Anatomy of a Population Model

Introduce the major components of population PK/PD models including structural models, fixed effects, random effects, residual variability, and covariates.
Tip

Big picture: Population models are built from layers. Structural models describe expected behavior, while statistical models explain why observations differ.

Learning Objectives

By the end of this lesson, you will be able to:

  • Describe the major components of a population model.
  • Distinguish fixed effects from random effects.
  • Explain interindividual and residual variability.
  • Describe the role of covariates.
  • Interpret a simple population model mathematically.
  • Recognize how model components work together.

Key Ideas

  • Population models combine structure and variability.
  • Fixed effects describe typical behavior.
  • Random effects describe subject-level differences.
  • Residual error captures unexplained observations.
  • Covariates explain predictable variability.

Building a Population Model

Population models are built by combining multiple layers that each explain a different aspect of the data.

Conceptually:

Observed Data
│
├── Structural Model
│     ├── CL
│     ├── V
│     └── ka
│
├── Fixed Effects (θ)
│
├── Random Effects (η)
│
├── Residual Error (ε)
│
└── Covariates
      ├── WT
      ├── AGE
      └── SEX

Each component answers a different scientific question.

Component Scientific Question
Structural model What biological pattern is expected?
Fixed effects What is typical in the population?
Random effects How do subjects differ?
Residual error Why do observations still differ from predictions?
Covariates Can some variability be explained?

Rather than thinking of a population model as one large equation, it is often more helpful to think of it as a system of connected components that together generate the observed data.

Throughout the course, we will gradually build these pieces until they become explicit parts of an nlmixr2 model specification.

Worked Example 1: Structural Model

A simple oral one-compartment model:

\[ C(t)=\frac{Dose\times ka}{V(ka-k)} \left(e^{-kt}-e^{-ka t}\right) \]

where:

\[ k=\frac{CL}{V} \]

Interpretation:

  • ka controls absorption
  • CL controls elimination
  • V controls distribution

Fixed Effects

Fixed effects represent typical parameter values.

Examples:

\[ CL_{typical} \]

\[ V_{typical} \]

\[ ka_{typical} \]

They answer:

What is typical?

Worked Example 2: Random Effects

A common population formulation:

\[ CL_i=CL_{typical}\times e^{\eta_i} \]

where:

  • \(CL_i\) = individual clearance
  • \(\eta\) = subject deviation

Interpretation:

Typical + Variability = Individual

Residual Error

Observation model:

\[ DV=PRED+\varepsilon \]

Residual error includes:

  • assay noise
  • timing mismatch
  • model simplification

Worked Example 3: Covariates

Example:

\[ CL_i= CL_{typical} \left(\frac{WT_i}{70}\right)^{0.75} e^{\eta_i} \]

Covariates explain predictable variability.

Examples:

  • weight
  • age
  • sex
  • renal function

Putting the Pieces Together

A population model can also be viewed as a sequence of layers that gradually transform a dosing event into observed data.

Dose
↓
Structural Model
f(time, θ)
↓
Add Random Effects (η)
↓
Individual Prediction
↓
Add Residual Error (ε)
↓
Observed Data

This process is often summarized mathematically as:

\[ DV = f(time,\theta,\eta) + \varepsilon \]

where:

  • \(DV\) = observed value, such as observed concentration
  • \(f(\cdot)\) = structural model describing expected behavior
  • \(\theta\) = fixed effects, or typical population parameters
  • \(\eta\) = random effects, or individual variability
  • \(\varepsilon\) = residual error, or remaining unexplained deviation

This equation captures the core philosophy of population modeling:

Expected Biology + Variability + Unexplained Noise = Observed Data

As the course progresses, each component of this equation will become an explicit part of the nlmixr2 model specification.

Strategies

  • Interpret structure first.
  • Add variability second.
  • Add covariates last.

Common Mistakes

  • Overcomplicating models.
  • Ignoring uncertainty.
  • Treating residual error as biology.

Practice Problems

  1. Explain fixed effects.
  2. Explain random effects.
  3. Give residual error examples.
  4. Explain covariates.

Fixed effects describe typical behavior.

Random effects describe unexplained subject differences.

Residual error captures unexplained observations.

Covariates explain predictable variability.

Summary

Population models combine:

  • structure
  • fixed effects
  • variability
  • residual error
  • covariates
  • Start simple.
  • Separate structure from variability.
  • Ask what each component means scientifically.