Worksheet: TOST Equivalence Calculation for Method Transfer

This is a ready-to-use calculation sheet for the TOST equivalence statistic behind an analytical method transfer’s assay-equivalence criterion. It reproduces the calculation step by step so a reviewer can check the arithmetic without redoing the statistics from scratch. Replace every <<FILL: ...>> placeholder. This content is educational reference, not legal or regulatory or statistical consulting advice; have a qualified statistician confirm the method and any assumptions for your actual data.

Header

Inputs

Calculation steps

1. Difference of means = RU mean - SU mean = <<FILL>>
2. Pooled standard deviation (or state the method used if not simple pooling) = <<FILL>>
3. Standard error of the difference = pooled SD x sqrt(1/n_SU + 1/n_RU) = <<FILL>>
4. Degrees of freedom = n_SU + n_RU - 2 = <<FILL>>
5. One-sided t-value at alpha, df (the multiplier for a 90% two-sided interval at alpha 0.05) = <<FILL>>
6. Margin = t-value x standard error = <<FILL>>
7. 90% confidence interval for the difference = difference of means +/- margin = <<FILL: lower>> to <<FILL: upper>>

Comparison to the predefined limit

Acceptance criteria

• The equivalence limit and alpha used here match the value predefined in the approved transfer protocol; if they do not match, stop and resolve the discrepancy before reporting a result.
• The result is PASS only if the entire calculated confidence interval falls within the predefined limit. A partial overlap is a FAIL.
• All inputs (n, means, standard deviations) trace to the raw data package, not to a rounded or summarized intermediate value.

References

USP General Chapter <1224>, Transfer of Analytical Procedures. A standard reference on two one-sided tests (TOST) for equivalence testing; confirm the specific statistical method and any assumptions (normality, variance homogeneity) with a qualified statistician for your data.

Filled specimen

A completed worksheet for the example transfer used elsewhere in this template set. Illustrative only.

1. Difference of means = 100.8 - 99.6 = 1.2%
2. Pooled standard deviation = 0.9% (pooled estimate from both labs’ variances)
3. Standard error = 0.9 x sqrt(1/6 + 1/6) = 0.9 x 0.577 = 0.52%
4. Degrees of freedom = 6 + 6 - 2 = 10
5. One-sided t at alpha 0.05, df 10 = 1.812
6. Margin = 1.812 x 0.52 = 0.94%
7. 90% CI = 1.2% +/- 0.94% = 0.26% to 2.14%

Note what this worksheet also shows if the limit had instead been predefined at +/-2.0%: the same interval’s upper bound of 2.14% would fall outside the limit, a FAIL, with an identical raw dataset. That is the reason the limit is entered as an input at the top of this sheet and locked before calculation, not chosen to fit the result.

Common inspection findings this worksheet prevents

• The reported confidence interval cannot be traced back to the underlying means, standard deviations, and n used to calculate it.
• The equivalence limit used in the pass/fail comparison does not match the one predefined in the approved protocol.
• A difference test (a t-test p-value) is substituted for the equivalence interval, misrepresenting a non-significant difference as proof of equivalence.

How to adapt this worksheet

1. Set the protocol reference, parameter, and predefined limit in the header before entering any data.
2. Enter the raw n, means, and standard deviations from the actual transfer data package.
3. If your statistical approach differs from simple pooled-variance TOST (for example a log-scale analysis for a bioassay), note the method used in step 2 and have a statistician confirm it is appropriate.
4. Have a second, independent reviewer recalculate and initial the result before it is cited in the transfer report.