Purpose: Despite effective antiretroviral therapy, rates of end-stage liver disease (ESLD) remain high. It is not clear whether contemporary antiretrovirals contribute to the risk of ESLD.
Methods: We included patients from cohorts with validated ESLD data in the North American AIDS Cohort Collaboration on Research and Design. Patients had to initiate antiretroviral therapy after 1 January 2004 with a nucleos(t)ide backbone of either abacavir/lamivudine or tenofovir/emtricitabine and a contemporary third (anchor) drug. Patients were followed until a first ESLD event, death, end of a cohort's ESLD validation period, loss to follow-up or 31 December 2015. We estimated associations between cumulative exposure to each drug and ESLD using a hierarchical Bayesian survival model with weakly informative prior distributions.
Results: Among 10 564 patients included from 12 cohorts, 62 had an ESLD event. Of the nine anchor drugs, boosted protease inhibitors atazanavir and darunavir had the strongest signals for ESLD, with increasing hazard ratios (HR) and narrowing credible intervals (CrI), from a prior HR of 1.5 (95% CrI 0.32-7.1) per 5 year's exposure to posterior HRs respectively of 1.8 (95% CrI 0.82-3.9) and 2.0 (95% CrI 0.86-4.7). Both backbones and efavirenz showed no signal. Hepatitis C coinfection was the most important covariate risk factor (HR 4.4, 95% CrI 2.6-7.0).
Conclusions: While contemporary antiretrovirals pose less risk for ESLD than hepatitis coinfection, atazanavir and darunavir had a toxicity signal. We show how hierarchical Bayesian modelling can be used to detect toxicity signals in cohort event monitoring data even with complex treatments and few events.
Keywords: Bayesian statistical methods; HIV; adverse events; end-stage liver disease; hepatocellular carcinoma; pharmacovigilance.
© 2021 John Wiley & Sons Ltd.