Issue: June 2021
Disclosures: Goss, Kotanko and Weinsteinreport no relevant financial disclosures.
June 17, 2021
7 min read

Artificial intelligence aids nephrologists in directing kidney care

Issue: June 2021
Disclosures: Goss, Kotanko and Weinsteinreport no relevant financial disclosures.
You've successfully added to your alerts. You will receive an email when new content is published.

Click Here to Manage Email Alerts

We were unable to process your request. Please try again later. If you continue to have this issue please contact

When nephrologists direct care for patients with kidney disease, many comorbid-related issues are addressed in a patient care plan, including fluid overload, access patency, anemia management and phosphorus control.

Artificial intelligence (AI) can not only help manage these conditions, nephrologist and researchers told Nephrology News & Issues, but also minimize the risk of recurrence.

“Fundamentally, a patient’s condition manifests itself to the health care provider (HCP) through data in the broadest sense,” Peter Kotanko, MD, head of research at Renal Research Institute, a subsidiary of Fresenius Medical Care, said. “Think about clinical information, imaging studies, laboratory data, etc. AI methods can be used to analyze and interpret them and complement the HCP’s assessment.”

Part of the aim of AI is to help clinicians look at stresses on body organs in a collective way, Kotanko said. No cell is spared when the kidneys fail.

“Today, we collect data from a wide variety of sources that can be used to create a high-dimensional mathematical representation of a patient, sometimes called a ‘mathematical twin,’” he said.

The notion behind models, Kotanko and others said, is to create mathematical representations of human physiology, showing how organs function together – and how they can fail together. A mathematical twin of body systems can help clinicians develop therapeutic interventions and predict when multiple organs systems fail.

“By harnessing individual data about the physiological makeup of patients’ bodies and predictive simulations, a virtual twin could speed up treatment processes and recovery times,” Atreyi Chakrabarty reported in an article in TechTribe.

Data application

Adam Weinstein, MD, who directs clinical IT services for DaVita Inc., sees AI as an adjunct to how he manages his patients with end-stage kidney disease. “Effective AI can be used to help predict the likelihood of a future event for a patient, such as progression of CKD to ESKD or the likelihood of a hospitalization. But to get the most out of predictions from an AI algorithm, we need clinical experts to interpret that information and apply processes within their practices to drive meaningful patient interventions.” Weinstein told Nephrology News & Issues. “Some practices are using AI in some specific ways; however, not every practice yet has the tools, resources and processes in place to maximally take advantage of the technology.”

AI tools can be used to improve home therapy, Weinstein said.

“DaVita developed a predictive algorithm using AI to help risk-stratify patients treating at home to help them stay successful at home longer. The tool can alert nephrologists and care teams in advance of potentially adverse events, which can assist in timely interventions.”

Machine learning

AI pioneer Alan Turing described the data collection method in 1950 as “the science and engineering of making intelligent machines, especially intelligent computer programs.”

“AI in kidney care and health care uses algorithms and software engineering principles to approximate the decisions made by clinicians in the analysis of complex health care data,” Len A. Usvyat, PhD, and colleagues wrote in a 2019 article on AI in Nephrology News & Issues.

Kotanko described the workings of AI in two integrated steps: machine learning and deep learning.

“Both machine learning (ML) and deep learning (DL) are methods from the larger AI domain,” he said. “One goal of these methods is to identify patterns in data that may come from various sources (eg, clinical records, images).”

He said, “Deep learning is a subset of machine learning, which deploys diverse methods. Many of them [are] well known from classical statistics (eg, regression analysis, decision trees), support vector machines, k-means clustering, and others.”

At Renal Research Institute, Kotanko and others conduct research into the use of DL to assess arteriovenous fistulas and aneurysms, and ML to predict intradialytic hypotension and discern patterns from metabolomic analyses.

“Fresenius Medical Care North America has predictive models that use algorithms that discern which in-center patients are the best candidates for home modalities or which patients currently dialyzed at home are at the highest risk of leaving home and may require some additional interventions,” Kotanko said. “During the COVID-19 pandemic, FMCNA also rolled out a COVID-19 predictive model that helps identify which patients may have an infection but have not yet presented with symptoms in the clinics,” he said.

CKD risk

A machine learning algorithm helped researchers at the Perelman School of Medicine at the University of Pennsylvania categorize patients with CKD into three distinct subgroups. Each subgroup was associated with different risks for disease progression, cardiovascular events and mortality.

“There are clear differences across individuals with CKD that can be captured by a comprehensive examination of phenotypic data, such as laboratory results, medical history, medications and social factors,” Zihe Zheng, MBBS, MHS, and colleagues wrote. “More sophisticated phenotyping may reveal different and more reliable CKD clusters and underlying disease pathologies, which could help better understand different mechanisms for disease pathways and progression.”

A study published in the American Journal of Pathology describes how researchers at Boston University School of Medicine (BUSM) are using AI to predict the grade of interstitial fibrosis and tubular atrophy (IFTA) in patients with kidney dysfunction.

“Having a computer model that can mimic an expert pathologist’s workflow and assess disease grade is an exciting idea because this technology has the potential to increase efficiency in clinical practices,” Vijaya B. Kolachalama, PhD, assistant professor of medicine at BUSM and co-author on the paper, said in a press release.

In the study, an international team of five practicing nephropathologists independently determined IFTA scores on the same set of digitized human kidney biopsies using a web-based software. Their average scores were taken as a reference estimate to build the deep learning model.

Kolachalama believes AI models that can automatically score the extent of chronic damage in the kidney can serve as second opinion tools in clinical practices. “Eventually, it may be possible to use this algorithm to study other organ-specific pathologies focused on evaluating fibrosis,” Kolachalama said in the release. “Such methods may hold the potential to give more reproducible IFTA readings than readings by nephropathologists.”

New AI device

The FDA recently granted breakthrough device designation to RenalytixAI’s artificial intelligence-enabled clinical diagnostic for kidney disease, known as KidneyIntelX.

The company said the device is meant to diagnose and improve clinical management of patients with type 2 diabetes and fast-progressing kidney disease. Complete with machine learning algorithms, the diagnostic will be able to assess the combination of predictive blood-based biomarkers and use electronic health record information to determine if a patient has progressive kidney disease.

According to Thomas Goss, PharMD, of Boston Healthcare Associates, the current method for risk stratification in diabetic kidney disease fails to identify about half of patients who ultimately have rapid kidney function decline. Furthermore, up to two-thirds of these patients initiate dialysis acutely due to a crash which, he argued, is a costly way to initiate dialysis.

“We know that many patients who are at high risk [for rapid kidney function decline and failure] often don’t get referred from primary care physicians to nephrologists early in the stages of the disease when the most opportunity for intervention is available,” he said. “We feel there is an important need for a predictive tool that will help us to identify patients at high risk and improve our intervention strategy.”

Following a hypothetical cohort of 10,000 patients for 5 years, researchers developed a budget impact model to compare potential cost savings with the diagnostic tool (which put patients into risk categories) vs. standard of care. Considered costs included test cost ($950), prescription medications, specialist and associated office visits, hospitalizations and other complications.

In the 5-year period, implementation of the diagnostic test led to a projected savings of $115 million (ranging from $83 million to $130 million). According to Goss, the savings stem from slowed progression, delayed/avoided dialysis and transplants, and fewer crashes.

Patient use of AI

Kotanko said patients can learn how AI enhances self-care.

“Ideally, AI systems should be so user friendly that no formal training is necessary,” Kotanko said. “They should be as seamless as using a smartphone or ordering food delivery.”

The comprehensiveness of AI may allow for generalists to take on the role of specialty care, such as nephrology. Weinstein and Kotanko said nephrologists can play a key role in maintaining ownership of AI-derived data on kidney disease.

“I would be deeply concerned if HCP, including nephrologists, would be replaced by AI. Fortunately, several bodies have developed ethical standards for the use of AI,” Kotanko said.

“There are several fields where AI will allow non-specialists to expand into areas currently reserved to highly trained experts. Image analysis in radiology is an example,” Kotanko said. “I can see situations in clinical medicine, including in nephrology, where AI and mathematical modeling will enable generalists to take care of patients who were in the past exclusively managed by specialists. Such examples are interpretation of lab results and dosing of drugs.

“This trend will be enhanced by workforce shortages, societal shifts and cost pressures on health care systems.”

Weinstein said nephrologists have a key opportunity to get onboard with AI.

“The understanding of AI predictions, the ability to communicate meaningfully with patients, and the processes needed to take action all require specialty training,” Weinstein said.

“AI could reshape the health care landscape in some ways; however, much like other technologies, at the end of the day you need specialists, such as nephrologists, care teams and a robust health care delivery system to take care of people. For the foreseeable future there is no AI algorithm that can counsel patients, help people identify the best choices in the context of their life, and tackle myriad small but important issues patients and their care teams face,” Weinstein said.