(AO) Heterogeneity Across European Electronic Healthcare Data Sources: Identification of Adverse Events of Special Interest (AESI) and Impact in Multi-Databases Assessment - An ACCESS Study

Publication Number: 1039

Background: The European electronic healthcare data source landscape is growing. In Europe, healthcare systems and healthcare practices vary broadly across countries which impacts the capture of medical records in electronic databases. It is of importance to consider heterogeneity in data capture for multi-databases assessment.

Objectives: To assess the impact of heterogenous diagnosis vocabularies and different provenance of data on the incidence rate of AESI across 10 European electronic healthcare databases.

Methods: A multi-database dynamic cohort study was conducted from 2017 to 2020 (2010 and 2014 for 2 databases) to generate background incidence rates (IR) of 41 AESI. We used 10 data sources from 7 European countries (DK, FR, GE, IT, ES, NL, UK) capturing data from different provenance: inpatient and/or outpatient settings, emergency room visits or general practitioner medical records (GP) and varying vocabularies (ICD9/ICD10, ICPC, SNOMED). Some data sources contained multi-data provenance for event diagnosis including simultaneously hospital-records linkage (in-outpatient) and GP data. To standardize the process for IR generation, syntactic harmonization was conducted by the Data Access Providers (conversion to ConcePTION CDM), semantic harmonization was part of the analytic R program. IR for each data source were computed by age and sex by dividing the number of incident cases by the total person-time at risk. Age-standardized rates (against Eurostat population) were pooled using random effect models according to the provenance of the events diagnosis: (1) Inpatient only, (2) In and outpatient, (3) Inpatient and emergency room, (4) GP and in-outpatient, (5) GP only.

Results: A total follow-up time of 211.7 million person-years from 63 million individuals was included. For most of the 41 AESI, rates varied according to the provenance of the event diagnosis. As example, rates for venous thromboembolism were most frequently reported in setting including in-outpatient data ranging from 211.9/100,000 person-years (PY) (95%CI[186.8-236.9]) in (4) to 229.7/100,000 PY (95%CI[207.5-251.8]) in (5) versus 40.7/100,000 PY (95%CI[38.3-43.1]/100,000 PY) in (1). AESI such as anosmia-ageusia were better captured in setting including GP data.

Conclusions: This study highlighted the appropriateness of electronic healthcare data sources for a correct identification of AESI and consequently, correct interpretation of IRs. Our results showed the importance of the nature of the event and the setting in which it is diagnosed which should be captured by the data sources that are used to calculate rates, otherwise underestimation is likely.