Dose-Normalized Bioequivalence With Unequal Doses

Why Normalize Exposure by Dose?

A direct comparison of exposure can be misleading when the test and reference treatments are administered at different doses. If pharmacokinetics are dose proportional, a higher dose is expected to produce proportionally higher exposure even when the formulations have the same exposure per unit dose. Dose normalization expresses exposure per unit administered dose, making a comparison of that quantity possible.

The normalization is straightforward arithmetically, but its interpretation depends on pharmacokinetics. Dividing exposure by dose does not itself correct for nonlinear absorption, dose-dependent clearance, or other dose effects. The key question is whether the exposure measure is proportional to dose over the doses being compared.

Definition: For an exposure measure $E$ and administered dose $D$, dose-normalized exposure is $E/D$. For AUC, its units are AUC units per dose unit, such as $\mathrm{ng\cdot h/mL/mg}$. For $C_{\max}$, its units are concentration per dose unit, such as $\mathrm{ng/mL/mg}$. The dose-normalized treatment ratio is the test treatment's dose-normalized exposure divided by the reference treatment's dose-normalized exposure.

When the dose is the same for both treatments, the dose terms cancel and the normalized ratio equals the raw exposure ratio. With unequal doses, they do not cancel. For test dose $D_T$, reference dose $D_R$, and corresponding exposures $E_T$ and $E_R$, the normalized ratio is

R_{\mathrm{norm}} = \frac{E_T/D_T}{E_R/D_R} = \frac{E_T}{E_R}\frac{D_R}{D_T}.

Thus a normalized ratio of 1 indicates equal exposure per unit dose, not equal total exposure after the administered doses. This distinction is essential when stating the conclusion.

Assumptions and Analysis on the Log Scale

A dose-normalized comparison is meaningful when the dose and exposure have been measured consistently and the exposure measure is approximately proportional to dose across the dose range in question. For AUC, this requires that the relevant processes governing bioavailability and clearance do not create a material dose-dependent change in AUC per unit dose. For $C_{\max}$, proportionality is a separate empirical assumption: a proportional AUC does not guarantee a proportional peak concentration.

For a randomized crossover comparison with both treatments observed in each subject, let $E_{Ti}$ and $E_{Ri}$ be subject $i$'s test and reference exposures, and let $D_T$ and $D_R$ be the corresponding doses. Define the subject-level log contrast

d_i = \ln\left(\frac{E_{Ti}}{D_T}\right) - \ln\left(\frac{E_{Ri}}{D_R}\right).

The mean of these contrasts estimates the log of the geometric-mean dose-normalized ratio. A paired $t$ confidence interval is appropriate when subjects are independent, the within-subject log contrasts are approximately normally distributed without severe influential observations, and the crossover comparison is not materially distorted by period effects or carryover. In a crossover design, period and sequence must be considered as specified by the study design; a simple paired calculation is not a substitute for the planned model when those effects matter.

Result: With $n$ complete independent paired contrasts, sample mean $\bar{d}$, sample standard deviation $s_d$, and $df=n-1$, a two-sided 90% confidence interval for the mean log normalized ratio is $$ \bar{d}\ \pm\ t_{0.95,df}\frac{s_d}{\sqrt{n}}. $$ Exponentiating both endpoints gives the 90% confidence interval for the geometric-mean dose-normalized ratio. For conventional average bioequivalence, compare that interval with the prespecified acceptance limits, commonly 0.80 to 1.25 for the relevant log-transformed exposure measure.

The interval is a confidence procedure for the population geometric-mean ratio under the model, not a statement that there is a 90% probability the fixed parameter lies in this particular interval. The conventional 90% interval criterion corresponds to two one-sided tests at the 5% level when the same limits and analysis are used. Acceptance limits and the metrics required for a particular assessment must be specified for the product and regulatory context.

Worked Comparisons

Worked Example: Normalizing AUC and Cmax

Suppose a hypothetical comparison gives a test treatment at 100 mg and a reference treatment at 200 mg. For illustration, assume the reported exposure values are the quantities being compared: test AUC is $2{,}400\ \mathrm{ng\cdot h/mL}$, reference AUC is $4{,}800\ \mathrm{ng\cdot h/mL}$; test $C_{\max}$ is $120\ \mathrm{ng/mL}$, and reference $C_{\max}$ is $220\ \mathrm{ng/mL}$. Calculate the dose-normalized exposures and ratios.

AUC. The test dose-normalized AUC is $2{,}400/100=24.0\ \mathrm{ng\cdot h/mL/mg}$. The reference value is $4{,}800/200=24.0\ \mathrm{ng\cdot h/mL/mg}$. Therefore,

R_{\mathrm{norm,AUC}} = \frac{24.0}{24.0} = 1.00.

The raw AUC ratio is $2{,}400/4{,}800=0.500$, but that is expected to differ from the normalized ratio because the test dose is half the reference dose.

$C_{\max}$. The dose-normalized test value is $120/100=1.20\ \mathrm{ng/mL/mg}$, and the reference value is $220/200=1.10\ \mathrm{ng/mL/mg}$. Thus

R_{\mathrm{norm},C_{\max}} = \frac{1.20}{1.10} = 1.091.

These calculations describe normalized exposure ratios only. Without subject-level variability and an inferential analysis, they do not establish bioequivalence. Nor does the equal normalized AUC prove that AUC is dose proportional across the dose range; that requires suitable supporting evidence.

Worked Example: A 90% Confidence Interval for Dose-Normalized AUC

Consider a hypothetical crossover study with 24 subjects, each receiving a 100 mg test dose and a 200 mg reference dose. Suppose the 24 paired log contrasts for dose-normalized AUC have mean $\bar{d}=0.0350$ and standard deviation $s_d=0.120$. Assume the complete paired contrasts are independent across subjects, reasonably compatible with a normal distribution, and not dominated by influential values. Also assume the crossover design and its planned analysis justify this paired comparison.

The standard error is

SE=\frac{s_d}{\sqrt{n}} =\frac{0.120}{\sqrt{24}} =0.02449.

With $df=24-1=23$, the two-sided 90% critical value is $t_{0.95,23}=1.714$. The margin of error is $1.714(0.02449)=0.04198$. Therefore, the log-scale interval is

0.0350\pm0.04198 = (-0.006981,\ 0.07698).

Exponentiating the estimate gives a geometric-mean normalized ratio of $\exp(0.0350)=1.036$. Exponentiating the interval endpoints gives

\left(\exp(-0.006981),\ \exp(0.07698)\right) = (0.9930,\ 1.080).

The 90% interval lies entirely within 0.80 to 1.25, so this hypothetical result meets that interval criterion for dose-normalized AUC. The conclusion is specifically about the normalized AUC comparison under the stated assumptions. It does not by itself establish equivalence for $C_{\max}$, demonstrate dose proportionality, or establish equivalence of the actual unequal-dose regimens.

Worked Example: A Dose-Normalized Difference Can Signal Nonlinearity

Suppose hypothetical AUC values are $2{,}000\ \mathrm{ng\cdot h/mL}$ after 100 mg and $3{,}000\ \mathrm{ng\cdot h/mL}$ after 200 mg of the same product, measured under otherwise comparable conditions. The dose-normalized values are

\frac{2{,}000}{100}=20.0,\qquad \frac{3{,}000}{200}=15.0 \quad \mathrm{ng\cdot h/mL/mg}.

The higher-dose normalized value is $15.0/20.0=0.750$ times the lower-dose value. If the AUC measurements and doses are reliable and the conditions are comparable, the data do not support constant AUC per unit dose across these doses. This could reflect dose-dependent pharmacokinetics, although this two-point calculation alone does not identify the mechanism.

In this situation, a formulation ratio based on dose normalization could confound formulation differences with dose dependence. The appropriate conclusion is not that one formulation is necessarily inferior, but that the normalized comparison needs additional justification or a design and model that account for the nonlinearity.

Derivation, Diagnostics, and Limits

The log contrast follows directly from the definition of a normalized ratio. Taking its natural logarithm gives

\ln(R_{\mathrm{norm}}) = \ln\left(\frac{E_T/D_T}{E_R/D_R}\right) = \ln(E_T)-\ln(D_T)-\ln(E_R)+\ln(D_R).

For fixed doses across subjects, $\ln(D_T)$ and $\ln(D_R)$ are constants. Hence the subject-level contrast is the raw log exposure contrast minus the log dose ratio:

d_i = \ln\left(\frac{E_{Ti}}{E_{Ri}}\right) - \ln\left(\frac{D_T}{D_R}\right).

This identity shows precisely what normalization does. It shifts the log exposure comparison by the known dose difference; it does not remove an unknown dose-dependent pharmacokinetic effect. If dose differs among subjects, use each subject's actual dose in the contrast and ensure that the design and model support the resulting comparison.

Diagnostics should examine the dose-normalized exposure across dose levels where supporting data exist, along with exposure-versus-dose patterns and relevant study conduct. A systematic trend in $E/D$, unusual observations, poor concentration-time profiles, or evidence of dose-dependent absorption or clearance weakens a simple normalized interpretation. AUC and $C_{\max}$ should be assessed separately because their dose relationships may differ.

The main alternative is to compare treatments at the same dose, when that is scientifically and practically appropriate. Equal-dose comparisons avoid the arithmetic dose adjustment and make the treatment contrast more direct. If equal doses cannot be used, a model that explicitly represents dose and any supported nonlinear dose relationship may be preferable to assuming proportionality. Neither alternative eliminates the need for an appropriate design, reliable exposure estimates, and a prespecified inferential analysis.

Check the comparison.
Confirm the actual administered doses, exposure units, study design, and whether AUC or $C_{\max}$ is the intended endpoint.

Justify normalization.
Assess whether exposure per unit dose is plausibly stable over the dose range; treat AUC and $C_{\max}$ as separate questions.

Estimate and interpret.
Analyze log normalized exposures using the design-appropriate model, exponentiate the estimate and interval, and compare with prespecified limits.

Limit the conclusion.
State which normalized metric was assessed and avoid extending the result to unequal-dose regimens or other doses without supporting evidence.

Common pitfall. It is incorrect to say, “The raw AUC ratio is 0.50, so the test treatment has half the exposure and fails bioequivalence,” when the test dose was half the reference dose and the target is exposure per unit dose. The correction is to compute $(E_T/D_T)/(E_R/D_R)$, then assess its uncertainty. The converse mistake is also important: a normalized ratio near 1 does not prove that the formulations are equivalent at every dose. That broader claim requires evidence that the normalization assumptions hold at those doses and that the relevant metrics meet their criteria.

Key takeaway: Dose normalization adjusts the scale of exposure for a known dose difference. A defensible bioequivalence conclusion additionally requires a suitable study and analysis, a confidence interval meeting prespecified limits, and evidence that dose-dependent pharmacokinetics do not invalidate the per-unit-dose comparison.

Check Your Understanding

Answer each question, then compare with the brief solutions below.

A test dose is 50 mg and its AUC is $1{,}000\ \mathrm{ng\cdot h/mL}$. A reference dose is 100 mg and its AUC is $2{,}200\ \mathrm{ng\cdot h/mL}$. Calculate the normalized ratio.
Why can equal dose-normalized AUC values coexist with unequal raw AUC values?
A 90% confidence interval for a normalized ratio is (0.78, 1.04). Does it meet the conventional 0.80 to 1.25 interval criterion?
What does a trend in AUC per mg across doses imply for a simple dose-normalized comparison?

Solutions. (1) The test value is $1{,}000/50=20.0$, and the reference value is $2{,}200/100=22.0\ \mathrm{ng\cdot h/mL/mg}$. The ratio is $20.0/22.0=0.9091$. (2) With dose-proportional exposure, raw exposure changes with dose even when exposure per unit dose is equal. (3) No. The lower endpoint, 0.78, is below 0.80, so the full interval is not contained within the limits. (4) It suggests that exposure per unit dose is not constant over the observed range; the simple proportionality assumption needs further assessment and may not justify normalization.

Dose-Normalized Bioequivalence With Unequal Doses

What You'll Learn