Nephrotoxic Acute Kidney Injury in Veterans

Acute kidney injury (AKI) occurs in 20-25% of inpatient Veterans and is associated with substantial increases in short-term mortality, and in long-term rates of chronic kidney disease (CKD), cardiovascular disease (CVD), and death. The use of nephrotoxic medication is involved in a quarter of all AKI events, and the combination of high incidence and poor outcomes makes nephrotoxin-associated AKI (NA-AKI) a major VHA health problem. This proposal seeks to address major gaps in our understanding of NA-AKI among Veterans, while also equipping the applicant with skills in machine learning (ML) including interpretable learning and incorporation of longitudinal modeling and translational testing to ensure that models link to important long-term outcomes and to underlying kidney pathophysiology, as well as skills to prepare the candidate for future multi-site deployment and surveillance of calibration drift. There are no specific treatments after AKI onset, so prevention is the primary means of addressing this disease. NA-AKI is the subtype most amenable to early intervention since medication use is entirely within the purview of the hospital team; however, given the ubiquity of nephrotoxic medication usage within VHA, it is important to identify those patients who are at highest risk. Machine learning is well-suited to parsing the enormous number of variables associated with NA-AKI. In Aim 1 we will use VHA data including patient factors, illness-related factors, and medication data to build machine learning algorithms that can accurately determine which Veterans are most likely to develop NA-AKI. While accurate ML models have been developed for other AKI subtypes, loss of performance at heterogenous sites and loss of calibration over time have hindered widespread adoption. Aim 1 will therefore prepare the candidate to use federated learning and data-driven calibration drift surveillance to overcome these barriers. Secondly, simply predicting a change in creatinine, the most commonly used definition of AKI, can be overly simplistic and not linked to long-term or clinically relevant disease. In Aims 2 and 3 we will use interpretable learning and incorporation of longitudinal modeling and translational testing to enhance the ML algorithm’s prediction of clinically meaningful AKI. In the short-term, completion of these Aims will position the applicant for a post-CDA grant proposal to develop a systematic approach to NA-AKI prevention that couples ML learning, AKI biomarkers, and clinical interventions to reduce nephrotoxin exposure in high-risk Veterans, and also a post-CDA grant expanding the use of AKI biomarkers so that ML algorithms can predict tubular injury specifically. My long-term goal is to improve outcomes among Veterans with kidney diseases by increasing the efficacy of VA nephrologists. Nephrology is a heavily lab-dependent specialty, and there is tremendous potential to use electronic medical record data, machine learning algorithms, and novel laboratory markers to increase the scope of care that can be provided. I would like to develop an approach to AKI and other kidney diseases that doesn’t await consultation, but rather one that proactively identifies and intervenes in the care of high-risk hospitalized Veterans. In order to accomplish these long-term goals, I will need a complementary set of skills in biomedical informatics and machine learning, translational medicine, and longitudinal modeling. The CDA-2 award period will build upon my subject matter expertise in NA-AKI and masters-level training in translational research, and will add expertise in health informatics with an emphasis on machine learning. These skills will be developed through focused didactic classwork through the Masters of Business Analytics (given the prominence of the University of Iowa Hospital system on campus, nearly half of all projects relate to clinical decision making) coupled with e-learning modules on how to successfully prepare data fr Project Number: 1IK2BX006525-01A1 | Fiscal Year: 2025 | NIH Institute/Center: Veterans Affairs (VA) | Principal Investigator: Benjamin Griffin | Institution: IOWA CITY VA MEDICAL CENTER, IOWA CITY, IA | Activity Code: IK2 | Study Section: Special Emphasis Panel[ZRD1 NEPH-N (01)] View on NIH RePORTER: https://reporter.nih.gov/project-details/11048705