Several of the EP Evaluator™ (EE5) statistical modules ask the user to input “Allowable Total Error”, ATE for short. One of the most frequent questions we hear from new users is  "What is Allowable Total Error?"

Answer: Allowable Total Error is a statement of your laboratory’s Performance Standards.

In addition to defining it, we are going to talk about where you might get a value for it, show you how to put that value into the right units for EE5, and present a calculator to help you do all of this.

What is Allowable Total Error?

• “The amount of error that can be tolerated without invalidating the medical usefulness of the analytical result.”
                Carey and Garber, 1989
       
• Alternate definition: An administratively acceptable error.

A simplistic, non-medical example

• You intend to fire a missile to destroy a tank. There is a school bus about 2 ½ miles from the tank.
• An error as large as 2 ½ miles means you may hit the bus instead of the tank.
• Your Allowable Total Error is 2 ½ miles.

A medical example

• Normal Fasting Plasma Glucose (FPG) is below 110 mg/dL.
• Individuals with FPG in the 110-126 range are considered to have impaired fasting Glucose.
• An FPG over 126 indicates hyperglycemia which, if confirmed, establishes a diagnosis of diabetes.
• At 110 mg/dL, an 14.5% error ([126-110]/110) results in an incorrect diagnosis.

Practical Limitations

• In EE5, if you set Allowable Total Error LOWER than 3X the SD from routine QC, you are likely to fail the program's accuracy test.
• If you set it HIGHER than your regulatory limits, you aren't really testing accuracy. And you may fail PT at examination time. This is a lot more serious than just failing an accuracy test in EE5!

Next questions …

• My lab hasn’t defined allowable total error for some analytes. How should we go about defining one?
• My lab uses the CLIA Limits for allowable total error. But CLIA doesn’t set standards for all the analytes. And some of the ones that are covered by CLIA are defined as ±3 SD. I can’t enter that into EE5.
• I use the CLIA PT limit of ±10% for Magnesium, and my low linearity sample almost always fails the accuracy test. What should I do?

There isn’t one “right” way to get a number for Allowable Total Error

• Different experts have proposed different approaches; these different approaches give different answers.
• An international conference was held in Stockholm in April 1999 to discuss whether a consensus could be reached on global strategies for setting quality specifications.
• The consensus statement from this conference ranked the strategies in a hierarchy of desirability

Hierarchy of Quality Models

• The best goals are based on medical decision-making.
  • Problem: you often don’t have that information.
     
• Goals based on biological variation are better than goals based on analyzing instrument performance (because they are closer to medical decision-making).
  • Allowable Total Error recommendations have been published for 200+ analytes (reference 1). AKA European Biological Goals, or Desirable Quality Specifications.
  • Problem: because error is expressed only as a percent, these goals may be unrealistic at low concentrations. The only way to pass accuracy at a zero concentration is to have no error at all!
     
• Goals based on peer group analysis are better than goals based on studies of just your own instruments.
  • Problem: peer group studies usually don’t cover the full reportable range.
  • Problem: SD at what concentration? Peer group SD varies with concentration of the PT samples.

Some of the more accessible approaches

• CLIA Limits (75 analytes); other regulatory limits  
• Limits based on biological variation (200+ analytes)
• Limits based on PT surveys: ATE = 3 x Peer Group SD
• Tonk’s Rule (earliest recommended general formula for ATE): Total Error (in percent) should not exceed ¼ of the normal range divided by the mean of the normal range.

Different approaches give different answers; answers may be in different units