Bland-Altman & MDP Guidelines
Bland-Altman Difference Analysis
The Bland-Altman analysis plots the difference between methods against the average of the two methods for each paired measurement. Two plots are generated — one for absolute differences and one for percentage differences:
Absolute Difference Plot
Difference = Y − X; Average = (Y + X) / 2
Plots the raw numerical difference (test minus reference) on the Y-axis against the average of the two methods on the X-axis. Best for analytes where bias is expected to be constant (fixed offset) across the measurement range.
Percentage Difference Plot
% Difference = (Y − X) / X × 100; Average = (Y + X) / 2
Plots the relative difference as a percentage of the reference value. Best for analytes where bias is expected to be proportional to concentration (the method tends to read higher or lower by a consistent percentage).
Each plot displays two reference overlays:
- Mean bias line — a horizontal line at the average difference across all paired measurements, showing the overall systematic offset between the two methods.
- Allowable systematic error bounds — lines derived from the TEa and configured bias allocation percentage, showing the maximum acceptable bias at each concentration. Points outside these bounds represent samples where the observed difference exceeds the allowable limit.
The Other Statistics panel below the charts reports:
- Bias allocation — the configured percentage of TEa allocated to systematic error, shown alongside the raw TEa value.
- Absolute mean bias — the average absolute difference between the test and reference method across all paired measurements.
- Percent mean bias — the average percentage difference between the test and reference method, expressed relative to the reference values.
Bias at Medical Decision Points (MDPs)
Medical Decision Points are specific concentrations of clinical importance — for example, a glucose threshold for diagnosing diabetes, or a troponin cutoff for ruling in myocardial infarction. Evaluating bias at these concentrations is more clinically informative than evaluating bias across the whole range, because errors at decision points directly affect patient management.
When configuring the experiment, you select the regression method to use for MDP analysis — Deming, Passing-Bablok, or Ordinary Least Squares. For each decision level, the module uses the selected regression equation to calculate the predicted test method value and the bias at that concentration, then constructs a 95% confidence interval for the bias as the primary acceptance metric per CLSI EP09C:
Predicted Test Method Value
Predicted Y = intercept + slope × X_c
The value the test method is expected to produce at the medical decision level X_c, derived from the selected regression equation. Both a confidence interval for the mean prediction and a prediction interval for individual observations are computed — the CI for the mean is narrower, while the prediction interval is wider and more conservative as it accounts for variability in individual new measurements.
Bias at a Decision Point
B_c = a + (b − 1) × X_c
The estimated bias of the test method at concentration X_c, where a = regression intercept, b = regression slope, and X_c = the medical decision level. Combines the constant bias (intercept) and proportional bias (slope deviation from 1.0) into a single estimate at the specific clinical threshold.
95% Confidence Interval for Bias (CLSI EP09C)
CI = B_c ± t(α/2, N−2) × SE(B_c)
The primary metric for bias assessment at each medical decision point per CLSI EP09C. Estimates the uncertainty in the bias with 95% confidence. The CI is compared against the Total Allowable Error for acceptance: if the CI falls within the TEa, the bias at that decision level is considered acceptable.
Acceptance Guidance
Unlike some other verification modules, the Quantitative Method Comparison module does not automatically assign a pass or fail status to the study. Method comparison results must be interpreted in context, and the final acceptance decision rests with the laboratory director. CLSI EP09C acknowledges that data points outside the allowable systematic error bounds do not automatically disqualify a method — factors such as pre-analytical variability, a single outlying sample, or the clinical consequences of the bias at a specific concentration all inform the decision.
The following CLSI EP09C-based guidelines can help frame the acceptance decision:
- Review the Bland-Altman plots — If all or nearly all data points fall within the allowable systematic error bounds across the full measurement range, the method is demonstrating acceptable bias and accepting the study is well supported.
- Evaluate medical decision points — Bias that falls outside the TEa at a clinically critical concentration (e.g., a diagnostic cutoff) is more consequential than bias at an extreme of the measurement range with limited clinical relevance. The 95% CI for bias at each MDP is the primary CLSI EP09C metric — if the CI falls within the TEa, the bias at that level is statistically acceptable.
- Assess the regression — A slope close to 1.0, an intercept close to 0, and a strong correlation coefficient (r ≥ 0.95) across a data set that covers the clinically relevant range all support acceptance.
- Investigate outliers before deciding — If one or more points fall outside the allowable bounds, determine whether a valid, documented cause exists (e.g., sample handling error, instrument flag, hemolysis). A well-characterized outlier may support acceptance; unexplained, systematic deviations outside the bounds generally do not.
- Consider the clinical impact — CLSI EP09C recommends that the magnitude of bias be weighed against its clinical consequences. A small bias that technically exceeds the TEa at only one concentration may be clinically inconsequential; the laboratory director must make this determination and document the rationale.