The increasing availability of large observational datasets of electronic health records give us the opportunity to address information management challenges facing modern clinicians. Machine learning solutions designed to optimise decision-making behaviour, such as reinforcement learning and treatment effects modelling, can be leveraged to provide recommendations for personalised treatment plans. From a methodological perspective, this work bridges the fields of RL and causal inference, with challenges such as learning entirely from purely observational data and from partially-observable representations, with potential sources of confounding. As an end-goal, I build on these methods to develop decision support frameworks for the intensive care.