Predicting how patients will actually respond to treatment

Clinical prediction models trained on observational data often conflate disease biology with clinician behavior, leading to poor real-world performance. This review proposes a unified framework for intervention-aware trajectory modeling that estimates how individual patients evolve under different treatment scenarios. The approach organizes the problem around six components: three decision tasks (factual forecasting, counterfactual estimation, policy evaluation) and three data mechanisms (disease evolution, treatment assignment, observation bias). It synthesizes existing methods—multistate models, temporal point processes, deep sequence architectures, and causal inference—and provides principled evaluation using overlap diagnostics, uncertainty quantification, and target-trial validation. The framework is designed for clinical practitioners and health systems seeking treatment-sensitive predictions and safer adaptive learning systems.