Viewics and Patient Safety & Quality Healthcare recently held a webinar with Dr. Navdeep Tangri, the Canadian nephrologist and medical researcher who developed the Kidney Failure Risk Equation for accurately predicting the risk of progression to renal failure in patients with chronic kidney disease (CKD). Below is an excerpt from his presentation.
Navdeep Tangri, MD, PhD, FRCP(C)
In 2011, we published a paper in JAMA which developed a predictive model for the progression of chronic kidney disease to kidney failure. I’m going to show you some data on how this model was initially developed, and its accuracy across the entire world — not just the United States. [The model we developed in this paper] was developed in patients with CKD stages 3 – 5. So, these were people with GFR 60 or less, followed by nephrologists in the province of Ontario and the province of British Columbia in Canada. We had 8,391 participants who had 1,563 kidney failure events. These were quite a few events, and it was certainly sufficient power to develop an accurate model. And, in fact, we developed multiple lab-based prediction models.
I really want to emphasize the lab-based part here, because I think lab-based models allow importing of data from the lab or from the EMR and allow you to instantaneously report risk. And that’s why GFR is widely reported; that’s why other lab-based models are widely used, where if any model that involves asking 20 questions from the patient or pulling out detailed pieces of information from the chart is never going to be universal; it’s never going to be used widely.
“Lab-based models allow importing of data from the lab or from the EMR and allow you to instantaneously report risk.”
Three models for risk prediction
We’ve developed three models here. One is a comparison, and two is our preferred use models. Model two includes age, sex, and GFR. That’s just meant to show you that this is a comparator in a GFR-based world. Model three adds urine ACR and is more accurate and substantially more accurate. And model six, which is the one with the two circles, adds albumin, phosphate, bicarbonate, and calcium. So, they’re all lab-based, and they’re all incrementally better. And all of them predict renal failure or progression to kidney failure with high degree of accuracy.
Global external validation
The results were from our initial two cohorts of Canadian patients referred to nephrologists. From 2011 – 2015, we implemented the equation using apps and platforms and online calculators, and I thought that it was a big success, but there were some concerns raised about validities in other ethnicities and health systems. For example, I used to get asked whether our models applied to African-Americans, who we know are at high risk of chronic kidney disease and progression, or do they apply to first-nation or aboriginal people or indigenous people, or do they apply to Hispanics, which are, again, other high-risk groups. Or do they apply in health systems, such as China and Japan, which are very different than Canada and the United States, which have different healthcare systems on their own. So, really we needed a global external validation prior to adoption.
So, we did this large study. It was a multi-national assessment of accuracy of equations for predicting the risk of kidney failure. This was published in 2016, and it included more than 700,000 participants across 30 countries and five continents. In every comprehensive way — you see the countries in yellow highlighted, and those are all the countries represented — and as you can see, short of the continent of Africa, where we only have South Africa, we have extremely wide coverage globally. And the equation’s been validated with the Kidney Failure Risk Equation’s three-, four-, and eight-variable models. In the initial work they had C statistics between 0.84 and 0.91 — three-variable model being age, sex, and eGFR; four-variable adding ACR, and the eight-variable adding calcium, phosphorus, bicarbonate, and albumin — and all of these models were then tested and proven highly accurate in these more than 700,000 individuals across 30 countries.
More than 90% accurate
[This figure is] representative of that. Here we have C statistics, or discrimination statistics, of the four-variable equation — that’s age, sex, GFR, and ACR — across 17 North American cohorts. The top one there is AASK, or the African-American Study of Kidney Disease, and the bottom one there is VA, or the Veterans Affairs healthcare system. As you can see, across the board our overall accuracy is 90%. That’s really quite remarkable. That means that if you take two randomly selected patients with CKD, the model will assign a higher probability of failure to the person who progresses 90% of the time or more. When I contrast this with C statistics from other models that you might be aware of, such as the Framingham Study Equation, or the CHADS2 score, they’re all around 70 – 72%. So really we’ve developed a highly accurate model that physicians, patients, and administrators can rely on to accurately predict the risk of kidney failure.
See Dr. Tangri’s entire presentation here: Revolutionizing Renal Care with Predictive Analytics for CKD.