Why \(Q\) and \(V_p\) Can Be Hard to Separate
In a two-compartment model, intercompartmental clearance \(Q\) describes exchange between the central and peripheral compartments, while \(V_p\) is the peripheral volume. Central concentration data reflect this exchange, but the peripheral-to-central return rate is determined by their ratio. As a result, different combinations of \(Q\) and \(V_p\) can imply similar return rates and, over a finite set of sampling times, similar predicted concentrations.
This is a question of practical separation, not a claim that \(Q\) and \(V_p\) are always structurally unidentifiable. With an appropriate model and sufficiently informative, ideal observations, parameters may be structurally identifiable; real data can nevertheless provide weak information for distinguishing them. The earlier material on structural and practical identifiability and on sampling coverage provides that distinction.
The quotient \(Q/V_p\) is important, but it is not the whole concentration profile. Changing \(Q\) also changes \(k_{12}\), and therefore can change the distribution process even when \(k_{21}\) is held constant. The practical issue is that these related effects may be difficult to distinguish from noisy central concentrations, especially when sampling does not adequately capture the distribution and terminal phases.
Model and Concentration Profile
Consider an IV bolus dose \(D\) entering the central compartment. Let \(A_c(t)\) and \(A_p(t)\) denote the amounts of drug in the central and peripheral compartments. Assume linear transfer between compartments, first-order elimination from the central compartment, fixed clearances and volumes, and no additional input after the bolus. The amount-balance equations are
with \(A_c(0)=D\), \(A_p(0)=0\), and observed central concentration \(C(t)=A_c(t)/V_c\). Eliminating \(A_p\) gives a second-order equation for the central concentration:
The two positive decay rates, conventionally written \(\alpha\) and \(\beta\) with \(\alpha>\beta\), satisfy
The initial concentration is \(D/V_c\), and its initial slope is \(-(k_{10}+k_{12})D/V_c\), since initially there is no drug in the peripheral compartment to return to the central one. These conditions determine the coefficients of the biexponential profile:
Thus \(k_{21}\) affects the profile’s coefficients as well as the decay rates. But the rates and coefficients also depend on \(k_{10}\) and \(k_{12}\). Because \(k_{12}=Q/V_c\) and \(k_{21}=Q/V_p\), changing \(Q\) and \(V_p\) together can leave one part of the profile unchanged while shifting another part only modestly.
Coordinated Changes in \(Q\) and \(V_p\)
Worked Example: Preserving the Return Rate
Suppose a hypothetical two-compartment model has \(Q=0.30\) L/h and \(V_p=1.50\) L. Calculate \(k_{21}\), then increase both parameters by 10% and check whether the return rate changes.
Initially,
After the coordinated increase, \(Q=0.330\) L/h and \(V_p=1.65\) L. Therefore,
The ratio, and hence \(k_{21}\), is unchanged. If \(V_c=1.00\) L, however, \(k_{12}\) changes from \(0.300\ \mathrm{h}^{-1}\) to \(0.330\ \mathrm{h}^{-1}\). This calculation shows why preserving \(k_{21}\) does not guarantee an identical concentration profile: the central-to-peripheral transfer rate also changes.
More generally, for small relative changes, the change in the ratio is described by
This follows by taking logarithms of \(k_{21}=Q/V_p\), giving \(\log(k_{21})=\log(Q)-\log(V_p)\), and differentiating. Similar proportional changes in \(Q\) and \(V_p\) can therefore leave \(k_{21}\) nearly unchanged. This is a useful way to understand why the two parameters may move together during estimation; it does not establish that any particular data set determines them poorly.
Comparing Two Hypothetical Profiles
Worked Example: A 10% Parameter Shift and Its Predictions
Consider a hypothetical 5 mg IV bolus with \(V_c=1.00\) L and \(CL=0.200\) L/h. Compare the following parameter sets: set A has \(Q=0.300\) L/h and \(V_p=1.50\) L; set B has \(Q=0.330\) L/h and \(V_p=1.65\) L. Both give \(k_{10}=0.200\ \mathrm{h}^{-1}\) and \(k_{21}=0.200\ \mathrm{h}^{-1}\), while \(k_{12}\) is \(0.300\ \mathrm{h}^{-1}\) for A and \(0.330\ \mathrm{h}^{-1}\) for B.
For this model, the resulting biexponential profiles, with coefficients and rates rounded for display, are
where \(t\) is in hours and the predicted concentrations are in mg/L. At \(t=0.5\) h, substitute into each expression:
The difference is about \(3.9248-3.8695=0.0553\) mg/L, or about \(100(0.0553/3.9248)=1.4\%\) of set A’s prediction at that time. At 8 h, the same substitutions give approximately \(0.74649\) mg/L for set A and \(0.73092\) mg/L for set B, a difference of \(0.01557\) mg/L.
| Time (h) | Set A (mg/L) | Set B (mg/L) | Absolute difference (mg/L) |
|---|---|---|---|
| 0.5 | 3.925 | 3.870 | 0.0553 |
| 8 | 0.74649 | 0.73092 | 0.01557 |
The predictions are close at these times despite the 10% increases in both \(Q\) and \(V_p\). They are not identical: \(k_{12}\) changed, as did the biexponential coefficients and rates. A difference that is small in concentration units may still matter if measurements are precise and sampling is informative; whether it is distinguishable depends on the observation error and the full sampling design.
What the Data Can and Cannot Tell You
Worked Example: Changing One Parameter at a Time
Start again from \(Q=0.300\) L/h and \(V_p=1.50\) L, so \(k_{21}=0.200\ \mathrm{h}^{-1}\). First hold \(Q\) fixed and increase \(V_p\) by 10% to 1.65 L:
Instead, hold \(V_p\) fixed and increase \(Q\) by 10% to \(0.330\) L/h:
These one-at-a-time calculations show how strongly the ratio responds when only one component changes. If fitted parameter estimates shift together instead, the quotient can be more stable than either parameter separately. That pattern is a reason to inspect how well the data distinguish the parameters, not proof that they are unidentifiable.
Useful checks include comparing predicted profiles across plausible parameter combinations, checking whether the estimates change substantially when early or late samples are removed, and examining whether parameter uncertainty is large relative to the estimated values. The interpretation should remain tied to the observation schedule: central concentrations measured only over a restricted time range may not provide enough distinct information about the distribution process and return from the peripheral compartment.
The main alternative is to report or estimate a parameterization that more directly reflects the information supported by the data, such as \(k_{21}\) alongside \(Q\), \(V_p\), or a suitably constrained combination. This does not create information that is absent. Additional informative sampling, reliable prior knowledge, or measurements beyond central concentration may help separate parameters, but each requires defensible assumptions and an appropriate model. The choice between parameterizations should preserve the model’s meaning and units.
A further limitation is that this comparison assumes the stated two-compartment structure and linear, time-invariant transfer. Similar predictions under that model do not establish that the model is physiologically correct, nor do they imply that all alternative parameter sets will fit all observation times equally well. Always distinguish a demonstrated similarity at selected times from equivalence across an entire profile.
Check Your Understanding
Answer each question, then compare with the short solutions below.
- If \(Q=0.24\) L/h and \(V_p=1.20\) L, what is \(k_{21}\), including units?
- If \(Q\) and \(V_p\) both increase by 5%, what happens to \(k_{21}\)? State the assumption behind your answer.
- Why can keeping \(k_{21}\) fixed still change the central concentration profile?
- Does a stable estimate of \(k_{21}\) prove that \(Q\) and \(V_p\) are each precisely estimated? Explain.
Solutions. (1) \(k_{21}=0.24/1.20=0.200\ \mathrm{h}^{-1}\). (2) The ratio is unchanged if both are multiplied by the same factor, since \(1.05Q/(1.05V_p)=Q/V_p\). (3) Changing \(Q\) changes \(k_{12}=Q/V_c\), which affects the distribution process and thus can alter the profile. (4) No. Many numerator-denominator combinations can give the same ratio, so precision for the ratio does not by itself establish precision for either component.