Forns J, Cainzos-Achirica M, Hellfritzsch M, Giner-Soriano M, Poblador-Plou B, Hallas J, Morros R, Prados-Torres A, Pottegard A, Cortes J, Aguado J, Castellsague J, Jacquot E, Deltour N, Perez-Gutthann S, Pladevall M. How valid are the codes used to identify acute liver injury (ALI)? A study in 3 European data sources. Presented at the 34th ICPE International Conference on Pharmacoepidemiology & Therapeutic Risk Management; August 26, 2018. Prague, Czech Republic. [abstract] Pharmacoepidemiol Drug Saf. 2018 Aug 17; 27(S2):371-2. doi: 10.1002/pds.4629

Background: Identifying valid cases of ALI in automated data sources is challenging. Most previous case‐identifying algorithms had positive predictive values (PPVs) around 50%. A case validation process was implemented in the framework of a PASS assessing the risk of ALI in users of agomelatine and other antidepressants

Objectives: To identify valid cases of ALI and evaluate the PPVs of algorithms used to detect potential ALI cases

Methods: Three data sources were used: EpiChron and SIDIAP in Spain (primary care and hospital cases) and the Danish NationalHealth Registers (inpatient and outpatient hospital cases) in Denmark. Algorithms for three ALI endpoints defined by literature review using ICD and ICPC codes were validated: primary (specific hospital discharge diagnosis codes), secondary (specific and non‐specific hospital discharge diagnosis codes), and tertiary (specific and non‐specific hospital and outpatient diagnosis codes). The validation strategy included review of patient profiles (SIDIAP and EpiChron) and of data abstracted from medical records (EpiChron and Denmark). ALI cases were confirmed when liver enzyme values met the consensus definition of an international DILI Expert Working Group

Results: In SIDIAP, EpiChron, and Denmark, respectively, 10, 19, and50 potential cases of the primary algorithm; 34, 59, and 489 of the secondary algorithm; and 2826, 268, and 1008 of the tertiary algorithm were identified. PPV (95% CI) of the primary algorithm was 60% (26%‐88%) in SIDIAP, 84% (60%‐97%) in EpiChron, and 74% (60%‐85%) in Denmark; of the secondary algorithm, 40%(19%‐64%) in SIDIAP, 65% (45%‐81%) in EpiChron, and 70% (64%‐77%) in Denmark; and of the tertiary algorithm, 8% (7%‐9%) in SIDIAP,25% (18%‐34%) in EpiChron, and 47% (42%‐52%) in Denmark. The overall PPVs were higher for specific than for non‐specific codes and for hospital discharge than for outpatient codes. The non‐specific code“Unspecified jaundice”had a high PPV in all data sources and was the largest contributor of confirmed cases in Denmark (75 of 79and 82 of 90 confirmed cases for the secondary and tertiary algorithms, respectively)

Conclusions: To maximise internal validity, studies on ALI in these and similar data sources should prioritize use of hospital discharge and specific codes. Conducting validation activities is required, especially when ALI outpatient potential cases are included. Case‐identifying algorithms should include ICD codes for unspecified jaundice

Share on: