Why Combined Effects Need a Model
A patient may differ from a reference patient in renal function, hepatic function, and body size at the same time. Applying three dose-adjustment rules independently can produce a plausible-looking number that is wrong: each rule may alter the same part of clearance, or a weight-related adjustment may repeat information already included in a renal-function measure. The central task is therefore to specify what each covariate changes, and on what scale, before predicting exposure or adjusting a dose.
This tutorial considers predictions based on a pharmacokinetic clearance model. The calculations assume linear pharmacokinetics over the relevant dose range. They do not, by themselves, account for changes in bioavailability, protein binding, active metabolites, or time-varying organ function; those require additional model components when relevant.
Start With the Clearance Structure
For parallel elimination pathways, total clearance is the sum of pathway clearances. Suppose renal, hepatic, and other pathways contribute to total clearance in a reference patient. Let \(CL_{R,ref}\), \(CL_{H,ref}\), and \(CL_{O,ref}\) denote those reference clearances, in L/h. Let \(r\) and \(h\) be dimensionless multipliers for renal and hepatic function, respectively, with 1 representing the reference condition. A basic component model is:
This formula assumes the pathways act in parallel, the multipliers apply to their specified pathways, and the reference component clearances and multipliers are appropriate for the patient. It does not assume that renal and hepatic impairment multiply the same total clearance. If body size has a common proportional effect on all three components, a model might instead specify:
where \(s\) is a dimensionless size multiplier. This version is appropriate only if the model supports applying that multiplier to every component. If a renal-function measure already represents size-adjusted renal function, applying \(s\) to the renal component as well may count size twice. The structure must match how the covariates were defined and how the model was estimated.
From Clearance to Exposure
For linear pharmacokinetics with constant clearance, the area under the concentration–time curve (AUC) after an intravenous dose is dose divided by clearance. For an oral dose, the same relation includes bioavailability \(F\), the fraction of the dose reaching systemic circulation. At steady state, average concentration over a dosing interval has the corresponding relation with dose rate.
Here, \(Dose\) is the dose, \(CL\) is total clearance, and \(\tau\) is the dosing interval. If \(F\) is unchanged, exposure is inversely proportional to clearance: a clearance ratio of 0.80 implies an AUC ratio of \(1/0.80=1.25\) for the same dose. A dose chosen to match a target AUC can be calculated from the same relation, but a target exposure and an appropriate route-specific value of \(F\) must be specified.
Worked Example: Combining Pathway and Body-Size Multipliers
Consider a hypothetical drug with reference renal, hepatic, and other clearances of 40, 50, and 10 L/h. For one patient, the renal multiplier is \(r=0.40\), the hepatic multiplier is \(h=0.60\), and a model-justified common size multiplier is \(s=1.15\). Assume these multipliers apply to separate, specified components as shown in the model, and that the patient receives 100 mg intravenously.
First calculate the clearance inside the size multiplier:
Now apply the common size multiplier once:
For intravenous administration, \(F=1\). The predicted AUC is:
The reference total clearance is \(40+50+10=100\) L/h, so at the same dose its AUC would be \(1.00\ \text{mg}\cdot\text{h/L}\). The patient's predicted AUC is therefore about 1.55 times the reference AUC under these assumptions. Multiplying the entire reference clearance by \(0.40\), \(0.60\), and \(1.15\) would give \(27.6\) L/h, but that calculation incorrectly applies both pathway-specific multipliers to every pathway.
Interactions Are Scale-Dependent
A model may include an interaction when the joint effect of two covariates is not captured by their separate main effects. One common form models log clearance relative to a reference value:
Here \(x_R\) and \(x_H\) are coded renal and hepatic impairment measures, each equal to 0 at its reference value; the \(\beta\) terms are model coefficients. With this coding, \(\beta_{RH}\) describes how the renal effect on log clearance changes as the hepatic covariate changes (and equivalently how the hepatic effect changes with renal status). Exponentiating the equation gives a clearance ratio. An interaction is therefore interpreted on the model's log-clearance scale, not as a fixed additive number of L/h.
Worked Example: Predicting Clearance With an Interaction
For a hypothetical model, suppose \(CL_{ref}=8.00\) L/h, \(\beta_R=-0.40\), \(\beta_H=-0.30\), and \(\beta_{RH}=-0.25\). A patient has \(x_R=1\) and \(x_H=1\). Assume these covariate values are within the model's supported range.
Substitute into the log-clearance model:
Exponentiate to obtain the clearance ratio and clearance:
At a 200 mg intravenous dose, the predicted AUC is \(200/3.09=64.7\ \text{mg}\cdot\text{h/L}\). Without the interaction term, the predicted ratio would be \(e^{-0.70}=0.4966\), giving \(CL=3.97\) L/h. In this hypothetical model, the negative interaction predicts lower clearance for the joint covariate values than the main effects alone predict. The interaction is not proof of a biological mechanism; it is a model feature whose support and interpretation depend on the data and coding.
Body Size and Renal Function: Check What Is Already Included
Renal-function measures do not all have the same size convention. For example, an estimated glomerular filtration rate reported per \(1.73\ \text{m}^2\) of body-surface area is indexed. Converting it to an absolute estimate uses the patient's body-surface area (BSA):
If the model's renal component is driven by absolute filtration, that conversion already makes the renal-function covariate patient-size-dependent. A separate size multiplier may still be appropriate for other clearance pathways, but should not automatically be applied again to the renal component. This is a modeling decision, not a universal correction: the model must use the same definition and scale of renal function on which its parameters were based.
Worked Example: Avoiding a Second Size Adjustment
Suppose a hypothetical patient has indexed \(eGFR=45\ \text{mL/min}/1.73\ \text{m}^2\) and \(BSA=2.30\ \text{m}^2\). The model uses absolute eGFR, with a reference value of \(90\ \text{mL/min}\). Its reference renal clearance is 3.00 L/h. It also assigns a size multiplier of 1.217 to the combined nonrenal clearance of 5.00 L/h, but not to the renal component, because absolute eGFR is the renal covariate.
Convert the indexed value:
The renal multiplier is \(59.8/90=0.664\), so renal clearance is \((3.00)(0.664)=1.99\) L/h. Nonrenal clearance is \((5.00)(1.217)=6.09\) L/h. Thus:
Applying the size multiplier to renal clearance again would instead predict \((1.99)(1.217)+6.09=8.51\) L/h. That differs because the extra multiplier changes the renal component a second time. Whether the stated model is scientifically appropriate depends on its development and validation; the calculation shows why covariate definitions must be checked before combining adjustments.
Assumptions, Diagnostics, and Alternatives
A combined prediction is defensible only if its assumptions are plausible for the intended patient and dosing question. Check that the reference clearances and covariate definitions are compatible; that pathway clearances are meaningfully separable if a component model is used; and that linear, time-invariant clearance is adequate for the exposure calculation. A fitted regression model additionally assumes its functional form and coefficient estimates adequately describe the relevant population. Predictions far outside the covariate ranges used to develop the model are extrapolations, even when the arithmetic is straightforward.
Useful checks include comparing predicted clearance with physiologically plausible values, plotting residuals against each covariate and their combinations, and examining whether the model systematically over- or under-predicts in patients with both impairments. For an interaction model, inspect the joint covariate data: if few patients have both renal and hepatic impairment, the interaction may be weakly estimated. Also verify units and transformations. A coefficient for a log-transformed clearance ratio cannot be interpreted as an absolute L/h change.
A common mistake is to assume that each impairment contributes a percentage reduction in total clearance, then multiply all reductions together. That reasoning is incorrect unless the model explicitly specifies those multiplicative effects on total clearance. If renal clearance is only 40% of baseline total clearance, a 50% reduction in renal clearance reduces total clearance by 20%, not 50%, when other pathways are unchanged. The correction is to model the affected component or use a validated total-clearance model, then calculate exposure from predicted total clearance.
| Approach | Useful when | Main caution |
|---|---|---|
| Pathway component model | Pathway contributions and their covariate effects are known or estimable. | Incorrect allocation or overlapping multipliers distort total clearance. |
| Regression model with main effects | Covariates predict clearance and joint effects are adequately represented without interactions. | Effects depend on coding and scale; extrapolation can be misleading. |
| Regression model with interactions | Data support a joint effect beyond the specified main effects. | Interactions may be imprecise when combinations of covariates are sparse. |
| Independent dose rules | A validated rule applies to the patient's specific circumstances. | Combining rules can double-count changes or ignore pathway contributions. |
No single structure is best for every drug. Component models can make pathway assumptions explicit; regression models can summarize observed relationships without claiming a mechanism. Both can fail if covariates are measured inconsistently, if patients lie outside the studied range, or if clearance is not the only determinant of exposure. For example, changes in \(F\) alter oral AUC even if clearance is predicted correctly.
Check Your Understanding
Answer each question before checking the brief solutions.
- A drug has reference renal clearance 2 L/h and nonrenal clearance 6 L/h. Renal function reduces the renal component to 50% of reference. What is total clearance, and what is the AUC ratio at the same intravenous dose?
- In a log-clearance model, what does a negative renal–hepatic interaction coefficient mean when both impairment covariates are coded as positive values?
- An indexed eGFR is converted to absolute eGFR using BSA. Why might applying another size multiplier to the renal clearance component be inappropriate?
- A model predicts \(CL/CL_{ref}=0.80\). For unchanged intravenous dose, what is the AUC ratio?
- Name one diagnostic concern when estimating an interaction between renal and hepatic function.
Solutions. (1) Total clearance is \((2)(0.50)+6=7\) L/h, compared with 8 L/h at reference. The AUC ratio is \(8/7=1.14\). (2) It predicts an additional decrease in log clearance for the joint covariate values, beyond the main effects; interpretation is conditional on the coding and model scale. (3) Absolute eGFR already incorporates BSA, so another multiplier may repeat the size-related change in the renal component. (4) The AUC ratio is \(1/0.80=1.25\). (5) If few patients have both impairments, the interaction may be imprecisely estimated; inspect joint covariate support and uncertainty.