Why Lump Parameters?
A PBPK model can contain several parameters that influence a prediction in similar ways. If the available outputs respond only to a particular combination of those parameters, estimating or varying them separately may add complexity without adding useful information about those outputs. A lumped parameter replaces multiple original parameters with one composite quantity that retains the relevant model behavior.
The word “relevant” matters. A composite can preserve a plasma concentration-time curve while discarding information about which metabolic route produced that curve. It can also work for one dose or endpoint but fail for a different tissue, time range, or physiological scenario. Lumping is therefore justified by the model structure and the intended predictions, not just by a pair of similar sensitivity rankings.
This definition makes lumping conditional on the output. A composite that is sufficient for total plasma concentration may not be sufficient for tissue concentrations or metabolite formation. The introductory material on sensitivity analysis provides the context for evaluating how model outputs change with parameters; here the focus is whether that information supports replacing parameters with a meaningful quantity.
Structural Conditions for a Composite
A strong basis for lumping is an equation in which the original parameters enter through a sum, product, ratio, or other identifiable combination. The composite must have a clear interpretation and consistent units. For example, two parallel clearances can be combined as a total clearance if the model uses only their sum to describe the measured parent-drug concentration. Two dimensionless fractions, by contrast, should not be added and called a clearance.
A useful general test is to write the selected model outputs as \(y=f(\theta_1,\theta_2,\boldsymbol{\eta})\), where \(\boldsymbol{\eta}\) denotes the other model inputs. Exact lumping is possible if there is a function \(g\) such that
for every allowed value of \(\theta_1\), \(\theta_2\), and \(\boldsymbol{\eta}\). The equality must hold for the outputs and conditions the analysis aims to preserve, not merely at one nominal parameter set.
Sensitivity columns can help identify a local candidate. For positive parameters and outputs, define the dimensionless log sensitivity of output \(y_i\) to parameter \(\theta_j\) as
If two columns are nearly equal across selected outputs, equal proportional changes in the two parameters have similar local effects. This can motivate checking a product composite, because \(\log(\theta_1\theta_2)=\log\theta_1+\log\theta_2\). But near-equal columns are local evidence, not proof of an exact composite. Sensitivities can vary with time, dose, or parameter values, and similarity for one endpoint does not imply similarity for another.
Worked Examples
Worked Example: Combining Parallel Clearances
Consider a hypothetical one-compartment model after an intravenous bolus. The parent drug is eliminated through two parallel pathways with clearances \(CL_1=2\) L/h and \(CL_2=3\) L/h. Assume linear elimination, a fixed volume \(V=20\) L, and that the only selected output is parent-drug plasma concentration. The model contains no pathway-specific metabolite output.
Under these assumptions, the elimination rate is the sum of the pathway rates. For concentration \(C\), the amount eliminated per unit time is \((CL_1+CL_2)C\), so the concentration equation is
The model depends on the two clearances through their sum. Define \(CL_{\mathrm{tot}}=CL_1+CL_2\). Substitution gives \(CL_{\mathrm{tot}}=2+3=5\) L/h. For a hypothetical 100 mg bolus, \(C(0)=100/20=5\) mg/L, and the concentration at time \(t\) is
At \(t=4\) h, \(C(4)=5\exp(-1)=1.839\) mg/L. Using \(CL_{\mathrm{tot}}=5\) L/h gives exactly the same concentration as retaining both clearances. This is an exact lump for the specified parent concentration output under the stated model. It is not an exact lump for pathway-specific metabolite formation: those outputs would depend on \(CL_1\) and \(CL_2\) separately.
Worked Example: A Composite in the Well-Stirred Liver Model
Suppose a hypothetical well-stirred liver has blood flow \(Q_h=90\) L/h, unbound blood fraction \(f_{u,b}=0.10\), and intrinsic clearance \(CL_{\mathrm{int}}=600\) L/h. Assume constant binding, linear metabolism, rapid mixing within the liver, and no transport limitation. Consider hepatic clearance of parent drug as the selected output.
In the well-stirred model, hepatic inflow is \(Q_h C_{\mathrm{in}}\). At steady state, outflow plus elimination balances inflow: \(Q_h C_{\mathrm{in}}=Q_h C_{\mathrm{out}}+f_{u,b}CL_{\mathrm{int}}C_{\mathrm{out}}\). Solving for the outlet-to-inlet ratio gives \(C_{\mathrm{out}}/C_{\mathrm{in}}=Q_h/(Q_h+f_{u,b}CL_{\mathrm{int}})\). The extraction ratio is one minus this ratio, and clearance is flow multiplied by extraction. Therefore,
The two parameters enter this clearance through the product \(\gamma=f_{u,b}CL_{\mathrm{int}}\), which has units L/h. Here \(\gamma=0.10(600)=60\) L/h, so \(CL_h=90(60)/(90+60)=36\) L/h.
For the hepatic clearance output under these assumptions, the pair can be replaced by \(\gamma=60\) L/h. For instance, \(f_{u,b}=0.20\) and \(CL_{\mathrm{int}}=300\) L/h also give \(\gamma=60\) L/h and \(CL_h=36\) L/h. This does not establish equivalence for every PBPK output: if unbound concentrations, distribution, binding, or a non-hepatic process is also an output, \(f_{u,b}\) may enter separately and the product may no longer preserve the predictions.
Worked Example: Similar Local Effects Do Not Guarantee a Valid Lump
In a hypothetical PBPK simulation, two positive parameters \(\theta_A\) and \(\theta_B\) have the following log sensitivities at one nominal parameter set. The four outputs are peak plasma concentration, plasma AUC, late plasma concentration, and tissue-to-plasma concentration ratio.
| Output | \(\theta_A\) sensitivity | \(\theta_B\) sensitivity |
|---|---|---|
| Peak plasma concentration | 0.42 | 0.40 |
| Plasma AUC | 0.80 | 0.76 |
| Late plasma concentration | 0.35 | 0.05 |
| Tissue-to-plasma ratio | 0.70 | 0.10 |
For peak concentration, the absolute difference in sensitivities is \(0.42-0.40=0.02\); for AUC it is \(0.80-0.76=0.04\). Near the nominal point, a 1% increase in either parameter is therefore associated with similar approximate percentage changes in these two outputs. This suggests a candidate local grouping if peak concentration and AUC are the only outputs of interest.
The evidence is not similar across all outputs: the sensitivity differences are \(0.35-0.05=0.30\) for late concentration and \(0.70-0.10=0.60\) for the tissue-to-plasma ratio. Thus, the parameters do not have interchangeable effects for the full output set. Moreover, the table alone does not show that the model equations contain a composite through which both parameters act. The defensible conclusion is that they have similar local effects on peak concentration and AUC at this nominal point, not that they can be replaced by a single parameter for the whole model.
Justifying and Checking a Proposed Lump
The central justification is algebraic: show that the chosen outputs factor through the proposed composite. The parallel-clearance example did this by deriving the rate equation in terms of \(CL_1+CL_2\). The liver example did it by deriving clearance in terms of \(f_{u,b}CL_{\mathrm{int}}\). In either case, the derivation states both the structural assumptions and the outputs for which the simplification holds.
Name the outputs, time range, dose conditions, and physiological scenarios that the reduced model must preserve.
Check whether the parameters enter through a single sum, product, or other interpretable composite for those targets. Confirm that the units of the proposed composite are appropriate.
Compare sensitivity effects for all selected outputs, and check whether the comparison changes across times, doses, or parameter values.
Run the original and lumped formulations over the intended conditions. Check the relevant concentration-time courses and endpoints, not only one summary statistic.
For an exact structural lump, step 4 should give identical target predictions up to numerical solver tolerance when the composite is held fixed. For a local approximate lump, differences are expected; they must be assessed against a prespecified, scientifically acceptable tolerance. A small difference in one endpoint does not establish a small difference in all outputs.
Alternatives, Diagnostics, and Limitations
Lumping is not the same as fixing a low-influence parameter. Fixing retains one parameter and assigns the other a chosen value; lumping replaces both with a composite that may still vary. Nor is lumping the same as averaging parameter values: an average has no general reason to preserve the equations or predictions. Retaining separate parameters is preferable when pathway-specific outputs matter or when the model uses the parameters separately. Local sensitivity similarity can help prioritize a candidate for investigation, but structural algebra and prediction checks are needed before making it a replacement.
Warning signs include a proposed composite with inconsistent units, a sensitivity relationship that changes sign across the time course, or a reduced model that matches AUC but not peak or trough concentrations. Another failure mode is to verify the composite only at one nominal input: nonlinear terms can make the agreement disappear at another dose or physiological condition. If two parameters affect binding, transport, and metabolism through distinct equations, a composite that captures one of those pathways may discard the others.
Exact lumping preserves only the outputs and conditions used to justify it. Even a valid composite can make it impossible to report the original parameters separately or to predict an output that distinguishes their mechanisms. A local lump has an additional limitation: it is an approximation whose accuracy depends on the neighborhood and outputs checked. Report the composite’s definition, units, structural assumptions, and scope so that its interpretation is not broader than the evidence.
Check Your Understanding
Answer each question, then compare your reasoning with the solutions below.
- Two parallel elimination clearances are 1.5 L/h and 2.5 L/h. Under a linear one-compartment model that predicts only parent-drug concentration, what is the total clearance? What additional output could make this lump invalid?
- In the well-stirred liver model, \(f_{u,b}=0.05\), \(CL_{\mathrm{int}}=800\) L/h, and \(Q_h=100\) L/h. Calculate the composite and hepatic clearance.
- Two parameters have nearly identical sensitivities for AUC but very different sensitivities for tissue concentration. Is a single lump justified for both outputs? Explain.
- Why does a close match at one nominal parameter set not prove exact lumping?
Solutions. (1) The total clearance is \(1.5+2.5=4.0\) L/h. A pathway-specific metabolite output could depend on the individual clearances and therefore invalidate the lump for that output. (2) The composite is \(f_{u,b}CL_{\mathrm{int}}=0.05(800)=40\) L/h. Thus \(CL_h=100(40)/(100+40)=28.57\) L/h. (3) No. The similar AUC sensitivities support, at most, a local candidate for AUC; the different tissue sensitivities show that the effects are not interchangeable for both outputs. (4) Agreement at one parameter set checks only that point. Exact lumping requires the outputs to depend on the original parameters only through the composite throughout the stated domain and conditions.