Abstract:

Concussion, the most common type of traumatic brain injury, has been identified as a public health issue. Recent research has exposed a troubling relationship between concussion and long-term health consequences to brain health, such as cognitive impairment, depression, and neurodegenerative disease. Appropriate concussion management is thought to play a critical role in improving long and short-term health outcomes for those with concussion. Unfortunately, the diagnosis and post-injury management of concussion remains challenging for many reasons, including: the lack of a gold standard diagnostic marker, the potential for strategic symptom reporting, and the need for guidelines built on rigorous analysis of large, observational clinical datasets. In this research, we develop two frameworks to address these issues. In the first framework, we formulate a multi-agent Partially Observable Markov Decision Process (mPOMDP) to model both the patient’s and doctor’s perspectives in sequential treatment decision problems. We analyze the role of strategic symptom-reporting on the optimal timing of return-to-play from sports-related concussion. While classical results for POMDPs do not hold for the mPOMDP, we derive conditions which ensure that the doctor’s optimal policy follows a threshold-type structure. In the second framework, we formulate the two-threshold problem (TTP) as a stochastic programming model to determine which patients should be diagnosed as positive, negative, or deferred due to a lack of conclusive evidence. We characterize the optimal solution to TTP and develop data-driven methodologies to solve and calibrate TTP. For both frameworks, we conduct numerical studies using multi-center data from the CARE Consortium – the largest available dataset on sports-related concussion. We show that our frameworks outperform existing methods commonly used in practice and use our findings to generate clinical insights for concussion management. The models developed in this research provide technical contributions to data-driven decision-making that can be applied broadly to other areas within and beyond healthcare.

Bio:

Gian-Gabriel Garcia is a PhD Candidate in the Industrial and Operations Engineering Department at the University of Michigan. He holds a Bachelor’s degree in Industrial Engineering from the University of Pittsburgh and a Master’s degree in Industrial and Operations Engineering from the University of Michigan. In his research, Gian is interested in developing data-driven frameworks for predictive and prescriptive analytics as motivated by high-impact problems in healthcare.  His current research focuses on (1) using large clinical datasets to gain patient-specific insights on disease progression and (2) combining these insights with stakeholders’ perspectives to improve diagnosis and treatment decisions. His research has been applied to concussion, glaucoma, and cardiovascular disease. It has been recognized by the National Science Foundation Graduate Research Fellowship, the INFORMS Bonder Scholarship for Applied Operations Research in Health Services, the Rackham Merit Fellowship, the SMDM Lee B. Lusted Prize in Quantitative Methods and Theoretical Developments, and first prize at the INFORMS Minority Issues Forum Poster Competition.