ELITE: Early Versus Late Intervention Trial With Estradiol
Another one to watch
The first problem is the name. ELITE is not comparing early with late starting of post menopausal hormone replacement therapy (HRT). Rather there are two separate trials, an “early” and a “late” one. Both compare HRT with placebo.
Participants in the “early” trial are women within six years of the menopause, and in the “late” one, women more than 10 years after it. Here is the text from the trial registration website here.
“A total of 643 (actual) (504, initially proposed) postmenopausal women were [sic] randomized according to their number of years since menopause, less than 6 years or 10 years or more, to receive either oral 17B-estradiol 1 mg daily or a placebo. Women with a uterus will also use vaginal progesterone gel 4% (or a placebo gel) the last ten days of each month.”
Two trials should have two hypotheses, two primary endpoints and two sample size calculations etc., but this has only one. Here it is.
“The primary hypothesis to be tested is that 17B-estradiol (estrogen) will reduce the progression of early atherosclerosis if initiated soon after menopause when the vascular endothelium (lining of blood vessels) is relatively healthy versus later when the endothelium has lost its responsiveness to estrogen.”
This makes no sense. Hopefully when the trial(s) get(s) analysed the statistician will prevent them comparing the early with the late group, and instead set up a new hypothesis for the “early” trial something like “starting HRT soon after the menopause reduces carotid artery intimal thickness, a presumed risk factor for later cardiovascular disease”. The late trial can test a similar hypothesis
The primary outcome is the rate of change of distal common carotid artery (CCA) far wall intima-media thickness (IMT). This is measured twice at baseline and then every 6 months on trial.
This is an ambiguous endpoint since the duration of participation is reported to be between 2 and 5 years. (There is similar ambiguity about the same endpoint in the KEEPS trial. I’m trying to persuade the chief investigator of that study to let me see the protocol/analysis plan to clarify that.)
Although there is not much detail on Clinicaltrials.gov, there is plenty else about the trial on the web. e.g. here.
“In ELITE, Hodis will compare estrogen’s effects in women who are six or fewer years past menopause to its effects in women who are 10 or more years beyond menopause. Out of 504 women who will enroll in the study, half will be randomly assigned to take 17beta-estradiol (a form of estrogen identical to women’s own estrogen) daily, while the rest will take a placebo.”
This is a newspaper article, and I guess the journalist may have misunderstood. But it sounds again like the comparison is going to be between the younger and older women, not between the groups as randomised. Elsewhere the triallists are reported as already believing that the active treatment will work.
Donna Shoupe, professor of obstetrics and gynecology, said … “I think the ELITE study will be a landmark study that will establish the real benefits of estrogen replacement,”
Howard N Hodis was also chief investigator for the Estrogen in the Prevention of Atherosclerosis Trial (EPAT). Registered here in 2005, which would be OK, except that the trial had been published four years earlier here! In that trial 111 women got estradiol and 111 placebo. Twenty one never had a follow-up carotid intimal thickness measure so they ended up comparing the rate of change in intimal thickness between 99 estradiol and 102 placebo women. According to the paper the trial was powered to show a “standardized difference in progression rates between the two treatment groups of 0.40 or greater”.
Here is the graph of what happened. The dotted line is placebo, and the solid estradiol. Low numbers are good.
The solid (HRT) line is lower but this is just a chance effect. To be more precise the difference on the left at baseline is chance. It is not statistically significant, and nor should it be, because the groups were selected at random. By chance the women randomised to placebo had slightly thicker carotid arteries before they started treatment.
The trial is testing a difference in the slope. It would be good to show a scatter plot of the raw values, or at least error bars on each line, but even without these you don’t need a statistician to tell you the slopes are the same – both essentially level. Any tiny trend at 24 months is much less than the chance difference at the start. But the authors analysed away as follows:
“The top panel of Figure 2 [the figure above] shows the time course of changes in common carotid artery intima–media thickness that was predicted by the mixed-effects model (goodness-of-fit P value = 0.2). In the placebo group, subclinical atherosclerosis progressed by 0.0036 mm/y. In contrast, the estradiol group experienced regression of subclinical atherosclerosis (negative average rate of change in intima–media thickness) at a rate of 0.0017 mm/y. The difference in the average rates of progression between the two treatment groups was 0.0053 mm/y (95% CI, 0.0001 to 0.0105 mm/y) (P = 0.046, and P = 0.045 after adjustment for oophorectomy status).”
But remember what the trial was designed to show – a “standardized difference in progression rates [...] of 0.40 or greater”. They observed a difference of 0.0053, so the result is negative. Let’s say it again – a NEGATIVE result.
But who cares? Someone has squeezed the magic P<0.05 out of the data. So the authors’ conclusion, trumpeted in hundreds of HRT drug company sponsored articles ever since:
“Overall, the average rate of progression of subclinical atherosclerosis was slower in healthy postmenopausal women taking unopposed ERT with 17b-estradiol than in women taking placebo.”
Perhaps I’m being harsh. Standards of trial reporting were lower in 2001. Let’s hope ELITE will be analysed and reported properly. But if it comes up with a positive result after some fancy statistical contortions, sceptics should look closely.