Assistant Professor Julius Center for Health Science and Primary Care, University Medical Center Utrecht, the Netherlands, Netherlands
Background: Real World Data (RWD) can produce essential evidence for medicines safety during pregnancy and lactation. As such, RWD quality and fitness for purpose must be assessed before generating such evidence. ConcePTION has created a pipeline for RWD quality assessment which has been implemented in several research networks.
Objectives: To characterize different data sources converted to the ConcePTION Common Data Model (CDM) in terms of completeness, plausibility, and reliability.
Methods: Setting. Data Access Providers (DAPs) extracted data on subjects aged 0 to 55y collected between 1995-2019 and convert it to the ConcePTION CDM. DAPs locally ran R scripts composed of 588 checks divided in three levels. Level 1: completeness & conformance, level 2: plausibility & uniqueness, and level 3: coverage & reliability. DAPs uploaded their outputs to the Digital Research Environment for centralized review. Main outcome measure. Detailed description of the data source population during the follow-up period (size, age distribution, birth dates), medicines, vaccine administration, procedures, and events of interest (code counts, incidence rates). Statistical analysis. Descriptive estimates for qualitative and quantitative variables and rates across time, supported by graphs.
Results: We provide an example of five out of 11 data sources (EFEMERIS, POMME10, POMME15, UOSL, FISABIO) who passed level checks 1 to 3. A total study population of 4,408,231 subjects, including mothers and children from 2004 to 2021 was captured. Out of 118 total variables present in the CDM, 14.4% of variables were missing (range: 15 [FISABIO] to 18 [POMME]). Most missing variables were found in EVENTS CDM table (range: 3 to 5). Less than 0.00001% records (n=308,575,254) were duplicates (excl. medicines). We found 929,133 end of pregnancy records, 312,864 interruptions of pregnancy, and no records for start and ongoing pregnancy. The level checks could not yet identify pregnancy status for FISABIO as their data CDM tables were filled differently. Level 3 checks showed that patterns of medicines use, and specific event rates were age dependent, but consistent over time. From 2004 to 2019, the use of antipyretic analgesics (N02B) in EFEMERIS showed a mean rate of 182.75 records per 100PY (range 156.52-190.35). The mean rate for depression from 2007 to 2021 in UOSL was 2.38 (range 2.26 to 3.16) per 100PY. Likewise, patterns of medicines use were comparable between data sources and against external benchmarks.
Conclusions: The ConcePTION quality pipeline can assist in the characterization of large RWD sources, facilitating fit-for-purpose evaluation of RWD for pharmacoepidemiological studies in childbearing-age and pregnant women, and their children.
