Identifying pregnancies in routinely collected health data: a scoping review of methods.

backgroundTo map and describe the methods used to identify pregnancy episodes in routinely collected health data, to report validation practices and assess the transparency and reusability of methods.

methodsThis study followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses for Scoping Reviews (PRISMA-ScR) guidelines. MEDLINE (Ovid), EMBASE (Ovid), OpenGrey, and Google Scholar were searched without time restrictions. Reference lists of relevant studies and reviews were screened for additional citations. All studies that utilised routinely collected health data to identify pregnancy episodes were included. Search results were imported into a reference management tool with duplicates removed. Title screening was conducted by one author. Two authors reviewed a subset (10%) of abstracts, with inter-rater agreement above 90% justifying the remainder of abstract review to be conducted by one author. This process was repeated at full-text review and during data extraction using a pre-piloted form.

resultsFrom 5,859 records screened, 31 studies were included. 29 used rule-based backward-looking algorithms anchored to outcome codes and 2 used forward looking logic from early pregnancy makers. Nine studies incorporated hierarchical logic to estimate pregnancy start date and 19 introduced biologically plausible gaps between outcomes to mitigate misclassification. 15 studies conducted direct validation using chart review with inconsistent reporting of algorithm sensitivity, specificity, and PPV. While 22 studies shared code lists, only three provided reusable code.

conclusionsFuture efforts should prioritise open-source algorithms, standardised validation protocols, and collaboration with clinical experts to ensure generalisability, reproducibility, and clinical relevance.