In the absence of a single, remarkably predictive marker of acute kidney injury (AKI), there is considerable interest in developing a multivariable model to predict AKI. We will develop a real-time updated risk model based on routinely collected data present in the Electronic Health Record (EHR) to predict incident AKI in hospitalized children. The predictive model will be developed for all hospitalized children including neonates and will automatically produce a risk score from conventional clinical factors, vital signs, medications, and laboratory data. We will curate 8,498 hospitalizations from Yale-New Haven Children’s Hospital and split the data into training and test datasets. We will use multiple feature selection strategies to create a parsimonious set of 10 variables which will form our risk prediction model.

     We will then prospectively validate our risk model with direct EHR integration. Implementing the best performing model into the live EHR, we will identify in real-time children at highest risk of developing AKI. These children will be enrolled in a prospective cohort study to evaluate the performance of targeted biomarkers as confirmatory tests for impending AKI.

     Lastly, we will conduct a pilot and feasibility study to develop and implement a standardized pre-AKI clinical support system. In collaboration with pharmacy, nursing, and pediatrician stakeholders, we will create a pre-AKI recommendation template that can be used to guide diagnostic, drug, and therapeutic management in children at high risk of impending AKI. We will further assess the degree to which providers adhere to these recommendations when presented at the point of care. Developing a real-time risk prediction model for AKI will allow us to identify a window of opportunity before creatinine rises and provide clinicians with immediately applicable management recommendations, transforming care for hospitalized children. This proposal may change how we think about treating hospitalized children with AKI from a reactive to a proactive paradigm.