New AI Model Predicts Acute Kidney Injury After Angiography
Key findings
- Using data from the Catheter Sampled Blood Archive in Cardiovascular Disease (CASABLANCA) trial and machine learning, cardiologists at Massachusetts General Hospital developed a model that was highly accurate at predicting acute kidney injury (AKI) in patients undergoing coronary angiography
- Diabetes, C-reactive protein, osteopontin and ratio of blood urea nitrogen to creatinine were positively associated with AKI risk; CD5 antigen-like and factor VII were negatively associated
- Reliable prediction of AKI could allow for adjustments in patient care that might prevent severe kidney dysfunction
Acute kidney injury (AKI) after coronary angiography is usually caused by the contrast agent or, less commonly, by atheroembolism. Changes in serum creatinine or estimated glomerular filtration rate (eGFR) are used to diagnose AKI, but they are only modestly useful for predicting it.
Subscribe to the latest updates from Cardiovascular Advances in Motion
Now, based on data from the CASABLANCA trial, researchers have used artificial intelligence to identify six easily assessed factors that predict the risk of AKI after angiography. Nasrien E. Ibrahim, MD, associate director, Resynchronization & Advanced Cardiac Therapeutics Program, Hanna Gaggin, MD, MPH, clinical investigator and cardiologist, James Louis Januzzi, MD, director of the Dennis and Marilyn Barry Fellowship in Cardiology Research at Massachusetts General Hospital, and colleagues describe their predictive model in Clinical Cardiology.
Background on CASABLANCA
The CASABLANCA trial prospectively enrolled 889 patients scheduled for coronary and/or peripheral angiography, with or without intervention and who did not have a history of renal replacement, between 2008 and 2011 at Mass General.
At the time of angiography, the team collected blood samples and recorded more than 100 clinical variables on a standardized form. For this study, they measured 109 biomarkers in the blood samples. Follow-up was complete for all patients.
Altogether, 4.8% of the patients developed AKI within seven days after contrast exposure, defined in this study as absolute increase in serum creatinine ≥0.3 mg/dL, ≥50% increase in serum creatinine or oliguria of <0.5 mL/kg per hour for >6 hours.
Developing a Predictive Model
The researchers used an iterative machine learning approach to select clinical variables and biomarkers that predicted AKI. They then developed a final prognostic model comprising six predictors:
- Positively associated with AKI risk: a history of type 1 or type 2 diabetes, C-reactive protein, osteopontin (a proinflammatory cytokine) and ratio of blood urea nitrogen (BUN) to creatinine
- Negatively associated with AKI risk: CD5 antigen-like and factor VII
Accuracy of the Model
The optimal score cutoff in the risk prediction model had 77% sensitivity, 75% specificity and a negative predictive value of 98% for AKI. An elevated score was predictive of AKI in all subjects (odds ratio, 9.87; P < .001).
The predictive power of the model held up in several subgroups of CASABLANCA subjects:
- Women
- Those ≥75 years old
- Those with eGFR <60 mL/min/1.73 m2
- Those with diabetes
- Those with heart failure
- Those with peripheral artery disease
It might be possible to prevent AKI in patients deemed to be at high risk, Dr. Januzzi and his colleagues point out. Appropriate changes in management could include minimizing exposure to the contrast agent, employing biplane angiography, pre-procedure hydration and avoiding nephrotoxins.
Some patients who develop AKI are at risk of progression of chronic kidney disease because of the presence of comorbidities such as diabetes and heart failure. In those cases, relevant interventions would be lifestyle changes, avoidance of nephrotoxins and delaying elective angiography until the comorbidities are under better control.
The researchers plan to validate their prediction model in an external cohort.
view original journal article Subscription may be required
Learn more about Mass General's Division of Cardiology
Refer a patient to the Mass General Heart Center