ICH M15: Why Model Risk, Not Model Sophistication, Sets the Validation Bar

On January 29, 2026, ICH adopted M15 at Step 4 — the first ICH guideline written entirely around model-informed drug development (MIDD). It does not tell a sponsor which platform, software, or modeling method to use. It tells sponsors how to justify trusting the output: a harmonized, six-element framework that grades a model by the regulatory decision it is asked to carry, not by its sophistication.

The fragmentation M15 is meant to fix

Model-informed approaches — population PK, physiologically based pharmacokinetic (PBPK) modeling, exposure-response, quantitative systems pharmacology — have been standard tools in drug interaction and dose-selection work for years, but sponsors answering to FDA, EMA, and PMDA have faced three different sets of expectations for how to document and justify that modeling evidence. M15 does not replace any agency's existing modeling guidance; it sits above them, giving sponsors one assessment language — the six elements and the assessment table — to bring into any of the three regions.

Model risk is the element that does the work

The first four elements describe the decision, not the model: what question is being asked, under what conditions the answer will be used, how much the model's output actually drives the decision, and what happens if that decision is wrong. Combine the third and fourth — model influence and the consequence of a wrong decision — and M15 derives model risk. Model risk is what sets the fifth element's practical weight: how rigorously the model needs to be verified and validated before a reviewer will rely on it. The sixth, model impact, is the resulting judgment on how much the evidence can actually carry in the regulatory decision.

1. Question of Interest. The specific scientific or regulatory question the model is meant to answer.
2. Context of Use (CoU). The precise conditions — population, dose range, decision point — under which the model's output will be applied.
3. Model Influence. How much weight the model's output carries relative to other evidence in the decision.
4. Consequence of a Wrong Decision. What happens — to patients, to the program — if the model-informed decision turns out to be wrong.
5. Model Risk. Derived from influence and consequence together; it sets the required rigor of verification and validation.
6. Model Impact. The resulting judgment on how much the model's evidence can carry in the overall regulatory argument.

A model is not high-risk or low-risk on its own merits. It is high-risk or low-risk relative to the decision it has been assigned — and M15 makes sponsors write that assignment down before the verification work begins, not after. Why Context of Use comes before model risk

Where this lands on a submission

For sponsors already running population PK or exposure-response analyses to support an IND or inform a marketing application, M15 is less a new obligation than a new vocabulary for work already underway — but the vocabulary matters, because the Model Analysis Plan is meant to be the document a sponsor brings into a Type B or scientific-advice meeting to align Context of Use and verification approach before the analysis is finalized, the same way an estimand gets agreed before the statistical analysis plan is locked. Sponsors who draft the MAP after the modeling is done, rather than before, lose the chance to agree the Context of Use with regulators while it can still change the verification plan rather than only the write-up.

None of this requires a new modeling platform or a new team. It requires writing down, early and in the same six-element language every region now recognizes, what the model is for and how wrong it is allowed to be. Programs that treat the assessment table as a CMC and nonclinical strategy document — not paperwork assembled after the analysis is already final — are the ones that get to use M15's harmonization as leverage rather than as one more thing to reconcile across three agencies.