Researchers have validated a software algorithm that can automatically scan through electronic medical records to identify people with chronic kidney disease (CKD) and classify them by CKD severity stage.
In a pair of validation studies the algorithm had a positive predictive value for identifying CKD of 95%-97%, and a negative predictive value of 89%, the investigators report in a recent publication in Nature.
“The algorithm is ready for clinical implementation,” declared Krzysztof Kiryluk, MD, the senior investigator on the project and a nephrologist at Columbia University Medical Center in New York City. “The algorithm has been validated extensively and is ready to use,” he said in an interview.
The software and its documentation sit in an online public catalog, available for free download on the website Phenotype Knowledgebase.
“Any interested health organization could implement the algorithm in their electronic medical records,” says Kiryluk.
Finding “Significantly Underrecognized“ CKD
The algorithm relies on information routinely included in the electronic medical record (EMR) of most patients, with special focus on two key kidney-related metrics: serum creatinine to calculate a person’s estimated glomerular filtration rate (eGFR), and a measure of proteinuria, such as a urinary albumin-to-creatinine ratio, albumin in a 24-hour urine specimen, or a dipstick protein test result.
“CKD is highly prevalent in adults, but unfortunately it is also significantly underrecognized. An algorithm that can reliably alert clinicians to the diagnosis would lead to improved recognition and is likely to impact patient care by improving management,” adds Sumit Mohan, MD, a coinvestigator on the study and a nephrologist at Columbia.
“We hope there will be widespread use of the algorithm in EMRs. We expect this will be valuable to primary care physicians as well as specialists who treat heart or liver diseases,” Mohan said in an interview.
Another validation reported by the researchers applied the algorithm to the EMRs of more than 1.3 million patients in Columbia’s health system who had at least one serum creatinine measure in their records. This resulted in successful CKD assessment of about half the patients. Assessments could not occur for patients without any entry for urinary protein.
Among the roughly 673,000 patients successfully assessed by the algorithm about 265,000 had some stage of CKD.
Only 45,000 of these patients (17%) had been previously diagnosed with CKD by conventional approaches.
Help for Primary Care Physicians
“Widespread adoption of the algorithm would definitely improve the rate of diagnosis of CKD, mainly by helping primary care physicians to diagnosis it sooner. Earlier diagnosis is even more critical today as more effective preventive strategies for kidney disease are available, such as the sodium-glucose cotransporter 2 (SGLT-2) inhibitors,” Kiryluk said.
The algorithm developed by the Columbia researchers is distinct from another new algorithm that assesses certain patients with CKD, the commercialized KidneyIntelX algorithm.
The two differ in significant ways, explains Kiryluk. The KidneyIntelX algorithm is designed to predict the progression of diabetic kidney disease specifically, while the Columbia algorithm is designed to detect any form of CKD. The KidneyIntelX algorithm uses special biomarkers that are not included in routine clinical care nor in most EMRs.
Perhaps the biggest difference is that KidneyIntelX is proprietary and commercialized; the Columbia algorithm is open access.
Kiryluk and Mohan have reported no relevant financial relationships.
NPJ Digit Med. Published online April 13, 2020. Abstract
Follow Medscape on Facebook, Twitter, Instagram, and YouTube.
Source: Read Full Article