. Long or multiple hospitalizations will elevate the probability of being in the hospital on a given day post randomization. More info is at https://www.fharrell.com/talk/cmstat/
From your description John it seems that recurrent hospitalizations were not counted and only the first hospitalization was used. This is problematic.
Great article Dr. Thank you
Why do investigators use composite endpoints at all? Why not pick one, like CV death or overall mortality?
Thanks Dr. Mandrola for another great article. Your TWIC articles on heart.org are also weekly required reading.
Trials have traditionally be done with “time-to-first-event”. There appears to be a small trend to doing “win-ratio” analyses to address this, but they all tend to be post hoc efforts. I wonder if future RCT will adopt more of this type of statistical analysis on a prospective basis.
When dealing with old people, asking a new drug to reduce total mortality or total hospitalizations is asking the wrong question. We all have to die of something and preventing one death or hospitalization among the many that take the life or cause a hospitalization in old people makes it impossible for one drug to effect improvement in all those events. The only rationale for another heart failure drug to be considered useful is if it improves the life of the patient who uses the drug.
Does the drug make you feel better , live better, be more active, useful or enjoy their life more.
These metrics are hard to quantify but at the end of their day they are often the only things that justify the use and the cost of the additional drug.
Right on with endpoints! Look at the Premarin/Provera rabbit hole. For decades we looked at cholesterol levels rather than death rates. Unfortunately we are no closer than we were 80 years ago to understanding the secrets of estradiol (E2) and real progesterone. Much damage has been done to patients both before and after the WHI study 22 years ago.
Very interesting, thank you. Of course the benefit does depend hugely on the cost. I understand these are very expensive in the US, although interestingly in the UK they are not particularly so. £30/month here, so they probably are cost-effective in terms of improving quality of life and symptoms. I appreciate that the latter were secondary endpoints.