A Unified Approach for Design and Analysis of Transfer Studies for Analytical Methods

Abstract

Various approaches are compared for the design and analysis of studies to assess the transfer of an analytical method from a research and development site to one or more other sites: comparison of observed bias and precision to acceptance limits, statistical quality control-type analysis, statistical difference tests, and statistical equivalence tests. These approaches are evaluated in terms of the extent to which the risks of incorrect decisions (consumer risk of failing to detect that a site is unacceptable, and producer risk of rejecting an acceptable site) are known and/or controlled. Comparison of observed accuracy and precision to acceptance limits is a flawed approach because both the consumer and producer risks are unknown and uncontrolled. For technology transfec where the objective is to demonstrate sufSicient acceptability or similarity, the statistical quality control and difference tests are well known to suffer from illogical characteristics (decreasing true acceptance probabilities as the sample size increases). The equivalence test is the preferred approach because it alone controls the more important consumer risk and performs in a scientifically logical manne,: Acceptance limits for accuracy and precision in the equivalence test should be based on need for intended use (ie, ensuring thar good batches will pass, and bad batches will fail, during future release testing and stability testing), and a rigorous method for selection of well-conceived limits is presented. Methods for sample size determination are also included. The proposed approach is illustrated with two examples.

Validation and Control of Bioanalytical Methods

“…The school of thought with this approach is that if both methods are valid to within FDA criteria using the same control, then the methods are also comparable within FDAestablished criteria. However, it has been shown that the nonstatistical approach of simply comparing the observed bias and precision between two laboratories to preset acceptance limits can result in both rejection of results that are truly equivalent and acceptance of results that are truly nonequivalent [25]. When using the Student's t-test, this approach controls the false positive error as the level of significance is fixed by performing the test.…”

Methodologies for the transfer of analytical methods: A review

“…The probability that one of these cases arises when deciding about the acceptability of a method transfer depends on the statistical test used [8,[18][19][20]. The consumer risk is the most important one [11].…”

“…Thus, when using the classical approaches both criteria (trueness and precision) should be evaluated before deciding about the acceptability of the transfer. For each of these two criteria different classical approaches are proposed: A descriptive one which compares the point estimates of the receiving laboratory intermediate precision relative standard deviation to an acceptance limit and the point estimates of the bias between the two laboratories to another limit [2,20,25].…”

Beyond the t-Test: Statistical Equivalence Testing

“…Because the required n for TOST is related to and to other parameters discussed earlier, it may be helpful at this early stage to assume a range of potential values as a fraction of the specification or standard that the test is designed to challenge or as a multiple of s; can be refined later. Then, with this series of potential values, along with values for ␣, ␤, ␦, and s, the relationship between these variables may be solved iteratively to yield an appropriate n. Although many approaches exist for determining n (1,10,11,18,19), Excel and the following equation (20) can be used to easily calculate a simplified approximation of the required n for each group: (5) in which the z values are the percentiles of the standard normal distribution, which are available in statistics tables or from the NORMSINV(␣) function in Excel.…”

“…If s is 1.0%, n ≥ 30 is necessary to use = 0.9%. After the appropriate sample size n is chosen, the third step is to take the first set of replicate measurements and estimate s. For better estimates of measurement precision, some approaches to equivalence testing have included multiple analysts and multiple days in data comparison studies (1). However, because of constraints on sample size, time, or resources, it is fairly common to use independent, replicate measurements from a single analyst or a single laboratory to estimate the precision of a measurement.…”