Keywords: Markov Processes, Infinitesimal Generators, Likelihood, Transition Matrices
More than 20,000 patients are monitored for over five years. Every time a patient comes to the hospital, the attending physician determines the patient's diabetes status. The status could be not diabetic, pre-diabetic, or diabetic. A plethora of measurements such as weight, diastolic and systolic blood pressures are recorded at each visit. The patients come random number of times and at random times. With all the visits taken into account, the data balloons into more than 500,000 rows. Answers to questions like what are the chances that the patient becomes diabetic in one year's time given that at the moment he is pre-diabetic are sought. Examine how the chances are modulated by the co-variate information on the patient. Find a way to identify patients at a very high risk for diabetes. A stochastic process model seems to be the way to have a crack at the data and answer questions. Computational difficulties cloud our attempt for a successful resolution of the problems. We will present how we resolved the computational problems to learn from the data.