6.1 Statistics for Densitometry: Mean, Standard Deviation, and Coefficient of Variation

6.1 Statistics for Densitometry: Mean, Standard Deviation, and Coefficient of Variation

Quick Answer: Mean is the arithmetic average and sets the reference center. Standard deviation (SD) measures spread around the mean and is the unit in which T-scores and Z-scores are expressed — a T-score of −2.5 literally means "2.5 standard deviations below the young-adult mean." Coefficient of variation (CV%) is SD ÷ mean × 100, expressing variability as a percentage of the value measured. The ARRT outline lists all three under Measuring BMD, and every other derived quantity in densitometry is built from them.

Why a Technologist Needs This

A DXA report contains one measurement and several statistics. The measurement is BMD in g/cm². The T-score, the Z-score, the precision error, the least significant change, and the percent-of-mean value are all statistical transformations of it. A technologist who can calculate them can spot an implausible report; one who cannot is dependent on the software being right.

The Mean

The arithmetic mean is the sum of values divided by their count:

$\bar{x} = \frac{\sum_{i=1}^{n} x_i}{n}$

Worked example. Five phantom scans return 1.005, 1.010, 1.008, 1.004, and 1.013 g/cm².

$\bar{x} = \frac{1.005 + 1.010 + 1.008 + 1.004 + 1.013}{5} = \frac{5.040}{5} = 1.008\ \text{g/cm}^2$

In densitometry the mean appears in three places: the phantom baseline mean established from the first set of QC scans, the young-adult reference mean used for T-scores, and the age-matched reference mean used for Z-scores.

The Standard Deviation

Standard deviation quantifies how widely values scatter around the mean. For a sample:

$s = \sqrt{\frac{\sum_{i=1}^{n}(x_i - \bar{x})^2}{n-1}}$

Worked example, continuing the five phantom scans with $\bar{x} = 1.008$:

$s = \sqrt{\frac{0.000054}{5 - 1}} = \sqrt{0.0000135} = 0.00367\ \text{g/cm}^2$

The divisor is $n-1$, not $n$, because a sample standard deviation loses one degree of freedom to the estimated mean. That same degrees-of-freedom logic reappears in the ISCD precision study requirement of 30 degrees of freedom.

The Normal Distribution

BMD in a healthy reference population approximates a normal (Gaussian) distribution, which is what makes standard-deviation scoring meaningful:

Two direct consequences for interpretation:

• A T-score of −1.0 places a patient at roughly the 16th percentile of the young-adult reference (half of the 32% falling outside ±1 SD lies below).
• A Z-score of −2.0 places a patient below approximately 2.5% of the age-matched population — the basis for the ISCD terminology "below the expected range for age."

The 95% figure is also the confidence level chosen for the least significant change, and the associated two-tailed critical value of 1.96 is what eventually produces the familiar 2.77 multiplier.

The Coefficient of Variation

The coefficient of variation expresses SD relative to the mean:

$\text{CV\%} = \frac{s}{\bar{x}} \times 100$

Continuing the phantom example:

$\text{CV\%} = \frac{0.00367}{1.008} \times 100 = 0.36\%$

A phantom CV of 0.36% is typical of a well-functioning scanner; stationary phantom precision is generally better than 0.5% because no repositioning is involved. In vivo precision — a real patient repositioned between scans — is several times worse, typically in the 1% to 2% range, because the technologist's positioning and analysis variability dominate.

Why CV% Is Useful

CV% is dimensionless, so it allows comparison across quantities of different magnitude: the precision of a spine measurement near 1.0 g/cm² and a forearm measurement near 0.6 g/cm² can be compared directly as percentages.

Why CV% Is Also Misleading

Because CV% divides by the mean, it rises as the mean falls, even when absolute error is identical. Consider two patients whose measurements both scatter with an absolute SD of 0.012 g/cm²:

The scanner and technologist performed identically. The osteoporotic patient's CV% is double purely because the denominator halved. If precision error were applied as a percentage, the patient with the lowest bone density — the one most in need of sensitive monitoring — would be assigned the largest threshold for detecting change.

This is exactly why the ISCD recommends expressing precision and least significant change in absolute units (g/cm²) rather than as percentages. It is a small point of arithmetic with a direct clinical consequence, and it appears on examinations.

Where Each Statistic Appears on a DXA Report