A novel test that estimates the risk of a patient with type 2 diabetes developing end-stage kidney disease significantly outperformed more conventional risk-prediction methods in studies that used data from more than 1100 patients.
The new tool, KidneyIntelX, is intended to identify patients with type 2 diabetes who are at high risk for progressive diabetic kidney disease in order that they can receive more aggressive renal-protective treatment. It is anticipated that the test would be of most use to primary care physicians.
KidneyIntelX is an in vitro test that uses a blood draw and medical record information. It utilizes a proprietary combination of three biomarkers and clinical data, which are analyzed by a machine-learning program, the company announced. The test has been given a breakthrough device designation by the US Food and Drug Administration and has been approved for commercial use in New York and California.
Details of the performance of KidneyIntelX using clinical specimens and data collected in two US biobanks appear in a report published April 2 in Diabetologia.
The company that markets KidneyIntelX, Renalytix AI, sponsored a survey of 401 US primary care physicians who see significant numbers of patients with type 2 diabetes. The survey gauged the physicians’ willingness to use a tool such as KidneyIntelX. The results were reported separately by Manasi Datar, PhD, in a talk during the National Kidney Foundation (NKF) 2021 Spring Clinical Meetings.
Predicting Progressive Decline of Renal Function
In the report published in Diabetologia, the overall dataset included 1146 patients with type 2 diabetes and associated kidney disease. The investigators divided the cohort into a derivation subgroup of 686 people and a validation subset of 460 people.
On the basis of findings from the derivation phase, the researchers divided the validation subgroup with respect to the predicted risk of their experiencing progression of kidney disease. Of the validation-group patients, 46% patients were considered to be at low risk, 37% at intermediate risk, and 17% at high risk.
During a median 4.3 years of follow-up, 241 patients (21% of the total 1146 dataset) experienced progressive decline of renal function, the endpoint that the developers designed the tool to predict.
Among patients in the low-risk subgroup of the validation set, the KidneyIntelX tool had a 90% negative predictive value (NPV) for correctly identifying patients whose condition would not ultimately progress. By comparison, the NPV of a control prognostic formula that relies on conventional analysis of clinical metrics and does not include the three biomarkers used in KidneyIntelX was 88%. This was a nonsignificant difference for one of the study’s primary endpoints.
Among patients in the high-risk subgroup of the validation set, the positive predictive value (PPV) for identifying patients who would later develop progressive renal dysfunction was 61% with the novel tool and 37% with the conventional tool, a significant difference for the study’s prespecified second primary endpoint, report Lili Chan, MD, a nephrologist at Mount Sinai Hospital, New York City, and co-authors.
The study’s third primary outcome was the area under the receiver operating characteristic curve (AUC, also known as the C-statistic), a measure of overall prognostic performance. In both the derivation phase and the validation phase, the mean AUC was 0.77. By comparison, the AUC for the conventional prognostic tool was 0.62 in the derivation phase and 0.61 in the validation phase.
The PPV and AUC results document “marked improvements in discrimination” over the comparator prognostic algorithm, conclude the report’s authors.
Additional results also showed that KidneyIntelX outperformed KDIGO, a simpler method of conventional risk stratification for progressive diabetic kidney disease based only on estimated glomerular filtration rate and degree of albuminuria.
“Somewhat Promising Numbers”
Among nephrologists not involved with this work, assessments of the findings were mixed.
The results were “reasonably good,” with the 0.77 AUC falling just shy of the level of 0.80 that “is generally considered good,” commented Carl P. Walther, MD, a nephrologist at Baylor College of Medicine, in Houston, Texas. An AUC of 1.0 corresponds to perfect performance, and an AUC of 0.50 corresponds to performance that is no better than a coin flip.
“These are somewhat promising numbers, and certainly better than the comparison tests they looked at,” Walther said in an interview.
A colleague of his at Baylor College of Medicine was more positive in his assessment.
The results “are exciting. This is a good start. This is what we are looking for,” commented Sankar D. Navaneethan, MD, a nephrologist and professor of medicine at Baylor College of Medicine.
The KidneyIntelX results “look better” in comparison with the usual algorithm currently used to assess risk for progression of diabetic kidney disease, Navaneethan added.
Interest Voiced by Primary Care Physicians
The test’s developers foresee the biggest role for KidneyIntelX in primary care.
“Low risk patients with diabetic kidney disease can continue with their existing providers and require less intense treatment,” write the authors.
Patients identified as being at high risk are prime candidates for referral to nephrologists and other specialists, such as dieticians, and for more intense monitoring and extensive medical treatment.
In the survey reported by Datar, 98% of the 401 primary care doctors said they would be extremely likely, very likely, or somewhat likely to order the KidneyIntelX test for some of their patients with type 2 diabetes.
The results also showed that roughly two thirds of the respondents agreed that it is currently difficult to detect diabetic kidney disease in its early stages and that no test (excluding KidneyIntelX) is currently available to accurately predict whether a patient will experience a decline in renal function, reported Datar, a researcher with Boston Healthcare Associates, a consulting company.
Of note, respondents of the survey were instructed to assume that the cost of the test was not a problem and that the test had been cleared by the FDA and adopted by the physicians’ practice administrators.
Early Prognostication “Could Be a Game Changer,” but Current Data Are Limited
“If patients who are highest risk for complications of kidney disease could be identified early and targeted with intensive kidney-directed therapy, it could be a game changer,” noted Walther.
“Newer kidney biomarkers [such as those utilized in KidneyIntelX] in combination with albuminuria measurements will likely enable targeted early interventions and kidney health preservation,” he said in an interview.
Navaneethan agreed that the new tool “definitely addresses an unmet need” and “would be helpful” for primary care physicians.
KidneyIntelX needs to be tested and validated in more diverse populations before it can be used in routine clinical practice, say the two doctors.
“I look forward to seeing how the test performs in other cohorts,” said Walther.
“I hope there are additional studies from other large healthcare systems” to confirm the findings in different patient populations, said Navaneethan.
Walther questioned the logistical requirements of the KidneyIntelX test.
The current algorithm’s reliance on three “nonstandard kidney biomarkers” makes “already incredibly complex primary care visits” even more complicated, said Walther. He suggests that a better algorithm would focus on more conventional clinical metrics already in the electronic health record.
The KidneyIntelX derivation and validation studies and the survey were sponsored by Renalytix AI, the company that markets the test. Chan has received consulting fees from GLG. Walther has disclosed no relevant financial relationships. Navaneethan has received personal fees from Bayer, Boehringer Ingelhein, REATA, and Tricida and research support from Keryx. Datar has received research support from Renalytix AI.
National Kidney Foundation (NKF) 2021 Spring Clinical Meetings: Abstract 327. Presented April 7, 2021.