Prediction of lupus nephritis treatment response may help doctors and patients preserve precious kidney function

Smiling African American woman gazes out window and into distance
credit: iStock

New web-based application adds power of machine learning to the SLE treatment toolkit

by Shawn Oberrath

While systemic lupus erythematosus (SLE) manifests in many ways and affects all organ systems, the complication of lupus nephritis is particularly devastating. This inflammatory kidney disease occurs in about half of patients with SLE, and nearly half of lupus nephritis patients will develop chronic and potentially end-stage kidney disease that requires dialysis or kidney transplant.

Many factors contribute to lupus nephritis, but most patients have lesions with inflammatory cells that invade the capillaries in the kidney, blocking these vessels and leading to kidney dysfunction. This damage can cause irreversible scarring and fibrosis, which in turn may lead to other health problems, such as heart attacks, strokes, infections and even death.

The standard treatment for someone diagnosed with lupus nephritis is one of two immunosuppressive drugs given as a trial for 6 months. At that point, the patient’s disease is assessed to see if they are responding to treatment. If they are not, the FDA has approved several second-line drugs that can be used instead, but these are expensive and are usually reserved for patients who have not responded to the full initial trial.

The challenge for clinicians is deciding when to add second-line drugs and other resources to help patients whose nephritis is not responsive to the initial treatment, because irreversible kidney damage can occur during the trial period.

To help physicians with this complex decision, Jim Oates, M.D., a professor of medicine and Director of the Division of Rheumatology and Immunology at MUSC Health, and his colleagues published a study in Lupus Science and Medicine along with a companion podcast episode to share a novel prediction tool created in their laboratory.

Because resources in rheumatology are scarce, be they expensive medicines, care coordinators or physician time, it is important to deploy those resources strategically.

“We want to find the patients who are most at risk for poor outcomes but whose outcomes can be reversed,” said Oates.

 

 

Screenshot of a computer program with test levels 
New, web-based clinical tool gives physicians streamlined prediction model for lupus nephritis treatment response. The tool needs further validation before it can be used for treatment decision-making, but it can indicate the likelihood that a patient will not respond well to treatment by the 1-year point. credit: Oates laboratory, MUSC

 

Because there is no single indicator that shows if a patient’s disease will respond to therapy, Oates and his colleagues developed the new tool by considering many variables and disease indicators. In a rheumatology clinic where physicians see lupus patients day in and day out, the specialists look at the whole picture to decide on treatment and how to monitor patients. The tool mirrors that holistic perspective and may provide a useful clinical aid for clinicians who do not encounter SLE and lupus nephritis as often.

The prediction tool is a web-based application with a simple interface that allows health care providers to submit specific pathology and laboratory values and then see an output that shows the likelihood that a patient will not respond well to therapy over the course of 1 year.

The MUSC team created the tool by looking at kidney biopsy indicator values and other selected laboratory values and then using a variety of machine learning models to test which indicators were most useful for prediction. For testing purposes, the study used values from 83 MUSC Health lupus patients who had renal biopsy information and 1-year follow-up data available. These patients were representative of the population at that clinic and were mainly African American women. While the tool is certainly relevant to this population, more testing is needed to make sure that it applies to other populations as well.

The research team narrowed the testing results to seven key indicators that they used to develop the web-based tool. These are the International Society of Nephrology/Renal Pathology Society biopsy scores for activity, chronicity, interstitial fibrosis and interstitial inflammation as well as the urine protein-to-creatinine ratio, white blood cell count and hemoglobin level. Based on these values, the tool predicts the likelihood that a patient will not respond to treatment, with statistical diagrams and values included to guide the clinician.

Because of the small test population and the retrospective nature of the study, Oates cautions that clinicians should not make treatment decisions based on the results until the tool is further validated. But physicians may still find the tool useful in the meantime. For example, if the results indicate a high chance of treatment failure for some patients, they may choose to monitor these patients more frequently or engage a care coordinator or social worker to make sure that patients are taking the medications, can afford them and can tolerate them.

“One of the things that just breaks my heart is when I prescribe something and a barrier to medication adherence pops up, unbeknownst to me,” said Oates. “When I discover that barrier months later, it is possible that irreversible damage has already occurred.”

Using the tool can help physicians understand which patients will benefit from extra assistance or more frequent monitoring. The ultimate goal for helping lupus nephritis patients is to preserve as much kidney function as possible. Knowing that a patient is not likely to respond to the initial therapy may mean lowering the threshold for when to change therapy, perhaps not waiting the full 6 months of a standard trial.

Now that the tool is established and validated with the initial data, Oates and his colleagues plan to expand the validation to different populations at both MUSC and other institutions. The laboratory is working on other prediction tools as well, but some of the biomarkers in those studies are not widely available or involve genetic markers that need further study.

Oates is gratified that even at this stage the new tool can begin to give immediate benefits. “I delved into making this tool because the variables are available and can make an impact now,” he said. “And I’ve been amazed by the high percentage of lupus patients who agreed to be part of this long-term study to help their fellow patients. They really make this work possible.”