Bland-Altman Plot

The Bland-Altman plot is the standard method for assessing agreement between two measurement methods. A high correlation between two methods does NOT mean they agree —two methods can be highly correlated while one consistently reads 20% higher than the other. The Bland-Altman plot reveals systematic bias and the range of variability in the difference between methods.

STEP 1 —When to Use

Use a Bland-Altman plot when: you are comparing two methods of measuring the same quantity (not comparing treatment effects); you want to assess whether a new method can replace a reference standard; you want to characterize the bias (mean difference) and the precision (limits of agreement) of a new device or assay.

Do not use when: you are comparing treatment effects between groups (use a strip/box/violin plot); you are assessing test-retest reliability (use an intraclass correlation coefficient with its CI, or a Bland-Altman plot on repeated measures —but disclose the difference); you want to show that two methods are "correlated" —correlation is not agreement.

[Image placeholder: Left —scatter plot of Method A vs. Method B (r=0.97, high correlation). Right —Bland-Altman plot of the same data. Y-axis: Method A ~Method B. X-axis: (Method A + Method B)/2. Mean difference line = +4.2 units (systematic bias). Limits of agreement at ﾂｱ1.96 SD of differences shown as dashed lines. Illustrates that r=0.97 竕 agreement: there is a systematic 4.2-unit bias.]

High correlation (r=0.97, left) does not mean agreement. The Bland-Altman (right) reveals a systematic 4.2-unit bias between the methods.

STEP 2 —Journal Requirement

Mean difference (bias) with its 95% CI must be reported.
Limits of Agreement (LoA: mean diff ﾂｱ 1.96 SD of differences) with 95% CIs must be shown.
A pre-specified clinical acceptability threshold must be stated if the plot is used to make a clinical acceptance claim.
n (number of pairs) must be stated.
If there is a proportional bias (difference correlated with mean), this must be disclosed and regression-based LoA reported.

STEP 3 —Licklider Implementation

Input

Two measurement columns (Method A and Method B) from the same subjects
Clinical acceptability threshold (user-defined, optional)
Repeated measures option: if each subject contributes multiple pairs, a modified Bland-Altman for repeated measurements is used

Output

Difference (A ~B) vs. average ((A+B)/2) scatter
Mean difference line (bias) with 95% CI shown as shaded band
95% Limits of Agreement as dashed horizontal lines with 95% CI bands
Proportional bias test: regression of difference on average (if p < 0.05, regression-based LoA auto-calculated)
Clinical threshold lines (if declared) for pass/fail assessment
Zero reference line (for no-bias reference)

[Image placeholder: Licklider Bland-Altman plot. X: mean of A+B, Y: A竏達. Blue dashed line at mean diff = +2.1 units with 95% CI shaded band. Red dashed lines at LoA: +8.4 and ~.2 with CI bands. Green dotted lines at ﾂｱ5 units (pre-declared clinical acceptability threshold). Proportional bias regression line (non-significant, noted in legend). n=40 pairs.]

Licklider Bland-Altman: bias line with CI band, limits of agreement with CI bands, clinical acceptability threshold, and proportional bias test result.

STEP 4 —Draft Output (Draft / Needs review)

Method agreement was assessed using the Bland-Altman method (n = 40
measurement pairs). Mean difference (Method A ~Method B) = +2.1 units
(95% CI [0.8, 3.4]), indicating a small systematic bias. 95% Limits of
Agreement: ~.2 to +8.4 units (95% CI for lower LoA: [~.1, ~.3];
upper LoA: [6.5, 10.3]). A pre-specified clinical acceptability criterion
of ﾂｱ5 units was used; the upper limit of agreement (8.4 units) exceeds
this threshold, indicating that the methods are not interchangeable for
clinical use. No proportional bias was detected (regression of difference
on average: ﾎｲ = 0.02, p = 0.43).