A great amount of statistical tools and methods have been applied in health care analytics to assist decision making and improve the quality of diabetes related health care services.However, limitations of existing methods, new types of data, and specific demands in different areas are challenging current statistical tools. These challenges further encourage developing new statistical methods or extending existing methods to better fit different demands and improve performance of models and methods in practice. In this dissertation, we developed, applied, and extended many innovative statistical models and methods to address practical issues in health care of diabetes related population. Firstly, we developed a novel automated event detection method for univariate time series, Continuous Glucose Monitoring (CGM) data, from diabetic patients. Secondly, we invented a low-dimensional framework to classify and track longitudinal glucose status of CGM users based on within-subject analysis and unsupervised variable selection methods. Thirdly, we investigated the influence of daily activities on glucose series by applying a nonparamentric multivariate two sample test with independence assumption relaxed. Moreover, besides focusing on diabetic population, we also developed predictive models to access the risk of diabetes for population with prediabetes in later two chapters. Two types of response variables, binary indicator and HbA1c values, were used to aid different demands in practical healthcare services.