BACKGROUND: The IMI-ConcePTION project aims to build an ecosystemto generate Real World Evidence to address the information gap ofmedication safety in pregnancy. Data sources do not come with an enu-meration of the pregnancies experienced in their populations, and differin terms of provenance. A novel algorithm to detect pregnancy episodeswas designed and tested, also in data sources participating in two EMA-funded studies using the ConcePTION Common Data Model:CONSIGNand retinoids/valproate risk minimization measures study.
OBJECTIVES: To describe a novel algorithm extracting pregnancies fromheterogeneous European data sources.
METHODS: Six partners extracted instances from the following datasources: ARS, CASERTA (IT); BIFAP, VID (ES); PHARMO Database Net-work (NL) and UOSL (NO). Time periods varied from 2005 to 2021.Records were first retrieved from all available provenances, including: irth registries (BR), primary care medical records, hospital dischargerecords, procedures, and others. To retrieve coded diagnoses, codeswere mapped from the Matcho algorithm and other sources to manyvocabularies (ICD9CM, ICD10CM, ICPC) via the Codemapper tool.Records of the same person were listed longitudinally, and grouped inepisodes of pregnancy. Using hierarchical rules we defined start date,end date (limited to pregnancies completed by the time of data extrac-tion) and type of end, categorized as livebirth, still birth, termination,spontaneous abortion, ectopic, unknown (UKN), ongoing (ONG).
RESULTS: Episodes of pregnancy were 566 050, 126 119, 1 954 078,7039 108 566, 885 415 in ARS, CASERTA, BIFAP, VID, PHARMO,UOSL, respectively. BR was the highest provenance in the hierarchyin 100% of cases in UOSL, and ranged from 95.1% in ARS to none inBIFAP.All data sources except UOSL could retrieve pregnancies whose endwas UKN: these constituted 13.2%, 4.5%, 35.2%, 24.4%, and 1.0% ofpregnancies identified in ARS, CASERTA, BIFAP, VID and PHARMO,respectively. ONG pregnancies could be retrieved from ARS, BIFAP,VID and PHARMO.
CONCLUSIONS: The algorithm leveraged heterogeneous provenances inEurope. Pregnancies could be retrieved that would have gone unno-ticed if querying only information recorded at the end of pregnancy,including pregnancies that were ONG at the time of data extraction.Detecting ONG and UKN pregnancies can substantially increasecohorts, with potential to extensively assess impact of regulatorydecisions, and to rapidly investigate emerging issues, such as drug uti-lization in pregnancies exposed to COVID, or exposure and safety ofCOVID vaccines during pregnancy.
