Predicting Cmax Ratios Under Time-Varying Interactions

Why a Peak Ratio Needs the Whole Profile

A perpetrator drug can alter victim-drug clearance or bioavailability over time. If that effect changes during a victim dosing interval, the interaction profile may rise or fall differently at different times. Its peak may therefore occur at a different time from the control peak. A comparison at one selected time is not necessarily a comparison of the two peak concentrations.

This matters because Cmax and AUC summarize different features of a concentration-time profile. Cmax is the highest concentration attained; AUC represents exposure accumulated over time. A longer-lasting concentration increase can raise AUC substantially without producing a similarly large peak increase. Conversely, changes during absorption can move or sharpen the peak even when the overall exposure change is more modest.

Definition: Let $C_{\mathrm{ctrl}}(t)$ denote the victim-drug concentration over the specified dosing interval under control conditions, and let $C_{\mathrm{DDI}}(t)$ denote its concentration under the drug interaction. The peak-concentration ratio is $$ R_{\mathrm{Cmax}} = \frac{\max_{t\in\mathcal{T}} C_{\mathrm{DDI}}(t)} {\max_{t\in\mathcal{T}} C_{\mathrm{ctrl}}(t)}, $$ where $\mathcal{T}$ is the same clinically relevant time interval for both profiles. A ratio above 1 indicates a higher peak under interaction conditions; a ratio below 1 indicates a lower peak. The percentage change is $(R_{\mathrm{Cmax}}-1)\times 100\%$.

The maxima are found separately. In general, the ratio is not the interaction concentration divided by the control concentration at the control Tmax. It is also not obtained by applying an AUC ratio directly to Cmax. Either shortcut requires additional assumptions that may not hold when perpetrator effects vary with time.

Linking the Peak to a Time-Varying Model

A simple model shows why the entire concentration history matters. Suppose victim drug is given orally into a one-compartment system with first-order absorption, constant apparent volume $V$, dose $D$, and absorption rate constant $k_a$. Let elimination be first-order with a possibly time-varying rate $k(t)$. Assume linear kinetics, no time-varying change in $F$, $V$, or $k_a$, and a concentration of zero before the dose. Then the concentration satisfies

\frac{dC(t)}{dt} = \frac{F D k_a}{V}e^{-k_a t}-k(t)C(t).

The first term is drug entering the systemic circulation per unit volume; the second is drug being eliminated. Solving this linear differential equation gives the concentration at time $t$ as the sum of the surviving contributions from input at all earlier times:

C(t)= \frac{F D k_a}{V} \int_0^t \exp\!\left[-k_a s-\int_s^t k(u)\,du\right]ds.

The inner integral, $\int_s^t k(u)\,du$, is the cumulative elimination effect between the time $s$ when drug entered and the later time $t$. Thus, the concentration at a candidate peak depends on the elimination history over that interval, not just on the perpetrator effect at the instant of the peak. If $t^*$ is an interior maximum and the profile is differentiable there, its slope is zero. The model therefore requires

\left.\frac{dC(t)}{dt}\right|_{t=t^*}=0 \quad\Longrightarrow\quad \frac{F D k_a}{V}e^{-k_a t^*}=k(t^*)C(t^*).

At an interior peak, input and elimination rates balance. A time-varying $k(t)$ can change both the concentration accumulated before that balance and the time when it occurs. A peak can also occur at a boundary of the interval; the zero-slope condition applies only to an interior peak.

For a complete profile that returns to zero, integrating the differential equation over time gives a further distinction. The time integral of the input rate equals the time integral of the elimination rate, so

\frac{F D}{V}=\int_0^\infty k(t)C(t)\,dt.

This is a weighted relationship involving concentration throughout time. It does not determine the maximum concentration from AUC alone, or AUC from the maximum alone. If bioavailability or absorption also varies under the interaction, those changes must be represented in the input term as well; the simple model above is not sufficient by itself.

Worked Example: A Higher AUC and a Higher Cmax

Consider hypothetical control and interaction profiles for the same dose and dosing interval. Concentrations are in mg/L and times are in hours. The table shows output from a hypothetical PBPK simulation. The displayed points are used for a trapezoidal AUC approximation; assume that evaluation of the full simulated profiles confirms their maxima occur at 2 hours.

Time (h)	Control (mg/L)	Interaction (mg/L)
0	0	0
1	7.2	8
2	9	10
4	6.3	7
8	3.6	5
12	1.8	4
24	0.45	2.4524

The control maximum is $9$ mg/L, and the interaction maximum is $10$ mg/L. Therefore,

R_{\mathrm{Cmax}}=\frac{10\ \mathrm{mg/L}}{9\ \mathrm{mg/L}} =1.111, \qquad (1.111-1)\times 100\% \approx 11.11\%.

The predicted peak is about 11.11% higher under interaction conditions. For the control profile, the trapezoidal AUC approximation is

\mathrm{AUC}_{\mathrm{ctrl}} = \frac{0+7.2}{2}(1) +\frac{7.2+9}{2}(1) +\frac{9+6.3}{2}(2) +\frac{6.3+3.6}{2}(4) +\frac{3.6+1.8}{2}(4) +\frac{1.8+0.45}{2}(12) =71.10\ \mathrm{mg\cdot h/L}.

For the interaction profile, the corresponding calculation is

\mathrm{AUC}_{\mathrm{DDI}} = \frac{0+8}{2}(1) +\frac{8+10}{2}(1) +\frac{10+7}{2}(2) +\frac{7+5}{2}(4) +\frac{5+4}{2}(4) +\frac{4+2.4524}{2}(12) =110.7144\ \mathrm{mg\cdot h/L}.

The approximate AUC ratio is $110.7144/71.10\approx1.5572$, or about a 55.72% increase. In this example, AUC rises more than Cmax. The peak calculation is $10/9$, so reporting a 10% increase would be incorrect: the increase is about 11.11%.

Worked Example: A Higher AUC but a Lower Cmax

Now keep the same hypothetical control profile but consider a different interaction profile, with concentrations $0, 6, 8, 5, 4, 3,$ and $2.6191$ mg/L at the times shown in the prior example. Assume the full simulated interaction profile has a maximum of $8$ mg/L. The peak ratio is

R_{\mathrm{Cmax}}=\frac{8}{9}=0.8889, \qquad (0.8889-1)\times100\%=-11.11\%.

Thus the predicted Cmax is about 11.11% lower, not 10% lower. Applying the trapezoidal rule to the interaction points gives

\mathrm{AUC}_{\mathrm{DDI}} = \frac{0+6}{2}(1) +\frac{6+8}{2}(1) +\frac{8+5}{2}(2) +\frac{5+4}{2}(4) +\frac{4+3}{2}(4) +\frac{3+2.6191}{2}(12) \approx88.7146\ \mathrm{mg\cdot h/L}.

The AUC ratio is approximately $88.7146/71.10\approx1.2477$, about a 24.77% increase, even though Cmax decreases. This is possible when the interaction produces a broader, more sustained profile without a correspondingly high peak. The example is a calculation from hypothetical profiles, not a claim that any specific mechanism must produce this shape.

Peak Timing and the Choice of Comparison

A delayed or time-varying perpetrator effect can move the interaction peak. The next hypothetical example illustrates why one must maximize each profile independently. Suppose the control profile has a maximum of $9$ mg/L at 2 hours, while the interaction profile has $7.5$ mg/L at 2 hours and a maximum of $10$ mg/L at 4 hours. The correct ratio is

R_{\mathrm{Cmax}}=\frac{10\ \mathrm{mg/L}}{9\ \mathrm{mg/L}}=1.111.

The ratio at the control Tmax would instead be $7.5/9=0.8333$. That value describes the concentration comparison at 2 hours; it is not the Cmax ratio. Here it would suggest the wrong direction of the peak change. In a time-varying interaction, the interaction profile may be affected differently before, near, and after the control peak. The relevant Cmax ratio is based on the two maxima, regardless of whether their times coincide.

AUC and Cmax should therefore be calculated as separate endpoints from the same appropriately defined profiles. AUC aggregates concentrations over a specified interval; Cmax selects the highest concentration in that interval. Neither endpoint can generally be recovered from the other without strong model-specific assumptions.

Practical Calculation and Diagnostics

Define comparable profiles.
Use the same victim dose, relevant dosing interval, units, and endpoint definition in control and interaction conditions. State whether profiles represent one dose or steady state.

Represent the changing interaction.
Use a time-resolved perpetrator effect and its consequences for victim-drug processes, rather than assuming a single constant clearance multiplier when that assumption is not supported.

Locate both maxima.
Search each full concentration-time curve over the same interval. Record the peak concentration and its time for each condition; do not assume the two Tmax values match.

Compute and interpret the ratio.
Divide the interaction maximum by the control maximum, then calculate $(R_{\mathrm{Cmax}}-1)\times100\%$. Retain concentration units for the maxima; the ratio and percentage change are dimensionless.

Check numerical resolution.
Refine the simulation time grid around each peak and confirm that the reported maximum is stable. When comparing to clinical observations, consider whether the sampling schedule could miss a narrow or shifted peak.

A practical diagnostic is to compare each predicted Tmax with the timing of perpetrator exposure and the expected onset or offset of its effect. A markedly shifted peak can be plausible, but it should prompt checks that perpetrator dosing, victim dosing, and the time course of the interaction were implemented as intended. Also check that the grid extends far enough to capture the relevant maximum and that numerical interpolation has not created an artificial spike or smoothed away a narrow one.

Sampling is a separate issue from the model’s continuous-time Cmax. The maximum of a sparse set of observed concentrations is only the maximum among those samples and may be lower than the true peak if sampling misses it. Conversely, predictions from an excessively coarse simulation grid may fail to resolve the modeled peak. Aligning the endpoint definition with the available sampling schedule is essential when comparing simulated and observed Cmax values.

Approaches, Assumptions, and Limits

A constant-effect approximation can be useful when the perpetrator effect is effectively stable over the victim dosing interval and the relevant victim-drug processes remain linear. In that restricted setting, a time-invariant model may be sufficient to calculate a profile and its maximum. But a single average effect can conceal a delayed onset, offset, or changing degree of inhibition or induction. When those time courses matter, a time-resolved PBPK prediction is more informative because it represents the full profile and permits the peak time to shift.

The convolution model above is a justification, not a universal PBPK model. It assumes a one-compartment victim, first-order absorption and elimination, and time-varying elimination as the only interaction-related change. Real predictions can involve multiple tissues, distribution, nonlinear processes, changes in absorption or bioavailability, and time-varying perpetrator concentrations. In those cases, use the relevant model’s predicted concentration profile and calculate the maxima from that profile rather than applying the simple equation outside its assumptions.

A further limitation is that a deterministic predicted ratio does not by itself describe uncertainty in parameters, individuals, or a clinical trial. It is a point prediction conditional on the model, inputs, and endpoint definition. The interpretation of Cmax also does not establish clinical consequence: clinical relevance depends on the therapeutic context and is not determined by the ratio alone.

Common Pitfalls

Comparing only at control Tmax. This gives a same-time concentration ratio, not necessarily the ratio of maxima. Find each profile’s maximum separately.
Assuming an AUC ratio is a Cmax ratio. The endpoints summarize different profile features. Calculate Cmax from the maximum concentrations.
Reporting a percentage with the wrong denominator or arithmetic. Use $R_{\mathrm{Cmax}}-1$, where the ratio is interaction maximum divided by control maximum. For example, $10/9=1.111$, which is an 11.11% increase.
Trusting a coarse grid without checking. A narrow peak can fall between time points. Refine the grid or use an appropriate peak-finding method on the continuous model output.
Attributing every profile change to clearance. Changes in input, distribution, or other modeled processes can also affect Cmax. Interpret the ratio in light of the actual model assumptions.

Key takeaway: Under a time-varying interaction, calculate Cmax as the maximum of each complete concentration-time profile, then take the ratio of those maxima. AUC and Cmax can change by different amounts or in opposite directions, and their relationship depends on the shapes and timing of the profiles.

Check Your Understanding

Answer each question, then compare your reasoning with the solutions below.

A control profile has Cmax $=12$ mg/L, and an interaction profile has Cmax $=15$ mg/L. What is the Cmax ratio and percentage change?
The control maximum is $8$ mg/L at 2 hours. The interaction concentration is $6$ mg/L at 2 hours, but its full-profile maximum is $9$ mg/L at 5 hours. What is the Cmax ratio? Why is $6/8$ not the answer?
Can AUC increase while Cmax decreases? Give the central profile-based reason.
In the time-varying elimination model, what equality holds at a differentiable interior peak?
Name one way sparse sampling or a coarse simulation grid can bias a Cmax estimate.

Solutions. (1) $R_{\mathrm{Cmax}}=15/12=1.25$; the predicted peak is 25% higher. (2) The ratio is $9/8=1.125$, an increase of 12.5%. The value $6/8=0.75$ compares concentrations at 2 hours only, not the two maxima. (3) Yes. A broader interaction profile can maintain elevated concentrations for longer while having a lower maximum, so its time-integrated exposure can rise even as its peak falls. (4) Input rate equals elimination rate: $(FDk_a/V)e^{-k_a t^*}=k(t^*)C(t^*)$. (5) A narrow peak between observation or simulation time points may be missed, causing the maximum among recorded points to underestimate the continuous-profile maximum.