Deming or Regular Linear Regression: Which is Best?

There are two types of linear regression that can be used to determine the slope and intercept of method comparison experiments:

• Regular Linear Regression makes the assumption that the data plotted on the X-axis has no error. This is most definitely FALSE for essentially all methods in the clinical laboratory.

• Deming Linear Regression (developed by the late QA guru Dr. W. Edwards Deming) makes the assumption that the data plotted on the X-axis has error. This more accurately represents the data in the clinical laboratory.

When does this difference matter?

When the relative errors for the two methods are similar and the correlation coefficient is greater than 0.8, the Regular Regression slope can be approximated as:

R = (Regular slope) / (Deming slope)

where R is the correlation coefficient. This means that the regular slope routinely underestimates the actual slope of the data. For R less than 0.8, the relationship no longer is as accurate. However differences of 20% and more continue to exist between the slopes calculated by the two methods.

For many clinical chemistry procedures R is greater than 0.995, and there is very little difference between Regular and Deming Regression. However, for analytes such as electrolytes and many hematology parameters (especially the white cells), R can easily be less than 0.95, and sometimes in the range of 0.2 to 0.8. In these cases, the use of Deming statistics makes a large difference in the results.

We show both Deming and Regular Regression statistics