Abstract
Background/Aims Drugs are a common cause of acute liver failure. However, confirming the diagnosis of drug-induced liver injury (DILI) is challenging due to its low incidence, lack of diagnostic markers and idiosyncratic nature. Our aim was to develop an automated scoring algorithm to identify potential DILI cases based on the Roussel Uclaf Causality Assessment Method (RUCAM) and test it in healthcare data.
Methods RUCAM includes seven criteria of causality assessment. Operational scoring definitions were developed for each of the criteria in collaboration with DILI experts and programmed into an automated algorithm. The automated algorithm was then tested on a retrospective cohort of patients at Kaiser Permanente Southern California. Study participants were 18 years old with twelve months of continuous membership plus drug benefit prior to exposure to one of 14 drugs commonly associated with DILI. Eligible patients filled at least one of the study drugs between January 1, 2003 and August 31, 2011. Patients were scored and cases were categorized into one of five categories ranging from ‘Highly Probable’ to ‘Excluded’. Rates of DILI events per 10,000 exposures were calculated and frequency counts of possible scores for each of the seven criteria were analyzed.
Results We identified 14,925 potential DILI events following 3,321,835 study drug exposures. Four antibiotics accounted for 89.4% of events. The average (SD) patient age was 60 (16) years and 54.5% of potential DILI events occurred in females. Cholestatic injury (65.0%) was the most common type followed by hepatocellular 29.4%, and mixed 5.6%. DILI events were categorized as probable or highly probable in 15.5% of cases, 59.6% were identified as possible, and 24.9% were unlikely or excluded. The overall rate of probable or highly probable DILI events was 6.9 per 10,000 exposures. The scores for most criteria spanned the entire range of possible scores and there were no obvious floor or ceiling effects. However, Criteria 5 (other non-drug causes of liver injury) showed poor discrimination of possible scores.
Conclusions An electronic causality assessment algorithm was developed and successfully tested on healthcare data. Further validation is needed comparing these results with the ‘gold standard’ medical record review by DILI experts.
