ICH M15 Model Evaluation Flexibility, Submission Requirements
Drug sponsors are not required to submit model analysis plans in advance under the new ICH M15 guideline on model-informed drug development (MIDD), and models that fail to meet predefined validation criteria are not necessarily disqualified from supporting regulatory decisions, international regulatory experts said during an 10/29 FDA-hosted webinar.
The clarifications came during a panel discussion and question-and-answer session on the finalized ICH M15: General Principles for Model-Informed Drug Development guideline, which establishes a harmonized framework for assessing modeling and simulation evidence used in drug development and regulatory submissions.
Responding to a question about whether sponsors must submit model analysis plans before generating modeling results, panelists said such plans represent good scientific practice but are not mandatory under the guideline. Regulators emphasized the value of defining technical acceptance criteria before conducting analyses, particularly for novel modeling approaches, but said sponsors do not necessarily need advance regulatory approval of those criteria.
The panel also addressed whether models classified as having high model risk or high model impact face a reduced likelihood of regulatory acceptance. Speakers said neither classification should be interpreted as a negative judgment about model quality or a predictor of rejection. Instead, the classifications help determine the rigor of model evaluation and the need for early regulatory engagement. High model risk generally calls for more robust supporting evidence and evaluation, while high model impact—reflecting a departure from established regulatory expectations—makes early discussions with regulators particularly important.
Validation Failures Not Necessarily Disqualifying
In another exchange, panelists discussed situations in which model predictions do not agree with validation data or predefined technical criteria are not met.
Scott Marshall, an industry representative (GSK MIDD head) and former rapporteur for the ICH M15 expert working group, said the framework is intended to support an assessment of modeling evidence rather than a simple pass-or-fail determination. Other panelists said sponsors should investigate whether discrepancies reflect model deficiencies, limitations of the validation data, differences in the conditions under which data were generated or uncertainty in the underlying evidence.
The assessment should determine whether the model remains fit for its intended regulatory purpose and whether additional analyses, modifications or supporting evidence are needed. Failure to meet a technical criterion does not automatically establish that a model is unusable.
The panel also encouraged sponsors to begin regulatory discussions early in development. FDA reviewer Xiaofei Wang specifically recommended preparing MIDD communication documents during preclinical development when modeling approaches are proposed to reduce animal studies. Regarding regulatory submissions, panelists said MIDD assessment tables may be submitted as standalone documents during early development or incorporated into appropriate sections of the Common Technical Document for marketing applications.
The webinar also identified FDA's MIDD meeting program, Type C meetings and Fit-for-Purpose program as potential avenues for discussing modeling approaches with the agency.
The ICH M15 guideline was finalized earlier this year. The webinar focused on its practical implementation, including how sponsors and regulators should evaluate modeling evidence and document its contribution to regulatory decisions.
The guidance defines MIDD as “the use of computational modeling and simulation methods that integrate nonclinical data, clinical data, prior knowledge, and information about drugs and diseases to generate evidence that informs development and regulatory decisions.
At the center of the guidance is a structured framework regulators and drug developers can use to determine whether model outputs are sufficiently reliable to support regulatory decisions. The framework requires sponsors to define a specific "question of interest" that a model is intended to answer and establish the model's "context of use," including its role in decision-making and the data used to build it.