Why Compare These Structures?
A model can reproduce plasma concentrations and still be unable to answer a question about a particular organ. That is not necessarily a poor fit: it may be a consequence of the model’s structure. A one-compartment model represents distribution with one homogeneous space; a two-compartment model separates a central space from a lumped peripheral space; an organ-resolved model represents selected organs separately. Each makes different predictions available, and each requires assumptions about what can be treated as a common pool.
The comparison below uses the same hypothetical intravenous-bolus observations throughout. The data are illustrative, not from a real study. Concentrations are measured in milligrams per litre (mg/L), and time is measured in hours.
| Time (h) | Plasma (mg/L) | Liver tissue (mg/L) | Muscle tissue (mg/L) |
|---|---|---|---|
| 0.5 | 7.80 | 1.80 | 0.25 |
| 2 | 4.60 | 3.90 | 0.55 |
| 6 | 1.10 | 3.50 | 1.20 |
The plasma concentration falls over time, while the two tissue profiles differ from each other and from plasma. A model that describes only plasma can address a plasma-prediction target, but it does not automatically supply a separate liver and muscle prediction. The earlier material on matching model detail to the prediction target provides the guiding principle: choose a structure whose outputs correspond to the quantity of interest.
What the Three Structures Represent
These structures are useful under different assumptions. A one-compartment model is appropriate when one apparent distribution space adequately describes the concentration behaviour relevant to the intended prediction. Its usual linear bolus form assumes a fixed apparent volume, first-order elimination, and no separately represented distribution delay. A two-compartment model adds a peripheral pool and assumes that exchange between the central and peripheral pools can be represented by the chosen linear exchange structure. It still treats all tissues assigned to that peripheral pool as one model state. An organ-resolved model instead assumes that separate organ states are needed and that the available physiology and data can support their separate parameterization.
A central concentration curve from a two-compartment model is often biexponential. That curve alone does not specify every tissue concentration: a peripheral state prediction also requires a defined exchange model and its parameter values. Likewise, an organ-resolved structure can output organ-specific concentrations only for the organs and outputs explicitly represented. Adding an organ label to a model does not, by itself, establish that the corresponding parameters are identifiable from the data.
One-Compartment Prediction for the Shared Data
For illustration, suppose a one-compartment exponential curve is used over the sampled interval beginning at 0.5 h. Its fitted concentration at 0.5 h is 7.827 mg/L, and its apparent decline rate is 0.3711 per hour. This is a post-0.5 h curve; it does not claim to describe the unobserved distribution immediately after the bolus.
Worked Example: Predicting Plasma at 2 and 6 Hours
Use the one-compartment curve to predict plasma concentrations at 2 h and 6 h, then compare each prediction with the illustrative observed plasma concentration.
At 2 h, \(t-0.5=1.5\) h:
The observed value is 4.60 mg/L, so the observed-minus-predicted difference is \(4.60-4.486=0.114\) mg/L.
At 6 h, \(t-0.5=5.5\) h:
The observed value is 1.10 mg/L, giving an observed-minus-predicted difference of \(1.10-1.017=0.083\) mg/L. The predictions are reasonably close at these sampled times, but this comparison alone does not establish that one compartment is adequate for every target. In particular, this curve has no liver-specific or muscle-specific state, so it cannot predict either tissue concentration separately.
The example assumes the chosen exponential form is meaningful over the interval being assessed. A diagnostic check is to examine residuals—observed concentrations minus predictions—across time, not just at one point. A systematic pattern, such as predictions consistently too high at early times and too low at later times, can signal that one exponential is not representing the observed time course. Sparse observations, measurement error, or a poorly chosen fitting interval can also obscure such patterns. The observed concentrations in a real analysis would need an appropriate measurement-error model; the simple differences above are descriptive, not a formal test.
What a Second Compartment Adds
A two-compartment model adds a distinct peripheral state, with exchange between central and peripheral pools. In a standard linear representation, amounts in the two pools change through elimination from the central pool and exchange between pools. The central plasma concentration can then have a rapid and a slower component. The model can represent distribution-related curvature that a single exponential cannot, while still keeping peripheral tissues lumped together.
For the same hypothetical plasma observations, consider this illustrative biexponential central concentration curve. The positive coefficients and rates are chosen to match the plasma values at 0.5 h and 6 h; the curve is then checked against the intermediate observation at 2 h.
Worked Example: Checking an Illustrative Two-Compartment Plasma Curve
At 2 h, substitute \(t=2\) into the curve:
The observed plasma value at 2 h is 4.60 mg/L, so the observed-minus-predicted difference is \(4.60-4.430=0.170\) mg/L. By construction, the curve gives approximately 7.80 mg/L at 0.5 h and 1.10 mg/L at 6 h. It therefore tracks these three plasma observations, but three points do not uniquely establish the model or its parameters.
The biexponential curve is a plasma output, not a separate estimate of liver and muscle concentrations. A specified two-compartment model can predict its peripheral state as well as its central state, but that peripheral state remains a pooled prediction. If liver and muscle behave differently, one pooled value cannot generally stand in for both. Whether a two-compartment model is useful therefore depends on the target: it adds a distributional state, but not organ-specific resolution.
What Organ Resolution Adds
An organ-resolved model can represent separate liver and muscle states, so it can address questions about their distinct concentration trajectories. The shared observations illustrate why that distinction may matter: at 6 h, liver concentration is 3.50 mg/L and muscle concentration is 1.20 mg/L. A single peripheral concentration cannot equal both values at that time. An organ-resolved model can, in principle, predict both separately, provided its inputs, structure, and parameters support those predictions.
Organ resolution also changes what must be checked. Predictions should be compared with observations for each organ over time, rather than only with plasma or a pooled tissue summary. A good plasma fit alongside persistent liver or muscle residual patterns indicates that the plasma agreement does not validate those organ predictions. Conversely, a few organ observations may be insufficient to distinguish organ-specific parameters, especially if the observation times do not capture different parts of the profiles. Predictions for unmeasured organs require additional assumptions and should not be presented as directly established by measurements from other tissues.
Worked Example: Calculating a Volume-Weighted Tissue Summary
Suppose, for illustration, the liver volume is 5 L and the muscle volume is 25 L. At 6 h, use the observed concentrations to calculate the combined amount in these two tissues and their volume-weighted mean concentration.
The liver amount is \(3.50\ \text{mg/L}\times5\ \text{L}=17.5\ \text{mg}\). The muscle amount is \(1.20\ \text{mg/L}\times25\ \text{L}=30.0\ \text{mg}\). The combined amount is \(17.5+30.0=47.5\) mg, and the combined volume is \(5+25=30\) L.
The unweighted arithmetic mean of the two concentrations would be \((3.50+1.20)/2=2.35\) mg/L, which is not the combined concentration for these unequal volumes. The weighted summary is useful for the aggregate amount in the two specified tissues, but it does not recover either organ’s concentration on its own. It also says nothing about organs excluded from the calculation.
Comparing Predictions, Diagnostics, and Limits
| Structure | Directly represented prediction | Key limitation for these observations |
|---|---|---|
| One compartment | One apparent concentration profile, such as plasma | No distinct distribution state or separate organ concentrations |
| Two compartments | Central and one lumped peripheral state, given a specified exchange model | Cannot distinguish liver from muscle within the same peripheral pool |
| Organ-resolved | Separate profiles for represented organs | Requires organ-specific structure and information; unmeasured organs are not automatically predicted reliably |
The main comparison is not “simple is wrong, detailed is right.” Each model is an approximation for a stated prediction target. A one-compartment model may be adequate for a plasma summary over a particular time range even when it cannot address tissue-specific questions. A two-compartment model may capture a central-versus-peripheral pattern while still averaging across heterogeneous tissues. Organ resolution is warranted when separate organ outputs matter and the available information can support them; it is not a guarantee of more accurate predictions.
Useful warning signs include structured residuals across time, different residual patterns across organs, and a pooled tissue prediction that appears to represent organs with visibly distinct concentrations. Also check that the comparison uses consistent units, time points, and definitions of the target. A model’s ability to reproduce the data used to fit it is not, on its own, evidence that it predicts a new organ, time range, or scenario accurately. With limited observations, apparent differences in fit can be uncertain, and an organ-specific prediction may be weakly identified.
A Practical Comparison Workflow
Specify whether the question concerns plasma, a pooled peripheral quantity, or one or more named organs.
Check that the model actually contains a state corresponding to each requested prediction.
Calculate residuals at shared times and, where available, separately for plasma and each tissue.
Distinguish a fitted plasma curve from evidence for organ-specific predictions, and report limitations from sparse or absent tissue data.
Check Your Understanding
Answer each question, then compare your reasoning with the solutions below.
- At 6 h, why can a one-compartment plasma curve not separately predict the observed liver and muscle concentrations?
- Using the one-compartment curve in this tutorial, what is the predicted plasma concentration at 6 h, to three significant figures?
- What does a two-compartment model add, and what organ-level distinction does it still lack?
- Why is the volume-weighted tissue concentration at 6 h lower than the unweighted mean in the worked example?
- If plasma residuals are small but liver residuals show a consistent time pattern, what should be questioned?
Solutions. (1) The one-compartment structure has one apparent concentration state and no separate liver or muscle states; one value cannot represent both 3.50 and 1.20 mg/L. (2) \(7.827\exp[-0.3711(6-0.5)]\approx1.017\) mg/L. (3) It adds a central and a lumped peripheral state, but does not distinguish organs grouped within that peripheral pool. (4) The larger muscle volume has the lower concentration, so it contributes more volume to the weighted summary; the calculation is \(47.5\ \text{mg}/30\ \text{L}=1.583\ \text{mg/L}\), below 2.35 mg/L. (5) The organ-specific prediction or model structure should be examined: good plasma agreement does not rule out tissue-specific structural mismatch, and the amount and timing of liver data should also be checked.