3.4 Precision Assessment & Least Significant Change (LSC)

Measure reproducibility under clinical conditions

Precision describes how closely repeated measurements agree. Accuracy describes agreement with the true quantity. In densitometry, bone ash weight provides a reference for bone mineral measurements. A scanner can return nearly identical values on repeat scans and still give biased results if, for example, an artifact is included in the measurement. A clinical precision study asks how much variation remains when patients are scanned and repositioned under normal operating conditions.

A phantom is useful for tracking scanner stability. Repeated phantom scans do not capture patient positioning, motion, anatomy, or analysis choices. The facility measures clinical precision in vivo using patients representative of the group it scans. Obtain participating patients' consent and follow applicable radiologic safety requirements. The International Society for Clinical Densitometry (ISCD) recommends assessing each technologist after basic training and roughly 100 patient scans. Repeat the assessment on a new system or if the technologist's skill level changes.

Reposition the patient after each acquisition. Remove and position the patient again so the repetitions measure the reproducibility of the clinical workflow. Choose the skeletal sites that the facility uses for monitoring. Evaluate each technologist's performance against the applicable standards. When multiple technologists meet the facility's acceptable performance range, combine their precision data to establish the facility precision error and least significant change. A manufacturer specification cannot substitute for local clinical measurements.

Pool within-patient variability

For each patient, calculate the sample standard deviation $s_i$ of that patient's repeat BMD measurements. With the same number of repeats per patient, square each patient's standard deviation, average the squares, and take the square root. Taking the standard deviation of all BMD values in the study would mix technical repeatability with differences between patients.

$$\text{RMS-SD} = \sqrt{\frac{1}{m}\sum_{i=1}^{m}s_i^2}$$

Here $m$ is the number of patients. For a two-scan study, the sample variance for an individual patient is the squared difference between the two scans divided by two. Thus for 30 patients scanned twice:

$$\text{RMS-SD} = \sqrt{\frac{1}{30}\sum_{i=1}^{30}\frac{(x_{i1}-x_{i2})^2}{2}}$$

The ISCD favors root-mean-square standard deviation (RMS-SD) in g/cm². To express relative precision, divide each patient's SD by their own mean BMD, multiply by 100, and take the root mean square of those percentages for the equal-repeat design. A single absolute precision value corresponds to different percentages at different patient BMDs. For example, 0.030 g/cm² is 3% of 1.000 g/cm² and 6% of 0.500 g/cm². These two ratios are not estimates of the facility's RMS-%CV.

ISCD requires retraining if precision is worse than the applicable benchmark. These benchmarks establish a performance standard. Use the locally measured precision error to calculate the actual comparison threshold.

Calculate and interpret the least significant change

The least significant change (LSC) is the minimum absolute difference that reaches statistical significance. When two measurements have independent random errors of equal variance, the standard deviation of the difference is $\sqrt{2}$ times the precision error. The two-sided 95% normal critical value of about 1.96 gives the multiplier $1.96\sqrt{2}\approx2.77$. For comparable scans with an absolute precision error in g/cm²:

$$\text{LSC}_{95\%} = 2.77 \times \text{RMS-SD}$$

Compare the magnitude of the BMD difference to the applicable LSC. A change that meets or exceeds the LSC is statistically significant. Keep the full precision of the values used for the comparison when rounding could change the decision. An observed difference can meet the LSC because of a systematic change, such as an evolving artifact. Review the scans before attributing a significant increase to therapy or a decline to bone loss.

Worked comparisons

• Spine after teriparatide. Comparable L1–L4 measurements rise from 0.720 to 0.765 g/cm². The local precision error is 0.012 g/cm², so the LSC is $2.77\times0.012=0.03324$ g/cm². The gain of 0.045 g/cm², or 6.25%, is statistically significant. The interpreter still checks the images and considers the clinical history before attributing the change to therapy.
• Hip during alendronate therapy. Comparable total hip measurements decline from 0.840 to 0.824 g/cm². The LSC from a precision error of 0.010 g/cm² is 0.0277 g/cm². The magnitude of the decline, 0.016 g/cm², is below that threshold. The report describes no statistically significant change at the total hip.

A different scanner requires a valid inter-system comparison before an LSC can be used. See cross-calibration after a scanner change.

ISCD Adult Official Positions · ISCD Precision Assessment FAQ