Synthesis of statistical and artificial intelligence modelling for pharmacometrics (SYSTAIMO)

Classical statistical models can be calculated from a small amount of data and provide good predictions on a broad scale in various fields of pharmacy, biology and ecology. However, they deteriorate with increasing levels of detail (e.g., individuals rather than total population). AI-based models can make much better predictions at these more detailed levels, but require significantly more data. For many practical applications, however, these data sets are not available.

The goal of SYSTAIMO is therefore to synthesize statistical and AI-based models in order to combine the complementary advantages of both models in a hybrid model. In this context, the statistical model can be computed from a small amount of data and is analytical, i.e., it can provide predictions for arbitrarily fine time intervals. However, these predictions are not as precise and personalized as possible with AI methods, but can serve as additional training data for the AI model, allowing much better and more precise predictions to be obtained from the synthesis of both models. To demonstrate the performance of this new approach, the prediction of the concentration of the active substance propofol in the blood of patients will be investigated as a use case.