← Back to the DEXA-machines question · Front page

Does the machine matter?

Do different DEXA machines report different results consistently?

Client name: Bill · 70 scans · 3 machines · back to the DEXA history

Answer to the presenting question

Yes, one of the three machines reads differently from machine 1 — based on the analysis below, and within its assumptions and limits.

• Machine 3 consistently reports more fat (1.5 lb), a higher body-fat percentage (0.9 points) and higher bone density than machine 1.
• Machine 2 is not clearly different from machine 1.

Do you have a similar question? Join the waitlist.

The clinic you visit has three DEXA scanners and uses whichever is free. Every report names the one it came from, in the file path printed at the foot of each page. Until 2024-03-23 every scan was on machine 1; after that the three rotated. Days scanned on each: machine 1 46, machine 2 14, machine 3 10 (a day with two scans, like the 2025-10-18 test-retest, counts once).

Which machine each scan was on

The codes in the reports are jr1 = machine 1, kh3 = machine 2, kj7 = machine 3. Which code is which machine comes from a diary note: 2025-10-18 was machine 2; 2025-10-20 was machine 3. machine 1 is by elimination, and is also the only code in use before the other two appear. Numbering follows the note: from the machine farthest from the lobby couch.

What the machines disagree about

• machine 2, against machine 1, shows no offset that survives correction and, short of the bar (q ≤ 0.10) but pointing the same way, fat mass · whole body +1.04 lb (q 0.06), body fat · whole body +0.6 % (q 0.06). Total mass shows no offset (q 0.96): the scale agrees; what differs is how the weight is divided between lean and fat.
• machine 3, against machine 1, reads fat mass · whole body +1.55 lb, body fat · whole body +0.9 %, bone mineral density · whole body +0.017 g/cm² and, short of the bar (q ≤ 0.10) but pointing the same way, lean mass · whole body -2.52 lb (q 0.06), lean mass · both arms -0.53 lb (q 0.06), lean mass · trunk -1.98 lb (q 0.06). Total mass shows no offset (q 0.64): the scale agrees; what differs is how the weight is divided between lean and fat.

Each cell: how much that machine's scans read above or below machine 1's, after taking out the slow change in your body (each scan measured against your other scans within 45 days). All machines are estimated together, so one machine's offset cannot leak into another's. Bold where it survives a permutation test corrected for the 18 comparisons (q ≤ 0.05). Only scans from 2024-03-23 on, when more than one machine was in use. Hover a cell for scan counts and the uncorrected p.

What that looks like

Lean mass · whole body (lb)

Fat mass · whole body (lb)

Dots are scans as reported, coloured by machine. The line is the same scans with every estimated offset subtracted, significant or not.

Scans days apart on different machines

The closest thing to a controlled comparison: consecutive scans no more than 7 days apart, on different machines. A body changes little in a few days; these changes are mostly the switch.

Does it change the history?

The main page's lean-mass findings, recomputed with the machine offsets removed:

The last column is the worst case: it takes out even the offsets that did not survive correction, so a finding cannot pass merely because nothing was adjusted. Each cell is a check below: the number must still say what the main page says.

What this cannot tell apart

• Anything that travels with a machine. If one machine was used at a different time of day or on different kinds of days, that shows up here as the machine.
• Machines 2 and 3 have 14 and 10 days of scans. The offsets are estimates from those, and would move with more.
• "Your body that week" is the average of your other scans within 45 days. A real change faster than that (a few days of water loss, say) that happened to land on one machine's scans would read as that machine.
• The report software's file path changed on 2025-02-23 (dashed line above). Whether calibration changed with it is not tested here.

Checked at build time

• ✓ the report for 2025-10-18 names kh3, the machine the diary says (machine 2)
• ✓ the report for 2025-10-20 names kj7, the machine the diary says (machine 3)
• ✓ every scanner code is one of the three known (none unknown)
• ✓ every scan's machine is known (0 without a report: none)
• ✓ significant offsets removed: '2024 minus 2023' in whole-body lean still says what the main page says (+5.6 lb raw, +5.6; floor 1.8)
• ✓ significant offsets removed: '2026 cut minus 2025' in whole-body lean still says what the main page says (-4.7 lb raw, -4.7; floor 1.8)
• ✓ significant offsets removed: '2025 minus 2024' in whole-body lean still says what the main page says (+1.7 lb raw, +1.7; floor 1.8)
• ✓ every estimated offset removed: '2024 minus 2023' in whole-body lean still says what the main page says (+5.6 lb raw, +6.2; floor 1.6)
• ✓ every estimated offset removed: '2026 cut minus 2025' in whole-body lean still says what the main page says (-4.7 lb raw, -5.3; floor 1.6)
• ✓ every estimated offset removed: '2025 minus 2024' in whole-body lean still says what the main page says (+1.7 lb raw, +2.4; floor 1.6)