Summary
A medRxiv preprint describes an algorithm that identified 880,926 pregnancies in Greater Manchester primary-care records. Validation results varied by outcome, with a substantial gap between the register’s miscarriage and termination rates and published UK estimates.
Researchers have developed an algorithm to identify pregnancy episodes in routine primary-care records held within the Greater Manchester Care Record (GMCR), a database integrating health and social care information for 2.8 million people. In a medRxiv preprint, the team reports identifying 880,926 pregnancies among 379,832 women aged 14–49, using records from March 2012 to August 2023.
The register is intended to make pregnancy information easier to use in research. Its validation results, however, varied across outcomes, an important consideration for researchers interpreting the records.
How the register was built and checked
The researchers used 1,448 pregnancy-related SNOMED clinical codes—standardised clinical terms recorded in health systems. The algorithm grouped codes by pregnancy outcome and applied a hierarchy when multiple outcomes appeared in one episode. It derived pregnancy start dates from coded events where possible, or estimated them using a biologically plausible duration based on the outcome.
For internal validation, the team compared register episodes with maternity admissions in Hospital Episode Statistics (HES). Nearly half of the register’s delivery records had a corresponding HES record: the reported positive predictive value was 48.1%. For matched records, the median difference between dates was 17 days, with an interquartile range of 1–30 days.
The authors also compared outcome rates with published UK estimates. Live births were reported at 927.8 per 1,000 deliveries in the register, compared with 995.6; stillbirths were 2.12 per 1,000 births, compared with 3.9. The gaps were larger for miscarriage—8.00 versus 15.3 per 100 pregnancies—and termination of pregnancy—5.36 versus 25.1 per 1,000 women. The denominators differ between these measures, so the rates should be read outcome by outcome rather than as directly comparable measures of overall coverage.
What the results mean for research
A pregnancy register assembled from routine records could help researchers study maternal health across a large regional population, including its relationship to other health and social-care information. The authors report close alignment with UK estimates for maternal age and deprivation: the register’s median maternal age was 31 years, compared with 29.8, and its median Index of Multiple Deprivation decile was 3, compared with 2.6. The share of people from ethnic minority groups was higher in GMCR than in the comparison estimate, at 37.0% versus 28.7%.
The outcome differences show why register-based research needs to account for how events are recorded and captured. In particular, the miscarriage and termination estimates differ substantially from the published comparators, while the HES comparison found a matching admission for 48.1% of register delivery records. The register may still support useful studies, but those results suggest that its suitability can depend on the outcome being studied.
This is a preprint describing retrospective algorithm development and validation, not a clinical trial. The authors say the code will be made available for reuse; the GMCR data themselves cannot be shared because of dataset access restrictions.