To optimally treat patients, a digital replica of the patient is needed to simulate the progression of a specific disease and predict how therapeutic interventions will affect this trajectory. This can be achieved by combining sequential data assimilation methods that simultaneously update both the state and parameters as new data continuously becomes available. These updates help to better describe the individual’s condition throughout the treatment process. The underlying model is a nonlinear mixed-effects model, with parameters fitted based on a small population, which introduces considerable variability. Additionally, an adaptive treatment protocol is determined by applying reinforcement learning (RL) before each dosing cycle. In this approach, a pre-trained, purely model-based Q-matrix is updated through extra training cycles, incorporating the newly updated set of states and parameters. We utilize Monte Carlo tree search with upper confidence bounds for the training procedure. For small cell lung cancer treatment, this digital twin approach demonstrates significant improvements in optimizing treatment protocols.