External Validation of an Electronic Health Record–Based Diagnostic Model for Histological Acute Tubulointerstitial Nephritis

What is it about?

Acute interstitial nephritis (AIN) is a treatable cause of acute kidney injury but requires a kidney biopsy for diagnosis. Incorrect assumption of AIN leads to discontinuation of potentially life saving medications such as antibiotics and cancer immunotherapy whereas missed or delayed diagnosis can result in permanent kidney damage. We previously developed a diagnostic model for AIN that provided probability of AIN on biopsy by combining four clinical tests. Here we validate this model in two cohorts showing that the model retained its ability to differentiate between those with and without AIN. We noted that the model predictions around exact probabilities were inaccurate; however, upon applying a correction factor we noted significant improvement in accuracy of these predicted probabilities. The model had a high negative predictive value meaning that it could be used to rule out AIN allowing continuation of life-saving medications. When combined with clinicians’ prebiopsy suspicion of AIN, this model also showed a high positive predictive value meaning that it could be used to diagnose and treat AIN in a subset of patients.

Featured Image