Troubleshooting / Common Issues
Bias exceeds the allowable threshold at one or more medical decision points
Possible causes
- –Calibration difference between the two methods — verify calibrator traceability
- –Reagent-related systematic offset — check reagent lot and formulation differences
- –Matrix-related interference at specific concentration levels
- –Constant or proportional bias introduced by pre-analytical differences (sample handling, tube type)
Recommended action
Examine the Bland-Altman plot to determine whether the excess bias is constant (offset present at all concentrations) or proportional (bias grows with concentration). Constant bias may indicate a calibration issue; proportional bias may indicate a reagent or method characteristic. Investigate and repeat the study after corrective action.
Poor correlation (r < 0.90) despite reasonable Bland-Altman bias
Possible causes
- –Samples clustered in a narrow concentration range — the measurement range does not cover low and high values
- –High within-method imprecision inflating scatter around the regression line
- –An insufficient number of samples to establish reliable regression parameters
Recommended action
Review the scatter plot for the concentration distribution of samples. CLSI EP09C requires samples spread across the full clinically relevant range. If most samples are clustered in the normal range, add samples from known pathological populations (high and low concentrations). A correlation coefficient alone is sensitive to range restriction — low r with good Bland-Altman agreement usually indicates a range problem, not a true method failure.
Outliers on the scatter plot or Bland-Altman plot
Possible causes
- –Transcription or data entry errors
- –Sample mix-up or mislabeling (sample tested on wrong day or replicate)
- –Interfering substance in a specific patient sample (hemolysis, lipemia, icterus)
- –Instrument error during a specific measurement
Recommended action
Review the raw data for the flagged point. If a valid, documented reason exists (instrument error, confirmed sample handling issue), the point may be excluded within the platform. All exclusions must be documented with justification. Do not exclude data solely because it exceeds the TEa — this is not a valid statistical reason for exclusion.
Large difference between Deming and OLS regression results
Possible causes
- –Significant imprecision in the reference method — OLS underestimates slope when X contains error
- –Outliers exerting disproportionate leverage on the OLS fit
Recommended action
When the reference method has meaningful imprecision, Deming regression is the appropriate choice and will yield a more accurate slope estimate. If the discrepancy is large, verify the precision of the reference method. Consider using Passing-Bablok regression as an independent non-parametric check — close agreement between Deming and Passing-Bablok supports confidence in the regression estimates.
Confidence interval for bias is very wide
Possible causes
- –Too few paired samples (below 40 or well below 100)
- –High between-sample variability in the difference (high SD of bias)
- –Concentration range too narrow — regression is poorly constrained
Recommended action
Increase the number of paired samples, ensure samples are spread across the full measurement range, and verify that replicates were handled identically on both methods. A wide CI means the bias estimate itself is imprecise — pass/fail determinations based on a wide CI are unreliable.