Researchers Use Electronic Diagnostic Model to Predict Acute Interstitial Nephritis in Patients

Researchers from Johns Hopkins Medicine and Yale University have collaborated on the development and application of a diagnostic model to detect acute interstitial nephritis (AIN) in patients, which could have a lasting impact on getting patients diagnosed earlier.

AIN is a common cause of acute kidney injury, or AKI, marked by swelling and inflammation of certain kidney tissues, and generally linked to use of medicines such as steroids, proton pump inhibitors and antibiotics. Early diagnosis and treatment are associated with reduced risk of permanent kidney damage in hospitalized patients.

The findings, which build on the researchers' prior development of the diagnostic model, are published in the Journal of the American Society of Nephrology. The model was developed using a panel of lab tests documented in electronic medical records.

Studies have shown that sudden loss of kidney function—also known as AKI—affects one in five hospitalized patients, according to the American Kidney Fund. One major challenge for clinicians caring for patients with AKI is differentiating AIN from other causes of AKI. This is largely due to more than 90% of patients with AIN having no signs or symptoms, and the fact that individual clinically available tests, such as urine eosinophils, urine microscopy, and imaging tests have poor accuracy for diagnosis of AIN.

Incorrectly diagnosing a patient with AIN could lead to a discontinuation of lifesaving therapies such as immune checkpoint inhibitors or antibiotics. Furthermore, a delayed or missed diagnosis could potentially lead to permanent kidney damage.

"AIN is one of the treatable causes of AKI and it is imperative that we identify these cases early in the course of disease," says Chirag Parikh, M.D., Ph.D., professor of medicine and the director of the Division of Nephrology at Johns Hopkins Medicine and key investigator for the study.

With AIN being so challenging to diagnose, often times, a kidney biopsy is required, which can be invasive and carry its own associated risks. That said, researchers have often attempted to determine how to diagnose AIN using alternative methods.

In this study, researchers developed a diagnostic model to predict AIN in patients by using a machine-learning selection technique called least absolute shrinkage and selection operator (LASSO). The laboratory tests using the LASSO feature included serum creatinine, blood urea nitrogen (BUN), urine protein, and the density of urine compared to water, which is known as a specific gravity test. .

"Although these laboratory tests were identified by machine learning from over 150 variables, it is interesting that these tests make biological sense as they differentiate AIN from other causes such as prerenal azotemia and acute tubular necrosis," Parikh says.

The study consisted of two patient cohorts. Both cohorts had previously undergone kidney biopsy procedures at either The Johns Hopkins Hospital (JHH) or Yale University. The JHH cohort identified 1,454 patients who underwent a native kidney biopsy between January 2019 and December 2022, while the Yale cohort examined 528 patients who were scheduled to undergo a clinical kidney biopsy between July 2020 and June 2023.

Both cohorts excluded patients who did not have a serum creatinine value within one year before the biopsy; were undergoing kidney allograft biopsy or being evaluated for kidney mass; and with known vasculitis or lupus nephritis before the biopsy.

In total, 1,982 patients were examined and a combined 22% of them were found to have been diagnosed with AIN. In both cohorts, those with AIN were more likely to be hospitalized, and had higher serum creatinine and blood urea nitrogen to creatinine ratio. While the diagnostic model was able to improve the accuracy of AIN diagnosis for clinicians to 77%, there were noted differences in the prevalence of AIN in both patient cohorts. After accounting for prevalence at the center, the calibration of the diagnostic model improved significantly, which led to more accurate AIN diagnoses.

The formula for predicting AIN is now available in MDCalc.

Researchers hope that with these study findings, the AIN diagnostic model could guide the decision of whether to perform a kidney biopsy in patients with AKI. This model could also be incorporated into clinicians' decisions on choosing the most appropriate treatment for patients with AIN.