BACKGROUND: Pharmacoepidemiologic multi-database studies (MDBS) provide opportunities to better evaluate the safety and effectiveness of medicines compared to single database projects. However, data may be partially or systematically missing in one or more of the contributing databases. This could result in bias or loss of precision, so specialised methods are required to deal with this, especially in the case of distributed data networks.
OBJECTIVES: To determine the extent to which missing data are reported and which methods are currently applied to account for missing data in MDBS.
METHODS: We conducted a systematic literature search in PubMed for MDBS evaluating the safety or effectiveness of medicines published between 01-01-18 and 31-12-19. This was supplemented by searches for eligible studies from established database networks. Studies were considered MDBS if two or more distinct databases were used to capture information on the same variables in different populations. We extracted information on the general study characteristics, the strategies for executing a MDBS (e.g. general common data model, distributed analysis), the reporting of missing data (variables missing, number of patients with missing data) and the methods used to account for missing data.
RESULTS: 2725 articles were identified, of which, 61 studies from 44 different journals were included in the analysis, with the majority using US (64%) and European (39%) data. Twenty-four (29%) used a common data model or common protocol, 19 (31%) carried out a distributed analysis and 10 (16%) shared raw data for a central analysis. Twenty-eight (46%) articles reported missing data, most of which were potential confounders such as lifestyle factors and 11 (18%) studies reported that this missing data could introduce bias. Eighteen (30%) studies addressed the missing data by using a complete case analysis (n=13, 21%), multiple imputation (n=3, 5%) or both methods (n=2, 3%).
CONCLUSIONS: Only 46% of MDBS published in the years 2018-19 report missing data and 30% reported applying methods to address it. The higher complexity of MDBS should be accompanied by an increased vigilance for database completeness and heterogeneity by reporting and addressing the missing data that could introduce bias.
