Forecasting Steady-State Troughs From Nonsteady Data

Why a Pre-Steady-State Sample Can Forecast a Steady-State Trough

An early concentration is not itself a steady-state trough. It reflects the dose just given, residual drug from earlier doses, the time since the latest dose, and the individual’s pharmacokinetic parameters. A forecast must account for all of these contributions and then ask what the same individual’s concentration would be after accumulation has stabilized.

The task considered here is to forecast a trough under a specified repeated-dose regimen from sparse concentrations collected before steady state. The concentration data inform an individual posterior distribution for pharmacokinetic parameters; each plausible parameter set is then used to project the concentration profile under repeated dosing. This separates two sources of uncertainty: uncertainty about the individual’s parameters and uncertainty in a future concentration due to residual variability.

Definition: A steady-state trough is the concentration immediately before a scheduled dose after repeated dosing at a fixed interval has reached its periodic dosing profile. A pre-steady-state trough is the concentration immediately before a dose earlier in that accumulation process. A steady-state trough forecast applies the intended regimen and individual parameter values to the limiting repeated-dose profile; it is not obtained by simply relabeling an observed early concentration.

Accumulation Under Repeated Dosing

Start with a one-compartment model with first-order elimination and instantaneous IV bolus doses. Let \(D\) be the dose, \(V\) the individual’s volume of distribution, \(CL\) clearance, \(k=CL/V\) the elimination-rate constant, and \(\tau\) the dosing interval. Assume doses are given at times \(0,\tau,\ldots,(n-1)\tau\), with no dose omitted or changed. Write \(t\) for elapsed time since the most recent dose, where \(0\leq t\leq\tau\). Under linear pharmacokinetics, each dose contributes independently to concentration.

One dose contributes \((D/V)e^{-kt}\) after \(t\) time units. The previous dose contributes \((D/V)e^{-k(t+\tau)}\), the dose before that contributes \((D/V)e^{-k(t+2\tau)}\), and so on. Summing the \(n\) contributions gives the finite-dose profile. At a trough after \(n\) doses, set \(t=\tau\).

C_n(\tau) = \frac{D}{V}e^{-k\tau}\sum_{j=0}^{n-1}e^{-jk\tau} = \frac{D}{V}\frac{e^{-k\tau}\left(1-e^{-nk\tau}\right)}{1-e^{-k\tau}}.

Because \(k>0\) and \(\tau>0\), \(0<e^{-k\tau}<1\). Thus the finite geometric sum converges as \(n\) increases, and the limiting trough is

C_{\mathrm{ss,trough}} = \frac{D}{V}\frac{e^{-k\tau}}{1-e^{-k\tau}}.

This derivation also gives a useful accumulation check: the finite-dose trough is the steady-state trough multiplied by \(1-e^{-nk\tau}\). It approaches the limiting trough from below in this model. The time required to approach steady state depends on \(k\tau\); the number of doses alone does not determine how much accumulation has occurred.

Worked Example: Correcting a Three-Dose IV Trough for Incomplete Accumulation

Suppose a hypothetical individual has an estimated post-bolus concentration increment \(D/V=5.000\) mg/L and \(k\tau=1.2\). Forecast the trough after three doses and the limiting steady-state trough.

For three doses, \(e^{-3k\tau}=e^{-3.6}=0.02732372\), and \(e^{-k\tau}=e^{-1.2}=0.3011942\). The three-dose trough is

C_3(\tau) = 5.000\frac{0.3011942(1-0.02732372)}{1-0.3011942} = 5.000\frac{0.292964\;}{0.6988058} \approx 2.096\ \mathrm{mg/L}.

The limiting trough is

C_{\mathrm{ss,trough}} = 5.000\frac{e^{-1.2}}{1-e^{-1.2}} = 5.000\frac{0.3011942}{0.6988058} = 5.000(0.4310128) \approx 2.155\ \mathrm{mg/L}.

The finite-dose result is also \(2.15506(1-e^{-3.6})\approx2.096\) mg/L, confirming the accumulation relationship. The difference does not mean the observed third trough is erroneous; it reflects that three doses have not yet produced the limiting profile.

Oral Dosing and the Absorption Contribution

For first-order oral absorption, each dose enters the central compartment over time rather than instantaneously. In a one-compartment model with absorption rate constant \(k_a\), elimination rate constant \(k\), bioavailability \(F\), and dose \(D\), the contribution from one dose at elapsed time \(t\) is

\frac{FD}{V}\frac{k_a}{k_a-k}\left(e^{-kt}-e^{-k_a t}\right), \qquad k_a\ne k.

For \(n\) doses given at the same interval, the contributions from the current and earlier doses are summed. At a trough one interval after the latest dose, the finite-dose expression is

C_n(\tau) = \frac{FDk_a}{V(k_a-k)} \left[ e^{-k\tau}\frac{1-e^{-nk\tau}}{1-e^{-k\tau}} - e^{-k_a\tau}\frac{1-e^{-nk_a\tau}}{1-e^{-k_a\tau}} \right].

Taking the limit as \(n\) increases gives the oral steady-state trough. The second term is the remaining absorption-related contribution; it should not be dropped merely because the sample is called a trough. When \(k_a\tau\) is large, that term may be small, but its size depends on the absorption rate and interval.

Worked Example: Oral Steady-State Trough With Accumulation

Consider a hypothetical oral regimen with \(FD/V=5.000\) mg/L, \(k=0.1000\ \mathrm{h}^{-1}\), \(k_a=1.000\ \mathrm{h}^{-1}\), and \(\tau=12.00\) h. Then \(k\tau=1.2\), \(k_a\tau=12\), and \(k_a/(k_a-k)=1/0.9\). The prefactor is

\frac{FDk_a}{V(k_a-k)} = 5.000\frac{1.000}{1.000-0.1000} = 5.55556\ \mathrm{mg/L}.

The elimination accumulation fraction is \(e^{-1.2}/(1-e^{-1.2})=0.4310128\). The absorption fraction is \(e^{-12}/(1-e^{-12})\approx0.00000614425\). Therefore,

C_{\mathrm{ss,trough}} = 5.55556(0.4310128-0.00000614425) \approx2.39448\ \mathrm{mg/L}.

The small absorption term lowers the predicted trough slightly relative to a calculation that treats the oral dose as an instantaneous input. The calculation assumes the stated one-compartment, linear, first-order absorption model and an unchanged 12-hour regimen.

Forecasting From an Individual Posterior

In individual Bayesian forecasting, the pre-steady-state observations update uncertainty about the individual’s parameters. Denote a parameter set by \(\theta\), which may include \(CL\), \(V\), \(k_a\), and \(F\), as appropriate to the model. The posterior distribution \(p(\theta\mid\text{data})\) represents parameter uncertainty conditional on the data and model. The steady-state trough is a function of those parameters and the planned dose and interval: \(g(\theta)\).

A practical calculation uses posterior draws \(\theta^{(1)},\ldots,\theta^{(M)}\). For each draw, compute \(g(\theta^{(m)})\) using the steady-state formula for that model. The resulting values form a posterior distribution of steady-state troughs. Its mean or median can summarize the forecast, and its quantiles can summarize parameter uncertainty. If the target is a future measured concentration, residual variability must also be simulated or otherwise included; parameter-only variation does not include that additional uncertainty.

Specify the regimen and timing.
Record dose, interval, route, dose history, and the exact time of each pre-steady-state sample. The forecast regimen must be explicit.

Obtain individual posterior draws.
Use the observed concentrations and the stated structural and residual error models to represent uncertainty in individual parameters.

Project each draw to steady state.
Apply the model-specific accumulation formula to each parameter draw, preserving parameter correlations within each draw.

Summarize and qualify the forecast.
Report the posterior summary and uncertainty interval with units, and state whether the interval concerns the latent steady-state trough or a future measured concentration.

Worked Example: Averaging Draw-Specific Oral Trough Forecasts

Suppose two equally weighted hypothetical posterior draws give the same \(k\tau=1.2\) and \(k_a\tau=12\) as above, but different values of the oral prefactor \(FDk_a/[V(k_a-k)]\). Draw A has prefactor \(5.55556\) mg/L, so its forecast is \(2.39448\) mg/L. Draw B has prefactor \(7.11242\) mg/L. Its forecast is

C_{\mathrm{ss,trough}}^{(B)} = 7.11242(0.4310128-0.00000614425) \approx3.06550\ \mathrm{mg/L}.

With equal weights, the posterior mean of the two draw-specific troughs is

\frac{2.39448+3.06550}{2} = 2.72999 \approx2.7300\ \mathrm{mg/L}.

This is a miniature illustration of draw-by-draw forecasting, not a useful interval estimate: two draws cannot characterize posterior quantiles reliably. In an actual analysis, many posterior draws would be transformed, and their empirical distribution would be summarized. Averaging parameters first and then applying a nonlinear trough formula generally need not give the same result as averaging the draw-specific troughs.

Assumptions, Diagnostics, and Alternatives

The formulas above require a model that adequately describes the dosing and concentration process. The IV formula assumes one-compartment disposition, linear elimination, fixed repeated boluses, and correct dose and sample times. The oral formula additionally assumes first-order absorption and the specified relationship between bioavailability, dose, and volume. Both rely on parameters remaining stable over the forecast horizon. For a posterior forecast, the individual posterior must also be based on a suitable likelihood and a correctly represented dose history; a precise formula cannot repair an inappropriate model or incorrect records.

Check the forecast process before interpreting it. Confirm that the observed samples are aligned to the actual dose times, that the trough is defined immediately before the next dose, and that the planned regimen matches the regimen used in the projection. Evaluate whether predicted concentrations under the posterior draws are compatible with the observed pre-steady-state samples. A forecast that changes sharply under small plausible changes in absorption or clearance may be weakly informed by sparse data; the resulting uncertainty should be retained, not concealed by a single point estimate.

A useful consistency check is that the finite-dose trough approaches the steady-state trough as the number of doses increases. In the IV model it must not exceed the limiting trough under the fixed-dose assumptions. For oral dosing, the complete profile includes both elimination and absorption terms, so do not use the IV geometric multiplier alone unless instantaneous input is an appropriate approximation. If the model has multiple compartments, nonlinear elimination, time-varying parameters, or changing doses, use that model’s forward prediction under the repeated regimen rather than forcing a one-compartment formula.

The main alternative to an analytic accumulation formula is direct forward simulation of the structural model over repeated dosing until the profile is periodic. Simulation is more flexible for complex models and dosing histories, while the closed-form formulas here are transparent and easy to verify for the stated simple models. A further distinction is whether the goal is a typical latent trough or a future observed concentration: the former transforms parameter uncertainty, whereas the latter must also include residual error. Neither alternative makes the forecast independent of its assumptions.

Common Pitfall

Wrong: Treat the latest pre-steady-state trough as the eventual steady-state trough because the dose and interval are unchanged.
Correction: The early trough includes only the doses administered so far. Sum the contributions from that finite history, or apply the limiting accumulation formula to forecast the steady-state value.
Wrong: Insert posterior mean parameters into a nonlinear trough formula and call the result the posterior mean forecast.
Correction: Transform each joint posterior draw, then summarize the resulting trough draws. Nonlinear transformations and parameter correlations can make these summaries differ.

Key takeaway: A steady-state trough forecast from nonsteady data is a model-based projection, not a direct reading of an early concentration. Apply the repeated-dose model to individual posterior draws, use the correct absorption and accumulation terms, and report uncertainty in the forecast.

Check Your Understanding

Answer each question, then compare with the brief solutions.

Under the IV bolus model, why does the finite-dose trough approach a limit as the number of doses increases?
For \(D/V=4.000\) mg/L and \(k\tau=0.8\), calculate the IV steady-state trough.
Why can an oral steady-state trough not always be calculated by using the IV formula with \(D\) replaced by \(FD\)?
What uncertainty is omitted if posterior parameter draws are transformed to troughs without adding residual variability?
Give two reasons a sparse pre-steady-state dataset might yield a highly uncertain steady-state forecast.

Solutions. (1) The contributions form a geometric series with ratio \(e^{-k\tau}\), which is between zero and one, so its sum converges. (2) \(e^{-0.8}=0.449329\), so \(C_{\mathrm{ss,trough}}=4.000(0.449329)/(1-0.449329)=4.000(0.815966)\approx3.264\) mg/L. (3) Oral input takes time; the first-order absorption contribution produces a second term involving \(k_a\), which the instantaneous-input IV formula omits. (4) It omits future residual variability in a measured concentration; the transformed draws represent parameter uncertainty in the model-predicted trough. (5) For example, sparse samples may not distinguish absorption from elimination well, or may provide limited information about clearance and volume; uncertainty in dose timing or adherence can also weaken the forecast.