In this post, Dr. Sevag Demirjian (SD) discusses a study recently published in the American Journal of Kidney Diseases on the development and validation of 4 new models to accurately predict acute kidney injury after cardiac surgery with Tejas P. Desai, eAJKD web advisory board member.

eAJKD: What was the impetus behind initiating this study?

SD: We have a high volume of surgical services, and assist in the care of many postoperative renal complications. The post-operative population is known to be at increased risk due to the high burden of comorbidities. Cardiovascular surgery is the most studied setting for acute kidney injury (AKI).

Previous predictive models of cardiac surgery-associated kidney injury have mostly confined their variable selection to preoperative risk factors, with the outcome of kidney failure requiring dialysis limiting their clinical utility.

We wanted to do this study using both preoperative and intra-operative information as well as doubling of serum creatinine. Both these variations would make the prediction model even more powerful in discriminating earlier AKI.

eAJKD: What do you feel is new in this investigation? What findings surprised you?

SD: The new features of this investigation are that we’ve included intra-operative parameters and doubling of serum creatinine. In addition, we were not shy of including many variables into the model to improve the accuracy of the model. The model remains easy to calculate with our web-based calculator, which is highlighted in the manuscript. In an ideal world, one’s electronic medical record would mine these variables and calculate a risk automatically.

I was most surprised by how accurate the models were with a C-statistics are in the 0.7 range. The ROC number predicts dialysis with pre- and intra-operative variables at 0.91.

eAJKD: What future studies are planned?

SD: The models that exist need to be externally validated. As with any model, it works well in the development cohort. Our model needs to be used by other centers to validate it. In addition, we tried to exclude variables that were based on decisions made by the clinician. This would limit the value of the model if practice patterns changes, or if a treatment variable isn’t offered at an institution.

eAJKD:Are there any changes that clinicians should make based on the findings of your investigation?

SD: Our study is not new in the sense that there are prior publications that looked at preoperative predictive models. Our model is a refinement of the existing models. I think people who are using predictive models clinically should continue to do so. With the use of intra-operative variables, we envision intensivists and nephrologists collaborating to use the model to make therapeutic decisions and offer prognostic counseling.

To view the article abstract or full-text (subscription required), please visit AJKD.org.