(295) Pregnancies with Unknown Outcome in CPRD Aurum: Investigation Conducted as Part of the Feasibility Assessment for a Potential Non-Interventional Post-Authorization Safety Study.

Background: The CPRD Aurum Pregnancy Register, adapted from an algorithm initially developed for CPRD GOLD, is a valuable and larger UK data source to evaluate medication safety during pregnancy. It includes all pregnancy episodes recorded in the primary care data regardless of their completeness or type of outcome, that can be reported as unknown. Understanding which pregnancy episodes with an unknown outcome represent true of false pregnancies is crucial to minimize the risk of bias for a post-authorization safety study (PASS).

Objectives: To examine the extent of the pregnancies with unknown outcome in a CPRD Aurum feasibility study cohort, quantify the potential reasons for their occurrence, and inform decisions on how to handle them for a potential PASS.

Methods: As part of the feasibility assessment, the study population included pregnancies in which the mother was a transplant recipient exposed to immunosuppressants during conception or pregnancy. Characteristics of pregnancy episodes with an unknown outcome were described. The possible scenarios leading to these pregnancy episodes were identified according to the criteria defined for the CPRD GOLD Pregnancy Register by Campbell et al, 2022, and quantified when the required elements were available in the extracted dataset.

Results: Out of 802 pregnancies identified in the study cohort, 395 (49.3%) had an unknown outcome. Over 75% of these pregnancy episodes were inferred from a single record of a pregnancy-related term (most often unspecific) and had an artificial duration of 28 days due to the sensitive and complex identification algorithm. Out of the 12 scenarios to explain pregnancy episodes with an unknown outcome, 9 could be assessed and the 3 others will require CPRD Aurum adapted code lists. Most pregnancy episodes were consistent with scenarios where the pregnancy record belongs to another pregnancy episode in the Register (respectively 29.4%, 13.7%, 1.5%, 1.3% for scenarios 4d, 4a, 4c, 4b). There were 24.3% pregnancy episodes where the outcome was captured in the linked hospital records but not in primary care data. Pregnancies that were still ongoing at the end of available follow-up were also identified (respectively 14.9%, 4.8% for scenarios 2a, 2b).

Conclusions: Using linked hospitalization data to retrieve missing outcomes and retaining pregnancies starting and ending during follow-up time can easily solve a large part of pregnancies with an unknown outcome. However, the correct ascertainment of true and false pregnancy episodes will also require hierarchical decision rules to sort out episodes compatible with multiple scenarios, as well as further validation and refinement of the Register identification algorithm.