Interpreting Upper and Lower Limits on the Ratio Scale

Reading Bounds as Ratios

A lower and an upper bioequivalence limit describe an allowed range for a test-to-reference exposure ratio. They do not describe a fixed number of exposure units. If a ratio is 0.80, the test exposure is 80% of the reference exposure; if it is 1.25, the test exposure is 125% of the reference exposure. The limits therefore express proportional comparisons, not equal absolute differences.

For a specified positive exposure metric, let \(\theta\) denote the test-to-reference ratio being evaluated. In standard ratio-based bioequivalence summaries, this is commonly a ratio of geometric means. The limits \(L\) and \(U\) define the interval \(L \leq \theta \leq U\). This tutorial concerns what those bounds mean and why they are usually centered multiplicatively. It does not, by itself, determine whether a study meets all regulatory criteria.

The ratio is meaningful only when numerator and denominator refer to comparable quantities: the same exposure metric, units, and intended comparison. For positive exposure measures, the ratio is positive. Reversing the order changes the interpretation: reference divided by test is \(1/\theta\), not \(\theta\). These basic checks matter before interpreting any limit.

What the Conventional Limits Say

The conventional limits are 0.80 and 1.25, often written as 80% and 125%. A ratio of 0.80 means that test exposure is 20% below reference exposure, because \(1-0.80=0.20\). A ratio of 1.25 means that test exposure is 25% above reference exposure, because \(1.25-1=0.25\). The unequal percentage differences are intentional: the limits are reciprocal.

Thus, if the ratio is reversed, the two conventional limits exchange places. A test-to-reference ratio at the lower limit, 0.80, becomes a reference-to-test ratio of \(1/0.80=1.25\). The reciprocal relationship gives a consistent proportional range whichever product is placed in the numerator, even though the upper and lower percentage differences from 1 are not equal.

Why the Center Is Multiplicative

On an ordinary arithmetic scale, a pair of limits is centered at 1 if the distances from 1 are equal. The arithmetic midpoint of 0.80 and 1.25 is 1.025, not 1. This is not the intended notion of symmetry for ratio limits. Ratios describe proportional changes, and a factor of 1.25 upward has the same magnitude as a factor of \(1/1.25=0.80\) downward. The relevant center is the geometric mean of the bounds.

The logarithm makes this multiplicative symmetry explicit. Taking natural logarithms turns multiplication into addition: the lower bound becomes \(\ln(0.80)\), and the upper bound becomes \(\ln(1.25)\). Since \(1.25=1/0.80\), their logarithms have equal magnitude and opposite signs. The interval is therefore centered at zero on the log-ratio scale, corresponding to 1 on the ratio scale.

This is the reason ratio-based analyses commonly use logarithms: equal proportional departures below and above 1 become equal distances on the log scale. It does not mean that the percentage departures measured from 1 are equal. Rather, the factors needed to move from 1 to either boundary are reciprocal.

Interpreting Other Ratio Limits

Not every product or endpoint necessarily uses the conventional pair. The same interpretation applies to any positive limits, but their relationship should be checked rather than assumed. If the intended criterion is centered multiplicatively at 1, the limits should satisfy \(LU=1\), or equivalently \(U=1/L\). When they do not, their geometric center is \(\sqrt{LU}\), which may differ from 1. Such limits may reflect a different prespecified criterion; do not silently replace them with reciprocal limits.

Common Errors, Diagnostics, and Scope

A reliable interpretation begins by checking the ratio’s direction and then translating each endpoint relative to 1. If the ratio is test/reference, a value below 1 means lower test exposure. If the calculation instead uses reference/test, the direction reverses and the bounds must also be reciprocated. A label such as “80%” is not enough without knowing which product is in the numerator.

• Calling the conventional bounds symmetric in percent. They are not: they represent 20% below and 25% above reference. Their symmetry is multiplicative, or equivalently logarithmic.
• Using the arithmetic midpoint as the center. An arithmetic midpoint answers an absolute-distance question. For reciprocal ratio limits, use the geometric mean to assess multiplicative centering.
• Treating a ratio limit as an exposure amount. A ratio is dimensionless when numerator and denominator use the same units. The limits do not specify an allowed absolute difference in, for example, concentration-time units.
• Reversing the comparison without changing the bounds. If \(\theta\) becomes \(1/\theta\), a bound \(L\) maps to \(1/L\), and \(U\) maps to \(1/U\); the interval endpoints switch order after reciprocation.
• Concluding that a study passes from a point estimate alone. The limits’ interpretation is distinct from applying the full regulatory procedure, including uncertainty and any other required criteria.

The logarithmic view is a useful diagnostic: for reciprocal limits, the log endpoints should be approximately equal in magnitude and opposite in sign, allowing for rounding. If they are not, check whether the limits are intentionally asymmetric, whether values were rounded, or whether the ratio direction has been confused. This check clarifies the scale; it does not justify changing a prespecified margin.

An arithmetic difference can still be useful for a separate question, such as how many exposure units separate two measurements. It is simply not interchangeable with a ratio criterion. Likewise, interpreting these bounds does not settle whether two products are therapeutically similar in every respect or whether all required endpoints satisfy their criteria. The bounds apply to the specific ratio and metric for which they were set.