Dan Dascalescu's Homepage
/reviews/fitness/how accurate is dxa

Body-Spec DXA reliability and accuracy

Dual-energy X-ray absorptiometry (DXA, formerly DEXA) is regarded as the gold standard for determining body fat percentage and lean mass. Several studies tout its excellent accuracy (1, 2, 3).

The cost for DXA has gone down to only ~$45 in California as of 2016, thanks to mobile truck services like Body-Spec. But how accurate is DXA? A number of different studies suggest DXA is not as reliably as commonly believed (1, 2, 3, 4):

This study shows that present interunit variability allows comparisons among laboratories, but only if highly standardized analysis procedures are used. [...] [B]oth at group and individual levels DXA did not present an expected accuracy in tracking changes in adiposity in elite male judo athletes.

Our results suggest that, even when using the same device, the between-centre variability remains a matter of concern, particularly where body composition is concerned. [...] High-quality operator performance is, however, a prerequisite for proper QA in any setting.

So how accurate is DXA in a truck setting, with operators who likely aren't trained to the level of a permanent lab?

In January 2016, I spent $90 (donations are welcome below) to test Body Spec on two consecutive days, and determine DXA variability. I tried to keep everything as constant as possible: truck, time of day, food, what I was wearing:

DXA tests on consecutive days
Date Truck Time of day Wearing Food / Hydration
2016-01-13 Parked at Google 10:48 am Shorts, t-shirt, socks Overnight fast
2016-01-14 -- same truck -- 12:24 pm -- same -- Overnight fast; might have had more water

I also tried to keep the day of the week constant, but that didn't quite work out. On a more serious note, I could't get an appointment for exactly the same time of day, so there was a ~90-minute difference. Still, I was fasting on both days. Hydration might have varied by a cup of water.

The DXA machine used was a GE Healthcare Prodigy. The operators were different - I got the usual girl the first day, and the Body-Spec founder the next. As you can see from how well they ensured my body was centered on the table, there was a difference in operator performance:

DXA is unreliable

The results of two DXA scans on consecutive days

To my surprise, the scan showed that I had lost 1.8lbs of lean tissue (0.8kg) overnight!

DXA is unreliable

This really undermines the trust I'm going to place in DXA measurements.

Before going in for the second test, I suspected that the machine's software could notice that I had two DXA scans one day after another, and compensate (aka fake data) for that. But they don't seem to do that, given how bad the difference is. Also, due to some technical glitch, I was scanned as a new patient the second time, then the operator moved the scan into my profile. I actually have a separate printout for the "orphan" scan.

Why this is a problem

Let's say this is the margin of error of DXA - ~2 lbs of fluctuation in lean mass overnight. If you look at the oldest two measurements in the chart, you'll see 117.3 lbs of lean mass on 4/16/2015 and 121.5 on 6/23/2015. A nice gain of four pounds of muscle after training for two months on the StrongLifts 5x5 program!

O not. This gain might actually be zero, if each measurement was within the margin of error, and at opposite ends. In other words, my 4/16 real value could've been 119.3 and the 6/23 one 119.5, so no progress. Working hard for two months without progress would upset me. I'm a beginner bodybuilder doing SL5x5, so muscle gain should happen.

DXA is unreliable. Where are the inaccuracies?

The Body-Spec scan report breaks down body composition per region: arms, legs, trunk, abdominal region. What changed from one day to the next?

Fat per body region, in lbs
Date Arms Legs Trunk Android (abdominal) Gynoid (hips, buttocks) Total
Jan 13 4 12.9 17.6 2.6 6.6 36.7
Jan 14 4 12.2 18.4 2.6 6.7 37.0

The fat percentage was essentially the same (0.2 - 0.3lbs of variation out of 37lbs is less than 1%). Now let's look at lean tissue.

Lean tissue per body region, in lbs
Date Arms Legs Trunk Android Gynoid Total
Jan 13 14.2 37.6 58.1 8.6 19.7 117.9
Jan 14 14.2 35.8 (-4.8%) 57.4 7.8 (-9.3%) 18.9 (-4.1%) 116.1 (-1.5%)

A few things to notice:

  1. The biggest lean tissue loss was in the abdominal region. We can chalk this off as due to the contents of my stomach (maybe I had an anti-gravity device in there).
  2. The second biggest muscle loss was in the legs (almost 5%). A little more difficult to ignore.
  3. The numbers don't add up! If you sum up the muscle loss in the four body sections that had muscle loss, that gives -4.1lbs. The Total column shows -1.8lbs. For the Fat table, the numbers do add up. I'm not sure what to make of this. The previous report I got, 1/13/2016 vs. 6/23/2015, also has body section mass change numbers that don't add up, this time for both fat (-0.3, -0.3, -4.0, -0.8, -0.4 vs. -4.5 total) and lean tissue (-0.4, -1.6, -1.8, -0.5, -0.9 vs. -3.6).


The best indicator of how well a fitness program works in the end is the mirror, but that can take months. I was hoping that DXA scans could reveal signs of progress (or lack thereof) faster, so I didn't have to waste time on programs that potentially didn't do much. I was also hoping that precise data could be used to compare results, but a loss of almost 2lbs of lean mass overnight blows any gains away. Beginner body builders can hope to gain two pounds in a month, and those who've been training for several years see much finer changes. From BodyBuilding.com:

Provided that they follow a sensible, structured diet and training program, a 150-pound beginner fitness enthusiast in Aragon's model can potentially gain 18-27 pounds of lean muscle per year.

Assuming this machine wasn't defective somehow, if you want to quantify muscle gain, DXA might be worth doing 4+ months apart - for an error rate of 25%, provided you train hard. But by then, the mirror will show results already. More often than that, DXA is probably a waste of money.