Using Novel Machine Learning Methods to Personalize Strategies for Prevention of Persistent AKI after Cardiac Surgery

My long-term goal is to integrate health informatics, data mining and machine learning to improve the care for patients with, and at risk for, acute kidney injury (AKI). I am dual trained in Nephrology and Critical Care Medicine. I am already developing my skills in health informatics. This proposal presents a five-year career development plan for NIH K08 award focused on training in advanced data mining, machine learning and their applications to critical care nephrology. To that effect, I have assembled a strong mentoring team with decades of experience in mentoring, research and leadership. The outlined career development plan in conjunction with intensive mentoring and hands-on training will provide me the perfect platform to become a leading independent investigator in the field. AKI is seen in over one-third of patients undergoing cardiac surgery. Several trials investigating various medications to prevent or treat AKI over the last two decades have proven futile. Management of AKI therefore focuses on its prevention, measures to reduce further progression and management of its complications. The strategy to prevent AKI and its progression relies on clinical interventions to optimize a patient’s fluid status, blood pressure and avoiding nephrotoxins and hyperglycemia. These clinical interventions when provided to patients requiring cardiac surgery as a care-bundle are associated with decreased incidence of AKI. This care- bundle, however, has very low compliance with implementation and lacks the ability to personalize care for patients. With prior work showing differential response to therapy in AKI phenotypes, there is a critical need to determine personalized strategies to prevent the development of persistent AKI. Personalization of treatment strategies based on dynamic clinical characteristics of patients will ensure that the right action is performed at the right time. As transient AKI resolves spontaneously within 48 hours, focusing interventions to those at high risk for developing persistent AKI will lead to further personalization of this approach. The overall objective of this project is to determine a personalized strategy using machine learning to prevent the development of persistent AKI after cardiac surgery. I will pursue following specific aims for this study: (1) Develop reinforcement learning (RL) based strategy to prevent the development of persistent AKI after cardiac surgery. (2) Develop digital biomarkers to predict patients at risk for persistent AKI after cardiac surgery. Completion of these aims will provide a structured framework to provide personalized care to prevent the development of persistent AKI after cardiac surgery. It will also provide me with preliminary data and experience necessary to apply for R01 applications as an independent investigator leading a data science research program in critical care nephrology.