ICH M15 in plain language
What ICH M15 changes for sponsors, and what to prepare.
This guide paraphrases ICH M15 for orientation. It is not a substitute for the guideline, and it is not regulatory advice for a specific program. M15 is a framework that sponsors can use to plan, document and discuss MIDD evidence. It does not make MIDD mandatory.
What changed
Agencies have reviewed population PK, exposure-response and PBPK for decades. What was missing was a shared way to judge how much weight a model can carry, and how to document it.
ICH M15 provides it. It reached Step 4 in January 2026, and FDA published it as final guidance in June 2026; other regions implement it through their own procedures.
In practice, a model is now judged against the question it answers and the decision it informs, and the documentation is expected to show that reasoning.
The assessment elements
M15 describes a risk-informed assessment for each use of a model. Below, each element is worked through for an example compound, PMX-101, and one question: which dose for Phase 3? Open a row for the definition and our reasoning.
Question of interestWhich dose of PMX-101 should go into Phase 3?
Meaning
The development question or decision the model is there to inform. Everything else in the table is judged against it.
For PMX-101
100 mg or 200 mg once daily, weighing response rate against grade 3+ adverse events.
Context of useE-R models inform the choice, next to the randomized Phase 2 data.
Meaning
How the model will be used, and what other evidence it sits next to.
For PMX-101
Efficacy and safety E-R models, built on population PK exposures, predict outcomes at both doses in the Phase 3 population. They add to the randomized two-dose comparison; they do not replace it.
Model influenceOne of two main lines of evidence.
Meaning
How much weight the model carries compared with the rest of the evidence.
Why medium
The randomized data carry part of the decision. The model adds the exposure dimension and the extrapolation to Phase 3, so it matters, but it is not the only basis.
Consequence of wrong decisionAvoidable toxicity, or a pivotal trial at risk.
Meaning
What happens to patients, or to the program, if the decision turns out to be wrong.
Why high
Too low a dose can cost the efficacy result. Too high a dose brings grade 3+ toxicity and dose reductions in a serious disease.
Model riskMedium influence, high consequence.
Meaning
Influence and consequence taken together. It decides how much credibility evidence the model needs.
Why high
The consequence dominates. We plan the most demanding verification and add an independent senior review.
Model impactSupports the dose justification at end of Phase 2.
Meaning
How much the model contributes to the regulatory decision: informing a design, supporting a dose, or replacing a study.
Why medium
It supports the rationale shown to the agency. It does not replace a study or carry a label claim on its own.
Technical criteriaFixed in the MAP before modeling starts.
Meaning
The verification and validation work, and the pass criteria, set in proportion to model risk.
For PMX-101
pcVPC and GOF within pre-set bounds, bootstrap CIs on key parameters, a pre-specified exposure metric, sensitivity analyses for confounding by disease severity, and independent re-execution of every reported run.
The point is proportionality: a model that informs a sampling schedule needs less than one that replaces a clinical study.
The model analysis plan (MAP)
M15 recommends that the analysis be planned before it is done. The plan states the question of interest and context of use, the initial assessment, the data to be used, the modeling methods, and the technical criteria the model must meet.
Choices once made during the analysis, such as the exposure metric, are now written down in advance, and the plan can be discussed with the agency before results exist.
In our packages, the MAP is the first deliverable and the first payment milestone. It is approved before modeling starts, and before any unblinded efficacy data are analysed.
The model analysis report (MAR)
The report documents what was done and what was found, against the plan. M15 describes recommended content: the question and context of use, data and data handling, methods, model evaluation, results, the final assessment, and any deviations from the plan with their justification.
It is also where the assessment is completed, so a reviewer can see why the model can, or cannot, carry the weight it is given.
Verification
M15 distinguishes between checking that a model is implemented correctly (verification), checking that it adequately represents the system (validation), and checking that it applies to the question asked (applicability). All three are scaled to model risk.
In our workflow every run is logged, reproduced in a clean environment, and the analysis dataset is double-programmed. How our verification works.
Agencies may ask for code and data. We deliver them as plain text, with definitions, in every package.
What to prepare before your next regulatory meeting
- Write the question of interest in one sentence. If you cannot, the model is not ready to be discussed.
- Draft the context of use. Say what the model will be combined with, and what it will not be used for.
- Draft the assessment table. Rate influence, consequence, risk and impact, with a justification for each. Disagreements inside your team are worth resolving before the meeting.
- Share the analysis plan, not only the results. An agency can still shape a plan.
- Choose the right venue. In the US, the FDA MIDD Paired Meeting Program is designed for exactly this discussion. In Europe, EMA scientific advice can address model-based approaches.
- Check that your code and data can be handed over. If the model lives on one person's laptop, fix that first.
Not sure where your program stands? Discuss your program with us in a 15-minute call.
Download the guide
Get this guide as a PDF, with a blank assessment-table template you can use in your own planning.
Thank you.
We will email you the guide and the template.