Regimen adherence is a common problem among chronic disease patients and has posed tremendous challenges to sustainable case management. Intervening on every single non-adherence case often creates unnecessary burdens for providers and considerable annoyance for patients, leading to wastes of resources and increasing patient churn rates. In current practice, mitigating the risk of non-adherence cases is a labor-intensive task that requires additional efforts from healthcare professionals to handle on a case-by-case basis. Previous work has investigated into the possibility of modeling patient adherence behavior, but left questions about the accountability of such models in services. With the prevalence of mobile devices and maturing cloud-based service models, more patient data are fed to cloud services from a variety of sources (e.g., health records, surveys, sensors, embedded GPS modules). In this paper, we propose a risk mitigation service that can utilize heterogeneous patient behavioral data sources to construct statistical models of adherence and estimate intervention need. We design evaluations to examine a number of dimensions in statistical models of patient adherence and their impacts on the task of determining critical cases and patient propensity to churn. Finally, we demonstrate how the new service is designed to assist adherence case management with models that can classify cases of different intervention needs and discuss its applications, limitations, and sustainability issues. © 2012 IEEE.