Why a DDI Exposure Ratio Needs a Precise Definition
A statement such as “the interaction increased exposure by 20%” is incomplete unless it identifies which exposure measure was compared, over what interval, under which treatment conditions, and for which population. A ratio based on total AUC is not automatically the same estimand as a ratio based on \(C_{\max}\), and an individual subject’s ratio is not the same quantity as a population summary of ratios.
In a drug-drug interaction (DDI) evaluation, the exposure ratio usually compares a victim drug given with a perpetrator drug against the victim drug given without the perpetrator. The comparison can be made for each subject, then summarized across a specified target population. Defining the estimand—the precise quantity the analysis is intended to estimate—before interpreting a result makes the comparison meaningful and reproducible.
Specify the Endpoint and Conditions
AUC is the area under the concentration-time curve over a stated interval. For a dosing interval of length \(\tau\), \(AUC_{0-\tau}\) denotes the area from the start to the end of that interval. \(C_{\max}\) is the maximum concentration over the stated interval or profile, as defined for the analysis. Both endpoints must refer to a clearly specified profile: for example, after one dose or during a dosing interval at steady state. The endpoint definition must be the same in both treatment conditions.
Let \(Y_{i,T}\) be a positive exposure measurement for subject \(i\) under the test condition and \(Y_{i,R}\) the corresponding measurement under the reference condition. For a typical DDI comparison, the test condition is the victim drug with the perpetrator and the reference condition is the victim drug without it. The individual subject ratio is \(R_i=Y_{i,T}/Y_{i,R}\). It is dimensionless when both measurements use the same units and endpoint. A value above 1 indicates higher exposure in the test condition for that subject; a value below 1 indicates lower exposure.
The dose interval is part of the endpoint, not a minor label. For instance, a ratio for \(AUC_{0-\tau}\) over a steady-state dosing interval and a ratio for AUC extrapolated to infinity after a single dose refer to different exposure summaries. Likewise, a partial AUC over a selected time window answers a different question from total AUC. The comparison should identify the interval explicitly rather than rely on the word “AUC” alone.
The treatment conditions also require sufficient detail to distinguish what changed. State the victim drug regimen and whether the perpetrator was present, including any regimen features needed to define the comparison. If victim doses differ between conditions, a raw exposure ratio combines the interaction contrast with the dose difference. A dose-normalized comparison may instead be intended, but it is a different estimand and must be stated explicitly; it should not be silently substituted for the raw ratio.
Worked Example: Writing an AUC Estimand in Full
Suppose a hypothetical study measures steady-state victim-drug concentrations in adults receiving 100 mg once daily, either alone or with a specified perpetrator regimen. The intended endpoint is the AUC during one 24-hour dosing interval. Write the estimand in words and identify the individual ratio.
The endpoint is \(AUC_{0-24\text{ h}}\) at steady state. The test condition is victim drug 100 mg once daily with the perpetrator regimen; the reference condition is the same victim regimen without the perpetrator. The target population is the adults represented by the study’s prespecified analysis population. These details define what is compared; “AUC ratio” alone would not.
For a hypothetical subject with \(AUC_{0-24\text{ h},T}=180\ \text{mg·h/L}\) and \(AUC_{0-24\text{ h},R}=120\ \text{mg·h/L}\), the individual ratio is \(180/120=1.50\). This subject’s measured interval AUC is 50% higher with the perpetrator than without it. This is a statement about that subject’s ratio, not yet a population-level conclusion.
The units cancel: \((\text{mg·h/L})/(\text{mg·h/L})=1\). The ratio is therefore dimensionless, while the two AUC measurements retain their units.
From Individual Ratios to a Geometric Mean Ratio
Exposure measurements are positive and are commonly compared multiplicatively. For \(n\) paired subjects, the sample GMR is the geometric mean of their individual ratios. Equivalently, it is the exponential of the average log ratio:
The paired structure matters: each \(R_i\) uses the test and reference exposure for the same subject. If the same complete set of subjects contributes both conditions, then \(\ln R_i=\ln Y_{i,T}-\ln Y_{i,R}\). Therefore, averaging the log ratios is equivalent to subtracting the mean reference log exposure from the mean test log exposure, and exponentiating gives the ratio of the corresponding geometric means.
Worked Example: Calculating a Paired AUC GMR
Three hypothetical subjects have paired \(AUC_{0-\tau}\) measurements in \(\text{mg·h/L}\). Their individual test-to-reference ratios are 1.50, 0.900, and 1.25. Calculate the GMR and interpret it.
The product of the ratios is \(1.50\times0.900\times1.25=1.6875\). Thus, \(\widehat{\mathrm{GMR}}=1.6875^{1/3}=1.19055\), or 1.191 to three decimal places. On a percentage scale, the GMR corresponds to an estimated \(100(1.19055-1)=19.055\%\), or approximately 19.1% higher geometric mean AUC in the test condition.
The calculation can be checked on the log scale: \(\ln(1.50)=0.405465\), \(\ln(0.900)=-0.105361\), and \(\ln(1.25)=0.223144\). Their sum is \(0.523248\); the mean is \(0.174416\); and \(\exp(0.174416)=1.19055\). Rounding the GMR to 1.190 would be incorrect to three decimal places.
This GMR summarizes the three paired ratios. It does not say that every subject’s AUC increased by 19.1%; one of the three individual ratios is below 1.
Geometric Mean Ratio Is Not the Mean of Ratios
Several quantities can be called a “mean ratio” informally, but they are not interchangeable. The GMR gives equal weight to each subject’s log ratio. The arithmetic mean of individual ratios gives equal weight to the ratios on their original scale. A ratio of arithmetic mean exposures first averages exposures in each condition, then divides those averages. These summaries can differ, especially when exposures or ratios vary substantially.
| Summary | Calculation | What it summarizes |
|---|---|---|
| Individual subject ratio | \(Y_{i,T}/Y_{i,R}\) | The multiplicative contrast for one subject |
| Geometric mean ratio | \(\exp\big(\frac{1}{n}\sum_i\ln R_i\big)\) | The average contrast on the log scale for the specified population |
| Arithmetic mean of ratios | \(\frac{1}{n}\sum_iR_i\) | The arithmetic average of subject-level ratios |
| Ratio of arithmetic means | \(\bar{Y}_T/\bar{Y}_R\) | The ratio of the two arithmetic mean exposures |
Worked Example: How Alternative Summaries Diverge
For three hypothetical subjects, let reference AUC values be 10, 20, and 40 \(\text{mg·h/L}\), and test-condition values be 20, 20, and 20 \(\text{mg·h/L}\). Calculate the individual ratios, their GMR, their arithmetic mean, and the ratio of arithmetic mean exposures.
The individual ratios are \(20/10=2.00\), \(20/20=1.00\), and \(20/40=0.500\). Their GMR is \((2.00\times1.00\times0.500)^{1/3}=1.00\). The arithmetic mean of the ratios is \((2.00+1.00+0.500)/3=1.167\).
The test arithmetic mean exposure is \((20+20+20)/3=20.0\ \text{mg·h/L}\). The reference arithmetic mean is \((10+20+40)/3=23.33\ \text{mg·h/L}\). Their ratio is \(20.0/23.33=0.8571\). Thus, these data produce a GMR of 1.00, an arithmetic mean individual ratio of 1.167, and a ratio of arithmetic means of 0.8571. Each is calculated correctly, but each answers a different question.
The GMR indicates no average multiplicative change on the log scale in this small set, even though the individual subjects differ. The ratio of arithmetic means is below 1 because the large reference exposure for the third subject contributes strongly to the reference arithmetic mean. It should not be reported as though it were the GMR.
Population, Assumptions, and Diagnostics
A population GMR refers to a defined target population, not automatically to every patient who might receive either drug. The analysis population must be described: for example, the eligible adults represented by a study, or a specified subgroup. If the intended population changes, the estimand can change. A result for one subgroup should not be generalized to another without justification.
The usual paired GMR calculation assumes positive, meaningful exposure measurements; a consistent endpoint and interval in both conditions; and comparable measurement definitions. Each included subject must have a valid test and reference measurement for the paired calculation. The sample GMR estimates a population quantity; it is not the exact population value. A confidence interval or other uncertainty analysis, when reported, addresses uncertainty in that estimate and must be interpreted separately from the point estimate.
Check the estimand and data before interpreting a number. Confirm the numerator and denominator labels, units, dose conditions, interval, and analysis population. Inspect individual ratios and log ratios for unusual values, missing pairs, or implausible measurements. Since logs require positive values, a zero or negative recorded exposure cannot be inserted into the standard log-ratio calculation; the data and endpoint handling must be resolved transparently rather than hidden by an arbitrary adjustment.
If the test and reference observations come from different subject sets, the paired subject ratios are unavailable for those unmatched observations. A ratio of separate geometric means might still be defined, but it is not automatically the same paired estimand, particularly if the subject composition differs. Any model-based or adjusted ratio likewise depends on its stated population and model assumptions. The choice of summary should follow the scientific question and be reported explicitly.
Specify AUC or \(C_{\max}\), including the time window, dose interval, and whether the profile is after a single dose or at steady state.
State what the victim drug regimen is in the numerator and denominator and what differs between them, including perpetrator presence.
Identify whose exposures are being summarized and whether the target quantity is an individual ratio, GMR, or another explicitly defined contrast.
Check paired observations, units, positivity, and whether the reported statistic is the intended ratio rather than a different average.
Interpretation and Common Pitfalls
A GMR of 1.191 means the test-condition geometric mean exposure is 1.191 times the reference-condition geometric mean exposure for the specified endpoint and target population, corresponding to about 19.1% higher exposure. A GMR of 0.80 corresponds to 20% lower geometric mean exposure. These relative comparisons do not establish the absolute concentration or clinical consequence; those require the relevant context and evaluation.
A common mistake is to calculate a ratio for every subject, take the arithmetic mean of those ratios, and label the result the GMR. The correction is to take the geometric mean—equivalently, average the log ratios and exponentiate—when the intended estimand is the GMR. Another mistake is to omit the interval and call both an \(AUC_{0-\tau}\) ratio and a partial-AUC ratio simply “the AUC ratio.” State the interval and endpoint each time so the numerical result has a clear interpretation.
Neither the GMR nor any other exposure ratio alone establishes a mechanism or proves a clinical outcome. It quantifies a defined exposure contrast under specified conditions. Its usefulness depends on a well-defined endpoint, appropriate comparison, and a population to which the result is intended to apply.
Check Your Understanding
Answer each question, then compare with the solutions below.
- A study compares \(C_{\max}\) after a victim drug with and without a perpetrator. Name four elements that should be specified to define the exposure-ratio estimand.
- Two paired subjects have test-to-reference ratios 1.44 and 0.64. Calculate their GMR.
- Why can the ratio of arithmetic mean exposures differ from the GMR?
- A reported GMR is 0.75. Interpret it in context without claiming that every subject had the same change.
Solutions. (1) Specify the \(C_{\max}\) profile or interval, the test and reference treatment conditions, the target population, and the intended summary measure. (2) The GMR is \((1.44\times0.64)^{1/2}=(0.9216)^{1/2}=0.9600\). (3) The ratio of arithmetic means weights exposure magnitudes before division; the GMR averages subject-level log ratios. Their calculations and therefore their estimands differ. (4) For the specified endpoint and population, the test-condition geometric mean exposure is 0.75 times the reference geometric mean, or 25% lower; individual subjects may have different ratios, including ratios above 1.