When fitting a two-compartment pharmacokinetic model to trough-only data while holding the peripheral volume (Vp) and inter-compartmental clearance (Q) fixed, the major determinant of the estimated total clearance (Cltotal) is the absolute magnitude of the trough concentration (Ctrough) — specifically the dose-normalized trough concentration, or steady-state drug exposure. [1] [2] [3] [4]
Trough data is captured at the end of a dosing interval (τ), long after the rapid distribution phase (α) has concluded. At this point the drug has reached a pseudo-equilibrium between the central and peripheral compartments, and the decline of serum concentrations is driven entirely by the terminal elimination phase (β). [4] [5] [6]
At steady state, the average concentration (Css,avg) is structurally defined as:
Because a trough concentration tracks very closely with the overall area under the curve (AUC) and Css,avg, the data-fitting algorithm adjusts Cltotal as the primary lever to slide the entire elimination curve up or down to match the height of the observed troughs. Higher observed troughs result in a lower estimated Cltotal, and lower troughs yield a higher estimated Cltotal. [4]
With Vp and Q held constant, the only two remaining parameters that could be estimated are the central volume (Vc) and Cltotal. [2] [4]
Note: Due to this heavy dependency and potential for bias, clinical protocols utilizing trough-only data (such as routine vancomycin therapeutic drug monitoring) frequently fix Vp, Q, and Vc to population averages, leaving Cltotal as the sole individually estimated parameter determined by the trough. [8] [9]
In commercial clinical decision support (CDS) platforms utilizing Bayesian forecasting — such as PrecisePK, InsightRX, and DoseMeRx — the software does not strictly "hold parameters flat at the mean" the way a frequentist standard regression does. Because it is a Bayesian MAP (Maximum a Posteriori) fit, all parameters with a defined inter-individual variability (ω²) are technically allowed to move. [10] [11] [12]
However, when handling trough-only data, commercial software effectively treats Q and Vp as fixed while primarily optimizing Cl and — to a lesser extent — Vc. [12]
| Parameter | Technical Software Behavior (With Trough-Only Data) | Practical Outcome |
|---|---|---|
| Clearance (Cl) | High movement / High update. The algorithm heavily shifts Cl away from the population prior to match the absolute magnitude of the trough. | Optimized to the patient's individual elimination profile. |
| Central Volume (Vc) | Minimal movement. The software mathematically allows Vc to shift, but because there is zero peak data (Cmax) to inform it, the Bayesian penalty (objective function) restricts it. It stays very close to the population mean or covariate-adjusted value. | Stays effectively anchored to the prior. |
| Inter-Compartmental Clearance (Q) | No movement. Q is strictly locked to the population model's typical value (or structurally adjusted by a global covariate such as continuous renal replacement therapy, if explicitly built in). | Stays at the population mean. |
| Peripheral Volume (Vp) | No movement. Like Q, there is essentially zero signal in a trough level to characterize peripheral tissue distribution. It is held constant at the population mean. | Stays at the population mean. |
Commercial tools differentiate between the parameters based on their structural variance settings:
Summary: If a peak and a trough are drawn, the package updates both Cl and Vc dynamically. If a trough only is entered, the package updates Cl significantly, allows Vc to shift minorly if the trough is drastically unexpected, and structurally forces Q and Vp to reside entirely at their population typical values. [12]
When fitting an individual patient's data using only a trough concentration, a one-compartment model is generally better, safer, and more structurally stable than a two-compartment model. [14] [15]
A single trough concentration provides only one distinct piece of information about the patient's individual drug exposure.
Trying to fit a two-compartment model using a single trough forces the software to guess how the drug distributes between the central and peripheral spaces with zero data from the distribution phase (α-phase). [17] [18]
| Metric / Attribute | One-Compartment Model | Two-Compartment Model |
|---|---|---|
| AUC Estimation Accuracy | Highly reliable. Multiple pharmacokinetic studies (e.g., Maung et al.) confirm that one-compartment models using trough-only data accurately approximate the true reference AUC. | Prone to overestimation. Two-compartment models using trough-only data frequently over-extrapolate the early distribution phase, leading to significant AUC overestimations (often 15–25%). |
| Parameter Identifiability | Structurally stable. The algorithm only has to modify Cl to fit the slope or position of the terminal line. | High collinearity. The software must mathematically balance Vc, Vp, and Q blindly, causing mathematical instability or severe Bayesian shrinkage to the population mean. |
| Clinical Utility | Safest and most practical choice for routine clinical tracking (e.g., standard vancomycin therapeutic drug monitoring). | Suboptimal unless utilizing highly specialized population priors designed specifically to lock peripheral parameters flat. |
If utilizing a commercial Bayesian decision support tool (e.g., InsightRX, DoseMeRx), the software may natively use a two-compartment population model as its default configuration. This configuration is acceptable only because the software does not actually try to fit the whole model to the trough. [19] [20]
Instead, the software locks Q and Vp entirely to the population mean, mathematically turning the two-compartment model into a pseudo-one-compartment model during the individual optimization step. However, if using standalone statistical software (such as R or Phoenix WinNonlin) to fit an individual's data directly via standard regression, a two-compartment model will produce uninterpretable and structurally biased results.
Links open in a separate, closable popup window. Ctrl/Cmd-click, middle-click, or right-click → "Open link in new tab" to open a reference in a new tab instead.