Why Dose Proportionality Can Fail
Dose selection often relies on the expectation that doubling a dose doubles exposure. That expectation is appropriate only under conditions such as linear pharmacokinetics: clearance and bioavailability remain constant over the dose range, and the same regimen and exposure metric are being compared. If elimination becomes capacity-limited as concentration rises, or the fraction of an oral dose reaching systemic circulation changes with dose, exposure need not scale proportionally. A dose chosen by multiplying an earlier dose by a target-to-observed exposure ratio can then miss an efficacy target or exceed a safety limit.
The practical task is to compare candidate doses using simulated concentration-time profiles under a specified pharmacokinetic model. The simulation should calculate the exposure measure that matters for the decision, such as AUC or peak concentration. The earlier material on choosing PK/PD indices for dose selection establishes why the selected metric must match the pharmacological target; here, the focus is how nonlinear pharmacokinetics changes that metric across doses.
A useful comparison is the dose-normalized exposure, \(AUC/D\), where \(D\) is the administered dose. If this quantity changes across doses, exposure is not proportional over that range. This comparison is descriptive: it does not by itself identify the mechanism. The mechanistic model and its assumptions must also be evaluated.
A Concentration-Dependent Elimination Model
Consider a hypothetical one-compartment model after a single intravenous bolus. Let \(C(t)\) be concentration in mg/L, \(V\) the apparent volume of distribution in L, \(D\) the dose in mg, \(V_{\max}\) the maximum elimination rate in mg/h, and \(K_m\) the concentration in mg/L at which the elimination rate is half of \(V_{\max}\). Elimination follows Michaelis–Menten kinetics, and there is no absorption or distribution phase in this simplified model.
At concentrations much lower than \(K_m\), the elimination rate is approximately \((V_{\max}/K_m)C\), which is proportional to concentration. At concentrations much higher than \(K_m\), elimination approaches \(V_{\max}\); increasing concentration then increases the elimination rate relatively little. These differences imply that higher doses can produce more-than-proportional exposure.
The result follows by rearranging the differential equation and integrating over the complete concentration decline. This derivation also shows why it is important to use the complete profile rather than infer exposure from peak concentration alone.
The integral uses the fact that concentration declines from \(C(0)\) to zero. At zero concentration, the elimination rate approaches zero, so the profile approaches zero over time; the integral remains finite. Substituting \(C(0)=D/V\) gives an AUC containing both a term proportional to dose and a term proportional to dose squared. That second term is the mathematical source of more-than-proportional scaling in this example.
Worked Example: Quantifying More-Than-Proportional AUC
Suppose a hypothetical drug has \(V=20\) L, \(K_m=2\) mg/L, and \(V_{\max}=10\) mg/h. Compare single intravenous doses of 50 mg and 100 mg. The stated model assumptions are taken to hold for both doses.
For 50 mg, \(C(0)=50/20=2.5\) mg/L. Substitution into the derived AUC expression gives:
For 100 mg, \(C(0)=100/20=5\) mg/L, so:
The dose doubled, but AUC increased by \(45.00/16.25=2.769\), not by 2. Dose-normalized AUC also increased: \(16.25/50=0.325\) and \(45.00/100=0.450\) h/L. These calculations indicate more-than-proportional exposure under the stated model; they do not establish that these parameter values describe any actual drug.
Selecting Among Candidate Doses
Dose selection is a decision against prespecified criteria, not simply a search for a dose with the largest exposure. For each candidate, simulate the concentration-time profile, calculate the relevant exposure metric, and assess any safety constraint on the profile. If the model is a differential equation, a numerical solver typically generates the profile; exposure can be calculated by numerical integration, with appropriate handling of the terminal part of the curve. The analytical result above provides a useful cross-check for this particular model.
The comparison below assumes that the model is adequate over the candidate range, the target and safety threshold were specified before comparing candidates, and doses use the same route and schedule. A simulated profile is conditional on the model and parameter values supplied. It does not automatically account for parameter uncertainty or patient-to-patient variability.
Worked Example: Choosing a Dose With an AUC Target and Peak Limit
Use the hypothetical model and parameters from the previous example. Suppose the decision rule is to meet a single-dose AUC target of at least 30 mg·h/L while keeping peak concentration at or below 4.5 mg/L. The candidates are 50, 75, 80, 90, and 100 mg. In this intravenous bolus model, peak concentration is \(C(0)=D/V\). The assumptions remain those stated for the model: fixed parameters and a single compartment with Michaelis–Menten elimination.
For 75 mg, \(C(0)=75/20=3.75\) mg/L, and:
For 80 mg, \(C(0)=4.00\) mg/L, giving \(AUC_{80}=2(8+8)=32.00\) mg·h/L. Its peak is 4.00 mg/L, below the 4.5 mg/L limit. Thus 80 mg meets both stated criteria.
For comparison, 90 mg gives \(C(0)=4.50\) mg/L and \(AUC_{90}=2(9+10.125)=38.25\) mg·h/L; it also meets both criteria, with its peak exactly at the limit. At 100 mg the AUC is 45.00 mg·h/L, but the peak is 5.00 mg/L, so that candidate violates the safety constraint. The 50 mg candidate has AUC 16.25 mg·h/L, and 75 mg has AUC 29.06 mg·h/L; both fall below the target. Among the listed candidates, 80 mg is the lowest dose meeting both criteria. This conclusion follows from the stated decision rule and candidate set, not from a claim that 80 mg is universally appropriate.
A proportional extrapolation would be misleading here. Starting at 50 mg, multiplying dose by \(30/16.25\) would suggest \(92.31\) mg to reach the AUC target. Under the nonlinear model, that dose would have a peak of \(92.31/20=4.615\) mg/L, exceeding the 4.5 mg/L limit. Simulating or calculating the candidate profiles reveals this conflict; a linear projection does not.
Dose-Dependent Bioavailability
Nonlinearity can also occur before a drug reaches systemic circulation. For an oral dose, the AUC under linear disposition is \(AUC=F(D)D/CL\), where \(F(D)\) is the fraction of the dose reaching systemic circulation and \(CL\) is clearance. If \(F\) changes with dose, the AUC may not be proportional to dose even when systemic elimination is linear. For example, a hypothetical formulation or absorption process could have increasing bioavailability over a specified dose range. A realistic model must keep \(F(D)\) between zero and one and be supported over the range where it is used.
Worked Example: Selecting an Oral Dose With Increasing Bioavailability
Consider a hypothetical oral drug with constant systemic clearance \(CL=5\) L/h and dose-dependent bioavailability \(F(D)=0.50+0.004D\), where \(D\) is in mg. This assumed relationship is used only from 50 to 100 mg, where it gives bioavailability values between zero and one. Assume linear systemic disposition, the same formulation and schedule for all candidates, and complete AUC to infinity. A simulator integrates each oral concentration-time profile; the resulting AUCs can be checked against \(F(D)D/CL\). Suppose the target is AUC of at least 15 mg·h/L.
At 50 mg, \(F(50)=0.50+0.004(50)=0.70\), so \(AUC_{50}=0.70(50)/5=7.00\) mg·h/L. At 100 mg, \(F(100)=0.90\), so \(AUC_{100}=0.90(100)/5=18.00\) mg·h/L. Doubling dose increases AUC by a factor of \(18/7=2.571\), because bioavailability rises as well as dose.
For a candidate of 90 mg, \(F(90)=0.50+0.004(90)=0.86\), and:
The 90 mg candidate exceeds the target. To find the model-based dose that just reaches the target, solve \((0.50+0.004D)D/5=15\). This gives \(0.004D^2+0.50D-75=0\); the positive root is \([ -0.50+\sqrt{0.50^2+4(0.004)(75)} ]/[2(0.004)]=88.02\) mg. A candidate of 90 mg therefore meets the target under this model, while 75 mg gives \(F=0.80\) and AUC \(=0.80(75)/5=12.00\) mg·h/L, below target.
If bioavailability were incorrectly held at its 50 mg value of 0.70, proportional scaling from 50 mg would suggest \(50(15/7)=107.1\) mg. That extrapolation both ignores the modeled change in bioavailability and goes outside the stated range of the dose-dependent relationship. The profile-based comparison supports 90 mg among these candidates, but does not validate the assumed bioavailability function beyond its stated range.
Checking Simulations and Understanding Their Limits
A dose simulation is useful only if its structure and outputs are credible for the intended decision. A sound analysis checks that the model represents the relevant route, dose range, and exposure time horizon; the chosen endpoint is calculated consistently for every candidate; and the numerical integration captures the profile, including its terminal tail. For the Michaelis–Menten example, the analytical AUC formula provides an independent check on numerical integration. For the oral example, the AUC calculation should agree with the assumed \(F(D)D/CL\) relationship when disposition is linear and the AUC is complete.
Investigate apparent nonlinearity rather than treating it as a numerical fact immediately. Check units, dose input, parameter values, solver tolerances, and whether the simulation has been run long enough. Compare dose-normalized exposure across candidates and inspect the concentration profiles for unexpected jumps, implausible peaks, or ordering that contradicts the model. If a numerical model produces a lower AUC at a higher dose, first confirm the equations and implementation; do not assume that every nonlinear process must produce a monotone exposure curve.
| Approach | What it assumes or uses | Useful role and limitation |
|---|---|---|
| Proportional extrapolation | Exposure scales in direct proportion to dose. | Simple when linear pharmacokinetics is justified; can mispredict exposure when clearance or bioavailability changes. |
| Mechanistic nonlinear simulation | A specified model describes the concentration- or dose-dependent process. | Can compare complete profiles and multiple metrics; conclusions depend on model structure and parameter validity. |
| Empirical exposure comparison | Observed exposure at tested doses is compared without requiring a particular mechanism. | Can reveal departures from proportionality in the observed range; provides limited guidance beyond that range. |
A common mistake is to see that one exposure metric increased more than proportionally and conclude that every clinically relevant metric must do so. That does not follow. AUC, peak concentration, and time above a target describe different features of a profile and can respond differently to the same change in pharmacokinetics. The correction is to simulate the profile and calculate the specific endpoint required by the decision, along with any separate safety constraints.
Another important limitation is extrapolation. A model that fits a tested dose range may not describe a much higher dose, especially if a different mechanism becomes important or the assumed function becomes implausible. Dose-dependent bioavailability must remain physiologically feasible, and a Michaelis–Menten model should not be assumed merely because exposure is nonlinear. When the mechanism or parameter estimates are uncertain, the resulting dose comparison is conditional on those assumptions; the simulation alone does not establish a clinical dose.
Check Your Understanding
Answer each question, then compare your reasoning with the short solutions.
- In the single-dose Michaelis–Menten model, what feature of the AUC formula causes exposure to rise more than proportionally with dose?
- For the hypothetical IV model with \(V=20\) L, what is the initial concentration after a 60 mg bolus?
- Under the oral model \(F(D)=0.50+0.004D\) and \(CL=5\) L/h, calculate the AUC at 75 mg.
- Why is proportional extrapolation from one dose not a reliable substitute for simulation when bioavailability changes with dose?
Solutions. (1) The dose-substituted AUC contains a positive dose-squared term, \(D^2/(2V V_{\max})\), in addition to the term proportional to dose. (2) \(C(0)=D/V=60/20=3.00\) mg/L. (3) \(F(75)=0.50+0.004(75)=0.80\), so \(AUC=0.80(75)/5=12.00\) mg·h/L. (4) Proportional extrapolation assumes a constant exposure-per-dose relationship, but dose-dependent bioavailability changes that relationship; it may therefore misstate both the dose needed for a target and the resulting profile.