Data Analytics And Stochastic Optimization Models For Decision Support In Chronic Disease Operations Management
Meeting the complex needs of patients with chronic illness is the single greatest challenge in medical practices. Chronic disease is a prevalent and high-cost issue in the United States healthcare systems. Efficient spending of healthcare funds and better management of healthcare operation costs lead to an enhanced access to high-quality healthcare services and reduces the overall healthcare cost. Thus, in this research, we have proposed a comprehensive framework for chronic disease operations management. Due to uncertainty in patient demand and workload, this framework consists of two predictive and prescriptive analysis phases. In the first phase, we have proposed a deep multi-task learning approach for predicting the required workload of patients. Then in the second phase, we have developed two stochastic optimization models for capacity planning and resource allocation for decision-making in strategic and tactical management levels where the scope of decision-making includes single and multiple facilities, respectively. One of the drawbacks of earlier studies in workload prediction is that the problem is not investigated for multiple facilities where the quality of provided services, equipment and resources used for provided services as well as diagnosis and treatment procedures may differ even for patients with similar conditions. Besides, the sparsity of chronic disease data is another challenge in workload prediction. To tackle the mentioned issues, we have considered patient-dependent and facility-dependent attributes as well as the relation between them into the proposed model and trained multiple related tasks simultaneously. In addition, we have transformed the data using multiple non-linear transformations through several hidden layers to capture data complexity and sparsity for providing a robust abstraction. The results of this study show that feature representation and training related instances jointly increase the performance of patient workload prediction. Moreover, we have addressed two critical issues in team-based healthcare strategic and tactical planning. The first issue is to determine the optimal number of providers for multiple facilities and eligible patients for pay-to-travel incentives where the demand and location of patients are unknown. The second issue is to minimize the number of different healthcare teams and balance their workload within every single facility. We have developed a stochastic workforce and workload optimization model under various scenarios to address this issue. The result of prescriptive analysis suggests considering the randomness rather than replacing the stochastic variables by their expected value significantly contributes in reducing the overall cost of healthcare and practically enhancing access to care.