Applying the EMA Limit Cap

Why the EMA Limit Cap Matters

EMA reference-scaling allows the equivalence limits to widen as within-subject variability for the reference product increases. But that widening is not unlimited: the maximum permitted ratio-scale limits are 69.84% and 143.19%. This tutorial focuses on applying that cap after calculating the variability-based limits.

The cap matters because the uncapped calculation and the limits actually used for an equivalence assessment can differ. When variability is sufficiently high, comparing an interval with the uncapped limits would make the allowable range too wide. The cap prevents that by fixing the effective limits at the stated maximum.

These are ratio-scale bounds. A ratio of 1.00 represents equal geometric means; 0.6984 and 1.4319 correspond approximately to 69.84% and 143.19%. The lower and upper cap values are reciprocal apart from rounding: \(1/1.4319\) is approximately 0.6984.

When Does the Cap Become Active?

The cap begins to bind when the variability-based upper limit exceeds 1.4319, or equivalently when the variability-based lower limit falls below 0.6984. Since the uncapped limits are symmetric on the log scale, both events occur at the same value of \(s_{WR}\). The EMA cap corresponds approximately to a within-subject reference CV of 50%.

Here \(CV_{WR}\) is expressed as a proportion, not as a percentage number: for example, a CV of 50% is entered as 0.50. The relation assumes the usual lognormal conversion between the coefficient of variation and log-scale standard deviation. At a CV of 50%, the corresponding log-scale standard deviation is approximately 0.4724. Multiplying by 0.760 gives an expansion of approximately 0.3590, which produces the cap values after exponentiation.

Thus, below roughly 50% CV, the calculated expansion is within the cap. At roughly 50%, the calculated limits reach the cap. Above that point, the uncapped limits continue to widen, but the effective limits remain 0.6984–1.4319. Because the published cap values are rounded, treat 50% as the practical threshold rather than using tiny rounding differences to distinguish cases immediately at the boundary.

Applying the Cap: Calculation and Justification

The application is a bound-by-bound comparison. The lower effective limit cannot be smaller than 0.6984, so use whichever is larger: the calculated lower limit or 0.6984. The upper effective limit cannot be larger than 1.4319, so use whichever is smaller: the calculated upper limit or 1.4319.

This max/min rule follows directly from the meaning of a maximum expansion. If the calculated lower limit is, for example, 0.67, it is lower than the allowed minimum, so it must be raised to 0.6984. If the calculated upper limit is 1.48, it is higher than the allowed maximum, so it must be reduced to 1.4319. When the calculated values are already inside the cap, neither bound changes.

What the Cap Changes in an Equivalence Assessment

In the earlier material on confidence intervals and the TOST procedure, equivalence is assessed by comparing the relevant confidence interval with prespecified equivalence limits. For a reference-scaled assessment, the limits must be calculated using the applicable scaling rule, including this cap. An interval must not be judged against the uncapped limits when those limits exceed the cap.

Consider the 55% CV calculation above and a hypothetical 90% confidence interval for a test-to-reference geometric mean ratio of 0.6900 to 1.450. Considering only interval containment, this interval is entirely inside the uncapped range 0.6766–1.478. It is not entirely inside the effective range 0.6984–1.4319: its lower endpoint is too small, and its upper endpoint is too large. The capped limits therefore do not support equivalence on this interval-containment check. This comparison does not replace any other requirements of the applicable regulatory procedure.

Common Errors, Diagnostics, and Alternatives

• Using the uncapped limits after they exceed the maximum. This incorrectly allows a wider range than the EMA cap permits. Apply the max/min rule to the calculated bounds before comparing an interval.
• Capping the CV or \(s_{WR}\) instead of the limits. The cap applies to the equivalence limits, not by replacing a high variability estimate with a lower one. Keep the estimated variability unchanged and cap the resulting ratio-scale bounds.
• Entering a percentage as a whole number. Substituting 55 rather than 0.55 into the CV conversion produces a nonsensical result. Convert percentages to proportions before calculating.
• Applying the cap at every level of variability. At a CV of 40%, the uncapped limits remain inside the cap and should not be widened to 0.6984–1.4319.
• Rounding too early. Intermediate rounding can shift a calculated limit, especially near the threshold. Retain several digits through the logarithm, square root, multiplication, and exponentiation; round the reported results at the end.
• Checking only one side of an interval. Interval containment requires both the lower endpoint to meet or exceed the lower limit and the upper endpoint to meet or fall below the upper limit.

A practical diagnostic is to inspect the uncapped limits before applying the cap. If the variability estimate is near the 50% CV threshold, verify the conversion and arithmetic with adequate precision. If variability is well above that threshold but the reported effective limits exceed 0.6984–1.4319, the cap has likely been omitted or applied incorrectly. If variability is well below it but the reported limits equal the cap, check whether the CV was mistakenly entered as a percentage number.

The main alternative to capped EMA expansion is to use conventional, fixed equivalence limits without reference scaling. Those limits do not widen with reference variability, so they answer a different regulatory question and may be difficult to satisfy for a highly variable drug. Uncapped expansion is not an appropriate alternative to the EMA cap: it is only an intermediate calculation used to determine whether the cap binds. Other reference-scaled approaches use different criteria; their formulas and decisions should not be substituted for the EMA limit calculation.

The cap has important limitations. It does not make an endpoint eligible for reference scaling, establish that the study design or variance estimate is appropriate, or guarantee that a confidence interval will fit within the effective limits. It also does not remove uncertainty in the variability estimate. The calculation here treats the estimated \(s_{WR}\) as the input required by the applicable EMA procedure; it does not model the sampling uncertainty of that estimate. Eligibility, estimation, and any additional regulatory conditions must be addressed separately.