Program

PO4-9-13

Data mining techniques to identify potential clinical presentation modulators in drug-induced liver injury

[Speaker] MI Lucena:1
[Co-author] A Gonzalez-Jimenez:1, K Mceuen:2, M Robles-Diaz:1, I Medina-Caliz:1, A Cueto-Sanchez:1, M Chen:2, A Suzuki:3, C Stephens:1, R J Andrade:1
1:UGC Aparato Digestivo y Servicio de Farmacologia Clinica, Instituto de Investigacion Biomedica de Malaga (IBIMA), Hospital Universitario Virgen de la Victoria, Universidad de Malaga, CIBERehd, Malaga, Spain, 2:Division of Bioinformatics and Biostatistics, National Center for Toxicological Research, US Food and Drug Administration, Jefferson, AR, USA, 3:Gastroenterology, Durham VA Medical Center, Duke University, Durham, NC, USA

Background: Drug-induced liver injury (DILI) often presents with diverse clinical phenotypes even when the same causative agent is involved, while some drugs cause a more distinct signature (e.g., acute hepatocellular injury by acetaminophen; acute cholestasis by anabolic steroids). The clinical phenotype diversity associated with most causal drugs may be partly explained by multi-phasic interplays between drug and host properties. We aimed to explore and characterize drug-host interplay in initial biochemical presentations in a large well-characterized Spanish DILI patient population.
Methods: We analyzed 695 cases enrolled in the Spanish DILI registry. Information on demography, causal drugs, comorbidities, co-medications, and clinical manifestations including initial biochemical presentations, was collected at the time of study enrollment. Information on causative drug properties was obtained from the literature as well as the Liver Toxicity Knowledge Base. Cases manifested as mixed biochemical injury pattern were excluded from this analysis. Logistic regression model (hepatocellular vs. cholesteric injury) was used to assess each host/drug factor pair along with their interaction term. Once a statistically significant interaction term was identified, separate logistic regression models were applied to determine whether a divergent relationship between liver injury and drug property exists among host factor subgroups.
Results: A total of 68 host factors and 71 drug properties were analyzed. Among 4828 host/drug factor pairs assessed, 26 pairs were identified as statistically significant interactions. The identified pairs included, but were not limited to, age-hepatic metabolism, hyperlipidemia-reactive metabolite, and allergic records-drug electronegativity. Further subgroup analyses revealed that older age (>60 years) was associated with cholestatic injury only for drugs which are not significantly metabolized in the liver (OR=3.9, 95%CI=2.0-7.7, p=0.0001), such as amoxicillin-clavulanate, nitrofurantoin or methotrexate, but not for drugs with significant hepatic metabolism (OR=1.6, 95%CI=0.9-2.9, p=0.09), such as isoniazid or flutamide.
Conclusions: Data-mining analysis using logistic regression models with an interaction term identified multiple potential combinations of host/drug factors, which may interact and modify initial biochemical presentations in DILI. The identified interactions may aid in better understanding the heterogeneity of clinical presentations in DILI cases. Further methodological implementation to discover latent associations among multiple factors (>2 factors) are warranted.
Funding: AEMPS, FEDER-PI15/01440, CIBERehd-ISCIII

Advanced Search