New Equations Predict Chronic Kidney Disease Risk

By Will Boggs MD

November 15, 2019

NEW YORK (Reuters Health) - Two new equations from the CKD Prognosis Consortium, one for diabetics and one for non-diabetics, accurately predict the risk of developing chronic kidney disease (CKD) over the next five years.

The lifetime risk of CKD in the US is estimated to be 59.1%. Identification of individuals at increased risk of CKD could allow more targeted surveillance strategies and possibly better management of risk factors, but relatively little work has been done to develop predictive tools to identify these individuals.

Dr. Robert G. Nelson from the CKD section of the National Institute of Diabetes and Digestive and Kidney Diseases, Phoenix, Arizona and colleagues used data from multinational cohorts from 28 countries to develop and evaluate risk prediction equations for CKD as defined by an incident estimated GFR (eGFR) below 60 mL/minute/1.73 m2.

They also compared the new models with two published equations: the Chien equation, developed from 5168 Chinese individuals, which included age, BMI, diastolic blood pressure, and history of type 2 diabetes and stroke; and the O'Seaghdha model, developed in the predominantly white population of Framingham, Massachusetts, which included age, hypertension, diabetes, eGFR, and albuminuria values.

Major risk factors for incident CKD included increasing age, female sex, black race, lower baseline eGFR values, higher baseline urine albumin:creatinine ratio, higher BMI, and hypertension in cohorts with and without diabetes. Additional risk factors in the diabetic cohorts included higher baseline hemoglobin A1c and the use of insulin versus oral diabetes medication.

The absolute risk of CKD was generally higher among persons with vs without diabetes and among older patients regardless of diabetes status, researchers reported November 8th online in JAMA and at the Kidney Week meeting in Washington, DC.

The models yielded good discrimination for the five-year predicted probability of incident CKD, with median C statistics of 0.845 in cohorts without diabetes and 0.801 in cohorts with diabetes.

Median C statistics were similar in 9 external validation cohorts (0.84 in the population without diabetes and 0.81 the population with diabetes).

In the absence of diabetes, the newly developed model had better discrimination than the Chien equation (0.094 higher C statistic) and the O'Seaghdha equation (0.020 higher C statistic). The new model also outperformed the Chien equation (0.107 higher C statistic) and the O'Seaghdha equation (0.037 higher C statistic) in the presence of diabetes.

"Further study is needed to determine whether use of these equations to identify individuals at risk of developing chronic kidney disease will improve clinical care and patient outcomes," the researchers note.

Dr. Michelle M. Estrella from University of California, San Francisco and San Francisco VA Health Care System, San Francisco, California, who co-authored an editorial related to this report, told Reuters Health in an email, "One of the most interesting finding for me was the equations' ability to distinguish patients at high versus low risk in cohorts of patients which were distinct from the cohorts used to develop the equation."

"The risk prediction equations did not utilize machine-learning methods, which are often viewed as a 'black box' to most clinicians; rather, the equations included factors which could be feasibly captured from health records," she said. "In a way, this approach will ease the equations' dissemination into day-to-day clinical practice."

"Kidney disease carries a tremendous health and financial burden," Dr. Estrella said. "Prevention of kidney disease through effective management of its risk factors is of the utmost importance to substantially lowering this burden. However, to enhance and focus our prevention efforts, we first need a way to predict which of our patients at highest risk of kidney disease. These new equations could facilitate these efforts."

She added, "With the recent Executive Order focused on improving kidney health within the U.S., we are poised to really shift clinical care and research efforts towards prevention and early detection and management of kidney disease. To successfully take advantage of these exciting times, buy-in and collaboration among multiple disciplines, particularly primary physicians and nephrologists, are absolutely critical."

Dr. Chava Ramspek from Leiden University Medical Center in The Netherlands, who recently reviewed models predicting kidney failure in CKD patients, told Reuters Health by email, "I believe the prediction models for patients with diabetes are of particular interest for clinicians. They were developed and validated on extremely large cohorts and showed good discriminatory performance (ability to distinguish those who will develop CKD from those who will not). As the prevalence of diabetes is rising, the importance of identifying patients at high risk of developing complications is important for prevention strategies as well as adequate allocation of resources."

"The most elegant form of use would be that these risk predictions are automatically calculated in the electronic health records," she said. "Counseling on risks and preventative strategies could be employed for patients with an increased risk of CKD. However, as physicians receive more and more data on their patients to consider, it remains to be seen if this calculation would truly improve clinical practice and patient outcomes. Before effort is put towards implementation it is of importance to perform an impact study."

Dr. Nelson was unable to provide comments in time for publication.


JAMA 2019.