Prior drug allergies are associated with worse outcome in patients with idiosyncratic drug-induced liver injury: A machine learning approach for risk stratification

Abstract

The impact of prior drug allergies (PDA) on the clinical features and outcomes of patients who develop idiosyncratic drug-induced liver injury (DILI) is largely unknown. We aimed to assess the clinical presentation and outcomes of DILI patients based on the presence or absence of PDA and explore the association between culprit drugs responsible for DILI and allergy. We analysed a well-vetted cohort of DILI cases enrolled from the Spanish DILI Registry. Bootstrap-enhanced least absolute shrinkage operator procedure was used in variable selection, and a multivariable logistic model was fitted to predict poor outcomes in DILI. Of 912 cases with a first episode of DILI, 61 (6.7%) had documented PDA. Patients with PDA were older (p = 0.009), had higher aspartate aminotransferase (AST) levels (p = 0.047), lower platelet count (p = 0.011) and higher liver-related mortality than those without a history of drug allergies (11% vs. 1.6%, p < 0.001). Penicillin was the most common drug associated with PDA in DILI patients (32%). A model including PDA, nR-based type of liver injury, female sex, AST, total bilirubin, and platelet count showed an excellent performance in predicting poor outcome in patients from the Spanish DILI Registry (area under the ROC curve [AUC] 0.887; 95% confidence interval [CI] 0.794 - 0.981) and the LATINDILI Network (AUC 0.932; 95% CI 0.884 - 0.981). Patients with suspected DILI should be screened for PDA as they would require a close monitoring for early detection of worsening clinical course.

Keywords: Acute liver failure; Allergy; Drug-induced liver injury; Machine learning; Prognosis.