Expert panel assessment of acute liver injury identification in observational data.

Background: Observational data are useful for studying drug safety; however, to be effective, accurate outcome measurement is paramount. Objectives: This study compared alternative outcome definitions for acute liver injury (ALI) and explored opportunities for improving ALI identification in observational data. Methods: The Truven MarketScan (R) Lab Database (MSLR) was used to identify patients meeting at least 1 of 4 ALI definitions, including definitions based on diagnosis codes, laboratory measures, or combinations of diagnoses, procedures, and/or laboratory measures. Expert panelists reviewed patient data using a Web dashboard. Panelists determined whether they believed the patient had ALI and identified factors influencing their decision. Logistic regression models explored which factors were influential in case determination. Results: Overall, only 37 of 208 reviewed patients (17.8%) were classified as cases. The diagnosis-based definition yielded no positive cases and the laboratory-based definition yielded the most positive cases (31 of 60). The most influential factors in case classification were occurrence of procedures after the index date (OR - 13.2, 95% CI - 5.3-32.9), no occurrence of drug treatments before the index date (OR - 4.6; 95% CI = 1.6-13.2), occurrence of drug treatments before the index date (OR = 0.3; 95% CI = 0.1-0.6), and no drug treatments after the index date (OR = 0.2; 95% CI = 0.0-0.5). Conclusions: Comparing ALI definitions illustrated tradeoffs between the number of plausible cases identified and the likelihood of cases being classified as positive. Future research should refine ALI case definitions, considering the import of laboratory results, procedures, and drugs in defining a case. (C) 2014 Elsevier Inc. All rights reserved.