Dec. 03, 2021
Acute kidney injury (AKI) remains a significant complication of acute illnesses, contributing to considerable morbidity and mortality and burdening health care systems. While there is no known treatment for AKI, its prevention and recognition plus timely intervention can decrease AKI incidence and related complications.
There have been considerable advances in digital health technologies over the last decade. These technologies include access to individual and population health care data through electronic health records, remote monitoring both inside and outside hospital settings, and analysis through machine learning and artificial intelligence to provide clinicians with semi- or fully automated clinical decision support systems (CDSS) for personalized management.
"These technologies have been implemented successfully in multiple domains, including chronic kidney disease, and are now being evaluated in a few centers for AKI with promising results," says Kianoush B. Kashani, M.D., M.S., nephrologist and critical care specialist, at Mayo Clinic in Rochester, Minnesota. "They provide tremendous opportunities to improve the care of patients at risk of or with AKI or its related complications, enabling clinicians to implement primary, secondary and tertiary preventive measures in a timely fashion and have the potential to improve the processes of care and patient outcomes.
Like other CDSS technologies, AKI CDSS has three components:
- Description (to identify when AKI diagnostic criteria are met)
- Prescription (to provide a set of recommendations based on the evidence or best practices at the right time, and individualized based on each patient's risk profile)
"At Mayo Clinic, progress on digital health pertaining to AKI has been palpable," says Dr. Kashani. Completed, ongoing and future projects include:
Prediction: Static models
- AKI prediction in collaboration with the University of California, San Diego; results published in Nephrology Dialysis Transplantation in 2017.
- AKI prediction using only information that is available before intensive care unit (ICU) admission; study published in Clinical Kidney Journal in 2021.
Prediction: Dynamic models
- Random forest model; study published in Mayo Clinic Proceedings in 2019.
- Gradient boosting model that is externally validated; study pending publication.
An AKI sniffer based on the latest AKI definition criteria is developed and validated; study results were published in the Journal of Critical Care in 2015.
"Our ongoing projects include a systematic review and a prospective Delphi process to identify the optimum AKI care bundle to be individualized for each patient," says Dr. Kashani. "In addition, a larger learning reinforcement machine-learning project to determine the best care decision according to each patient risk profile is being conducted. We also use simulated learning to generate many simulated scenarios for each patient, to identify the best care plan with the lowest complication and optimized results."
Other AKI-related projects
Dr. Kashani explains other projects underway at Mayo Clinic:
- The AIDEx open-source platform model predicts overdosing or underdosing drugs eliminated by the kidney and provides appropriate drug dosing recommendations. The model is already developed and validated for vancomycin.
- Mayo Clinic developed a model to predict baseline serum creatinine for those who do not have measured serum creatinine available before ICU admission; the research was published in the American Journal of Nephrology in 2021.
- Mayo Clinic developed a patient-agnostic model that identifies the density of use of some drugs in association with AKI. "This model allows us to monitor drug use patterns in each institution and alert if any new medications have any potential to be nephrotoxic," says Dr. Kashani.
- An artificial intelligence model identifies patients with advanced AKI who are discharged from the hospital with a prescription for a nephrotoxin.
- The AID-ART project, sponsored by the National Institutes of Health, intends to validate a model that predicts hemodynamic instability among patients who receive dialysis in ICUs.
- Prediction of contrast-associated AKI by CT images
- Prediction of hyperkalemia among patients with AKI stage 3
- Prediction of progressive metabolic acidosis among patients with AKI stage 3
- Prediction of volume overload among patients with AKI stages 2 to 3
- Prediction of kidney replacement therapy independence after hospital discharge among patients in the ICU who are on continuous kidney replacement therapy
- Prediction of estimated glomerular filtration rate using contrast and non-contrast studies
- Artificial intelligence model to enhance CT images to decrease the dose of contrast media
Dr. Kashani concludes: "While there has been significant interest in the development and validation of AKI CDSS, its full implementation requires a significant effort by all members of our medical community. Like any other critical care syndrome, AKI is a million-piece puzzle that requires tremendous and long-term efforts to solve. Thus, we have a long road ahead of us before identifying the positive impact of these technologies on the quality of care and our patients' outcomes."
For more information
Malhotra R, et al. A risk prediction score for acute kidney injury in the intensive care unit. Nephrology Dialysis Transplantation. 2017;32:814.
Shawwa K, et al. Predicting acute kidney injury in critically ill patients using comorbid conditions utilizing machine learning. Clinical Kidney Journal. 2021;14:1428.
Chiofolo C, et al. Automated continuous acute kidney injury prediction and surveillance: A random forest model. Mayo Clinic Proceedings. 2019;94:783.
Ahmed A, et al. Development and validation of electronic surveillance tool for acute kidney injury: A retrospective analysis. Journal of Critical Care. 2015;30:988.
Ghosh E, et al. Estimation of baseline serum creatinine with machine learning. American Journal of Nephrology, 2021;52:753.