1. What Is Interindividual Variability?
Interindividual variability (IIV) describes differences in pharmacokinetic parameters between individuals in a population. Two patients receiving the same dose can have different concentrations because their clearance, volume of distribution, absorption, or other PK characteristics are not identical.
In a traditional PK analysis, it is common to summarize a study using a single set of parameter estimates. Population PK models go further by explicitly recognizing that individuals can deviate from the population's typical PK behavior.
A population PK model describes typical PK behavior while allowing individual parameters to vary around the population values.
2. Why Do Individuals Have Different PK?
PK parameters can differ substantially between individuals because drug disposition depends on biological, demographic, clinical, and treatment-related factors.
| Source of variability | Possible PK consequence | Example |
|---|---|---|
| Body size | Differences in clearance or volume | Body weight, body surface area, or other size descriptors |
| Organ function | Altered elimination | Renal or hepatic impairment |
| Age | Changes in disposition or physiology | Age-associated changes in clearance or body composition |
| Genetics | Differences in metabolic or transporter activity | Variation in drug-metabolizing enzymes |
| Concomitant medications | Changes in clearance or bioavailability | Enzyme induction or inhibition |
| Disease state | Changes in absorption, distribution, or elimination | Critical illness or organ dysfunction |
| Unmeasured factors | Remaining unexplained variability | Biological differences not represented by measured covariates |
Some of these factors can be measured and incorporated into a model as covariates. Other differences remain unexplained and are represented statistically as between-subject variability.
3. From Individual Parameters to Population Parameters
Suppose a one-compartment model is used and clearance differs between individuals. A population PK analysis might estimate a typical clearance, often denoted by \(CL_{\mathrm{pop}}\), rather than assuming that every individual has exactly that value.
The population parameter represents the central tendency of the population under the chosen model. Individual parameters then vary around that typical value.
Here, \(CL_i\) is the clearance for individual \(i\), \(CL_{\mathrm{pop}}\) is the typical population clearance, and \(\eta_{CL,i}\) represents that individual's deviation from the typical value on the log scale.
This formulation is commonly used because it guarantees a positive individual clearance when \(CL_{\mathrm{pop}}>0\).
4. What Is a Random Effect?
In population PK models, the individual deviation from a typical parameter is commonly represented by a random effect. For example, \(\eta_{CL,i}\) describes how individual \(i\)'s clearance differs from the population value.
A simple model assumes:
The mean of the random effect is zero. This means that, on the log scale, positive and negative deviations are centered around the typical population parameter.
The variance \(\omega^2_{CL}\) quantifies the amount of between-subject variability in clearance. Its square root, \(\omega_{CL}\), is the standard deviation of the random effect.
| Quantity | Meaning |
|---|---|
| \(CL_{\mathrm{pop}}\) | Typical population clearance |
| \(\eta_{CL,i}\) | Individual deviation from typical clearance |
| \(\omega^2_{CL}\) | Variance of the clearance random effect |
| \(\omega_{CL}\) | Standard deviation of the random effect |
| \(CL_i\) | Individual clearance |
5. Why Is Exponential Variability So Common?
PK parameters such as clearance and volume of distribution must generally be positive. A common population PK formulation therefore models variability exponentially.
If \(\eta_{CL,i}=0\), the individual has the typical population clearance:
If \(\eta_{CL,i}>0\), the individual's clearance is above the typical value. If \(\eta_{CL,i}<0\), it is below the typical value.
The exponential formulation also makes the random effect multiplicative rather than additive. This is often biologically and statistically useful because a given proportional difference is represented similarly across the range of positive parameter values.
6. From \(\omega\) to Between-Subject Variability
The magnitude of the random-effect variance is often summarized as a coefficient of variation (CV) for a log-normal parameter distribution.
If:
then the exact CV of the corresponding log-normal distribution is:
When \(\omega\) is relatively small, a useful approximation is:
when both quantities are expressed as proportions rather than percentages.
| Random-effect SD \(\omega\) | Approximate CV | Interpretation |
|---|---|---|
| 0.10 | ≈ 10% | Relatively modest variability |
| 0.30 | ≈ 31% | Moderate variability |
| 0.50 | ≈ 53% | Substantial variability |
| 0.70 | ≈ 75% | Large variability |
The exact conversion should be used when reporting a CV from a log-normal random-effect variance rather than relying on the small-\(\omega\) approximation.
7. Variability in Clearance and Volume
Interindividual variability can be modeled on multiple PK parameters simultaneously. For example:
These equations allow each individual to have their own clearance and volume while retaining population-level estimates for the typical values.
Importantly, the random effects need not be independent. An individual with unusually high clearance may also tend to have an unusually high or low volume of distribution. Such relationships can be represented through the omega covariance matrix.
where \(\Omega\) contains the variances and covariances of the individual random effects.
8. Covariates Explain Part of the Variability
A population PK model can use measured patient characteristics, called covariates, to explain systematic differences in PK.
For example, clearance might depend on body weight:
Here, \(WT_i\) is the individual's weight, \(WT_{\mathrm{ref}}\) is a reference weight, and \(\theta_{WT}\) describes how clearance changes with body size.
The random effect remains in the model because weight may explain only part of the differences between individuals. Other biological factors can remain unexplained.
Covariates can explain systematic differences between individuals, while remaining unexplained differences can still be represented by random effects.
9. Interindividual Variability vs. Residual Variability
One of the most important concepts in population PK is distinguishing between-subject variability from residual unexplained variability.
| Type | What varies? | Typical representation |
|---|---|---|
| Interindividual variability | PK parameters between individuals | Random effects such as \(\eta_{CL}\) and \(\eta_V\) |
| Residual variability | Observed concentrations around model predictions | An observation/error model |
| Interoccasion variability | PK parameters for the same individual across occasions | Occasion-specific random effects |
For example, two patients may have different clearances. That difference is IIV. Even after accounting for those patient-specific clearances, an observed concentration may not fall exactly on the predicted concentration. That remaining discrepancy is residual variability.
10. A Simple Population PK Model
Consider a one-compartment IV bolus model with clearance as the only variable PK parameter.
For individual \(i\):
Now allow clearance to vary between individuals:
Substituting gives:
The population model therefore generates a family of possible concentration-time profiles rather than one single curve.
11. Worked Example: Two Individuals With Different Clearance
Consider a hypothetical IV bolus dose of 500 mg. Suppose the typical population parameters are:
- \(CL_{\mathrm{pop}}=5\) L/h
- \(V=25\) L
- \(\omega_{CL}=0.50\)
Assume two individuals have random effects of \(\eta_{CL,1}=-0.50\) and \(\eta_{CL,2}=+0.50\).
Step 1: Individual clearance
The two individuals therefore have substantially different clearances even though they belong to the same modeled population.
Step 2: Individual elimination rate constants
Step 3: Initial concentration
Because the volume is the same in this example:
Step 4: Concentration after 5 hours
Step 5: Exposure
The same dose therefore produces very different exposure profiles when clearance differs between individuals.
12. What Does the Omega Matrix Represent?
When several PK parameters have random effects, their variability and relationships can be represented in a covariance matrix.
For clearance and volume:
The diagonal elements represent the variances of the individual random effects. The off-diagonal element represents their covariance.
The corresponding correlation can be written as:
| Element | Meaning |
|---|---|
| \(\omega^2_{CL}\) | Between-subject variance in clearance |
| \(\omega^2_V\) | Between-subject variance in volume |
| \(\omega_{CL,V}\) | Covariance between clearance and volume random effects |
| \(\rho_{CL,V}\) | Correlation between clearance and volume random effects |
Whether covariance terms should be estimated depends on the information available in the data and the complexity that can be supported by the study design.
13. What Is Shrinkage?
Population PK models can produce individual-level estimates of random effects, often called empirical Bayes estimates or individual random-effect estimates. These estimates combine the population model with the individual's observed concentration data.
When an individual has little information in their own observations, the estimated random effect may be pulled toward the population mean of zero. This phenomenon is known as shrinkage.
For example, a patient with only one sparse concentration measurement may not provide enough information to estimate a precise individual clearance. The model therefore relies more heavily on the population distribution.
14. Why Sampling Design Matters for IIV
The ability to estimate interindividual variability depends strongly on the information contained in the concentration data.
Rich sampling can provide substantial information about each individual's concentration-time profile. Sparse sampling can still support population PK analysis when data are collected across many individuals and the sampling design is informative, but individual parameters may be estimated less precisely.
| Sampling situation | Potential consequence |
|---|---|
| Rich individual sampling | More information about individual PK parameters |
| Sparse sampling | Greater reliance on the population model |
| Samples concentrated in one phase | Limited information about other PK processes |
| Very similar sampling times for everyone | Reduced information about some components of variability |
| Large number of individuals | More information about the population distribution |
Consequently, population PK is not simply a statistical technique applied after data collection. The sampling design affects what the model can identify and how precisely variability can be estimated.
15. From Unexplained Variability to Covariate Relationships
A useful population PK model often attempts to explain some of the observed variability using clinically meaningful covariates.
For example, suppose renal function is related to clearance:
The covariate component describes a systematic relationship, while \(\eta_{CL,i}\) captures remaining between-subject differences not explained by the covariate model.
This distinction is important. A covariate does not have to explain all variability to be scientifically useful. A model may substantially reduce unexplained variability while still retaining a nonzero IIV term.
16. What Does a Large IIV Estimate Mean?
A large estimate of between-subject variability means that individuals differ substantially in the corresponding PK parameter under the fitted model.
It does not automatically mean that:
- the model is wrong;
- the drug is unsafe;
- a specific covariate must explain the variability;
- the population is biologically heterogeneous for only one reason; or
- individual parameter estimates are precise.
A large IIV estimate can reflect genuine biological differences, omitted covariates, model misspecification, limited data, or difficulty separating different sources of variability.
Interpretation therefore requires examining the full model, including diagnostics, sampling design, covariate relationships, parameter uncertainty, and the residual error model.
17. IIV, Interoccasion Variability, and Residual Error
Several types of variability are easy to confuse.
| Concept | Unit of variation | Question being modeled |
|---|---|---|
| IIV | Between individuals | Why do different people have different PK? |
| IOV | Between occasions within an individual | Why can the same person's PK differ from one occasion to another? |
| Residual error | Between observations and predictions | Why does an observed concentration differ from the model prediction? |
These sources of variability can coexist. For example, an individual can have unusually high clearance compared with the population, that same individual's clearance can change somewhat between treatment occasions, and individual measured concentrations can still differ from the model predictions because of residual error.
18. A Practical Workflow for Modeling IIV
- Specify the structural PK model. Decide whether the data support one compartment, two compartments, absorption processes, and other structural components.
- Estimate typical population parameters. Establish the population-level PK behavior.
- Introduce plausible IIV terms. Allow important PK parameters to vary between individuals.
- Specify the random-effect distribution. For example, use exponential variability for positive parameters.
- Evaluate the magnitude of IIV. Examine whether the estimated variability is supported by the data.
- Investigate covariates. Determine whether clinically meaningful patient characteristics explain systematic differences in PK.
- Evaluate residual variability. Make sure residual error is not being used to compensate for an inadequate structural or IIV model.
- Assess diagnostics and uncertainty. Examine parameter precision, diagnostic plots, and model behavior.
- Interpret individual predictions cautiously. Consider the amount of individual information and the possibility of shrinkage.
- Use the final model for prediction or simulation. Distinguish typical population predictions from individual predictions and population variability.
19. Key Takeaways
- Interindividual variability describes differences in PK parameters between individuals.
- A population PK model separates typical population PK behavior from individual deviations around those typical values.
- Random effects such as \(\eta_{CL}\) represent individual deviations from population parameters.
- Exponential random-effects models are commonly used for positive PK parameters such as clearance and volume.
- The omega matrix describes between-subject variances and covariances among random effects.
- Covariates can explain systematic components of interindividual variability, while unexplained differences can remain in the random-effect distribution.
- Interindividual variability is distinct from residual unexplained variability in observed concentrations.
- Interoccasion variability describes changes within the same individual across occasions and is conceptually different from IIV.
- Sparse sampling can increase reliance on the population model and can limit the information available for individual parameter estimation.
- Shrinkage can pull individual random-effect estimates toward the population mean when individual data are weakly informative.
- A large IIV estimate does not by itself identify the biological cause of variability.
- The goal of population PK modeling is not to eliminate variability but to characterize it, explain important components of it, and use it appropriately for prediction and simulation.
Where to Go Next
A natural progression is to study fixed effects and random effects in population PK, followed by covariate modeling, residual error models, empirical Bayes estimation, shrinkage, model diagnostics, and nonlinear mixed-effects estimation.
The next tutorial can build directly on the ideas introduced here by showing how population PK models formally combine fixed effects, random effects, and residual error to describe both typical PK behavior and individual differences.