1. What Is Interoccasion Variability?
Interoccasion variability (IOV) describes changes in a patient's pharmacokinetic parameters between different occasions, periods, or study visits. The key idea is that the same individual does not necessarily have exactly the same clearance, volume, or other PK parameter on every occasion.
In a population PK model, variability is often introduced at more than one level. Interindividual variability (IIV) describes persistent differences between individuals, whereas IOV describes deviations that occur within the same individual from one occasion to another.
Population PK can represent variability hierarchically: individuals differ from one another, and the same individual can also differ across occasions.
2. Why Does Interoccasion Variability Matter?
If repeated PK observations are available for the same patient, they may show that the patient's clearance or other parameters are not completely stable over time. A model that assumes one fixed individual parameter for all occasions may then attribute some of that structured variation incorrectly to residual error.
IOV can be particularly relevant when physiological state, disease status, concomitant treatment, adherence, organ function, formulation, or other time-varying factors can change drug disposition.
| Source of variation | Question answered | Typical interpretation |
|---|---|---|
| Interindividual variability | Why do different patients have different PK? | Persistent differences between subjects |
| Interoccasion variability | Why can the same patient have different PK on different occasions? | Within-subject changes between occasions |
| Residual variability | Why does an observation differ from the model prediction at a particular time? | Measurement error, model misspecification, and other unexplained observation-level variation |
Separating these components can make the model more realistic when the study design contains repeated occasions with enough information to estimate the additional variability.
3. What Is an Occasion?
An occasion is a predefined period over which a patient's PK parameters are assumed to be approximately stable, while parameters are allowed to differ between occasions.
The precise definition depends on the study. An occasion might correspond to a dosing period, treatment cycle, study visit, day, or another scientifically meaningful interval.
For example, suppose a patient is studied during three treatment cycles. If clearance is assumed to be relatively stable within each cycle but potentially different across cycles, each cycle could be represented as an occasion.
| Patient | Occasion | Example interpretation |
|---|---|---|
| Patient 101 | 1 | Cycle 1 PK |
| Patient 101 | 2 | Cycle 2 PK |
| Patient 101 | 3 | Cycle 3 PK |
4. Interindividual Variability vs. Interoccasion Variability
The distinction between IIV and IOV is fundamental to population PK.
Interindividual variability
IIV represents differences among patients. For example, one patient may have a typical clearance of 6 L/h while another has a clearance of 9 L/h.
Interoccasion variability
IOV represents changes within a patient. The same patient might have clearance of approximately 6 L/h on one occasion and 8 L/h on another.
The equation above is conceptual. The actual model is usually formulated on a transformed scale, most commonly the logarithmic scale for positive PK parameters.
| Feature | IIV | IOV |
|---|---|---|
| Level | Between subjects | Within subject |
| Changes across occasions? | Usually no | Yes |
| Example | Patient A vs. Patient B | Patient A, Occasion 1 vs. Occasion 2 |
| Primary source of information | Variation across subjects | Repeated observations within subjects |
5. How Is IOV Incorporated Into a Population PK Model?
Because clearance and other PK parameters are positive, a common approach is to use an exponential model.
For clearance, a model with IIV alone might be written as:
where \(CL_{ij}\) is the clearance for subject \(i\), \(CL_{pop}\) is the typical population clearance, and \(\eta_i\) represents the subject-specific deviation.
When IOV is included, an occasion-specific random effect can be added:
Here, \(\kappa_{ij}\) represents the IOV for subject \(i\) during occasion \(j\).
6. What Do the Variance Terms Mean?
Suppose the random effects have mean zero and variances:
Then \(\omega^2\) represents the variance associated with IIV and \(\pi^2\) represents the variance associated with IOV on the modeled random-effect scale.
For the exponential model, the corresponding coefficients of variation can be expressed approximately as:
This provides a useful way to translate random-effect variance into a more interpretable measure of relative variability.
7. Which PK Parameters Can Have IOV?
IOV can theoretically be associated with any PK parameter for which meaningful within-subject changes are plausible and estimable from the data.
| Parameter | Possible source of occasion-to-occasion change |
|---|---|
| CL | Changes in organ function, disease status, concomitant medications, enzyme activity, or other time-varying factors |
| V | Changes in fluid status, body composition, disease state, or other distribution-related factors |
| F | Changes in absorption or first-pass processes |
| ka | Changes in gastrointestinal conditions or formulation-related processes |
| Other model parameters | Depends on the structural model and scientific context |
In practice, IOV should not automatically be added to every parameter. The parameter should have a plausible scientific mechanism for within-subject variation, and the dataset must contain enough repeated information to estimate that variation.
8. Can IIV and IOV Be Correlated?
Population PK models can include covariance structures among random effects when supported by the data and scientific rationale.
For example, the model might allow the IIV in clearance and volume to be correlated. Similarly, more elaborate models can consider relationships among occasion-specific effects.
However, correlation structures add parameters and therefore require sufficient information.
9. Why Repeated Occasions Are Important
IOV is fundamentally a within-subject concept. Therefore, repeated observations from the same individuals across multiple occasions are important for distinguishing IOV from IIV.
Repeated occasions allow the model to distinguish persistent subject-level differences from occasion-specific changes.
With only one occasion per subject, there is generally no direct within-subject replication from which to identify a separate IOV component. Repeated occasions therefore provide critical information for estimating IOV.
10. IOV Is Not the Same as Residual Variability
A concentration observation can differ from a model prediction for several reasons. Residual variability operates at the observation level, whereas IOV changes the underlying PK parameter for an entire occasion.
| Component | Example | Level |
|---|---|---|
| IOV | Clearance is higher during Cycle 2 than Cycle 1 | Occasion / parameter level |
| Residual error | An assay result is slightly above the model prediction | Observation level |
If IOV is ignored when it is genuinely present, the residual error model may have to absorb some of the systematic within-subject variation. Conversely, introducing IOV when the data do not support it can produce an unnecessarily complex or poorly identified model.
11. IOV and Time-Varying Covariates
A particularly important modeling question is whether an apparent occasion effect can be explained by a measurable covariate.
Suppose clearance changes between occasions because renal function changes. If renal function is measured, a covariate model may explain part of the apparent variability:
In this example, the time-varying covariate explains a predictable component of clearance, while the remaining random effects represent unexplained variability.
12. How Should Occasions Be Defined?
The definition of an occasion should reflect the scientific and operational structure of the study.
- Identify when PK could plausibly change. Consider treatment cycles, disease progression, organ function, concomitant therapy, or other relevant factors.
- Define a period of relative stability. Observations within an occasion should reasonably share the same underlying PK parameters.
- Use the same occasion definition consistently. The model should have a clear and reproducible rule for assigning observations to occasions.
- Avoid arbitrary fragmentation. Creating many short occasions can increase model complexity without adding identifiable information.
- Check the data structure. There must be enough observations per subject and enough repeated occasions to support estimation.
Occasion definitions are therefore part of model specification rather than merely a formatting choice in the dataset.
13. Worked Example: Clearance With IIV and IOV
Consider a population PK model with a typical clearance of 8 L/h. Suppose the estimated IIV and IOV are represented by random effects on the exponential scale.
For one patient, suppose:
- Population clearance: \(CL_{pop}=8\) L/h
- Subject-specific effect: \(\eta_i=0.10\)
- Occasion 1 effect: \(\kappa_{i1}=-0.05\)
- Occasion 2 effect: \(\kappa_{i2}=0.15\)
Step 1: Clearance on Occasion 1
Step 2: Clearance on Occasion 2
Step 3: Interpret the difference
The patient has the same subject-level effect \(\eta_i\) on both occasions, but the occasion-specific effect changes. Consequently, the model predicts higher clearance on Occasion 2.
14. How Is IOV Estimated?
IOV is estimated as part of the population PK model. The estimation procedure uses the repeated concentration-time observations to determine how much variability is better represented by subject-level effects, occasion-level effects, and residual variability.
As with other random effects, estimation depends on the amount and structure of information in the dataset.
Important sources of information include:
- Number of subjects.
- Number of occasions per subject.
- Number and timing of samples within each occasion.
- Variation in PK parameters across occasions.
- Covariate information.
- Strength of the structural PK model.
- Residual error specification.
When the dataset contains little repeated information, IOV can be difficult to distinguish from IIV or residual variability.
15. Why Can IOV Be Difficult to Estimate?
IOV adds an additional layer of random effects to an already hierarchical model. The model must distinguish among population variation, persistent individual differences, occasion-specific deviations, and observation-level error.
These components can compete to explain the same observed variation.
| Problem | Potential consequence |
|---|---|
| Few repeated occasions | Limited information about within-subject variability |
| Sparse sampling | Weak information about individual PK within each occasion |
| Highly variable residual error | IOV can be difficult to distinguish from observation-level noise |
| Too many random effects | Unstable or imprecise variance estimates |
| Strongly correlated parameters | Difficulty separating different sources of variability |
16. Evaluating a Model With IOV
Adding IOV should improve the scientific representation of the data without producing an unnecessarily complicated model.
Evaluation can include:
- Parameter estimates. Examine the magnitude and precision of the IOV variance.
- Model diagnostics. Assess whether predictions and residuals improve appropriately.
- Visual predictive checks. Determine whether the model reproduces important features of the observed data across occasions.
- Biological plausibility. Consider whether the estimated within-subject variability is scientifically reasonable.
- Model comparison. Compare models with and without IOV using appropriate statistical and scientific criteria.
- Covariate assessment. Determine whether measurable predictors explain some of the apparent occasion-to-occasion variability.
A lower objective function or improved diagnostic plot is not, by itself, sufficient justification for an IOV term. The additional complexity should have a meaningful interpretation and adequate support from the data.
17. Common Mistakes When Modeling IOV
Mistake 1: Treating IOV as IIV
IOV is within-subject variation across occasions. IIV is between-subject variation. They answer different questions.
Mistake 2: Treating IOV as residual error
IOV changes the underlying PK parameter for an occasion, whereas residual error operates on individual observations.
Mistake 3: Defining occasions arbitrarily
The occasion definition should have a scientific rationale and should correspond to periods during which the underlying PK parameters can reasonably be considered stable.
Mistake 4: Adding IOV to every parameter
More random effects are not automatically better. IOV should be considered where within-subject variation is plausible and estimable.
Mistake 5: Ignoring time-varying covariates
Some apparent IOV may be explained by measurable changes in patient characteristics or treatment conditions.
Mistake 6: Interpreting the IOV variance as a direct percentage
For exponential models, the random-effect variance must be transformed appropriately before expressing variability as a coefficient of variation.
18. The Three Levels of Population PK Variability
A useful conceptual framework is to separate the population PK model into three levels.
| Level | Symbol | Meaning |
|---|---|---|
| Population | \(CL_{pop}\) | Typical clearance in the population |
| Individual | \(\eta_i\) | Persistent deviation of subject \(i\) from the population |
| Occasion | \(\kappa_{ij}\) | Deviation for subject \(i\) during occasion \(j\) |
| Observation | \(\epsilon_{ij}\) | Residual discrepancy between observation and model prediction |
Conceptually, this can be written as:
This hierarchy is one of the most important ideas for understanding population PK models with IOV.
19. When Is IOV Particularly Useful?
IOV can be useful in situations where PK is expected to change within an individual over time.
- Repeated treatment cycles: PK may change between cycles while remaining relatively stable within a cycle.
- Chronic disease: disease progression or changing organ function may alter drug disposition.
- Therapeutic drug monitoring: repeated measurements may reveal within-patient changes that are relevant to dosing.
- Drug-drug interactions: starting or stopping a concomitant medication can change clearance between occasions.
- Time-varying physiology: renal or hepatic function may change over time.
- Longitudinal pharmacokinetic studies: repeated PK sampling creates an opportunity to distinguish persistent and occasion-specific variability.
Whether IOV should actually be included depends on the study question, data structure, and model diagnostics.
20. How Does IOV Affect Prediction?
Including IOV changes the way uncertainty is represented when predictions are made for future occasions.
For a future occasion, the model can recognize that a patient's clearance is not necessarily identical to the clearance estimated from a previous occasion. The occasion-specific random effect represents this additional source of uncertainty.
This can matter for simulation, exposure prediction, dose selection, and evaluation of variability in longitudinal treatment settings.
21. A Practical Workflow for Assessing IOV
- Understand the study design. Identify repeated occasions and determine what each occasion represents.
- Inspect the concentration-time data. Look for systematic within-subject changes across occasions.
- Build an appropriate base population PK model. Establish the structural model, IIV, and residual error before adding unnecessary complexity.
- Consider plausible sources of within-subject change. Include relevant time-varying covariates where appropriate.
- Introduce IOV selectively. Start with parameters for which occasion-to-occasion changes are scientifically plausible.
- Evaluate identifiability. Check whether the dataset can support the additional random-effect parameters.
- Assess diagnostics and parameter precision. Determine whether the IOV model provides a meaningful improvement.
- Interpret IOV in context. Distinguish unexplained within-subject variability from predictable covariate effects.
- Use the final model for simulation or prediction. Carry the estimated uncertainty and variability structure into downstream applications when appropriate.
22. Key Takeaways
- Interoccasion variability (IOV) describes within-subject changes in PK parameters from one occasion to another.
- Interindividual variability (IIV) describes persistent differences between subjects, whereas IOV describes changes within a subject.
- IOV is distinct from residual variability, which operates at the observation level.
- A common exponential model represents an occasion-specific parameter as a population value multiplied by subject- and occasion-level random effects.
- Repeated occasions provide the within-subject information needed to estimate IOV.
- IOV can apply to clearance, volume, bioavailability, absorption parameters, or other model parameters when scientifically justified.
- Time-varying covariates can explain part of what might otherwise appear as unexplained IOV.
- Adding IOV increases model complexity and can create identifiability problems when the dataset is sparse.
- The occasion should be defined using a scientifically meaningful rule rather than arbitrary data partitioning.
- IOV can be particularly important for longitudinal studies, repeated treatment cycles, changing disease states, and other settings where PK may change within individuals over time.
- A well-specified IOV model can provide a more realistic representation of uncertainty when predicting PK on future occasions.
Where to Go Next
A natural next step is to study fixed effects and random effects in population PK, followed by covariate modeling, residual error models, shrinkage, model diagnostics, and nonlinear mixed-effects estimation.
Once IIV and IOV are understood, the next major question is how to determine which patient characteristics explain the variability—and how those relationships are incorporated into a population PK model.