1. What Are Fixed and Random Effects in Population PK?
Population pharmacokinetic (PopPK) models are designed to describe both the typical pharmacokinetic behavior of a population and the differences between individuals. This requires a model that can represent more than a single set of PK parameters.
A useful conceptual decomposition is:
The fixed effects describe systematic features that are shared across the population or explained by measured predictors such as body weight, age, renal function, or treatment group. The random effects describe unexplained individual-to-individual deviations around those typical values.
2. Why Population PK Models Are Called Mixed-Effects Models
A population PK model commonly combines two kinds of information:
- Fixed effects: population parameters and systematic covariate effects.
- Random effects: subject-specific deviations from the population predictions.
Because the model contains both components, it is commonly described as a nonlinear mixed-effects model, or NLME model, when the PK relationships are nonlinear.
For example, suppose clearance is modeled as a function of body weight. The population model might estimate a typical clearance and a weight effect, while a subject-specific random effect allows two patients with the same modeled weight to have different individual clearances.
A population PK model combines population-level fixed effects with subject-level random effects to obtain individual PK parameters.
3. What Is a Fixed Effect?
A fixed effect represents a parameter or relationship that is estimated for the population as a whole. The estimated value is treated as a common model parameter rather than as a random realization that differs across subjects.
Typical fixed effects in a population PK model include:
| Fixed effect | Example | Interpretation |
|---|---|---|
| Typical CL | \(\theta_{CL}\) | Typical clearance for the reference population or covariate values |
| Typical V | \(\theta_V\) | Typical volume of distribution |
| Weight effect | \(\theta_{WT}\) | Systematic relationship between body weight and a PK parameter |
| Renal function effect | \(\theta_{RF}\) | Systematic relationship between renal function and clearance |
| Formulation effect | \(\theta_{F}\) | Systematic difference associated with a formulation or treatment condition |
The term fixed does not mean that the parameter is known or cannot change. It means that the model estimates one population-level parameter rather than a separate random value for each subject.
4. Typical Values as Fixed Effects
Suppose a population PK model estimates a typical clearance of 5 L/h. This value represents the population-level parameter around which individual clearances vary.
A simplified model could therefore begin with:
Here, the value 5 L/h is a fixed effect, while \(\eta_i\) is a subject-specific random effect.
If \(\eta_i=0\), the individual is at the population-typical value:
If \(\eta_i\) is positive, the individual's clearance is above the typical value. If it is negative, the individual's clearance is below the typical value.
5. What Is a Random Effect?
A random effect represents unexplained variability among individuals. In population PK, the most familiar random effect is between-subject variability (BSV)interindividual variability (IIV).
For example, two patients can have the same dose, similar measured covariates, and similar sampling schedules but still have different clearance values. The random effect allows the model to represent that difference.
A common model for clearance is:
where:
- \(\theta_{CL}\) is the typical population clearance.
- \(\eta_{CL,i}\) is the individual-specific random effect.
- \(CL_i\) is the resulting individual clearance for subject \(i\).
The random effect is commonly assumed to have a mean of zero:
The variance \(\omega^2_{CL}\) describes the magnitude of between-subject variability on the model's random-effect scale.
6. Why Is the Random Effect Often Exponential?
Population PK models frequently use an exponential random-effects model:
This formulation has an important practical property: if \(\theta_P\) is positive, then \(P_i\) remains positive for every finite value of \(\eta_i\). That is useful for parameters such as clearance and volume, which are generally constrained to be positive.
The exponential formulation also means that random effects act multiplicatively rather than additively.
For example, if:
and:
then:
The individual therefore has a clearance 20% above the typical value.
7. Where Do Covariates Fit?
Covariates explain systematic differences between subjects. Common population PK covariates include:
- Body weight
- Age
- Sex
- Renal function
- Hepatic function
- Concomitant medications
- Disease status
- Treatment or formulation
Suppose clearance increases with body weight according to an allometric model:
The components have different roles:
| Component | Role |
|---|---|
| \(\theta_{CL}\) | Typical population clearance at the reference weight |
| \(WT_i\) | Individual body weight |
| 70 | Reference body weight in kg |
| 0.75 | Fixed-effect allometric exponent |
| \(\eta_{CL,i}\) | Individual random deviation not explained by the model |
This illustrates an important feature of population PK: covariates and random effects answer different questions. A covariate explains a systematic pattern; the random effect represents remaining individual variability after accounting for modeled systematic effects.
8. Explained and Unexplained Variability
Suppose two patients have different clearance values. A population PK model might determine that some of the difference is associated with body weight. After accounting for body weight, however, the patients may still differ.
Conceptually:
The covariate component is represented through fixed effects. The remaining individual difference is represented through random effects.
9. What Does the Random-Effect Variance Mean?
Suppose the model specifies:
The quantity \(\omega^2_{CL}\) is the variance of the random effect. Its square root, \(\omega_{CL}\), is the standard deviation on the log scale for the exponential model.
For example, if:
then:
For an exponential random-effects model, the corresponding variability is naturally interpreted on a multiplicative scale rather than as a simple additive number of L/h.
This is one reason it is important to distinguish the random-effect variance from the variability of clearance itself.
10. Random Effects Can Be Correlated
Population PK models often include random effects on several parameters. For example, a model may include between-subject variability on both clearance and volume:
The variance-covariance matrix might be written as:
The off-diagonal element represents covariance between the individual deviations in clearance and volume.
Equivalently, the model can describe this relationship using a correlation:
A correlation allows the model to represent situations in which subjects with higher-than-typical clearance also tend to have higher- or lower-than-typical volume, after accounting for the fixed effects.
11. Fixed and Random Effects Are Not the Only Sources of Variability
Population PK models generally distinguish between-subject variability from residual unexplained variability.
The random effects \(\eta_i\) describe differences between individuals in their underlying PK parameters. A separate residual error model describes differences between observed concentrations and the concentrations predicted for an individual.
Conceptually:
Thus, there are at least two different levels of variability:
| Variability | What it represents | Typical notation |
|---|---|---|
| Between-subject variability | Differences in individual PK parameters | \(\eta\), \(\Omega\) |
| Residual variability | Differences between observations and model predictions | \(\epsilon\), \(\Sigma\) |
Keeping these sources of variability conceptually separate is essential when interpreting a population PK model.
12. Worked Example: Clearance With a Weight Covariate
Consider a hypothetical population PK model with a typical clearance of 5 L/h. Assume clearance follows an allometric weight relationship with a fixed exponent of 0.75 and includes an individual random effect:
Step 1: Consider a 70-kg subject
For a subject weighing 70 kg:
If the subject's random effect is zero:
This is the typical clearance at the reference weight.
Step 2: Consider a 100-kg subject
Ignoring the individual random effect for the moment:
The fixed-effect weight relationship therefore predicts a higher typical clearance for the 100-kg subject.
Step 3: Add an individual random effect
Suppose the 100-kg subject has:
Then:
The interpretation is important: body weight produces a systematic change in the predicted clearance through the fixed effect, while the random effect moves the individual away from that covariate-adjusted prediction.
13. What Does an Individual Random Effect of Zero Mean?
An individual random effect of zero does not mean that the individual has no variability or that the individual is exactly average in every respect.
For an exponential model:
when \(\eta_i=0\):
Therefore, \(\eta_i=0\) means that the individual's parameter is equal to the model-predicted value after accounting for the fixed effects and covariates.
It does not mean that the person's observed concentrations will equal the population prediction at every sampling time. Residual variability and other sources of uncertainty remain.
14. Fixed Effects vs. Random Effects
| Feature | Fixed effect | Random effect |
|---|---|---|
| Primary role | Describe typical population behavior or systematic relationships | Describe subject-specific deviations |
| Example | Typical CL, weight exponent, renal function effect | Individual deviation in CL |
| Varies by subject? | Not directly | Yes |
| Estimated as | Population parameter | Distribution of subject-level deviations |
| Typical notation | \(\theta\) | \(\eta\) |
| Population quantity | Typical value or covariate coefficient | Variance/covariance structure \(\Omega\) |
| Scientific question | What systematic factors explain PK? | How much do individuals differ after accounting for those factors? |
15. What Happens When a Covariate Explains Variability?
Suppose a base model contains substantial between-subject variability in clearance:
If body weight explains part of the variability, the model can be expanded:
The new model explicitly explains part of the systematic difference in clearance using body weight. The remaining unexplained variability is still represented by \(\eta_{CL,i}\).
If the covariate meaningfully explains between-subject differences, the estimated random-effect variance may decrease. This is one reason covariate modeling is an important part of population PK development.
16. Population Predictions vs. Individual Predictions
The distinction between fixed and random effects also affects prediction. Population predictions use the estimated fixed effects and measured covariates, without requiring an individual random effect.
An individual prediction can additionally incorporate information about the individual's observed concentrations and estimated random effects.
Conceptually:
This distinction is fundamental to population PK and therapeutic drug monitoring. A population model can provide an initial expectation, while individual data can provide information that shifts predictions toward a specific patient's observed PK behavior.
17. A Caution About Estimated Individual Random Effects
After fitting a population PK model, individual random effects can be estimated from each subject's data. These estimates are often called empirical Bayes estimates or post hoc estimates, depending on the modeling framework.
They can be useful for understanding individual PK, but they should not automatically be treated as if they were measured quantities.
The amount of information available for an individual depends on factors such as:
- Number of concentration measurements
- Sampling times
- Assay precision
- Magnitude of between-subject variability
- Magnitude of residual variability
- How informative the individual's observations are about the parameters
When individual data are sparse, the estimated individual random effects may be pulled toward the population distribution. This phenomenon is commonly referred to as shrinkage.
18. Common Misunderstandings
Mistake 1: Thinking “fixed” means known
Fixed effects are estimated from data. The term refers to their role as population-level parameters, not to certainty.
Mistake 2: Thinking random effects are measurement error
Random effects describing between-subject variability are different from residual error. Random effects represent differences in underlying individual PK parameters, whereas residual error represents discrepancies between observations and model predictions.
Mistake 3: Assuming a covariate explains all variability
A covariate can explain systematic differences without eliminating the need for between-subject variability.
Mistake 4: Interpreting \(\eta=0\) as “no variability”
An individual random effect of zero means that the individual's modeled parameter equals its covariate-adjusted typical value. It does not mean that the individual's concentrations are identical to population predictions.
Mistake 5: Treating individual random effects as observed data
Individual random effects are inferred from the population model and the individual's observations. They are not directly measured biological quantities.
19. A Practical Population PK Modeling Workflow
- Define the structural PK model. Specify compartments, absorption, elimination, and other structural features.
- Estimate typical population parameters. These form the fixed-effect portion of the model.
- Add an appropriate random-effects structure. Allow relevant PK parameters to vary between individuals.
- Specify the residual error model. Separate between-subject variability from unexplained observation-level variability.
- Explore covariates. Determine whether measured patient characteristics explain systematic differences in PK.
- Evaluate the model. Use parameter estimates, variability estimates, diagnostic plots, objective function changes where appropriate, predictive checks, and scientific plausibility.
- Interpret population and individual predictions separately. Distinguish typical population behavior from subject-specific inference.
20. Key Takeaways
- Population PK models describe both typical population behavior and between-subject differences.
- Fixed effects represent population-level parameters and systematic relationships such as covariate effects.
- Random effects represent subject-specific deviations from the population prediction.
- A common individual parameter model is \[ P_i=\theta_Pe^{\eta_i} \] where \(\theta_P\) is the typical population value and \(\eta_i\) is the individual random effect.
- Covariates explain systematic differences between subjects; random effects represent remaining individual variability.
- Between-subject variability and residual variability are distinct components of a population PK model.
- Multiple random effects can be correlated, allowing the model to represent relationships between individual deviations in parameters such as clearance and volume.
- An individual random effect of zero means that the individual's parameter is equal to the model's covariate-adjusted typical value.
- Individual random effects are model-based estimates rather than directly observed biological measurements.
- The distinction between fixed and random effects is central to understanding how population PK models move from a typical population description to individualized PK predictions.
Where to Go Next
A natural progression is to study between-subject variability and residual variability in greater detail, followed by covariate model building, allometric scaling, interoccasion variability, parameter correlations, and nonlinear mixed-effects estimation.
The next tutorial can build on these concepts by showing how a population PK model is translated into an actual NLME model and how typical parameters, random effects, residual error, and covariates are estimated together from concentration-time data.