|Subject: Machine Learning Derived Probability Score for Rapid Kidney Function Decline|
|Document #: LAB.00041||Publish Date: 07/06/2022|
|Status: Reviewed||Last Review Date: 05/12/2022|
This document addresses the use of a machine learning derived probability score (i.e., artificial intelligence) which may combine a variety of clinical characteristics such as, biomarkers, genetics, gender or race, to generate prognostic information with the end-goal of facilitating a more personalized approach to the management of chronic kidney disease (e.g., KidneyIntelX™). This document does not address the standard use of blood-based biomarkers, estimated glomerular filtration rate (eGFR) or urinary albumin and creatinine levels in the prognostic evaluation of newly diagnosed kidney disease.
Investigational and Not Medically Necessary:
Use of a machine learning derived probability score (e.g., KidneyIntelX) to predict rapid kidney function decline in chronic kidney disease is considered investigational and not medically necessary for all indications.
Chronic kidney disease (CKD) is defined by the Kidney Disease Improving Global Outcomes (KDIGO) organization as abnormalities of kidney structure or function, present for > 3 months. In the KDIGO Clinical Practice Guidelines for the Evaluation and Management of Chronic Kidney Disease, factors associated with CKD progression to inform prognosis include the etiology of CKD (e.g., diabetes, hypertension, etc.), level of GFR, level of albuminuria, age, sex, race/ethnicity, elevated blood pressure, hyperglycemia, dyslipidemia, smoking, obesity, history of cardiovascular disease and ongoing exposure to nephrotoxic agents (ungraded recommendation; Stevens, 2012). A standardized system for integrating sociodemographic risk factors with clinically relevant biomarkers to accurately identify those most at risk for progression is not yet available in most practice settings, potentially hampering clinicians’ timely intervention in CKD management. Recently, the use of machine learning approaches that can combine biomarkers and electronic health record data to produce prognostic risk scores have been explored. One such approach is the KidneyIntelX, a proprietary artificial intelligence-enabled algorithm which combines blood-based biomarkers, genetics and personalized data from electronic health records to generate a unique risk score which is then used to develop a prediction of progressive kidney function decline in diabetes-related CKD.
To date, the only published peer-reviewed study evaluating the clinical utility of KidneyIntelX is a large retrospective cohort by Chan and colleagues (2020), which enrolled 1146 individuals with diabetes-related CKD age 21-81. During the follow-up period (median of 4.3 years) 241 study enrollees (21%) experienced progressive decline in kidney function. KidneyIntelX stratified 46%, 37% and 16.5% of validation cohort (n=460) into low-, intermediate- and high-risk groups, respectively, with a positive predictive value (PPV) of 62% (PPV of 37% for the clinical model and 40% for KDIGO; p<0.001) in the high-risk group and a negative predictive value (NPV) of 91% in the low-risk group. The net reclassification index for events into the high-risk group was 41% (p<0.05). In this retrospective, exploratory validation study, KidneyIntelX scores accurately classified more cases into the KidneyIntelX-defined low, intermediate and high-risk strata (p-value<0.05) relative to KDIGO risk strata. The study authors conclude, “A machine learned model combining plasma biomarkers and EHR [electronic health record] data improved prediction of progressive decline in kidney function within 5 years over KDIGO and standard clinical models in patients with early DKD [diabetes-related CKD].” Given the retrospective study design and marginal statistical significance, further investigation in the setting of a large, ideally randomized, trial is warranted to establish whether use of KidneyIntelX materially improves net health outcomes compared to established alternatives, such as the KDIGO guideline’s specified sociodemographic risk factors, pertinent health history and clinically relevant biomarkers.
In 2022, Lam and colleagues published results from a retrospective study of samples collected during conduction of a prospective randomized controlled trial, CANagliflozin cardioVascular Assessment Study (CANVAS). In total, 1,325 CANVAS participants with diabetic kidney disease and baseline plasma samples were enrolled into the study. Kidney IntelX risk scores were generated from the available samples at baseline and years 1, 3, and 6 of study follow-up. The study’s primary aim was to assess the association of changes from baseline in KidneyIntelX scores with progression of diabetic kidney disease; composite outcomes included, (1) rapid kidney function decline, (2) a sustained 40% decline in eGFR, or (3) kidney failure. During the mean follow-up of 5.6 years, 131 study participants (9.9%) experienced a composite kidney outcome. Using risk cutoffs established from previous validation studies, KidneyIntelX stratified participants into low- (42%), intermediate- (44%), and high-risk (15%) groups with cumulative incidence for the outcomes of 3%, 11%, and 26%, respectively (risk ratio=8.4; 95% confidence interval [CI], 5.0-14.2) for the high-risk versus low-risk groups. Changes in KidneyIntelX score within the first year were significantly associated with future risk of a composite outcome (odds ratio [per 10 unit decrease]=0.80; 95% CI, 0.77, 0.83; p < 0.001). Study authors conclude that “KidneyIntelX risk-stratified a large multinational external cohort for progression of DKD [diabetic kidney disease]…”. Given the retrospective design and unclear clinical significance of these findings, further study is warranted to determine the impact of KidenyIntelX on net health outcomes.
Currently, there is no guidance from specialty medical societies addressing the use of machine learning to generate prognostic information in the treatment of CKD. The published peer-reviewed medical literature has not established KidneyIntelX, or any technology like it, as a proven method that materially improves net health outcomes nor has any benefit been established beyond currently available alternatives (e.g. KDIGO guidelines).
In 2019, approximately 37 million Americans reportedly had chronic kidney disease (CKD), with nearly 118,000 requiring initiation of treatment for kidney failure, also known as end stage renal disease (ESRD). There was a steady rise in the rate of ESRD from 1980 to 2011, since then, the incidence rate of ESRD has started to decline. The most prevalent cause of kidney disease is diabetes, which accounts for approximately 38% of ESRD cases (CDC, 2019). On average, 50,000 individuals with diabetic kidney disease progress to kidney failure annually in the United States (Chan, 2020).
Predicting which newly diagnosed diabetic kidney disease cases may progress to ESRD has proved challenging for clinicians, potentially resulting in delayed diagnosis of individuals and the subsequent need for life-saving dialysis or kidney transplants. Typically, prognosis is achieved through integration of established sociodemographic risk factors (i.e., smoking, obesity, and race/ethnicity) along with clinically relevant biomarkers, such as glycemic levels, eGFR, and lipid levels. KidneyIntelX is described by the manufacturer (RenalytixAI) as a validated machine-learned, prognostic risk score which combines data from EHRs and circulating biomarkers to predict diabetic kidney disease progression. The risk score is purported to help clinicians manage individuals with diabetes-related CKD in a streamlined fashion with the end goal of slowing the progression of kidney disease and potentially preventing the occurrence of progressive kidney function decline such as kidney failure and the resultant need for long-term dialysis or kidney transplant (Chan, 2020).
Artificial Intelligence (AI): A science of computer simulated thinking processes and human behaviors, which involves computer science, psychology, philosophy and linguistics.
Chronic renal disease: The permanent loss of kidney function.
End stage renal disease: Persistent decline in renal function as documented by falling creatinine clearance in an individual diagnosed with a renal disease whose natural history is progression to renal impairment requiring renal replacement (dialysis or transplant).
Glomerular filtration rate (GFR): A test used to check how well the kidneys are functioning by estimating how much blood passes through the glomeruli each minute.
Glomeruli: A cluster of nerve endings, spores, or small blood vessels, in particular a cluster of capillaries around the end of a kidney tubule, where waste products are filtered from the blood.
The following codes for treatments and procedures applicable to this document are included below for informational purposes. Inclusion or exclusion of a procedure, diagnosis or device code(s) does not constitute or imply member coverage or provider reimbursement policy. Please refer to the member's contract benefits in effect at the time of service to determine coverage or non-coverage of these services as it applies to an individual member.
When services are Investigational and Not Medically Necessary:
For the following procedure code, or when the code describes a procedure indicated in the Position Statement section as investigational and not medically necessary.
Nephrology (chronic kidney disease), multiplex electrochemiluminescent immunoassay (ECLIA) of tumor necrosis factor receptor 1A, receptor superfamily 2 (TNFR1, TNFR2), and kidney injury molecule-1 (KIM-1) combined with longitudinal clinical data, including APOL1 genotype if available, and plasma (isolated fresh or frozen), algorithm reported as probability score for rapid kidney function decline (RKFD)
Peer Reviewed Publications:
Government Agency, Medical Society, and Other Authoritative Publications:
|Websites for Additional Information|
The use of specific product names is illustrative only. It is not intended to be a recommendation of one product over another, and is not intended to represent a complete listing of all products available.
Medical Policy & Technology Assessment Committee (MPTAC) review. Updated Rationale, References and Websites sections.
MPTAC review. Initial document development.
Federal and State law, as well as contract language, including definitions and specific contract provisions/exclusions, take precedence over Medical Policy and must be considered first in determining eligibility for coverage. The member’s contract benefits in effect on the date that services are rendered must be used. Medical Policy, which addresses medical efficacy, should be considered before utilizing medical opinion in adjudication. Medical technology is constantly evolving, and we reserve the right to review and update Medical Policy periodically.
No part of this publication may be reproduced, stored in a retrieval system or transmitted, in any form or by any means, electronic, mechanical, photocopying, or otherwise, without permission from the health plan.
© CPT Only - American Medical Association