Tutorials › Pharmacometrics › Introduction to Model-Informed Drug Development
Pharmacometrics · Drug Development

Introduction to Model-Informed Drug Development

Learn how quantitative models integrate pharmacokinetics, pharmacodynamics, disease biology, clinical trial data, and prior knowledge to support drug development decisions from first-in-human studies through late-stage development.

Beginner MIDD Foundations Pharmacometrics Drug Development
01 · The big picture

1. What Is Model-Informed Drug Development?

Model-informed drug development (MIDD) is an approach to drug development in which mathematical and quantitative models are used to integrate information and support decisions throughout the development process.

The central idea is straightforward: drug development produces many different types of information—drug concentrations, pharmacodynamic responses, disease measurements, clinical outcomes, prior studies, and information about patient characteristics. Models provide a structured way to connect these sources of information.

Data PK · PD · clinical Models mechanisms · variability Decisions dose · design · strategy learning from predictions and new observations

MIDD creates a quantitative feedback loop: data inform models, models generate predictions, and predictions can inform development decisions and future data collection.

Core idea: MIDD is not a single model or software package. It is a development strategy that uses quantitative models to integrate evidence, make predictions, evaluate uncertainty, and inform decisions.
02 · Why models matter

2. Why Use Models in Drug Development?

Drug development is fundamentally an information problem. Decisions often need to be made before all of the desired clinical data are available. A model can help organize what is already known and quantify what can reasonably be inferred about conditions that have not yet been directly observed.

Development question Model-informed contribution Example use
What dose should be studied? Predict exposure and response over a range of doses Dose selection
How does exposure vary between patients? Characterize between-subject variability and covariates Population PK
What exposure is associated with efficacy? Quantify exposure-response relationships Exposure-response analysis
What exposure is associated with toxicity? Relate exposure to safety endpoints Exposure-safety analysis
How should a clinical trial be designed? Simulate alternative designs and assumptions Trial design optimization
What happens under an untested scenario? Generate model-based predictions Scenario simulation

The value of MIDD therefore comes from connecting quantitative evidence to a specific development question. A model should not be built simply because a model is available; it should have a clearly defined purpose.

03 · The MIDD toolbox

3. What Models Are Used in MIDD?

MIDD encompasses a family of related modeling approaches rather than one universal methodology. Different models answer different questions.

Modeling approach Primary role
Population PK Characterize typical PK, between-subject variability, residual variability, and covariate effects.
PK/PD modeling Connect drug exposure to pharmacologic response.
Exposure-response modeling Quantify relationships between exposure and efficacy or safety outcomes.
Disease progression models Represent changes in disease biomarkers or clinical measures over time.
Quantitative systems pharmacology Represent interacting biological mechanisms and pathways.
Physiologically based PK Use physiological and drug-specific information to predict concentration-time behavior.
Trial simulation Explore the operating characteristics and expected outcomes of alternative development strategies.

These approaches can also be combined. For example, a population PK model can provide individual exposure estimates that are subsequently related to efficacy or safety endpoints in an exposure-response model.

04 · PK foundation

4. Pharmacokinetics as a Foundation for MIDD

Pharmacokinetics (PK) describes the time course of drug concentrations in the body. It provides the quantitative link between dose and systemic exposure.

A simple one-compartment IV model illustrates the idea. For a bolus dose \(D\), volume of distribution \(V\), and clearance \(CL\):

$$ C(t)=\frac{D}{V}e^{-(CL/V)t} $$

Although real development programs often require considerably more sophisticated models, this simple equation captures the central concept: a model translates dose and drug properties into a predicted concentration-time trajectory.

MIDD builds on this foundation by connecting the concentration trajectory to additional information such as pharmacodynamic response, patient characteristics, disease progression, or clinical outcomes.

05 · Exposure and response

5. From PK to Pharmacodynamics

Pharmacodynamics (PD) describes the relationship between drug exposure and pharmacologic effect. A PK/PD model combines the two.

$$ \text{Dose} \rightarrow \text{PK} \rightarrow C(t) \rightarrow \text{PD} \rightarrow E(t) $$

For example, a simple maximum-effect model can be written as:

$$ E(C)=E_0+\frac{E_{\max}C}{EC_{50}+C} $$

Here, \(E_0\) represents baseline response, \(E_{\max}\) represents the maximum drug-associated effect in the model, and \(EC_{50}\) represents the concentration associated with half of the maximum drug-associated effect.

More complex models can incorporate indirect effects, delays between concentration and effect, tolerance, biomarkers, disease progression, or multiple interacting mechanisms.

Key distinction: PK asks what the body does to the drug, while PD asks what the drug does to the system. MIDD frequently connects both.
06 · Variability

6. Why Population Models Matter

Patients are not identical. Drug clearance, volume of distribution, absorption, and pharmacodynamic response can differ substantially between individuals.

A population PK model explicitly represents this variability. A simplified individual parameter model can be written as:

$$ CL_i=CL_{\mathrm{pop}}e^{\eta_i} $$

where \(CL_{\mathrm{pop}}\) is the typical population clearance and \(\eta_i\) represents the deviation of individual \(i\)'s clearance from the typical value.

Covariates can then be incorporated to explain some of the systematic variability. For example:

$$ CL_i=CL_{\mathrm{pop}} \left(\frac{WT_i}{70}\right)^{\theta_{WT}} e^{\eta_i} $$

This type of model can help distinguish predictable variability associated with patient characteristics from unexplained between-subject variability.

Source of variability Typical interpretation
Between-subject variability Differences among individuals in model parameters
Residual variability Difference between observations and model-predicted concentrations or responses
Covariate effects Systematic relationships between patient characteristics and model parameters
07 · Exposure-response

7. Linking Exposure to Efficacy and Safety

One of the central questions in development is whether differences in drug exposure are associated with differences in clinical response.

Exposure-response analysis can examine relationships such as:

  • Exposure and biomarker response.
  • Exposure and probability of clinical response.
  • Exposure and time-to-event outcomes.
  • Exposure and adverse-event risk.
  • Exposure and laboratory abnormalities.

For a binary efficacy endpoint, a logistic exposure-response model might take the form:

$$ \operatorname{logit}(P_i) = \alpha+\beta E_i $$

where \(E_i\) represents an exposure measure and \(P_i\) represents the modeled probability of response.

The exposure measure might be AUC, average concentration, trough concentration, peak concentration, or another model-derived quantity. The appropriate exposure metric depends on the mechanism and scientific question.

Important: an observed exposure-response association is not automatically proof that exposure causes the clinical outcome. Study design, confounding, disease severity, and model assumptions all affect interpretation.
08 · Disease models

8. Incorporating Disease Progression

Many development questions cannot be answered by a drug model alone. The underlying disease may change over time, even in the absence of treatment.

A disease progression model can separate the natural trajectory of a disease from the treatment effect.

A simple linear disease progression model might be represented as:

$$ Y(t)=Y_0+kt $$

A treatment effect can then be added to the model. For example:

$$ Y(t)=Y_0+kt-\text{TreatmentEffect}(t) $$

Real disease models can be considerably more sophisticated. They may incorporate turnover processes, delayed treatment effects, multiple biomarkers, disease states, or feedback mechanisms.

Disease models are especially useful when the clinical endpoint changes slowly and when the treatment effect needs to be distinguished from the underlying trajectory of disease.

09 · Biological mechanisms

9. Quantitative Systems Pharmacology

Quantitative systems pharmacology (QSP) extends the modeling framework toward biological mechanisms. Rather than representing only the relationship between concentration and response, QSP models can represent interacting biological pathways, feedback loops, cell populations, biomarkers, and disease mechanisms.

Drug exposure Target engagement Pathway biology Disease outcome feedback and biological interactions

A conceptual QSP model can connect drug exposure to targets, biological pathways, and disease-level outcomes while representing interactions and feedback.

QSP is therefore particularly useful when the development question depends on mechanistic understanding rather than only empirical exposure-response relationships.

10 · Physiology

10. Physiologically Based Pharmacokinetic Modeling

Physiologically based pharmacokinetic (PBPK) models represent drug disposition using physiological characteristics of the organism together with drug-specific properties.

Instead of treating the body only as abstract compartments, a PBPK model can represent organs and tissues with physiological quantities such as blood flow, organ volume, tissue composition, and partitioning behavior.

Feature Traditional compartment model PBPK model
Representation Abstract kinetic compartments Physiologically defined organs and tissues
Parameters Estimated kinetic parameters such as CL and V Physiological and drug-specific parameters
Primary strength Efficient description of observed PK Mechanistic extrapolation across conditions
Typical applications Population PK, dose characterization, exposure analysis DDI, special populations, species translation, mechanistic extrapolation

PBPK models are one example of how MIDD can incorporate prior biological and physiological knowledge rather than relying entirely on parameters estimated from a single clinical dataset.

11 · Simulation

11. Why Simulation Is Central to MIDD

A fitted model describes what has been learned from available data. Simulation uses that model to ask what might happen under alternative conditions.

For example, a model can simulate concentration-time profiles for several candidate doses:

$$ C(t\mid D,CL,V) = \frac{D}{V}e^{-(CL/V)t} $$

Repeated simulation can then quantify the expected distribution of outcomes rather than producing only a single deterministic prediction.

Model parameters Simulate many scenarios Compare development options dose · population · trial design · endpoint · uncertainty

Simulation allows a model to be used prospectively to compare alternative development scenarios before all of those scenarios are tested experimentally.

12 · Trial design

12. Using Models to Inform Clinical Trial Design

MIDD can be used before a clinical trial begins to explore how different design choices may affect the information generated by the study.

Potential questions include:

  • What dose levels should be evaluated?
  • What sampling schedule provides useful PK information?
  • How many subjects are needed to characterize exposure?
  • What range of exposures is expected in the target population?
  • How likely is a proposed study to distinguish competing hypotheses?
  • How might dropout, variability, or adherence affect the study?

Trial simulation can combine a PK/PD model with assumptions about patient variability, treatment allocation, endpoint distributions, dropout, and other features of a proposed trial.

Simulation does not replace a clinical trial. It evaluates the consequences of assumptions before the trial is conducted and can help identify designs that are poorly informative under plausible scenarios.
13 · Dose selection

13. Model-Informed Dose Selection

Dose selection is one of the most visible applications of MIDD. The objective is often to identify doses that produce exposures likely to provide adequate efficacy while maintaining an acceptable safety profile.

Conceptually, the process can be represented as:

$$ \text{Dose} \rightarrow \text{Exposure} \rightarrow \text{Efficacy} $$

alongside:

$$ \text{Dose} \rightarrow \text{Exposure} \rightarrow \text{Safety} $$

A development program may therefore evaluate several candidate doses using different sources of evidence, including PK, PD biomarkers, efficacy endpoints, safety observations, and prior studies.

The resulting decision is inherently quantitative but is also dependent on the quality and relevance of the underlying data and assumptions.

14 · Worked example

14. Worked Example: Comparing Two Candidate Doses

Consider a hypothetical drug described by a simple one-compartment model with:

  • Clearance \(CL=5\) L/h
  • Volume of distribution \(V=25\) L
  • Linear PK
  • Two candidate IV doses: 250 mg and 500 mg

Step 1: Elimination rate constant

$$ k=\frac{CL}{V} = \frac{5}{25} = 0.20\text{ h}^{-1} $$

Step 2: Initial concentration after 250 mg

$$ C_0=\frac{250}{25}=10\text{ mg/L} $$

Step 3: Initial concentration after 500 mg

$$ C_0=\frac{500}{25}=20\text{ mg/L} $$

Step 4: AUC for each dose

$$ AUC_{0-\infty} = \frac{D}{CL} $$

Therefore:

$$ AUC_{250} = \frac{250}{5} = 50\text{ mg·h/L} $$
$$ AUC_{500} = \frac{500}{5} = 100\text{ mg·h/L} $$

Under the assumed linear model, doubling the dose doubles both the initial concentration and AUC.

This simple example illustrates the role of a model in dose selection: rather than considering dose as an isolated quantity, MIDD connects dose to predicted exposure and, in a broader model, potentially to efficacy and safety.

15 · Uncertainty

15. Models Are Predictions With Uncertainty

An important feature of MIDD is that model-based predictions are not exact statements about the future. They depend on estimated parameters, variability, data quality, and assumptions.

Suppose a model predicts an exposure measure \(E\). Rather than considering only a single value, simulation can generate a distribution:

$$ E_1,E_2,\ldots,E_N $$

The resulting distribution can be summarized using quantities such as medians, percentiles, prediction intervals, or probabilities of exceeding a target.

Source of uncertainty Example
Parameter uncertainty Uncertainty in estimated clearance or treatment-effect parameters
Between-subject variability Patients differ in PK or response
Residual variability Observed measurements vary around model predictions
Model uncertainty Alternative structural models may provide different predictions
Scenario uncertainty Future populations or treatment conditions differ from development data

Good MIDD practice therefore requires attention not only to the model's central prediction but also to the uncertainty surrounding that prediction.

16 · Decision framework

16. From Model to Development Decision

The final purpose of MIDD is not the model itself. The model is a tool for supporting a development decision.

A useful conceptual workflow is:

  1. Define the decision. What development question must be answered?
  2. Identify the evidence. What PK, PD, clinical, physiological, and prior information is available?
  3. Build an appropriate model. Represent the processes relevant to the question.
  4. Estimate and evaluate. Quantify parameters and evaluate model adequacy.
  5. Characterize uncertainty. Determine which assumptions or parameters drive the prediction.
  6. Simulate relevant scenarios. Examine alternative doses, populations, trial designs, or assumptions.
  7. Interpret the results. Connect model outputs to the original scientific question.
  8. Make the development decision. Use model-based evidence alongside other available evidence.
The model is part of the evidence, not the entire evidence base. MIDD is most useful when quantitative model predictions are integrated with clinical, pharmacological, biological, and safety information.
17 · Across the development lifecycle

17. Where Does MIDD Fit in Drug Development?

MIDD can contribute at multiple stages of development. The specific modeling question changes as the evidence base becomes richer.

Development stage Potential model-informed questions
Preclinical How might animal or mechanistic information translate into human exposure?
First-in-human What doses and sampling strategies are appropriate for initial clinical studies?
Early clinical development What exposure is associated with pharmacodynamic activity and tolerability?
Proof of concept What dose range and exposure-response relationship support further development?
Phase III How should dose, population, endpoints, and trial design be informed by accumulated evidence?
Regulatory applications How can integrated quantitative evidence support dosing and other development conclusions?
Post-approval How should dosing or exposure be considered in additional populations, indications, or scenarios?

The same general principle applies throughout: the model should address a specific decision using the evidence available at that point in development.

18 · Interpretation

18. Common Misconceptions About MIDD

Because MIDD combines mathematics, pharmacology, statistics, and clinical development, several misconceptions are common.

  • MIDD is not synonymous with population PK. Population PK is one important component of MIDD, but MIDD encompasses a much broader range of models and applications.
  • MIDD is not simply computer simulation. Simulation is one activity within a broader modeling-and-decision framework.
  • A complex model is not automatically a better model. Additional complexity is useful only when it is supported by data and relevant to the scientific question.
  • Model fit is not the same as predictive performance. A model can describe existing observations well while performing poorly when extrapolated to new conditions.
  • Model-based evidence does not eliminate uncertainty. Models organize and quantify uncertainty; they do not make uncertainty disappear.
  • Models do not replace clinical judgment. MIDD integrates quantitative evidence with pharmacological, clinical, and development expertise.
19 · Practical workflow

19. A Practical MIDD Workflow

  1. Start with the decision. Define the development question before choosing the modeling method.
  2. Assemble the evidence. Identify relevant clinical, PK, PD, preclinical, physiological, and prior-study information.
  3. Choose the model appropriate to the question. Use the simplest model that adequately represents the processes required for the decision.
  4. Specify assumptions. Clearly document structural assumptions, parameter assumptions, variability, covariates, and extrapolation conditions.
  5. Estimate parameters. Use appropriate statistical or mechanistic methods to quantify the model.
  6. Evaluate model adequacy. Use diagnostics, predictive checks, biological plausibility, sensitivity analysis, and other appropriate evaluation methods.
  7. Quantify uncertainty. Distinguish parameter uncertainty, variability, and structural or scenario uncertainty.
  8. Simulate relevant scenarios. Explore alternative doses, populations, trial designs, and assumptions.
  9. Connect results to the decision. Explain exactly how the model changes or informs the development strategy.
  10. Update the model as evidence accumulates. MIDD is iterative: new data can refine the model and change future predictions.

20. Key Takeaways

  • Model-informed drug development uses quantitative models to integrate evidence and support drug development decisions.
  • MIDD is a framework rather than a single modeling technique. It can include population PK, PK/PD, exposure-response, disease progression, PBPK, QSP, and clinical trial simulation.
  • Pharmacokinetic models connect dose to concentration and exposure, providing a foundation for many model-informed analyses.
  • PK/PD models connect exposure to pharmacologic response and can provide a quantitative basis for exposure-response analysis.
  • Population models account for variability between patients and can identify covariates associated with differences in drug disposition or response.
  • PBPK and QSP models can incorporate physiological or mechanistic knowledge to address questions that extend beyond empirical PK descriptions.
  • Simulation allows models to be used prospectively to explore alternative doses, populations, trial designs, and development scenarios.
  • Model-based predictions are conditional on model assumptions, estimated parameters, data quality, and uncertainty.
  • A more complicated model is not automatically a better model. The appropriate model is one that is sufficiently informative for the scientific question and supported by the available evidence.
  • MIDD is most valuable when model outputs are connected explicitly to a development decision rather than treated as an end in themselves.
  • MIDD is iterative: as new data become available, models can be evaluated, updated, and used to inform subsequent development decisions.
Next step

Where to Go Next

A natural progression after this introduction is to study the major components of MIDD individually. A useful sequence is population PK modeling, followed by PK/PD modeling, exposure-response analysis, clinical trial simulation, PBPK, QSP, and finally integrated model-based drug development.

The next tutorial can build directly on these ideas by introducing Population Pharmacokinetic Modeling and showing how nonlinear mixed-effects models represent typical PK, between-subject variability, residual variability, and covariate effects.

← Back to Pharmacokinetics Tutorials