Expectation Tolerance Limits Are they Practical for Setting Qc

							Expectation Tolerance Limits: Are they Practical for Setting Qc Acceptance Criteria?
T. Thway, C. Ray, T. Graves, M. Ma, J. Lee, B. DeSilva
Amgen Inc
Purpose:

The total error (TE) approach was recommended at the Crystal City III conference as the preferred approach for establishing
QC acceptance criteria. Another approach is the 90% expectation tolerance limits/interval (ETL). Both of these parameters
can be derived from method validation experiments. In this study, we evaluated the utility and feasibility of using 90% ETL
for QC acceptance.

Methods:

A retrospective analysis of three methods (two toxicokinetic and one clinical study) was conducted to determine the
suitability of using 90% ETL for QC acceptance criteria. Smith et al. developed a program to estimate both the TE and 90%
ETL. The 90% ETL were computed using data obtained from both the pre-study validation runs (n 6) and the first six in-
                                                                                                    ≥
study runs. Also, a simulation was conducted using a total of 10,000 experimental runs, each consisting of 3 QC levels to
determine the impact of increasing inter-assay coefficient of variation (CV) on QC acceptance and rejection with both TE and
90% ETL.



Results:
A 7.5% change in bias from pre-study (multiple QC preparations) to in-study (single/multiple QC preparations) led to
changes in the 90% ETL and subsequently an increase in the assay failure rate. This could lead to rejection of suitable runs.
The imprecision from pre-study to in-study decreased on average of 2.9% CV, which led to a decrease in the number of failed
runs. This could lead to increased risk of accepting a failed run. From the simulation data, increasing inter-assay %CV of the
method had no effect on QC failure rate when 90% ETL was used. However, increasing inter-assay %CV for the TE, set at
20%, led to increased rejection of QCs.

Conclusion:
The 90% ETL is dependent on the pre-study validation design. Any changes in bias or imprecision will lead to invalid 90%
ETL and subsequently increase the risk of accepting failed QCs or rejecting valid QCs. Without industry recommendations
on acceptable 90% ETL, difficulty in implementation, and risks associated with accepting false positive and false negative
results, we have concluded that TE in the 4-6-X rule is still the most practical and valid approach for setting assay acceptance
criteria for QCs.