FDA and EMA Expectations for Model-Informed Drug Development

1. What Is MIDD From a Regulatory Perspective?

Model-informed drug development (MIDD) uses quantitative models to integrate information from preclinical, clinical, pharmacological, and other sources to inform drug-development and regulatory decisions.

MIDD is broader than population pharmacokinetics. Depending on the question, a development program may use population PK, exposure-response models, physiologically based pharmacokinetic (PBPK) models, pharmacodynamic models, disease or drug-trial-disease models, quantitative systems pharmacology (QSP), clinical trial simulations, or combinations of these approaches.

The central regulatory question is not simply whether a model is technically sophisticated. It is whether the model provides sufficiently credible evidence for its intended regulatory use.

This context-of-use perspective is now explicitly reflected in the harmonized ICH M15 framework adopted by both the FDA and EMA.

2. The Most Important Development: ICH M15

The regulatory landscape changed substantially in 2026 with the finalization and implementation of ICH M15, General Principles for Model-Informed Drug Development.

FDA issued the final M15 guidance in June 2026. EMA identifies the current ICH M15 guideline as a Step 5 guideline with a legal effective date of July 23, 2026.

ICH M15 establishes a common framework for planning, evaluating, documenting, reporting, and discussing MIDD evidence. It is intended to facilitate a multidisciplinary understanding of MIDD evidence rather than prescribe one modeling method for every development program.

3. Start With the Question and Context of Use

A common mistake is to begin with a preferred modeling technique rather than with the regulatory decision. A stronger MIDD strategy begins by defining what decision needs to be supported.

Step 1: Define the question of interest

Examples include:

• What dose or dosing regimen should be evaluated in the next clinical study?
• Can exposure-response relationships support dose selection?
• Can a PBPK model predict the effect of an intrinsic or extrinsic factor?
• Can a model support a pediatric dosing recommendation?
• Can simulations help determine an appropriate trial design or duration?
• Can model-based evidence support an alternative dosing regimen?

Step 2: Define the context of use

The context of use (COU) describes how the model will be used to answer the question and support the decision.

For example, the same population PK model could have a relatively limited COU such as describing typical clearance and identifying important covariates, or a more consequential COU involving simulation of an alternative dose in a special population.

Step 3: Determine how influential the model will be

A model that provides supportive information alongside a large clinical dataset may have a different regulatory risk profile from a model that is the principal evidence supporting a dose recommendation.

4. What Makes MIDD Evidence Credible?

Regulatory credibility is built through the connection between the scientific question, the data, the model, the evaluation strategy, and the decision that will ultimately be made.

A visually attractive fit to observed data is therefore only one component of model credibility. A model can reproduce the data well and still have limited predictive value for a different population, dose range, endpoint, or physiological condition.

5. Model Risk and Consequences of a Wrong Decision

MIDD is inherently associated with uncertainty. Regulatory evaluation therefore asks not only whether a model is plausible, but also what could happen if the model-supported decision is wrong.

Consider two hypothetical uses:

This does not mean that a high-impact model must always use one particular validation technique. Rather, the evaluation strategy should be appropriate for the model, its intended use, and the consequences of model-informed decisions.

6. Model Analysis Plans: Plan Before You Analyze

ICH M15 recommends pre-defining and documenting intended model analyses in a Model Analysis Plan (MAP). The MAP typically describes the objectives, data, methods, and planned model evaluation activities.

A useful MAP can include:

1. Background and rationale. Explain the development question and why modeling is being used.
2. Question of interest. State the scientific or regulatory question explicitly.
3. Context of use. Explain how the model output will influence the intended decision.
4. Data sources. Identify clinical, nonclinical, literature, formulation, physiological, or other data used.
5. Model structure. Describe the proposed mathematical and biological structure.
6. Parameterization. Define parameters, distributions, covariates, and relevant assumptions.
7. Estimation and simulation methods. Specify the computational methods and simulation framework.
8. Model evaluation. Predefine diagnostics, validation procedures, sensitivity analyses, and relevant acceptance considerations.
9. Decision criteria. Explain how model results will inform the development decision.
10. Limitations. Identify important assumptions and anticipated limitations.

7. How Should a Model Be Evaluated?

There is no single universal model-validation checklist that applies to every MIDD application. Evaluation should be tailored to the model and context of use.

Structural evaluation

The structural model should be scientifically justified. For mechanistic models, this includes consideration of the biological and pharmacological basis of model structure and parameters. For population models, the structural PK or PK/PD assumptions should be appropriate for the data and scientific question.

Parameter evaluation

Parameters should be sufficiently identifiable for the intended use. Important considerations include parameter uncertainty, correlations, boundary estimates, implausible values, and sensitivity to modeling assumptions.

Goodness-of-fit and diagnostics

Observed-versus-predicted plots, residual diagnostics, prediction-corrected visual predictive checks, individual fits, and other diagnostics may be informative depending on the modeling approach.

Predictive evaluation

When the regulatory question depends on prediction, predictive performance is particularly important. Evaluation may include external validation, temporal validation, simulation-based assessment, sensitivity analysis, or other methods appropriate to the model.

Scenario and sensitivity analysis

Regulators may need to understand whether the decision changes when important assumptions or uncertain inputs change. Sensitivity analysis can therefore be an important part of demonstrating robustness.

8. FDA Expectations for MIDD

FDA's final ICH M15 guidance establishes a harmonized framework for MIDD planning, model evaluation, documentation, regulatory interaction, reporting, and submission.

FDA also maintains a dedicated MIDD program and has extensive experience reviewing population PK, exposure-response, PBPK, clinical trial simulation, and other model-informed evidence.

Regulatory interaction

FDA encourages sponsors to discuss important MIDD questions with the Agency when prospective regulatory feedback could materially improve the development strategy. FDA's MIDD Paired Meeting Program specifically provides opportunities for sponsors to discuss MIDD approaches with Agency staff.

The FDA program identifies examples such as:

• Dose or dosing-regimen selection and refinement.
• Clinical trial simulation.
• Predictive or mechanistic safety evaluation.
• Prediction of outcomes or trial duration.
• Selection of response measures.
• Identification of critical biomarkers.

For an MIDD meeting request, FDA asks sponsors to clearly describe the product and development context, question of interest, proposed MIDD approach, context of use, meeting objectives, and key assessment elements such as model influence, consequences of a wrong decision, and model risk.

9. FDA Expectations for Pharmacometric Submissions

FDA provides specific expectations for submission of pharmacometric data, models, scripts, and supporting documentation.

For NDAs and BLAs, pharmacometric analysis reports may be submitted in eCTD Module 5.3.3.5. FDA identifies population PK, exposure-response, PBPK, and other pharmacometric analyses within this submission framework.

FDA also describes expectations for datasets, analysis scripts, model code, control streams, outputs, and documentation that allow reviewers to understand and reproduce major analyses.

This reflects a broader regulatory principle: a model submitted for regulatory decision-making should be sufficiently documented that reviewers can understand how the analysis was constructed, evaluated, and used.

10. EMA Expectations for MIDD

EMA's current framework is also centered on ICH M15. The EMA identifies the final M15 guideline as the current Step 5 guideline, with an effective date of July 23, 2026.

EMA has also developed substantial experience with model-informed approaches through population PK, pharmacometrics, PBPK, exposure-response analyses, pediatric development, and model-based approaches to alternative dosing regimens.

EMA's existing population PK guidance emphasizes sufficiently detailed reporting to allow regulatory authorities to conduct a secondary evaluation of the analysis and the conclusions drawn from it.

Model-based approaches in regulatory submissions

EMA's modelling-and-simulation Q&A materials illustrate that model-based approaches can, in appropriate circumstances, provide major evidence for regulatory decisions. For example, EMA describes circumstances in which modelling and simulation of PK and dose-exposure-response relationships can support alternative posology or routes of administration for certain monoclonal antibodies.

The key point is not that modeling automatically replaces clinical evidence. Rather, the acceptability of a model-based approach depends on the quality of the data, model implementation, validation, simulations, and the regulatory question being addressed.

11. PBPK, QSP, PBBM, and Other Mechanistic Models

Mechanistic MIDD models can integrate physiological, biochemical, pharmacological, biopharmaceutical, and population information. Examples include:

• Physiologically based pharmacokinetic (PBPK) models
• Physiologically based biopharmaceutic models (PBBM)
• Quantitative systems pharmacology (QSP) models

EMA has been developing a dedicated guideline framework for assessment and reporting of mechanistic models in the context of MIDD. A 2025 EMA concept paper identified issues including model structure, parameter identifiability, mechanistic justification, parameter plausibility, uncertainty quantification, and communication between developers and regulators.

As with other MIDD approaches, the model should be evaluated relative to the regulatory decision it is intended to support rather than simply judged by its mathematical complexity.

12. FDA and EMA: What Is Harmonized and What Remains Regional?

For sponsors developing products in both regions, ICH M15 substantially reduces the need to maintain fundamentally different conceptual frameworks for MIDD.

13. When Should Sponsors Talk With Regulators?

Regulatory interaction is particularly valuable when the proposed MIDD analysis could materially influence development strategy or when the acceptability of a novel model-based approach is uncertain.

Examples include:

• A model will be used to select a pivotal dose.
• A simulation will influence the design or duration of a major clinical trial.
• A PBPK model is proposed to address an interaction or special-population question without a dedicated clinical study.
• A model is expected to support a pediatric dose recommendation.
• A model-based approach may replace or substantially reduce a dedicated clinical study.
• A novel mechanistic model will be used to support a regulatory decision.
• The model will provide a major component of evidence for a labeling recommendation.

A useful regulatory interaction is not simply a presentation of the model. It should contain clearly formulated questions for the Agency.

Weak question

“Does the Agency agree with our model?”

More actionable question

“Does the Agency agree that the proposed context of use, model evaluation strategy, and planned sensitivity analyses are appropriate to support selection of the proposed dose for the Phase 3 program?”

14. What Should an MIDD Report Contain?

A strong model analysis report should allow a reviewer to understand the question, data, assumptions, methods, model evaluation, results, limitations, and regulatory interpretation.

If a MAP was developed, ICH M15 recommends that it be provided as an appendix to the associated MAR.

15. Worked Example: Using MIDD to Support Dose Selection

Consider a hypothetical development program in which the sponsor has completed Phase 2 and intends to select a dose for Phase 3.

Step 1: Question of interest

The question is:

Step 2: Context of use

A population PK and exposure-response model will be used to characterize exposure, describe relationships between exposure and efficacy/safety, and simulate candidate Phase 3 dosing regimens.

Step 3: Model development

The analysis integrates PK observations, dose information, demographic characteristics, relevant clinical endpoints, and safety measurements.

Step 4: Model evaluation

The sponsor evaluates goodness of fit, parameter uncertainty, covariate relationships, predictive performance, and sensitivity to important modeling assumptions.

Step 5: Simulation

The final model is used to simulate candidate dosing regimens and quantify expected exposure distributions and relevant efficacy and safety outcomes.

Step 6: Regulatory interpretation

The sponsor does not simply report that one simulated dose produced a desirable exposure. The sponsor explains how the model was evaluated, how uncertainty was characterized, what assumptions drive the predictions, and how the resulting evidence informs the Phase 3 dose-selection decision.

16. MIDD and Regulatory Labeling

MIDD can contribute to labeling-related decisions when the model provides credible evidence relevant to dosing, exposure, special populations, interactions, or other clinically meaningful questions.

Examples can include:

• Selection or refinement of recommended doses.
• Adjustment of dosing for renal or hepatic impairment.
• Pediatric dosing recommendations.
• Alternative dosing intervals.
• Drug-drug interaction predictions.
• Exposure-response interpretation.
• Support for alternative routes or formulations in appropriate circumstances.

The stronger the connection between the model output and the proposed labeling decision, the more important it becomes to demonstrate that the model is adequate for that specific use.

Model-based labeling decisions should therefore distinguish clearly between directly observed clinical evidence and model-based inference.

17. Common Problems in Regulatory MIDD Programs

1. Starting with the model instead of the decision

A sophisticated model does not automatically answer an important regulatory question. Define the question and COU first.

2. Treating model fit as validation

A model can fit existing observations while performing poorly when used for prediction in a new population or scenario.

3. Under-specifying uncertainty

Decision makers need to understand how uncertainty in parameters, assumptions, and model structure affects the resulting recommendation.

4. Building the regulatory package too late

Retrospectively assembling model documentation can make it difficult to reconstruct why decisions were made and how model evaluation was performed.

5. Failing to distinguish exploratory and confirmatory uses

A model used for hypothesis generation may require a different evidentiary framework from a model whose output directly determines a regulatory recommendation.

6. Insufficient reproducibility documentation

Missing code, data descriptions, software versions, package dependencies, or execution instructions can make regulatory review more difficult.

7. Overclaiming beyond the COU

A model evaluated for one population, dose range, formulation, or clinical question should not automatically be assumed to be reliable for a substantially different application.

18. A Regulatory MIDD Workflow

1. Define the development decision. Identify the decision that needs quantitative evidence.
2. Define the question of interest. State precisely what the model needs to answer.
3. Define the context of use. Describe how the model will influence the decision.
4. Assess model influence and decision consequences. Determine how much the decision depends on the model and what happens if it is wrong.
5. Develop the Model Analysis Plan. Predefine objectives, data, methods, and evaluation activities.
6. Build the model. Use an appropriate structural and statistical or mechanistic framework.
7. Evaluate the model. Assess fit, parameter behavior, predictive performance, uncertainty, and sensitivity as appropriate.
8. Perform simulations or predictions. Apply the model only within a scientifically justified domain.
9. Interpret the evidence. Connect model outputs to the original regulatory question.
10. Discuss with regulators when appropriate. Obtain prospective feedback when the MIDD strategy is consequential or novel.
11. Prepare the Model Analysis Report. Document the analysis, evaluation, assumptions, limitations, and conclusions.
12. Prepare reproducibility materials. Organize datasets, code, models, software information, and supporting files.
13. Submit according to regional requirements. Apply the common ICH M15 principles while following FDA or EMA-specific submission procedures.

19. MIDD Regulatory Readiness Checklist

Where to Go Next

A natural progression from this tutorial is to examine how regulatory expectations apply to specific MIDD methods, including population PK modeling, exposure-response analysis, PBPK, clinical trial simulation, QSP, and model-informed dose selection.

The next level is to learn how to construct a regulatory-ready Model Analysis Plan and Model Analysis Report, including the relationship between the question of interest, context of use, model evaluation, uncertainty, and the final regulatory decision.

References

Regulatory guidance evolves. Sponsors should consult the current FDA, EMA, and ICH documents applicable to the product, development stage, model type, and regulatory procedure when preparing an actual submission or regulatory interaction.