The Danger of Using Total Error Models to Compare Glucose Meter Performance

Area of Science:

• Clinical Chemistry
• Medical Device Performance
• Biomedical Engineering

Background:

• Glucose meter performance specifications often define total error as the limit for 95% of results.
• A widely used model for total error is (average bias) + 2 * (imprecision).
• This model has guided the specification of acceptable bias and imprecision combinations for glucose meters.

Purpose of the Study:

• To evaluate the completeness and reliability of a popular total error model for glucose meters.
• To investigate the impact of interferences, such as hematocrit, on the accuracy of this total error model.
• To compare the model-based approach with direct total error estimation methods.

Main Methods:

• Analysis of a common total error model: Total Error = (average bias) + 2 * (imprecision).
• Examination of glucose meter performance under interference conditions, specifically hematocrit interference.
• Comparison with the direct total error estimation method outlined in CLSI standard EP21-A.

Main Results:

• The popular total error model is incomplete and its conclusions are questionable.
• When interferences like hematocrit occur, the model fails to differentiate between glucose meters with varying performance.
• CLSI standard EP21-A directly estimates total error, avoiding the limitations of the model-based approach.

Conclusions:

• The (average bias) + 2 * (imprecision) model for glucose meter total error is inadequate, particularly in the presence of interferences.
• Hematocrit interference highlights the model's inability to distinguish performance differences between glucose meters.
• Direct total error estimation, as per CLSI EP21-A, provides a more robust method for assessing glucose meter accuracy.