The scale of the problem is considerable. Approximately 50 million major noncardiac surgeries are performed annually in the United States, and an estimated 12 percent of them are complicated by postoperative acute kidney injury. The condition can lengthen hospital stays and, as previous Mount Sinai-led research has shown, increase the risk of chronic kidney disease and death. Because there is currently no direct treatment for the condition, predicting risk and preventing kidney injury are especially important.

During and after surgery, the body undergoes significant stress, and patients may experience fluctuations in blood pressure, heart rate, and breathing, as well as bleeding, according to the investigators. The kidneys are particularly sensitive to these changes, making them among the first organs to show signs of injury. The risk is higher in patients with pre-existing kidney or heart conditions, those undergoing major surgery, or those experiencing other complications.

Andrew Leibowitz, MD, Chair of the Department of Anesthesiology, Perioperative and Pain Medicine at the Icahn School of Medicine at Mount Sinai, said that postoperative acute kidney injury is a common and serious complication after surgery, yet reliable ways to identify which patients are most likely to develop it remain lacking. He said the research brings together data from multiple hospitals to develop a more precise approach to predicting risk and, ultimately, to help clinicians tailor preventive strategies to the individual patient.

The Mount Sinai-led team will analyze data from more than 320,000 patients at 15 hospitals across three health systems to develop and test an artificial intelligence model that predicts which patients may develop acute kidney injury after surgery. Researchers will also examine whether different patient groups may benefit from different prevention strategies, with the long-term goal of tailoring prevention to each patient. The study has two aims: first, to develop and test models that predict whether a patient may develop postoperative acute kidney injury and how severe it may be; and second, to identify patient groups that might benefit from different prevention strategies.

A central technical feature of the project is federated learning, a method that allows hospitals to develop a shared AI prediction model without transferring individual patient records between institutions. Each hospital retains its own patient data. Researchers from Mount Sinai, Vanderbilt University, and the University of California, Los Angeles, will contribute expertise in anesthesiology, nephrology, biomedical informatics, and health data science.

The federated approach addresses a known weakness in existing tools. Many current postoperative acute kidney injury prediction models were developed with relatively small patient populations at individual hospitals, which, according to the researchers, can make it difficult to determine whether a model will work reliably at other hospitals or for different patient populations. Girish N. Nadkarni, MD, MPH, Chair of the Windreich Department of Artificial Intelligence and Human Health at Mount Sinai and Director of the Hasso Plattner Institute for Digital Health at Mount Sinai, said federated learning offers a way to learn from a much larger and more diverse patient population while keeping patient-level data within the health systems where it originated, allowing the team to ask whether a prediction model can work reliably across different hospitals and patient populations rather than building a tool around the experience of a single institution.

Ira Hofer, MD, Associate Professor of Anesthesiology, Perioperative and Pain Medicine at Mount Sinai and the lead investigator for the study, said that today there is limited ability to predict who will develop kidney injury after surgery, and that prevention strategies are often applied broadly rather than tailored to individuals. He said the goal is to use the wealth of information available in electronic health records to identify patients at risk before surgery and move toward prevention strategies that are better matched to each patient. Hofer added that the ultimate aim is to move beyond a one-size-fits-all approach to preventing kidney injury after surgery, by combining expertise across institutions and using data in a way that protects patient privacy.

In evaluating the new models, researchers will compare the federated learning models with existing prediction approaches and with a model built using pooled data from participating institutions. They will also assess performance across different patient populations, including groups that may be underrepresented in existing models. The findings are intended to inform future clinical trials that will test whether personalized prevention strategies can reduce postoperative acute kidney injury in patients undergoing surgery; the grant itself funds model development and testing, not the demonstration that such strategies reduce injury.

The research builds on the team’s prior work developing perioperative risk models and applying federated learning across hospitals, including research using the approach to predict COVID-19 mortality. The investigators also have experience using data to identify personalized treatment strategies for sepsis-associated kidney injury. Nadkarni noted that if successful, the work could provide a model for using collaborative, privacy-conscious data science to improve prevention of complications beyond kidney injury, particularly in patients undergoing surgery. The work is supported by the National Institutes of Health under award number 1R01DK144051-01A1.

Subject of Research: Prediction and prevention of postoperative acute kidney injury using federated AI models

Article Title: Mount Sinai receives $4 million NIH grant to address risk of acute kidney injury after surgery

Article References: Mount Sinai receives $4 million NIH grant to address risk of acute kidney injury after surgery. Original publication (Bibliographic details unavailable.)

Image Credits: AI Generated

DOI: Not provided

Keywords: acute kidney injury, surgery, Mount Sinai, NIH grant, federated learning, artificial intelligence, risk prediction, anesthesiology, nephrology, electronic health records, personalized prevention, postoperative complications