Longitudinal health records contain data on patients' visits, condition, treatment, and test results representing progression of their health status over time. In poorly understood patient populations, such data are particularly helpful in characterizing disease progression and early detection. In this work we developed a deep learning algorithm for temporal pattern discovery over Rochester Epidemiology Project data. We modeled each patient's records as a matrix of temporal clinical events with ICD9 and HCUP CSS diagnosis codes as rows and years of diagnosis as columns. Patients aged 18 or younger at the time of diagnosis were selected. A deep Boltzmann machine network with three hidden layers was constructed with each patient's diagnosis matrix values as visible nodes. The final weights of the network model were analyzed as the common features among patients' records.