1. What Is Model-Informed Drug Development?
Model-Informed Drug Development (MIDD) is an approach in which computational modeling and simulation are used to integrate information from multiple sources and generate evidence that supports drug development decisions.
The underlying idea is broader than simply fitting a pharmacokinetic model. MIDD can integrate nonclinical data, clinical data, prior knowledge, drug characteristics, disease biology, and trial information into quantitative models that address a specific development question.
MIDD is best understood as a decision-oriented framework: models are developed and evaluated because they can provide evidence relevant to a specific development question.
2. Where Does MIDD Fit in Drug Development?
MIDD can contribute at many stages of development. The specific model and the decision it supports change as evidence accumulates.
| Development stage | Typical question | Examples of MIDD approaches |
|---|---|---|
| Discovery / preclinical | What exposure or biological behavior should be expected in humans? | PBPK, QSP, mechanistic models, translational PK/PD |
| First-in-human | What starting dose and escalation strategy are appropriate? | PK modeling, translational PK/PD, exposure prediction, simulation |
| Phase I | How does exposure vary across doses, subjects, and populations? | Population PK, dose-exposure modeling, PBPK, exposure-response |
| Phase II | Which dose and regimen should advance? | Exposure-response, PK/PD, disease models, trial simulation |
| Phase II/III | How should the confirmatory trial be designed? | Clinical trial simulation, disease progression, longitudinal models |
| Regulatory submission | What does the totality of model-based evidence imply? | Population PK, exposure-response, PBPK, QT modeling, M&S |
| Post-approval | How should dosing or use be optimized in additional populations? | PBPK, population PK, exposure-response, real-world model integration |
The important point is that the role of MIDD evolves. Early models may be used primarily to reduce uncertainty and guide experimentation. Later models may support dose selection, trial design, regulatory decisions, or optimization of product use.
3. MIDD Starts With a Development Question
A common mistake is to begin with a modeling method rather than a decision. Instead of asking, “Can we build a PBPK model?” a development team should first ask what decision the model is intended to inform.
Examples include:
- What dose should be used in the next clinical study?
- What dose range should be evaluated?
- Is a dedicated drug-drug interaction study necessary?
- Can a clinical trial population be informed by simulations?
- What exposure is associated with efficacy?
- What exposure is associated with a safety signal?
- How should dosing be adjusted in renal or hepatic impairment?
- What trial duration is needed to characterize a treatment effect?
- Can information from one population or formulation be extrapolated to another?
4. Define the Context of Use
A model does not have to be “valid” in an abstract sense for every possible purpose. Instead, model evaluation should be connected to the intended context of use.
For example, a model might be intended to:
- Predict exposure over a specific dose range.
- Support selection of a dose for a Phase II study.
- Predict drug-drug interaction magnitude.
- Characterize exposure-response relationships.
- Simulate alternative clinical trial designs.
- Support dosing recommendations for a special population.
The same underlying model may have different levels of credibility for different applications. A model that is adequate for exploratory dose selection may require additional evaluation before being used for a high-consequence regulatory decision.
5. The MIDD Modeling Toolbox
MIDD encompasses a family of quantitative approaches rather than a single modeling discipline.
| Approach | Primary information represented | Common development applications |
|---|---|---|
| Population PK | Drug concentration-time behavior and variability between individuals | Dose optimization, covariate effects, sparse-data analysis |
| PK/PD | Relationship between exposure and pharmacologic response | Dose selection, exposure-response, biomarker interpretation |
| PBPK | Drug disposition using physiological and mechanistic information | DDI, special populations, formulation and route questions |
| QSP | Mechanistic relationships among drug, disease, biology, and pathways | Translational questions, target biology, combination therapy |
| Disease progression models | Natural history and treatment effects over time | Trial design, endpoint interpretation, duration |
| Clinical trial simulation | Interaction among design, disease, treatment, variability, and outcomes | Design optimization and operating characteristics |
| Exposure-response models | Relationship between exposure and efficacy or safety | Dose selection, benefit-risk characterization |
| Mechanistic safety models | Biological mechanisms associated with adverse outcomes | Safety prediction and biomarker evaluation |
These approaches can also be combined. For example, a PBPK model may predict concentrations that become inputs to a PK/PD or exposure-response analysis, while a disease model can provide the disease trajectory used in a clinical trial simulation.
6. MIDD Before the First Human Dose
MIDD can begin before clinical development. At this stage, the central challenge is translation: available evidence comes primarily from in vitro, animal, mechanistic, and physicochemical information, while the ultimate decision concerns humans.
Potential objectives include:
- Characterizing expected human pharmacokinetics.
- Connecting preclinical exposure to pharmacologic effects.
- Exploring dose ranges for first-in-human studies.
- Identifying important physiological or biological uncertainties.
- Evaluating potential drug-drug interactions.
- Exploring translational hypotheses before committing to clinical studies.
Mechanistic approaches such as PBPK and QSP can be particularly useful when physiology, biological pathways, and drug properties provide information that can be incorporated into the model.
7. MIDD and First-in-Human Dose Selection
The first-in-human study is an important transition from predominantly preclinical evidence to human clinical data. Quantitative modeling can help connect the two.
A simplified translational workflow might look like:
Model-based predictions can be used alongside other nonclinical and clinical considerations. Once human data become available, the model can be updated and evaluated against the observations.
This creates an iterative process rather than a one-time prediction:
Model-based development becomes increasingly informative as clinical observations are added to the evidence base.
8. MIDD in Phase I
Phase I studies provide early human information about exposure, variability, tolerability, food effects, formulation effects, and sometimes pharmacologic biomarkers.
Population PK can help characterize typical pharmacokinetic parameters and between-subject variability. Exposure-response analyses can begin to connect exposure with biomarkers or early measures of pharmacologic activity.
At this stage, modeling may help answer questions such as:
- How does exposure change with dose?
- Is pharmacokinetics approximately linear over the studied range?
- How much between-subject variability is present?
- Which covariates appear important?
- What exposure range is being achieved?
- What doses should be studied next?
The model is therefore both an analytical tool and a way of organizing knowledge for the next development decision.
9. MIDD in Phase II: Choosing the Dose
Dose selection is one of the most important applications of MIDD because clinical development often requires decisions about doses that have not all been directly tested.
An exposure-response model can connect exposure to efficacy:
A corresponding safety model might relate exposure to the probability or magnitude of an adverse outcome. Together, these models can help characterize the exposure range over which efficacy and safety are expected to change.
Conceptual exposure-response curves can help characterize the region in which increasing exposure may provide diminishing efficacy gains while safety responses continue to change. The exact shapes and interpretation are drug-specific.
MIDD does not automatically identify a single “correct” dose. Rather, it provides quantitative evidence that can be considered alongside clinical, statistical, safety, and operational information.
10. Clinical Trial Simulation
MIDD can also be used to simulate alternative clinical trial designs before the trial is conducted.
A clinical trial simulation can combine:
- A disease progression model.
- A treatment-effect model.
- A dropout or missing-data mechanism.
- Inter-individual variability.
- A proposed dosing regimen.
- An endpoint model.
- A statistical analysis plan.
The simulated trials can then be analyzed using the proposed analysis method. Repeating the simulation many times allows investigators to explore how different design choices affect quantities such as bias, precision, power, estimability, or the probability of reaching a particular decision.
This approach is especially useful when the trial is expensive, long, or difficult to conduct and when important design choices can be evaluated quantitatively before enrollment begins.
11. MIDD in Confirmatory Development
By Phase III, the evidence base is typically much larger. MIDD can be used to integrate accumulated information rather than treating each study as an isolated source of evidence.
Potential applications include:
- Refining the dose and dosing interval.
- Characterizing exposure-response relationships across studies.
- Supporting population-specific dosing.
- Understanding variability in efficacy or safety.
- Evaluating trial duration and sampling strategies.
- Simulating alternative design assumptions.
- Supporting interpretation of pharmacometric endpoints.
The stakes of modeling decisions also increase. As the model begins to influence a pivotal development decision, model evaluation, documentation, traceability, and sensitivity analysis become increasingly important.
12. MIDD for Special Populations
One of the major strengths of model-based approaches is the ability to integrate information across populations when dedicated clinical studies are limited or when physiology provides useful mechanistic information.
Examples include:
- Renal impairment.
- Hepatic impairment.
- Pediatric populations.
- Older adults.
- Body-weight or body-size extremes.
- Pregnancy-related physiological changes.
- Drug-drug interaction scenarios.
For example, a PBPK model can represent physiological processes that differ between populations. Population PK models can characterize covariate effects using observed clinical data.
13. MIDD and Regulatory Decision-Making
MIDD can contribute to regulatory submissions by providing quantitative evidence that complements clinical and nonclinical findings. Examples include population PK analyses, exposure-response analyses, PBPK analyses, and other pharmacometric approaches.
The regulatory question is not simply whether a model fits the observed data. The more important question is whether the model provides sufficiently credible evidence for the particular decision for which it is being used.
The current ICH M15 guideline establishes a harmonized framework for planning, evaluating, and documenting MIDD evidence. Its framework emphasizes several elements, including:
- Question of interest — What development or regulatory question is being addressed?
- Context of use — How will the model be used?
- Model influence — How much does the model affect the decision?
- Consequence of a wrong decision — What are the implications if the model-supported conclusion is incorrect?
- Model risk — What uncertainties or limitations could affect the reliability of the model for its intended use?
14. Understanding Model Risk
Every model contains assumptions. Model risk is the possibility that those assumptions, limitations, data deficiencies, or structural choices could lead to an inappropriate conclusion for the intended use.
Consider two models that have similar predictive performance over the observed data range. If one model will be used only for exploratory visualization and the other will be used to support a high-consequence dosing decision, the implications of model uncertainty are different.
| Question | Why it matters |
|---|---|
| How much does the model influence the decision? | A model that determines the decision requires more attention than one providing supplementary information. |
| What happens if the model is wrong? | The consequences help determine how much uncertainty is acceptable. |
| Are the data informative? | Weak or sparse data can make important parameters poorly identified. |
| Are the assumptions appropriate? | Incorrect structural or mechanistic assumptions can produce misleading predictions. |
| Is the model evaluated for the intended context? | Evidence from one application may not automatically establish suitability for another. |
Model risk does not mean that modeling should be avoided. It means that uncertainty should be explicitly considered and managed in relation to the decision being supported.
15. How MIDD Evidence Accumulates
A useful way to think about MIDD is as an evidence-generation cycle. The model begins with assumptions and available information. New observations are then used to evaluate and refine those assumptions.
This is particularly important in longitudinal development programs. A model developed before the first human study should not be expected to remain unchanged throughout development. Its parameters, covariate relationships, structural assumptions, and intended use may evolve as evidence accumulates.
The important requirement is traceability: the development team should be able to explain what information entered the model, what assumptions were made, how the model was evaluated, and how its results affected the decision.
16. Worked Example: MIDD Across a Hypothetical Program
Consider a hypothetical oral drug being developed for a chronic inflammatory disease. The development team wants to determine an appropriate dose for Phase III while minimizing unnecessary exposure.
Step 1: Preclinical translation
Preclinical PK/PD experiments suggest that increasing exposure produces a saturable pharmacologic response. A translational model is used to define a plausible human exposure range.
Step 2: First-in-human study
The initial clinical study evaluates several dose levels. Observed concentration data are analyzed using a population PK model.
The model estimates typical clearance and volume of distribution and characterizes between-subject variability.
Step 3: Exposure-response analysis
Clinical biomarker data are linked to predicted exposure. The analysis suggests that the incremental pharmacologic benefit becomes smaller at higher exposure.
Step 4: Phase II simulation
A disease-response model is combined with the exposure-response model. Candidate doses are simulated under several assumptions about disease progression and variability.
Step 5: Phase III design
The selected exposure range is used to inform the dosing regimen and simulation of the proposed confirmatory study.
Step 6: Regulatory evidence
The final population PK and exposure-response analyses are integrated with clinical efficacy and safety results to characterize the dose-exposure- response relationship.
This is the central idea of lifecycle MIDD: the value comes from the integration of quantitative evidence with development decisions, not from modeling for its own sake.
17. MIDD After Approval
MIDD does not necessarily stop when a product receives regulatory approval. The approved population is often only the beginning of a much broader evidence base.
Post-approval modeling may support:
- Dosing in additional populations.
- New indications.
- New formulations or routes of administration.
- Drug-drug interaction evaluation.
- Special-population dosing.
- Exposure-response analyses using additional clinical experience.
- Integration of new clinical or real-world evidence.
The same general principle applies: define the question, identify the context of use, evaluate the available evidence, quantify uncertainty, and determine how the model should influence the decision.
18. The MIDD Lifecycle Map
MIDD is a lifecycle strategy rather than a single analysis performed at one point in development.
19. A Practical MIDD Workflow
- Define the decision. Identify the specific development question the analysis is intended to address.
- Define the context of use. Specify how the model's output will be used and under what conditions.
- Assemble the evidence. Identify relevant nonclinical data, clinical data, prior knowledge, biological information, and study-design information.
- Select an appropriate modeling strategy. Choose the simplest approach that can adequately address the question while representing the important scientific features.
- Develop the model. Specify structural assumptions, parameters, variability, and residual error where appropriate.
- Evaluate the model. Assess predictive performance, diagnostics, parameter plausibility, sensitivity, and other evidence appropriate to the context of use.
- Assess model risk. Consider model influence, consequences of an incorrect conclusion, and important sources of uncertainty.
- Simulate or predict. Use the model to generate evidence relevant to the development question.
- Integrate with other evidence. Model outputs should be interpreted alongside clinical, statistical, nonclinical, and operational information.
- Document the reasoning. Record the question, context of use, assumptions, data, model development, evaluation, limitations, and decision impact.
- Update as evidence accumulates. A lifecycle MIDD strategy should evolve as new information becomes available.
20. Common MIDD Mistakes
Starting with the model instead of the decision
A sophisticated model is not automatically useful. The model should have a clear connection to a development question.
Confusing model fit with model adequacy
A model can reproduce observed data while still being inappropriate for its intended predictive application. Evaluation should reflect the context of use.
Ignoring extrapolation
Predictions outside the observed data range may depend strongly on structural assumptions. Extrapolation should therefore be explicitly justified.
Overcomplicating the model
Additional compartments, parameters, mechanisms, or covariates do not automatically improve the decision. Complexity can make a model difficult to identify, evaluate, communicate, or apply.
Treating uncertainty as a nuisance
Uncertainty is part of the evidence. Sensitivity analyses and simulation can help determine whether uncertainty materially changes the development decision.
Failing to document decision impact
A regulatory or development reviewer should be able to understand not only what the model estimated, but how the result affected the development strategy.
21. What Should an MIDD Analysis Document?
A strong MIDD analysis should make the chain from question to decision transparent.
| Element | Key question |
|---|---|
| Question of interest | What decision is the analysis intended to inform? |
| Context of use | How will the model output be used? |
| Data | What information was used and how was it prepared? |
| Model structure | What biological, pharmacological, or statistical assumptions were made? |
| Parameters | How were model parameters estimated and identified? |
| Evaluation | What evidence supports model adequacy for the intended use? |
| Uncertainty | Which assumptions or parameters contribute most to uncertainty? |
| Model risk | What could happen if the model-supported conclusion is incorrect? |
| Decision influence | How did the model affect the development decision? |
| Limitations | Where should the model not be extrapolated or applied? |
This documentation becomes particularly important when model-based evidence is included in regulatory interactions or submissions.
22. From Individual Models to Model-Informed Development
The most powerful use of MIDD is not necessarily a single highly detailed model. It is the creation of a quantitative development framework in which models are used at the points where they can reduce uncertainty or improve a decision.
A development program might therefore contain several connected models:
Each model has a specific role. The models do not need to be identical, and they do not all need the same level of complexity. Their value comes from being fit for their intended purpose and integrated into the development strategy.
23. Key Takeaways
- MIDD uses modeling and simulation to integrate nonclinical data, clinical data, prior knowledge, and mechanistic information into evidence for drug development decisions.
- MIDD can contribute throughout the drug development lifecycle, from discovery and translational research through first-in-human studies, dose selection, confirmatory development, regulatory review, and post-approval optimization.
- The starting point should be the development question, not the modeling technique.
- The context of use specifies how the model will be used and provides an important basis for evaluating whether the model is fit for purpose.
- Population PK, PK/PD, PBPK, QSP, disease models, exposure-response models, and clinical trial simulation are complementary components of the MIDD toolbox.
- Early MIDD can help translate preclinical knowledge into human development hypotheses and inform first-in-human studies.
- During Phase I and Phase II, MIDD can help characterize exposure, variability, exposure-response relationships, and candidate dose ranges.
- Clinical trial simulation can evaluate alternative trial designs before conducting an expensive or lengthy clinical study.
- As model influence and the consequences of an incorrect decision increase, model evaluation, uncertainty assessment, and documentation become increasingly important.
- A good model fit does not by itself establish that a model is adequate for a particular predictive or regulatory application.
- Model risk should be considered explicitly, including the influence of the model on the decision and the consequences of a wrong conclusion.
- MIDD is iterative: models should evolve as new clinical and nonclinical evidence becomes available.
- The goal is not maximum model complexity. The goal is a model that is appropriate for the scientific question, available evidence, intended use, and decision consequences.
References
- U.S. Food and Drug Administration. M15 General Principles for Model-Informed Drug Development: Guidance for Industry. Final Guidance, June 2026.
- International Council for Harmonisation of Technical Requirements for Pharmaceuticals for Human Use. ICH M15: General Principles for Model-Informed Drug Development. ICH Harmonised Guideline, 2026.
- U.S. Food and Drug Administration. Model-Informed Drug Development Paired Meeting Program. FDA Drug Development Resources.
- U.S. Food and Drug Administration. MIDD Paired Meeting Program Frequently Asked Questions.
- U.S. Food and Drug Administration, Center for Drug Evaluation and Research. Division of Pharmacometrics.
- U.S. Food and Drug Administration. Model-Informed Product Development. Focus Area for Regulatory Science.
These references provide regulatory and scientific background for the principles described in this tutorial. Regulatory guidance should be consulted directly for current expectations for a specific development program.
Where to Go Next
A natural progression is to study the individual MIDD disciplines in greater depth: population PK, exposure-response modeling, PBPK, QSP, disease progression modeling, clinical trial simulation, and model-informed dose selection.
The next tutorial can focus on Introduction to Model-Informed Drug Development, establishing the terminology, evidence framework, and decision-oriented principles that underlie the lifecycle approach described here.