Oral Single-Ascending-Dose Pharmacokinetics

Why Oral Single-Ascending-Dose Data Need a Population Model

An oral single-ascending-dose (SAD) study gives concentration-time observations after one dose at each planned dose level, often in different participants or cohorts. The analysis must describe both the shape of each concentration-time profile and the differences among participants. A population pharmacokinetic (popPK) model provides a common structural description while allowing individual parameters to vary around typical population values.

This tutorial uses a one-compartment model with first-order absorption and elimination. It is a useful starting point, not a universal description of oral pharmacokinetics. The model assumes linear, time-invariant disposition over the studied range, a fixed fraction of dose reaching systemic circulation, first-order input from the absorption compartment, and no lag in absorption. Within each participant, clearance, volume, and absorption rate are constant across the profile. Dose levels must be recorded correctly, sampling times must be measured from dosing, and observations from different participants must retain their participant identifiers.

Definition: In the one-compartment oral model, $D$ is dose, $F$ is the fraction of dose reaching systemic circulation, $k_a$ is the first-order absorption-rate constant, $CL$ is clearance, $V$ is apparent distribution volume, and $k=CL/V$ is the elimination-rate constant. For $t\geq 0$ and $k_a\ne k$, the predicted concentration is $$ C(t)=\frac{F D k_a}{V(k_a-k)}\left(e^{-kt}-e^{-k_a t}\right). $$ When $k_a=k$, the finite limiting form is $$ C(t)=\frac{F D}{V}kt e^{-kt}. $$

With oral concentration data alone, $F$, $CL$, and $V$ are generally not separately identified: multiplying $F$, $CL$, and $V$ by the same positive constant leaves the predicted concentrations unchanged. It is therefore usual to estimate apparent parameters $CL/F$ and $V/F$, with $k=(CL/F)/(V/F)$. The concentration equation can then be written using $F/V=1/(V/F)$. Interpreting an estimated $CL/F$ as true systemic clearance requires independent information about $F$.

Population Structure and Exposure Quantities

A basic population model represents an individual parameter as a typical value multiplied by a participant-specific deviation. For example, for participant $i$,

\left(\frac{CL}{F}\right)_i = \left(\frac{CL}{F}\right)_{\mathrm{typ}}\exp(\eta_{CL,i}), \qquad \eta_{CL,i}\sim N(0,\omega_{CL}^{2}).

Here the typical value is the median under this log-normal model, and $\omega_{CL}$ is the standard deviation of the participant deviations on the log scale. Similar deviations may be specified for $V/F$ and $k_a$; a model may allow them to be correlated. Observed concentrations also differ from predictions because of residual error, which combines measurement error and unexplained within-participant variation. A common formulation is $C_{\mathrm{obs}}=C_{\mathrm{pred}}(1+\epsilon_p)+\epsilon_a$, where the proportional and additive errors have mean zero under the model. The selected residual-error structure should match the scale and pattern of the data; it should not be assumed correct merely because it is familiar.

Under linear disposition and constant bioavailability, the area under the concentration-time curve from zero to infinity (AUC) is $FD/CL$, or $D/(CL/F)$ in apparent-parameter form. Thus, at a fixed parameter value, AUC is proportional to dose. Peak concentration and the full curve are also proportional to dose under this model if all kinetic parameters and $F$ remain unchanged. These are model implications to check against the data, not guarantees about a drug.

Worked Example: Predicting an Oral Concentration and Peak Time

Suppose a hypothetical participant receives $D=100$ mg. Let $F=0.80$, $V=50$ L, $CL=5.0$ L/h, and $k_a=1.2$ h$^{-1}$. The assumed values imply $k=CL/V=5.0/50=0.100$ h$^{-1}$, so $k_a\ne k$ and the first concentration formula applies.

At $t=2.0$ h, substituting the values gives $$ C(2)=\frac{(0.80)(100)(1.2)}{(50)(1.2-0.100)} \left(e^{-(0.100)(2)}-e^{-(1.2)(2)}\right) =1.74545(0.81873-0.09072) =1.271\ \text{mg/L}. $$

For $k_a\ne k$, the time of the maximum concentration is $t_{\max}=\ln(k_a/k)/(k_a-k)$. Here, $$ t_{\max}=\frac{\ln(1.2/0.100)}{1.2-0.100} =\frac{\ln(12)}{1.1} =2.259\ \text{h}. $$ The predicted concentration at 2.0 h is therefore near, but just before, the model peak. These are predictions from stated hypothetical parameters, not measured clinical results.

Why the Formula and Peak-Time Result Hold

The concentration-time expression follows from first-order input into a one-compartment system with first-order elimination. After oral dose $D$, the amount entering the systemic compartment per unit time is $F D k_a e^{-k_a t}$. The amount in the compartment is removed at rate $k$ times the amount present. Solving this linear input-elimination equation with zero drug amount at time zero gives the stated concentration after dividing amount by $V$.

The equal-rate case deserves explicit treatment because direct substitution into the usual expression gives a zero denominator. Put $k_a=k+\delta$. As $\delta$ tends to zero, $$ e^{-kt}-e^{-(k+\delta)t} =e^{-kt}(1-e^{-\delta t}) \mathrel{\sim} e^{-kt}\delta t, \qquad \frac{k+\delta}{\delta}\mathrel{\sim}\frac{k}{\delta}. $$ The product in the concentration formula therefore tends to $(FD/V)kt e^{-kt}$, a finite value. This is the limiting expression, not a failure of the underlying model.

To locate the peak when $k_a\ne k$, differentiate the concentration with respect to time. A maximum at a positive time satisfies $k_a e^{-k_a t}=k e^{-kt}$. Rearranging gives $e^{(k_a-k)t}=k_a/k$, and hence $t_{\max}=\ln(k_a/k)/(k_a-k)$. If $k_a=k$, differentiating $t e^{-kt}$ gives a maximum at $t=1/k$. These results presume positive rate constants and the model’s immediate, first-order absorption assumption.

Reading Dose-Related Exposure Across Cohorts

In a SAD analysis, a practical sequence is to check the data and dose records, fit a shared structural model to all dose levels, inspect individual and population predictions, and then summarize exposure by dose with uncertainty. Dose is included as the actual administered dose for each participant; it is not treated as a covariate effect merely because dose levels differ. Cohort differences can also reflect participant-to-participant variability, so dose-related patterns should be interpreted alongside the design and model diagnostics.

Verify the observations.
Check units, dose amounts, dose times, sample times, concentration records, and participant-to-cohort assignments before fitting.

Fit a defensible structural model.
Use the stated oral model as a candidate, estimate typical apparent parameters and between-participant variability, and specify a residual-error model.

Review fit and assumptions.
Compare observed concentrations with predictions over time and dose, inspect residual patterns, and check whether parameter estimates are plausible and sufficiently informed by sampling.

Summarize exposure with uncertainty.
Report model-based AUC and peak-related summaries by dose, distinguishing typical predictions from population ranges and uncertainty in the estimated parameters.

Worked Example: Calculating Dose-Proportional AUC

Assume, hypothetically, that the estimated typical apparent clearance is $CL/F=6.25$ L/h and that this parameter does not change with dose. The model-implied AUC is $D/(CL/F)$. For dose levels of 25, 50, 100, and 200 mg, the calculations are:

At 25 mg, $AUC=25/6.25=4.00$ mg·h/L. At 50 mg, $AUC=50/6.25=8.00$ mg·h/L. At 100 mg, $AUC=100/6.25=16.0$ mg·h/L. At 200 mg, $AUC=200/6.25=32.0$ mg·h/L. Each dose-normalized AUC is $0.160$ h/L.

This is a dose-proportional prediction from fixed linear parameters. In an analysis, the corresponding observed and model-estimated exposure summaries should be compared across cohorts, with uncertainty and participant variability shown. A systematic rise or fall in dose-normalized exposure may indicate that constant bioavailability or dose-independent clearance is not an adequate description, but it can also arise from sampling variation or cohort differences.

Population Variability and Model Diagnostics

Population summaries should not collapse individual differences into one typical curve. For log-normal apparent clearance, the 10th and 90th percentiles are the typical value multiplied by $\exp(-1.2816\omega_{CL})$ and $\exp(1.2816\omega_{CL})$, respectively. Because AUC is inversely related to apparent clearance under the linear model, the corresponding AUC percentiles reverse order. These are model-based population percentiles, not confidence intervals for the typical parameter and not guarantees about future individuals.

Worked Example: Translating Clearance Variability into Exposure Variability

Suppose a hypothetical popPK fit gives a typical $CL/F=6.25$ L/h and $\omega_{CL}=0.25$, and consider a 100 mg dose. The typical AUC is $100/6.25=16.0$ mg·h/L. The 90th percentile of apparent clearance is $$ 6.25\exp(1.2816\times0.25) =6.25\exp(0.3204) =8.61\ \text{L/h}. $$ At that clearance, AUC is $100/8.61=11.6$ mg·h/L. The 10th percentile of clearance is $6.25\exp(-0.3204)=4.54$ L/h, giving AUC $100/4.54=22.0$ mg·h/L. Thus, the model’s central 80% clearance range maps to an AUC range of approximately 11.6 to 22.0 mg·h/L, with the order reversed. This illustration holds other parameters fixed and does not include uncertainty in the estimated model parameters.

Diagnostics assess whether the model is an adequate description, not simply whether an optimizer returned estimates. Plot observed concentrations against time with population and individual predictions, and inspect residuals against time, predictions, and dose. Residuals should not show a persistent time trend, dose-specific bias, or changing spread that the error model fails to capture. Check that individual profiles are not systematically missed around the rise, peak, or decline. Examine whether estimates are stable, plausible, and supported by the sampling times; a parameter at a boundary, a very wide uncertainty interval, or strong dependence among parameters can signal weak information or model misspecification.

Worked Example: Checking Residuals in a Hypothetical Profile

For the 100 mg model in the first example, the predicted concentrations at 0.5, 2, 8, and 24 h are approximately 0.702, 1.271, 0.784, and 0.158 mg/L. Suppose a hypothetical participant has observed values 0.74, 1.20, 0.86, and 0.13 mg/L at those times. With a proportional residual error of 10% and no additive component, a simple standardized residual is $(C_{\mathrm{obs}}-C_{\mathrm{pred}})/(0.10C_{\mathrm{pred}})$.

The four standardized residuals are approximately $(0.74-0.702)/0.0702=0.54$, $(1.20-1.271)/0.1271=-0.56$, $(0.86-0.784)/0.0784=0.97$, and $(0.13-0.158)/0.0158=-1.77$. These values do not alone establish model adequacy, particularly with only four observations. A repeated pattern across participants—for example, consistently positive late residuals—would be more concerning and could point to a structural or error-model problem.

Alternatives, Limitations, and a Common Pitfall

A pooled analysis that fits one curve per dose or one mean profile can summarize dose trends, but it does not naturally separate typical kinetics from between-participant variability and may obscure differences in sampling. A population model uses the individual observations and represents participant-level parameter variation explicitly. Conversely, a more complex structural model may fit particular profiles better, but added parameters require informative data and a biological rationale. Model choice should be supported by diagnostics and design, not by fit statistics alone.

The one-compartment, first-order model can be inadequate if concentration profiles show systematic structure that it cannot reproduce, if absorption or bioavailability changes with dose, or if disposition is nonlinear. Sparse or poorly timed samples may leave absorption, peak concentration, or terminal decline weakly determined. Oral data alone also do not identify absolute bioavailability separately from clearance and volume. Predictions for doses beyond the observed range are extrapolations and should be presented with that limitation, especially in first-in-human planning. Model-derived exposure is an input to broader dose decisions, not a standalone safety or escalation rule.

Pitfall: treating the fitted typical curve as every participant’s curve. A typical prediction does not describe the full population distribution. Use individual predictions and variability summaries, and distinguish population variability from uncertainty in the estimated parameters.
Pitfall: interpreting $CL/F$ as $CL$. Without independent bioavailability information, oral estimates are apparent parameters. State which quantity was estimated and avoid an unsupported claim about systemic clearance.
Pitfall: assuming proportionality because the doses were planned to ascend. Ascending dose is a design feature, not evidence of linear PK. Compare dose-normalized exposure and diagnostics, and consider uncertainty and cohort differences before interpreting a trend.

Key takeaway: For an oral SAD study, a popPK model links concentration-time profiles across participants and doses. Its dose-exposure summaries are conditional on the structural, variability, and residual-error assumptions; apparent parameters, uncertainty, and diagnostic limitations must accompany the interpretation.

Check Your Understanding

Try each question, then compare with the short solutions.

For $CL=4.0$ L/h and $V=40$ L, calculate $k$. If $k_a=1.0$ h$^{-1}$, which concentration expression applies?
A hypothetical oral model has $CL/F=8.0$ L/h. What AUC does it predict after 80 mg under linear PK?
Explain why oral concentration data alone generally identify $CL/F$, rather than $CL$ and $F$ separately.
What might a persistent pattern of positive residuals late in the profiles suggest?

Solutions. (1) $k=CL/V=4.0/40=0.100$ h$^{-1}$. Since $k_a=1.0$ h$^{-1}\ne k$, use the expression for unequal rates. (2) $AUC=D/(CL/F)=80/8.0=10.0$ mg·h/L. (3) The oral prediction depends on the ratio $F/CL$; without separate information about bioavailability, different $F$ and $CL$ values can give the same concentration predictions. (4) A systematic late bias may indicate that the structural model does not describe the decline adequately, or that the error model or data require review; one should investigate across participants rather than infer a cause from a single residual.