Lipid lowering - a Bempedoic Acid Subset Analysis – To be right, you have to also look at what is left.
A guest post from James McCormack, BSc (Pharm), Pharm D, Faculty of Pharmaceutical Sciences, University of British Columbia
Dear Readers, This guest post from Professor McCormack is the type of content we strive for here at Sensible Medicine.
Bempedoic Acid is a cholesterol-lowering drug. In the CLEAR Outcomes trial, it was found effective in modestly reducing cholesterol and cardiovascular outcomes in statin-intolerant patients.
The authors have now published a subanalysis of the main CLEAR Outcomes study. More than 80 health news sites covered the study—most with “positive” headlines and leads.
Professor McCormack finds serious nuances in this paper. We are pleased to publish his analysis. Let us know what you think in the comments, and thanks for your support. JMM
By James McCormack
On June 24, 2023 a subset analysis of primary prevention participants (n=4200) from the CLEAR Outcomes trial was published in JAMA. However, the unreported results of the larger subset of secondary prevention participants (n=9800), are arguably much more interesting.
Let’s first look at the original trial.
Bempedoic acid is a medication similar to statins (it targets cholesterol synthesis up-stream of 3-hydroxy-3-methylglutaryl coenzyme A reductase as do statins) and, in fact, both similarly lower LDL-cholesterol.
In April of this year, the New England Journal of Medicine published the overall results from the CLEAR Outcomes study (bempedoic acid vs placebo in patients with statin intolerance).
In short, the 41-month, placebo-controlled trial of nearly 14,000 statin-intolerant patients trial found:
LDL decreased by 21% more in the bempedoic arm at 6 months
4-point MACE (CV death, MI, stroke, or coronary revascularization) was reduced by 1.6% (13.3% placebo vs 11.7% bempedoic acid) ≈ 13% relative difference
No significant difference in cardiovascular death
No significant difference in overall mortality – 6.0% P vs 6.2% B
Safety findings – higher incidence in the bempedoic group for kidney impairment (11.5% vs 8.6%) and elevated liver enzymes (4.5% vs 3.0%)
These data were for the entire trial population. About 30% of patients in CLEAR Outcomes were at high risk for heart disease (primary prevention) and about 70% of patients already had heart disease (secondary prevention).
The June 24, 2023 results for the primary prevention subset (30%) of the CLEAR trial were as follows:
4,206 participants – average age 68, 59% female, mean LDL 143 mg/dL(3.7mmol/L). They were followed for 40 months.
In this pre-specified subgroup analysis the results were:
4-point MACE – reduced by 2.3% (7.6% P vs 5.3% B) – an ≈ 30% relative difference
A 1.3% absolute reduction in cardiovascular death – an ≈ 39% relative reduction
All-cause mortality – reduced by 1.6% (5.2% P and 3.6% B)
Safety findings – in the bempedoic group a higher incidence of gout (2.6% vs 2.0%), cholelithiasis (2.5% vs 1.1%), kidney impairment (10.3% vs 8.1%), hyperuricemia (12.1% vs 6.3%), and elevated liver enzyme levels (4.5% vs 2.6%
The lead investigator, Dr. Steve Nissen, stated in a Medscape article:
"These results are frankly striking" and "These are really large reductions. These results are telling us that high-risk primary prevention patients, although their absolute event rate is lower than secondary prevention patients, can have very impressive relative risk reductions in major cardiovascular events with lipid-lowering therapy.”
When I looked at the primary prevention subset analysis numbers, I wondered if the results were “so good” in this subset then what about what was left in the secondary prevention group - the other 2/3rds of the participants?
Unfortunately, these results weren't reported. (Ed. Note: This is curious and unusual. Common practice in presenting subgroup analyses is to present the results of both subgroups.)
So, I took the numbers from the primary prevention subset analysis and subtracted them from the original overall trial to give me the results from just the secondary group. I link here to a synopsis of all these numbers on my website MyStudies.org.
When you remove the benefit seen in the primary prevention subset, and just look at those participants with a history of CVD:
The 4-point MACE primary outcome loses statistical significance, despite this being a much larger group with more events. The actual numbers are reported in the original article (supplement Figure S3) HR 0.91 (0.82-1.01) - 15.7% P and 14.5% B – a 9% relative difference.
In the original trial there was no overall mortality benefit, so I wondered about the overall mortality in the secondary prevention subset given that the primary prevention subset had shown a difference. Well, in the secondary prevention subset, the overall mortality is higher (15% relative) in the bempedoic arm (7.3% vs 6.4%) but one edge of the confidence interval still suggests this is not quite statistically significant HR 1.15 (0.99-1.31) - so the argument on this endpoint will likely go both ways given the arbitrariness of statistical significance thresholds.
In contrast to the cardiovascular mortality benefit seen in the primary prevention subset, there was actually a 0.8% absolute increase in cardiovascular mortality in the secondary prevention group: HR 1.16 (0.99-1.33).
This is curious because we now have a medication that appears to have an effect on MACE in primary prevention (~30% relative difference) but possibly none or at least less (~8% relative difference) in secondary prevention. This occurs despite the fact the medication reduced LDL equally in both subsets.
In addition, in the primary prevention group all-cause mortality is decreased ( ≈ 27% relative) but in the secondary prevention group certainly there was no benefit and possibly an increase ( ≈ 15% relative) driven by an increased increase in cardiovascular mortality.
Comments:
First, subset analyses are challenging to interpret. The authors write that “such analyses can result in false-positive findings due to the testing of multiple subgroups and may represent the play of chance.”
In fact, a 2017 review suggests that “attempts to corroborate statistically significant subgroup differences are rare; but when done rigorously, the initial claims for subgroup findings are not reproduced.”
I have other concerns:
It is unusual that a medication lowers CVD risk in primary prevention but not at all, or at least a lot less, in secondary prevention.
LDL was lowered by 21% in both the primary and the secondary group yet the hazard ratios for the primary outcome were very different - 0.70 (0.55-0.89) and 0.91 (0.82-1.01) – respectively.
Overall mortality was lowered in the primary care group and possibly increased in the secondary group. So, for this medication, it appears that either the degree of LDL lowering is pretty much irrelevant with regard to benefits, or lack thereof, or there is some inherent cardiovascular toxicity in the secondary prevention group.
(Ed. Note: The CLEAR Outcomes authors did not explain why the survival curves in the active arm vs placebo separate immediately. You would expect a cholesterol lowering effect to take months to years.)
The overall benefit in the primary prevention subset was a 2.3% reduction in MACE and a 1.6% reduction in all-cause mortality over ≈ 3.5 years – but that benefit needs to be balanced against a higher incidence of gout, cholelithiasis, kidney impairment, uric acid, and elevated liver enzyme levels and also a cost of ≈ $4-5,000 USD a year.
Finally, the limitations of my analysis are the numbers are simply based on subtracting the subset group from the full group and I’m assuming this provides a reasonable estimate of what was seen in the secondary prevention group.
I am perplexed by the absence of publishing the secondary analysis results. When reporting subgroups, it seems like a standard to report both groups.
I hope the authors will publish their findings for the secondary prevention subset and if they do, it will be interesting to see what they report and what conclusions they state.
No way do I want to lower my cholesterol since the body makes 80-85% of its own (in at least a dozen different varieties) and for a million good reasons. Why mess with this natural process? And I still won't worry about it after I switch to a carnivore diet.
But go ahead and invent more drugs. After 50 years, the state of America's health is still going downhill despite all these life saving pharmaceuticals.
I love this very clever analysis! My takeaway is we should only be paying attention to the original study as designed, which showed no mortality benefit and a modest reduction in MACE which feels outweighed by the rates of kidney and liver toxicity. I think the most likely explanation of the subset analysis which confusingly has different effects on primary and secondary prevention is that this is noise due to a methodologically flawed approach to interpreting trials.