Background: Acute liver injury (ALI) is a potential adverse drug reaction that often leads to regulatory action. A postauthorization safety study (PASS) conducted in five European data sources used combinations of ICD codes to identify ALI as a safety endpoint among antidepressant users. Validation of potential cases was not possible in the GePaRD, so this external validation study was conducted.
Objectives: Estimate the positive predictive value (PPV) of ICD‐10‐GM discharge and outpatient hospital codes used to identify ALI. Compare the PPV estimated with or without applying exclusion criteria used in the PASS.
Methods: A cross‐sectional study of routine data from a single academic hospital was used. Adult (≥18 years) potential ALI cases reported in the hospital information system between 01 Jan 2007 and 31 Dec 2016 were included. Case definitions from the PASS were applied to extract potential cases from the hospital information system. Specific endpoint: Discharge diagnosis of ICD‐10‐GM toxic liver disease codes (K71.0, K71.1, K71.2, K71.6, K71.9, K72.0, K72.9, K75.9, K76.2). Less specific endpoint: as above but adding outpatient diagnoses and nonspecific codes suggestive of liver injury (K76.8, K76.9, R16, R16.2, R17, R74.0, Z94.4). True cases were defined based on liver enzymes from electronic medical records and the hospital laboratory database. All clinical data were assessed using double extraction by trained personnel. Exclusion criteria, applied for the secondary objective, were cancer, chronic liver and pancreatic disease, congestive heart failure, and alcohol‐related disorders. PPVs (95% confidence intervals [CI]) were estimated with the exact method.
Results: Overall, 154 (485) potential cases were identified for the specific endpoint (less specific endpoint). Sufficient information for analysis was available for 143 [93%] (357 [74%]) cases. Based on liver enzymes and clinical information, 71 (100) cases were classified as true ALI. This resulted in PPVs (95% CI) of 49.7% (41.2‐58.1%) for the specific endpoint and 28.0% (23.4‐33.0%) for the less specific endpoint. After applying the PASS exclusion criteria, PPVs (95% CI) increased to 62.7% (50.0‐74.2%) for the specific endpoint and 45.7% (37.2‐54.3%) for the less specific endpoint.
Conclusions: Substantial misclassification must be considered when studying acute liver injury based on routine data using ICD‐10‐GM discharge codes. PPV increased when using only specific ALI codes with inpatient cases and following exclusion of competing diagnoses.