Also relevant for Early Phase Trials: EMA Reflection paper on the use of Artificial Intelligence (AI) in the medicinal product lifecycle.
Besides the great opportunities for improved drug development, there is also concern about an increase in intransparency. Here is a brief summary of relevant topics:
The paper differentiates between the validation requirements depending on the area of application for artificial intelligence and machine learning in the different phases of clinical development.
The ‘regulatory impact’ and the ‘patient risk’ are taken into account.
In the drug discovery phase, regulatory impact and patient safety are classified as less relevant, while the risk is seen more on the developer side.
A relevant benefit for AI and ML in non-clinical development is seen in the avoidance of animal testing. As soon as GLP comes into play here, the requirements increase accordingly due to interference with patient safety, for example in first-in-human trials. In the early phase, AI and ML are still classified as low risk, but this changes as soon as treatment assignment or dosing is involved.
This applies to all cases in which patient safety is affected - and therefore also in Clinical Development: If a ‘method has not been previously qualified by the EMA for the specific context of use, the full model architecture, logs from model development, validation and testing, training data and description of the data processing pipeline would likely be considered parts of the clinical trial data or trial protocol dossier and thus may be requested for comprehensive assessment at the time of market authorisation, clinical trial application or GCP inspection.‘
‘When AI/ML systems are used for clinical management of an individual patient, they may be considered medical devices according to MDR or IVDR.‘
Further, the use in precision medicine is discussed.‘
Interestingly, concerning the Product Information it is emphasised that AI/ML applications used for drafting, compiling, editing, translating, tailoring, or reviewing medicinal product information documents should be used under close human supervision.‘
Without medical writers and qualified scientists, it is still not possible!
The document also refers to technical aspects such as data acquisition and augmentation, training, validation, and test datasets, model development, performance assessment, interpretability and explainability and model deployment.
Finally, aspects of governance, integrity, data protection and ethical considerations are discussed.
#AI; #Modelling; #EarlyPhase
