The HMB-FA randomized trial: Not Sure If….
On first glance, I haven’t seen data this promising about a supplement since…ever. HMB (beta-hydroxy-beta-metahybutyrate) has been so _unpromising_ since its emergence back in the late 1990’s/early 2000’s, that I had basically stopped reading about it. And then Andrew Vigotsky mentioned me on Twitter and…well, consider me initially genuinely excited. Which is rare. Like, the next thing that might happen is a rain of frogs. But then I dug a little bit.
Wilson JM, Lowery RP, Joy JM et al. The effects of 12 weeks of beta-hydroxy-beta-methylbutyrate free acid supplementation on muscle mass, strength, and power in resistance-trained individuals: a randomzied, double blind, placebo-controlled study. European Journal of Applied Physiology, DOI: 10.1007/s00421-014-2854-5, Mar, 2014.
So what’s the deal with HMB? The reality is that there hasn’t been much of one. HMB is a metabolite of the amino acid leucine, and while there is still speculation as to the exact mechanisms by which it acts, the theorized function of HMB is to stimulate protein synthesis but also prevent its breakdown (though, probably, not at the exact same time.) The studies up until this one, have generally been quite lackluster with issues such as ‘untrained individuals’ to ‘really short study times’ as well as measurement validity issues being the main reasons why it’s been really hard to get behind this supplement.
HMB is usually supplied as a calcium salt. If you look at the labels of HMB commonly found online or in your supplement store, you’ll likely see “as Calcium-HMB”, which means the HMB molecule is bound to calcium, which, when ingested, dissociates from the HMB so that the HMB can circulate and act. However, a new form of HMB, HMB free-acid (or HMB-FA) was developed and shown to have higher bioavailability (i.e. more of what you take becomes used by your body) and faster peaking levels in the blood. HMB-FA has only been around since around 2011, and I’m not aware of a commercially available formulation yet (at least, none of the major supplement outlets; online or bricks-and-mortar have listed any that I’ve seen.) HMB-FA is served in gel format, as opposed to pill or powder form.
The primary research question of this study was, “Does a 12-week supplementation of HMB-FA with a monitored, periodized resistance-training program improve muscle strength and power more than a placebo?” Although my version of question disagrees with the published purpose of the study, the reporting of the methods, results and analysis suggest that power, hypertrophy/body composition, and ‘muscle decay prevention’ are secondary outcomes—though the data is equally interesting. In the original protocol on the registered trial, the primary outcomes were defined as, “muscle strength, force and power after intense weight training,” with secondary outcomes listed as, “Increased muscle hypertrophy.”
My own private beef with the methods is more in the reporting than anything else. The CONSORT statement has been out since 1996/7 and most journals require authors to meet a fairly easy checklist requirement. I don’t think the study fails because it didn’t follow the checklist, but there _are_ missing elements that detract from the quality of the study as a result. Given the fact that the authors went through the trouble of registering the trial at ClinialTrials.gov in 2012, it would not have been a big feat to report this trial completely.
From the registration page on ClinialTrials.gov, this trial appears to be only a subset of the collected data, which is mildly concerning because either the protocol was changed after data collection started, or data is actively being suppressed or delayed (Edit: Looks like there’s an ahead of pub in JSR on the rest of the data). This has analysis implications on future reports on this dataset, but I’m still pretty excited about what’s here. Final data collection took place in October of 2012.
This trial was designed as a randomized controlled trial. According to the registration page, people were considered eligible for this trial if they met the following criteria:
-18-30 years of age
-at least 3 years of weight training experience
-Squat 150% of their bodyweight
-Bench press 100% of their bodyweight
-Deadlift 150% of their bodyweight
-Had not used creatine or HMB for 6 weeks prior to the study
-No amino acid supplements (it’s not clear if this included protein powder)
-Not on anabolic or catabolic hormones
-Not on medication that would interfere with study measurements
Randomization and blinding
Subjects were divided into quartiles based on their lean body mass and strength (I’m assuming the just added the two together, since it’s quite difficult to arrange a two-dimensional quartile. Each quartile was randomized into one of the two treatment groups (HMB or placebo.) This is basically a way to stratify the randomization by baseline lean body mass and strength to ensure that not all the strong and high-muscle guys end up in one group. While this is fine in principle, the number of subjects involved in the study does make it slightly questionable because there were only 24 subjects in the study, which means each quartile for each of the two stratified factors only had 6 people in it; which then means that once you take the _other_ factor into consideration, you’re randomizing 3 guys in each strata.
The authors don’t say how the random sequence was generated other than by “computer-generated random numbers”.
Packets were made and sent to the research group by an outside researcher who was not involved in data collection or statistical analysis and never met the research subjects. This outside researcher was the only one with the group assignment code, and therefore both subjects and assessors were blinded in this study. The analysis was also sent to an outside researcher who was blinded to the treatment groups.
Presumably the sequence was generated by outside researcher #1. It’s not clear who assigned subjects to groups and whether _that_ was blinded.
Strength testing was a 1RM max test (exact protocol not published, but I don’t think that’s too important, since this is a comparison between groups) for squat, bench press and deadlift.
Peak power was assessed with a modified Wingate and vertical jump test.
Body composition was by DXA scan. Subjects were in a fasted state. Hypertrophy was measured by muscle thickness using ultrasound on the vastus lateralis and vastus intermedius muscles.
Measurements were taken at 0, 4, 8 and 12 weeks.
The authors reported intraclass correlation coefficients as a measurement of test-retest validity for their measurements, but did not mention how these were measured, over how long a period of time, so it’s not ascertainable as to whether these are valid or not.
Subjects either got 3g of HMB (1g, 3 times per day) or placebo. On training days, subjects took their one serving 30 minutes prior to training, and then the remaining two servings at mid-day and evening (training occurred in the mornings.) On non-training days, subjects took their servings with their 3 meals of the day. Empty packets were collected at the beginning of each training day.
Subjects were place on a diet with a 25/25/50 split of protein/fat/carbohydrate. This is presumed to be 3 meals in a day (indirectly inferred from the supplementation schedule, but not verifiable from the paper.) Three-day recall food records were used at the beginning, middle and last week of the protocol.
The full training protocol is available as supplemental material (which is also open access). Basically, weeks 1-8 were 3 days of lifting per week (MWF) where days 1 and 3 were full body workouts (including squat, bench and deadlift) and day 2 was just squat, bench and deadlift. The workout scheme was described as daily undulating, where reps (and thus loads) and rest were varied for each workout. Weeks 9 and 10 were 5 days of lifting per week (M-F), with not as much undulation. Bench press was trained every day, Deadlifts were alternated with Military Press and Squats alternated with Leg Press. Every lifting day was a relatively full body day except Friday which was the 1RM testing day. Weeks 11-12 went back to 3 days per week, with squats, bench and deadlifts present in all three workouts and a closer structure to the daily undulation in weeks 1-8.
All sessions were monitored. If a workout was missed in the morning, it could be made up within 24 hours.
Bloodwork was taken for a bunch of stuff. I don’t think this is quite as relevant in this study, but it’s a nice-to-have.
I am personally not a big fan of ANOVA’s. I think they’re less informative than formal regression, even though they’re pretty much two sides of the same coin. I just think regression is easier to interpret. However, the authors chose to use repeated measures ANOVAs, with the values at week 0 used as covariates. Main effects were time (ie. the week of the protocol), treatment group, and the time X group interaction term. Significant effects were then tested with the Least Squares Mean method.
This does mean a significance test for each outcome variable of bench, squat, deadlift, total strength (the sum of all three), peak power, vertical jump, weight, lean body mass, fat mass, quadricep depth (it looks like they combined the two thicknesses?) This makes 10 tests of significance, which could be problematic without adjustment. This doesn’t include the additional tests performed on just weeks 8-10 to look at the over-reaching effects.
I’m also a little disappointed that they chose to report adjusted means alone without the raw means. The adjusted mean indicates the mean after baseline values were taken into consideration. The fact that no raw values were published makes it difficult to determine whether baseline values were, in fact, substantially different between the two groups. The non-significant baseline one-way ANOVA does not protect against this as has been pointed out in several papers decrying the use of probability testing on baseline values in randomized data sets.
The authors stated that a ‘power analysis’ of lean body mass differences by Kraemer et al (2009) was how they came up with their sample size.
I have an issue with this since their main outcome was strength, but I don’t think this affects the conclusions in any major way.
Of the 24 guys recruited to the study, 4 of them dropped out of the study. Three came out of the placebo group (2 for injury and 1 for time commitment) and one out of the HMB group for injury. All dropouts happened within the first 4 weeks of the study.
On average, the remaining 20 subjects were aged 21.6 (SD 0.5) years, weighed 87.1kg (SD 4.8) in the placebo group and 83.1kg (SD 2.8kg) in the HMB group, and had an average squat, bench and deadlift of 1.7 (SD 0.04), 1.3 (SD 0.04) and 2 (SD 0.05) times their bodyweight. Average height was 180.9cm (no variation reported) in the placebo group and 179 (SD 2.1) in the HMB group. So, roughly 175-200 pounds and a height of roughly 5’9” to 5’11”.
Primary outcome- Strength
If we consider total strength (bench + squat + deadlift) to be the primary outcome of this study, the placebo group gained 25.3kg (SD 22) compared to 77.1kg (SD 18.4) over 12 weeks.
Where things fall down a little bit is that the biggest gains where seen in the squat compared to the placebo group, with the second biggest difference in the deadlift and then bench. Most of the change in total strength, therefore, is probably accounted for by changes in the squat, though the final bench strength between the groups is nothing to sneeze at (116.7kg vs 125.2kg or 255ish pounds vs 275ish pounds on average, but with a fairly large standard deviation which, although statistically significant, makes for a less dramatic practical relevance as the ranges almost overlap.
Secondary outcomes – Body composition
Secondary outcome tests and results are generally considered exploratory. This is because even though they’re planned, they’re not generally adjusted for the multiple testing phenomenon. I’m still excited by the results, but guardedly so as the further down the list we go, the greater the chance of finding a significant result in the sample that wouldn’t otherwise be present.
Both groups gained total weight. Both groups gained lean body mass and both groups lost fat. How much of this is attributable to diet vs exercise vs HMB is not entirely clear, but in the current experimental framework, the only thing that varied was the HMB.
I think the most impressive change was the increase in lean body mass, which was a little over 7kg on average in the HMB group vs a little over 2kg in the placebo group. There was a concommitant loss of fat (on average 2.5kg in the placebo group vs 5.4kg in the HMB group)
There were other outcomes (as mentioned above) but I think I’m going to stop there because I think these are the most salient and interesting points. Besides, the article itself is open access, so you can read more if you want to.
All in all, this study as it’s reported as a pretty well-designed study. However, I have 3 issues with it, one of which does, in my opinion, break it.
When designing a trial, there is an art to sample size estimation. You want to have enough power to detect a relevant difference between groups (i.e. you don’t want to miss a right answer), but sometimes, that difference is so high, that you can find out that you only need a very small number of subjects. This is great from a feasibility point of view, but then detracts from your ability to generalize your results to a wider population. In the case of this study, there were only 24 subjects, only 20 of whom finished the protocol. What we know of them is that they were quite young (20-23 years old), but with moderate training ages (at least 3 years.) We know they were around 20% body fat, and on average, 147lbs of lean body mass when they started the trial. And on the whole, these subjects were pretty strong to begin with—not _super_ strong, but definitely not beginners.
This does limit the generalizability of this data to this population. We can make conjecture as to why 25 years old aren’t that different than 23 year olds and continue this up to 50 year olds and with a variety of training ages, but ultimately, that really is extrapolating beyond the data, and should be done with caution.
Intraclass correlation coefficients are great tools to demonstrate test-retest reliability (ie. that the tool you’re using to measure something gives roughly the same value when you measure the same thing and it hasn’t changed.) I’m just not sure how they measured it in this study. If 1RMs were performed just at the beginning of the study and then once a week in weeks 9-10, I would expect there to be a significant change in weight, thus throwing off the ICC. So the question is when was the second testing done, and how do the authors know that there was no change in strength (or other measures?)
I’m going to flip this around a little and talk about the secondary outcome first.
A change in 15 pounds of lean mass over 12 weeks seems…slightly implausible. If we look at the Bhasin (1996) trial of testosterone + weights in untrained men—a group that should see the most gains over 10 weeks, we see an increase in lean body mass of (on average) 6.1kg or 13ish pounds. This study is suggesting that in _trained_ men, that the gain in lean body mass with HMB-FA use is comparable to (and possibly exceeding) using testosterone in supraphysiological doses (600mg per week) with resistance training. The authors of this study report that there were no changes to free or total testosterone during the study, but have explicitly left this data (though it was collected) out of both the main manuscript and the supplemental material.
It doesn’t matter how you slice this data, (whether by percentage or absolute value), there appears to be an overwhelming response in strength improvements, particularly with an average starting total strength of 426.7 (SD 14.5), which is a respectable 938lbs (roughly and average.) That means that the range of values added in the HMB group was 90lbs at the smallest, and 248 at the greatest (roughly). If we generate 95% confidence intervals for these values, we get 8.4-42.2 kg for the placebo group and 64.7-89.5 kg for the HMB group, which, though pretty wide for both groups, puts the plausible real value of how much improvement one could expect in another experiment to be about 64kg at the worst, if this study is representative of what’s happening in all populations.
If we look at the range of squat values, the Bhasin trial, the testosterone + weights group gained an average of 38kg (83lbs) in their squat (starting from an average of 224lbs). The HMB-FA group in this study gained an average of 36kg (79lbs) in their squat (but from a starting average of 314lbs.) This number mirrors the result found in a previous HMB-Ca (calcium HMB) in 2009 by Kraemer et al, (the direct numbers were not reported, just a figure) where it looks like the average gain in the HMB group was almost 40kg, but with supplementation of HMB-Ca of 1.5g per day, suggesting that if all of this data is to be believed, that there’s either a) no dose-response relationship between the amount of HMB, given a near-identical increase in squat strength whether the protocol used 1.5g of a less bioavailable version of HMB with a slower peaking rate, or 3g of a higher bioavailable HMB with a higher peaking rate; or b) there is absolutely nothing special about free-acid form of HMB compared with the forms of HMB currently available on the market; which have not panned out since HMB isn’t flying off the shelves anywhere close to the same non-flying rate of creatine (which _does_ have a reproducible effect in responders.)
What’s interesting is that most of the other trials on HMB-Ca (which design-wise, are nowhere near this one in quality) show very modest effects (3.2kg over 7 weeks in squat, 7.5kg in bench in 4 weeks) If the supplement is as effective as testosterone, even in a poor trial, the results should be astounding.
The Bottom Line: While the design of this trial is quite good (almost _too_ good), the context of the results as well as some of the reporting quirks (i.e. why both adjusted and non-adjusted means were not reported) lead me to be very guarded about the results of this study. A part of me really wants to believe, because it would be an incredible breakthrough—a FOOD that acts as potently as testosterone? However, I would definitely want to see replication of these results by a different group.
Bhasin S et al. The effects of supraphysiological doses of testosterone on muscle size and strength in normal men. NEJM, 335(1): 1-7, 1996.
Kraemer WJ et al. Effects of amino acid supplement on physiological adapations to resistance training. MSSE, 41:1111-1121, 2009