(A-232) High-throughput screening using self-controlled tree-based scan statistics to identify medications potentially associated with hospitalization for severe acute liver injury

Background: Drugs associated with acute liver injury (ALI) are primarily identified by case reports. No systematic approach has been used to identify hepatotoxic drugs. High-throughput screening of real-world data could detect hepatotoxicity signals from drug-specific effects.

Objectives: To apply tree-based scan statistics in real-world data to identify drugs associated with hospitalization for severe ALI among patients without liver/biliary disease.

Methods: We implemented a self-controlled case-crossover design in a sample of patients without liver/biliary disease diagnoses who had a hospitalization for severe ALI in the Veterans Health Administration between April 1, 2000 and April 30, 2023. Hospitalization for severe ALI was defined by either inpatient: 1) alanine aminotransferase >120 U/L + total bilirubin >2.0 mg/dL, or 2) international normalized ratio ≥1.5 + total bilirubin >2.0 mg/dL, recorded within the first two days of admission. All drugs newly dispensed within 365 days prior to a hospitalization for severe ALI were identified by the Anatomic Therapeutic Classification System. We used the conditional Bernoulli tree-based scan statistics for fixed risk window analyses (1-133 days after each incident drug fill). Monte Carlo simulation was used to estimate log-likelihood ratio test statistics and p-values. We classified drugs with p< 0.3 as “alerts,” indicating potential association with severe ALI. We performed sensitivity analyses to enhance our ability to identify alerts, whereby we repeated the analysis: 1) evaluating risk windows of 1-28 and 1-84 days, and 2) removing drugs known to be potential treatments for ALI symptoms (i.e., antiemetics, antispasmodics).

Results: We identified 12,860 patients without liver/biliary disease who had a hospitalization for severe ALI. Among 450 incident drugs screened for possible association with severe ALI, those identified as alerts were from the following classes: acid-suppressives (ranitidine [p < 0.001], omeprazole [p=0.004]), antiemetics (ondansetron [p < 0.001], promethazine [p=0.06]), antibiotics (amoxicillin/clavulanate [p=0.008], ciprofloxacin [p=0.02], mupirocin [p=0.032], ethambutol [p=0.275]), anticoagulants (heparin [p=0.015]), and chemotherapy (pazopanib [p=0.275]). Sensitivity analyses evaluating shorter risk windows additionally identified hydrocodone/acetaminophen (1-84 days: p=0.005; 1-28 days: p< 0.001), and those conducted after removing possible ALI treatments also identified dicyclomine (p=0.15), prochlorperazine (p=0.17), and isoniazid (p=0.28).

Conclusions: High-throughput screening using tree-based scan statistics detected potentially hepatotoxic drugs for investigation in future pharmacoepidemiology studies.