Active Pharmacovigilance of Drug-Induced Liver Injury Using Electronic Health Records

Drug-induced liver injury (DILI) is relatively rare but frequently associated with serious morbidity and mortality compared to other adverse drug reactions (ADRs).¹ Severity ranges from asymptomatic mild elevation in hepatic enzymes (aspartate aminotransferase [AST] and alanine aminotransferase [ALT]) or bilirubin to liver failure and death. Thus, monitoring DILI is essential for the pharmacological treatment with hepatotoxic drugs. Furthermore, assessment of the risk of DILI in newly developed drugs is important part of pharmacovigilance. However, spontaneous reports of DILI often lead to under-report and under-estimation of the risk of DILI. While prospective observational studies could result in high-quality evidence of the risk of DILI in a specific drug in narrow conditions, a great body of resources and time is needed to draw a reasonable conclusion.

Active surveillance of ADR has been attempted to overcome the disadvantages of spontaneous reporting and post-marketing surveillance.² In the advent of active pharmacovigilance in various ADR, DILI has been examined for the usefulness of detection algorithms using electronic health records (EHR).³ While the clinical course of DILI varies according to the culprit drugs or comorbidities of the subjects, the diagnosis of DILI is suspected based on the laboratory findings, including liver function tests, and exclusion of the other conditions affecting liver function abnormality. Thus, before clinical manifestations such as gastrointestinal symptoms or generalized symptoms due to liver injury or hepatic failure, surveillance of liver function tests could be used for early detection of DILI and discontinuation of the causative drugs. Unsurprisingly, laboratory data has been utilized in the detection algorithm for DILI from early days of active pharmacovigilance for DILI.⁴

In this issue of the Allergy, Asthma & Immunology Research, Kang and colleagues⁵ have reported a multicenter observational study that investigated DILI developed during hospitalization. They enrolled 256,598 subjects who were hospitalized for 1 year in 3 tertiary university hospitals in Korea. The algorithm consisted of 2 steps using both laboratory criteria and diagnostic codes. First, the algorithm excluded subjects with no laboratory data or high level of serum ALT > 3 times upper normal limit (UNL) of or total bilirubin > 2 times UNL and included subjects with high serum ALT > 3 × UNL plus total bilirubin > 2 × UNL or ALT > 5 × UNL afterward during hospitalization. Next, they excluded subjects with diagnostic codes recorded at the time of discharge suggesting liver diseases other than DILI. Applying this algorithm, 1,100 subjects (0.43%) were screened for suspected DILI. Further review of EHR by the specialists confirmed 365 cases of DILI. As the causative drugs for DILI, they reported antibiotics, such as piperacillin-tazobactam, and chemotherapeutic agents.

The utility of algorithms detecting DILI based on laboratory findings and/or diagnostic codes has been tested in several studies.⁶,⁷,⁸ The novelty of these studies is that the algorithms were applied in a larger number of inpatients with higher positive predictive value than the previous studies. Use of previously developed clinical data warehouse (CDW) in the study institutions enabled the application of these algorithms. The incidence of DILI in hospitalized patients was 0.14%, which is very lower than in previous studies.⁹ This might be attributed to the application of higher values of liver enzymes and bilirubin for selecting DILI. Additionally, they excluded patients with a large number of patients without laboratory data within 48 hours of admission (42.3%) and abnormally high values of liver enzymes and bilirubin (12.4%). A meta-analysis on the performance of detection algorithms for DILI showed the exclusion of specified diagnosis and designation of the drugs of interest.³ It is well acknowledged that a diagnostic test with a high cutoff value and diagnostic criteria with strict clinical findings would lower false negative sensitivity. Thus, it would be reasonable to perform a sensitivity analysis to select the diagnostic algorithm for DILI.

The performance of the detection algorithm for DILI could be improved in the future. In addition to laboratory findings and diagnostic codes, the other signals have been tested for their usefulness in detecting DILI. For instance, searching text in EHR improved the performance of the algorithm by the elimination of text suggesting liver diseases and inclusion of text suggesting DILI.¹⁰ Additionally, the inclusion of exposure to drugs in the algorithm also improved the diagnostic yield.¹¹ Incorporating the automatic causality assessment system into the algorithm can also be used for quick onsite surveillance.¹² While CDW is a useful database on active surveillance of ADR and DILI, it is limited to a single institution. Therefore, to assess the risk of DILI, a rare event of ADR, in a large population, the detection algorithm needs to be applied in multiple institutions.¹³ Given the genetic susceptibility for a specific DILI, such as N-acetyltransferase 2 variants in antituberculosis drug-induced liver injury,¹⁴ active surveillance of DILI could utilize both EHR and genetic databases in the future.¹⁵

Active pharmacovigilance based on EHR using a detection algorithm is useful in assessing of DILI in a retrospective way. A better and more sophisticated algorithm for specific drugs in specific conditions should be developed in the future. In addition, real-time onsite application of this algorithm in clinical practice would enhance early detection of DILI and decrease the morbidity and mortality of DILI.