Why Candidate Doses Need Direct Simulation
If pharmacokinetics are linear over the dose range of interest, multiplying a dose by a factor usually multiplies exposure by the same factor, provided the regimen and relevant conditions remain unchanged. That shortcut can fail when a PK process is nonlinear. With saturable elimination, for example, a modest increase in input can produce a much larger increase in concentration. If bioavailability falls as dose rises, exposure may instead increase less than proportionally.
The practical consequence is that a dose judged adequate by linear extrapolation might not reach an exposure target, or a seemingly small dose increase might move some individuals across a target threshold. Candidate doses should therefore be run through the nonlinear model itself, using the same virtual-subject parameters and regimen assumptions at each dose. The aim here is to understand exposure predictions and their implications; formal population target-attainment analysis is a separate step.
A Saturable-Elimination Model
Consider a one-compartment model with concentration \(C\) in mg/L and a constant intravenous infusion rate \(R\) in mg/h. Suppose elimination follows Michaelis–Menten kinetics, with maximum elimination rate \(V_{\max}\) in mg/h and Michaelis constant \(K_m\) in mg/L. The concentration changes according to the difference between infusion and elimination rates:
where \(V\) is the one-compartment volume in L. At a finite steady state, concentration is constant, so the rate of change is zero and input equals elimination. Solving that balance gives:
The condition \(R<V_{\max}\) matters: maximum elimination cannot keep up with an infusion rate equal to or above \(V_{\max}\), so this model has no finite steady-state concentration in that case. Below that limit, concentration rises faster than proportionally as the rate approaches \(V_{\max}\). Differentiating the steady-state expression gives \(dC_{\mathrm{ss}}/dR=K_mV_{\max}/(V_{\max}-R)^2\); the denominator shrinks as \(R\) increases.
Worked Example: A Small Increase in Infusion Rate
Suppose a hypothetical drug has \(V_{\max}=12\) mg/h and \(K_m=2\) mg/L. Compare constant infusion rates of 6 and 8 mg/h. Assume the stated model applies and both rates are below \(V_{\max}\), as required for a finite steady state.
At 6 mg/h:
At 8 mg/h:
The infusion rate increased by one-third, from 6 to 8 mg/h, but steady-state concentration doubled. With a hypothetical target of at least 3 mg/L, the first rate is below target and the second is above it. Proportional scaling from the 6 mg/h prediction would give \(2.00\times(8/6)=2.67\) mg/L, incorrectly predicting that the second rate remains below target.
Dose-Dependent Bioavailability
Nonlinearity can also occur before drug reaches the systemic circulation. For a simple illustration, suppose the fraction absorbed at oral dose \(D\) is described by \(F(D)=F_{\max}/(1+D/K_F)\), where \(F_{\max}\) is a maximum fraction and \(K_F\) is a dose scale in the same units as \(D\). This is a hypothetical model in which bioavailability declines as dose increases. If elimination is linear with clearance \(CL\), then at steady state the AUC over a dosing interval of length \(\tau\) is \(F(D)D/CL\). This follows from mass balance: over an interval, the absorbed amount is \(F(D)D\), and linear elimination gives total elimination \(CL\times AUC_\tau\); at steady state those amounts are equal.
Worked Example: Exposure With Declining Bioavailability
Suppose a hypothetical oral drug has \(F_{\max}=0.90\), \(K_F=200\) mg, and linear clearance \(CL=10\) L/h. Compare total doses of 100 and 200 mg per dosing interval. Assume the dose-dependent bioavailability model and steady-state mass balance apply.
For 100 mg, \(F(100)=0.90/(1+100/200)=0.60\). The absorbed amount is \(0.60\times100=60\) mg, and:
For 200 mg, \(F(200)=0.90/(1+200/200)=0.45\). The absorbed amount is \(0.45\times200=90\) mg, and:
The dose doubled, but AUC increased by a factor of \(9/6=1.50\), not 2.00. The calculation depends on the specified bioavailability model and linear elimination; a different absorption or elimination model could give a different dose–exposure relationship.
Comparing Candidate Doses Across Virtual Subjects
A population simulation applies the model to subject-specific parameter values rather than assuming that everyone has the same PK. For each candidate dose or rate, the same virtual subjects can be simulated again under that regimen. Holding the virtual cohort fixed across candidates makes differences in predicted exposure easier to attribute to the regimen rather than to a different set of sampled subjects. This example uses a deliberately small, fixed hypothetical cohort to show the calculation; it is not a clinical estimate.
Worked Example: Nonlinear Target Attainment in a Small Cohort
Consider four hypothetical virtual subjects receiving constant infusion. Each has \(K_m=2\) mg/L, and their \(V_{\max}\) values are 9, 12, 15, and 18 mg/h. Compare rates of 6 and 8 mg/h, and suppose the illustrative concentration target is at least 3 mg/L. Both rates are below every subject's \(V_{\max}\), so the steady-state formula applies to all eight predictions.
At 6 mg/h, each concentration is \(C_{\mathrm{ss}}=12/(V_{\max}-6)\) mg/L. The resulting values for \(V_{\max}=9, 12, 15,\) and 18 mg/h are 4.00, 2.00, 1.33, and 1.00 mg/L. One of the four subjects is at or above 3 mg/L.
At 8 mg/h, each concentration is \(C_{\mathrm{ss}}=16/(V_{\max}-8)\) mg/L. The corresponding values are 16.0, 4.00, 2.29, and 1.60 mg/L. Two of the four subjects are at or above 3 mg/L. In this small cohort, increasing the rate changes the count meeting the illustrative target from one to two.
For comparison, linear scaling each 6 mg/h concentration by \(8/6\) would predict 5.33, 2.67, 1.78, and 1.33 mg/L. Only the first of those predictions reaches 3 mg/L. Linear scaling therefore predicts one subject meeting the threshold instead of the two obtained by applying the nonlinear model at 8 mg/h. The example shows why dose-specific simulation matters; the four-subject count is only an illustration, not a precise population probability.
How to Simulate and Check Candidate Doses
For an intermittent regimen, a nonlinear concentration-time profile generally cannot be obtained by multiplying a reference profile by the dose ratio. The model must be evaluated with the candidate regimen. For a one-compartment Michaelis–Menten example, this means numerically solving the concentration differential equation between doses, applying each dose as the model specifies, and calculating the exposure metric from the resulting profile. For repeated dosing, the encoded dose history must match the intended regimen. If steady state is required, verify that the simulated profile has stabilized rather than assuming that one interval is sufficient.
The simulated output should be checked against the model's assumptions and dose range. Useful checks include confirming units, verifying that all rates and parameters are physically admissible, and comparing predicted concentrations with the expected direction of change as dose increases. For the infusion model, a rate approaching \(V_{\max}\) should produce a sharply increasing predicted concentration; a rate at or above \(V_{\max}\) must not be reported as having a finite steady state. For an ODE simulation, repeat the calculation with a suitably finer time step or tighter solver tolerance and confirm that the exposure summary is stable to the required precision.
Where a mass-balance identity is available, it provides an additional check. Under the stated steady-state infusion assumptions, elimination rate must equal infusion rate. Under the oral steady-state example with linear elimination, \(CL\times AUC_\tau\) must equal \(F(D)D\). A mismatch may indicate a coding error, inconsistent units, or that the assumptions behind the identity do not hold for the simulated model.
State which process is nonlinear, the dose or rate, the interval, and whether the comparison concerns a single dose, repeated dosing, or steady state.
Use the virtual subjects' relevant parameter values and covariates. Keep the cohort consistent across candidate doses when the goal is a controlled dose comparison.
Use the nonlinear equations or solve the concentration-time model for each regimen; do not substitute proportional scaling unless proportionality is justified over the range.
Calculate the exposure measure of interest, check units and model constraints, and inspect whether numerical solutions and mass balance are consistent with the assumptions.
Alternatives, Diagnostics, and Limitations
Linear dose scaling is a useful alternative when dose proportionality is supported over the candidate range and the regimen, population, and exposure metric are comparable. It is simpler than rerunning a nonlinear model, but a local appearance of proportionality does not establish proportionality at higher doses. An empirical power model, such as exposure proportional to \(D^b\), can summarize curvature over observed doses; it is descriptive and should not automatically be extrapolated beyond that range. A mechanistic nonlinear model represents a specified process, such as saturable elimination, but its predictions depend on the adequacy and precision of that model.
Several patterns can signal trouble. Predicted exposure that decreases when dose increases may be possible under some models, but should prompt a check of the model specification, dose encoding, units, and numerical solution. Abrupt or implausibly large changes near a saturation limit may be mathematically consistent with the model yet highly sensitive to uncertain parameters. Predictions that vary greatly when \(V_{\max}\), \(K_m\), or the assumed bioavailability function is changed should be interpreted as model-dependent, not as a precise dose recommendation. Compare model predictions with available observations across the relevant dose range and investigate systematic departures rather than relying on one summary curve.
A simulation cannot resolve uncertainty that the model or data do not identify. Sparse observations near the nonlinear region may leave the location and severity of saturation uncertain. A model may also omit dose-dependent absorption, time-varying physiology, or other processes. In particular, a fixed-parameter example does not represent parameter uncertainty, and a small illustrative cohort does not provide a stable estimate of a population target-attainment probability. These issues matter when interpreting candidate-dose predictions, even when the calculations themselves are correct.
Check Your Understanding
Answer each question, then compare your reasoning with the solutions below.
- For \(V_{\max}=10\) mg/h, \(K_m=2\) mg/L, and a constant infusion of 5 mg/h, calculate the finite steady-state concentration. State why the formula is valid.
- With \(V_{\max}=10\) mg/h and \(K_m=2\) mg/L, what does the model imply about a constant infusion rate of 10 mg/h?
- If dose doubles while dose-dependent bioavailability falls, can AUC still double under linear elimination? Explain what must be compared.
- Why is it informative to use the same virtual subjects at each candidate dose?
- Name one numerical or model diagnostic for a nonlinear dose simulation and explain what it can reveal.
Solutions. (1) \(C_{\mathrm{ss}}=(2\times5)/(10-5)=2.00\) mg/L. The rate is below \(V_{\max}\), and the constant-infusion steady-state model assumptions are stipulated. (2) The rate equals \(V_{\max}\), so maximum elimination cannot balance input at any finite concentration; the model has no finite steady state. (3) It is possible only if the absorbed amount also doubles, since at steady state \(AUC_\tau=F(D)D/CL\). Compare \(F(D)D\) at the two doses rather than dose alone. (4) A fixed cohort makes the comparison less affected by differences in sampled subject characteristics, so changes in predictions are more directly attributable to dose. (5) For example, tightening solver tolerances tests numerical stability; a material change in the exposure summary suggests the solution is not yet sufficiently accurate. Checking units or mass balance can also reveal implementation errors.