Presented in this paper is a Dynamic Bayesian Network (DBN) approach to predict glycaemia levels in intensive care patients. The occurrence of hyperglycaemia is associated with increased morbidity and mortality in critically ill patients. Due to the large inter-patient and intra-patient variability, the sparse nature of observations, inaccuracies in the data and the large number of factors that influence glycaemia, the system being modelled contains several sources of uncertainty. In the context of this uncertainty, the DBN-based system presented here performs extremely well. By using a DBN we integrate multiple strands of temporal evidence, arriving at varying time intervals, to determine the most probable underlying explanations. A key contribution of this work is that it presents a principled technique for recalibration of Model parameters from general population-level values to patient-specific values, based entirely on standard real-time measurements from the patient. While in this paper we apply our approach to the glycaemia problem, this approach is equally applicable to other applications where unseen variables must be assessed and individualized in real time.