Worked Example: Joint AUC and Cmin Attainment

Why Joint Attainment Requires Paired Exposure Results

A regimen can meet an AUC target for many simulated individuals and meet a trough-concentration target for many of them, yet the two marginal percentages do not tell us how many meet both targets. That number depends on whether the simulated individuals who attain one target also tend to attain the other. The calculation must therefore preserve the pairing: each individual’s AUC and \(C_{\min}\) are evaluated together.

Here, \(C_{\min}\) means the minimum concentration during a dosing interval at steady state, which for the model below occurs immediately before the next dose. The example uses a one-compartment model with first-order absorption and elimination after repeated oral dosing. The exposure targets are hypothetical and illustrate the calculation; they are not clinical recommendations.

Definition: For simulated individual \(i\), let \(A_i\) indicate whether that individual’s AUC over one dosing interval is within its target range, and let \(M_i\) indicate whether the individual’s \(C_{\min}\) is within its target range. The joint attainment indicator is \(J_i=A_iM_i\). It equals 1 only when both targets are met. With \(N\) simulated individuals, estimated joint probability of target attainment is \(\widehat{P}_{\text{joint}}=\sum_{i=1}^{N}J_i/N\).

Exposure Quantities for a Simulated Individual

For an individual receiving an oral dose \(D\) every \(\tau\) hours, with bioavailability \(F\) and clearance \(CL\), steady-state AUC over one dosing interval is:

AUC_{\tau}=\frac{F D}{CL}.

If \(D\) is in mg and \(CL\) is in L/h, \(AUC_{\tau}\) is in mg·h/L. This expression assumes linear pharmacokinetics over the dose range: clearance and bioavailability do not change with concentration or dose. For a one-compartment model with volume \(V\), first-order absorption rate constant \(k_a\), and elimination rate constant \(k_e=CL/V\), the steady-state concentration at time \(t\) after a dose is:

C_{\mathrm{ss}}(t)= \frac{F D k_a}{V(k_a-k_e)} \left[ \frac{e^{-k_e t}}{1-e^{-k_e\tau}} - \frac{e^{-k_a t}}{1-e^{-k_a\tau}} \right], \qquad 0\leq t<\tau .

This model assumes repeated identical doses, linear disposition, first-order absorption and elimination, and dosing continued long enough to reach steady state. It does not represent a lag time, multiple compartments, time-varying clearance, or nonlinear elimination. With the stated model, \(C_{\min}\) is the concentration just before the next dose, calculated as the limit of \(C_{\mathrm{ss}}(t)\) as \(t\) approaches \(\tau\) from below.

Worked Example: Calculating One Individual’s AUC and Trough

Consider a hypothetical individual receiving \(D=400\) mg orally every \(\tau=12\) h. Let \(F=0.80\), \(CL=5.0\) L/h, \(V=50\) L, and \(k_a=1.2\) h\(^{-1}\). Assume the individual is at steady state and the linear one-compartment model applies.

First, \(k_e=CL/V=5.0/50=0.10\) h\(^{-1}\). The interval AUC is:

AUC_{\tau}=\frac{0.80(400\ \mathrm{mg})}{5.0\ \mathrm{L/h}} =64.0\ \mathrm{mg\cdot h/L}.

At the end of the 12-hour interval, \(e^{-k_e\tau}=e^{-1.2}=0.3010\), while \(e^{-k_a\tau}=e^{-14.4}\), approximately \(5.57\times10^{-7}\). Substitution in the steady-state concentration equation gives:

C_{\min}= \frac{0.80(400)(1.2)}{50(1.2-0.10)} \left[ \frac{0.3010}{1-0.3010} - \frac{5.57\times10^{-7}}{1-5.57\times10^{-7}} \right] \approx 3.01\ \mathrm{mg/L}.

For hypothetical targets of \(50\)–\(80\) mg·h/L for \(AUC_{\tau}\) and \(2.0\)–\(4.0\) mg/L for \(C_{\min}\), this individual meets both. The two results belong to the same parameter draw and must remain paired when calculating joint attainment.

Estimating Joint and Marginal PTA

A simulation for dose justification draws or otherwise specifies individual pharmacokinetic parameters, calculates exposures under the regimen, and evaluates the target conditions for each simulated individual. For interval targets, attainment means falling within both stated bounds, including the bounds if the protocol defines them as inclusive. The target definitions should be fixed before counting results.

For clarity, suppose there are 10 hypothetical simulation outputs under one regimen. The exposure values below are illustrative model outputs, not observations from a real study. Apply the same two targets to every row.

Individual	\(AUC_{\tau}\) (mg·h/L)	\(C_{\min}\) (mg/L)	AUC target met?	\(C_{\min}\) target met?	Both?
1	64	3.01	Yes	Yes	Yes
2	48	2.4	No	Yes	No
3	72	1.8	Yes	No	No
4	55	2.2	Yes	Yes	Yes
5	82	3.5	No	Yes	No
6	60	4.2	Yes	No	No
7	45	1.7	No	No	No
8	75	2.8	Yes	Yes	Yes
9	68	3.9	Yes	Yes	Yes
10	52	1.9	Yes	No	No

Worked Example: Counting Joint Attainment

Using the table, the AUC target is met by individuals 1, 3, 4, 6, 8, 9, and 10: \(7\) of \(10\). The \(C_{\min}\) target is met by individuals 1, 2, 4, 5, 8, and 9: \(6\) of \(10\). Both targets are met by individuals 1, 4, 8, and 9: \(4\) of \(10\).

\widehat{PTA}_{AUC}=7/10=0.70,\qquad \widehat{PTA}_{C_{\min}}=6/10=0.60,\qquad \widehat{PTA}_{\mathrm{joint}}=4/10=0.40.

Thus the estimated joint attainment is 40% in this small illustrative simulation, compared with marginal attainment of 70% for AUC and 60% for \(C_{\min}\). The joint percentage cannot exceed either marginal percentage, because meeting both necessarily means meeting each one individually.

Why Multiplying Marginal PTAs Usually Fails

Let \(A\) and \(M\) be the binary indicators of meeting the AUC and trough targets for a randomly selected individual from the simulated population. The identity connecting the joint probability with the marginals is:

P(A=1,M=1)=P(A=1)P(M=1)+\operatorname{Cov}(A,M).

This follows directly from the definition of covariance. Since \(AM=1\) only when both indicators equal 1, \(E[AM]=P(A=1,M=1)\), while \(E[A]E[M]=P(A=1)P(M=1)\). Subtracting the latter from the former gives \(\operatorname{Cov}(A,M)\). Therefore, multiplying marginal PTAs gives joint PTA only when the two attainment indicators are independent, so their covariance is zero. Exposure endpoints from the same pharmacokinetic profile are generally related, so independence should not be assumed without justification.

In the 10-individual example, multiplying the marginal estimates gives \(0.70(0.60)=0.42\), not the observed joint estimate of \(0.40\). The difference is small here, but it is not a valid substitute for counting paired results. Depending on the pattern of co-attainment, joint PTA may be greater or smaller than the product of the marginals. It must still be no greater than the smaller marginal PTA.

Worked Example: A Larger Simulation and Monte Carlo Precision

Suppose a hypothetical simulation contains \(N=10{,}000\) individuals. In the paired results, 7,200 meet the AUC target, 6,500 meet the \(C_{\min}\) target, and 5,600 meet both. The estimated joint PTA is \(5{,}600/10{,}000=0.560\), or 56.0%. The marginal estimates are \(0.720\) and \(0.650\); their product is \(0.468\), which differs from the paired joint estimate.

If simulation draws are independent and identically distributed, the joint indicators are Bernoulli with success probability estimated by \(\hat{p}_{\mathrm{joint}}=0.560\). The Monte Carlo standard error is:

SE_{\mathrm{MC}}= \sqrt{\frac{\hat{p}_{\mathrm{joint}}(1-\hat{p}_{\mathrm{joint}})}{N}} = \sqrt{\frac{0.560(0.440)}{10{,}000}} \approx 0.00496.

A rough normal-approximation Monte Carlo interval is \(0.560\pm1.96(0.00496)\), or approximately 0.550 to 0.570. This quantifies simulation sampling error conditional on the model and its inputs; it does not include uncertainty in the pharmacokinetic model, parameter estimates, or target definitions. Nor does it mean that the true population joint PTA has a 95% probability of lying in this particular interval.

Implementation Checks, Alternatives, and Limitations

A dependable calculation keeps each simulated individual’s exposures together from parameter draw through target classification. A useful implementation records the input parameters, regimen, calculated \(AUC_{\tau}\), calculated \(C_{\min}\), and both attainment flags for each draw. Summary counts can then be reproduced from the individual-level flags.

Define both targets.
Specify the units, lower and upper bounds, and whether boundary values count as attainment.

Generate paired exposures.
For each simulated individual, use one consistent set of parameters and the same regimen to calculate \(AUC_{\tau}\) and \(C_{\min}\).

Classify and count.
Create separate AUC and \(C_{\min}\) indicators and a joint indicator that is 1 only when both individual indicators equal 1.

Report estimates and checks.
Divide each count by the same simulation denominator, verify that joint PTA is no larger than either marginal PTA, and assess Monte Carlo precision.

Common failure modes include evaluating \(C_{\min}\) at the wrong time, using a concentration from a single dose instead of the steady-state profile, mixing dose intervals or units, and pairing an AUC from one individual with a trough from another. Such errors can produce plausible-looking percentages while answering a different question. Checks should include a few hand-calculated profiles, confirmation that the regimen and steady-state assumptions are consistent, and a review of the simulated exposure distributions for impossible values or unexpected concentrations.

Direct paired simulation is the primary approach when individual parameters are already being simulated: it estimates the joint event without requiring an independence assumption. Multiplication of marginal PTAs is simpler but valid only under independence. A third possibility is to calculate the joint probability analytically by integrating a specified joint distribution of exposures or parameters over the region where both targets are met. That can avoid Monte Carlo error, but it requires a tractable and credible joint distribution and correct implementation of the target region. These approaches should not be treated as interchangeable when dependence matters.

Joint PTA is conditional on the model, population parameter distribution, covariate assumptions, regimen, and target definitions used in the simulation. It does not establish that the targets predict clinical benefit, and it does not remove model uncertainty. If within-individual variability or measurement error is included, decide whether the target applies to latent model exposure or to a noisy observed measure, and include each source only once. The earlier material on avoiding double counting of variability is relevant to that distinction.

Common Mistake

A frequent mistake is to report \(0.70\times0.60=0.42\) as joint PTA merely because the marginal PTAs are 70% and 60%. This calculation silently assumes that AUC target attainment and trough target attainment are independent. The correction is to form both indicators for every same simulated individual and count the rows in which both equal 1; use the product only if independence is justified for the indicators, not simply because both marginal estimates are available.

Key takeaway: Joint PTA is the fraction of simulated individuals who meet every specified exposure target in the same simulation draw. Preserve individual-level pairing, report marginal PTAs alongside the joint estimate, and do not infer the joint value by multiplying marginals unless independence is justified.

Check Your Understanding

Answer each question, then compare with the solutions below.

In a simulation of 200 individuals, 150 meet the AUC target, 120 meet the \(C_{\min}\) target, and 100 meet both. What are the two marginal PTAs and the joint PTA?
Why is multiplying marginal PTAs not generally a valid way to estimate joint PTA?
If joint PTA is 0.38 and marginal AUC PTA is 0.42, what can be said about marginal \(C_{\min}\) PTA?
For \(N=2{,}500\) independent simulated individuals and estimated joint PTA 0.64, calculate the Monte Carlo standard error using the Bernoulli approximation.

Solutions. (1) The AUC marginal PTA is \(150/200=0.750\); the \(C_{\min}\) marginal PTA is \(120/200=0.600\); and joint PTA is \(100/200=0.500\). (2) Multiplication assumes independence of the two attainment indicators. Paired simulation directly accounts for their dependence. (3) Joint PTA cannot exceed either marginal PTA, so marginal \(C_{\min}\) PTA must be at least 0.38; its exact value cannot be determined from the two figures given. (4) \(SE_{\mathrm{MC}}=\sqrt{0.64(0.36)/2{,}500}=\sqrt{0.00009216}\approx0.00960\), or about 0.960 percentage points.