Utility of a bayesian-based assesment tool for the early diagnosis of drug induced liver injury

L Llanos1, R Moreu1, T Ortin1, S Pascual2, JF Horga1, P Zapater1. 1Clinical Pharmacology Section, Hospital general Universitario, Alicante, Spain, 2Liver Unit, Hospital General Universitario, Alicante, Spain

Background: The CIOMS algorithm, considered the present gold standard in the diagnosis of drug induced liver injury (DILI), has limited clinical applicability when hepatotoxicity is firstly detected because it was validated to be applied a posteriori. Bayesian-based probabilistic methods can be applied in the initial phases of the diagnostic process, even if case information is incomplete, and are less influenced by observer subjectivity.

Aim: to evaluate the utility of a bayesian-based assessment tool for the early identification of DILI, comparing its performance with the CIOMS algorithm.

Methods: Cases of suspected DILI were identified from inpatients in the Liver Unit of Hospital General Universitario de Alicante. Probability of the causal role of the suspected drug was calculated using a Bayesian approach. A priori probability (PrP) was calculated using the incidence of abnormal liver enzymes in patients receiving a certain drug from placebo randomized controlled clinical trials when available. A Posteriori (PsP) probability was compared with CIOMS algorithm scores at two time-points: detection of DILI (PsP1) and the end of the diagnostic process (PsP2). Primary analysis was to determine the number of registries classified as probable by the CIOMS algorithm that had a PsP1 >0.50.

Results: After screening, 44 patients were included, corresponding to 66 registries (1 registry= 1 patient+1 drug). The early use of a bayesian-based tool for causality assessment in DILI identifies 67% of the cases (20 out of 30 registries) classified as probable by the CIOMS algorithm applied a posteriori. Mean PrP and PsP1 values were significantly higher for registries classified as probable than for possible and unlikely ones.

Conclusion: Bayesian-based probabilistic methods are an alternative tool to CIOMS algorithm in the diagnosis of DILI in the initial phases of the diagnostic process when case information is incomplete.