Can Prenatal Services Learn from Pregnancy Outcomes? A 16-Year Feasibility Study of Data Availability Across...

Prenatal services generate increasingly dense antenatal data, but their ability to evaluate what those data predict depends on outcome information arising in delivery services they may not control. Routinely collected data may fail to support a longitudinal clinical study for fundamentally different reasons: the required event may never occur, the relevant quantity may never be measured, or information generated elsewhere may never reach the record. Record linkage addresses only the last of these. In a retrospective descriptive feasibility and data-availability study, we assessed whether the information required for an intended longitudinal obstetric study could be reconstructed from routinely collected records, distinguishing these mechanisms and examining how availability differed across 16 years of index examinations. The intended biological hypothesis was not tested.

We analysed routinely collected health data from the obstetric clinical database of a private maternal-foetal medicine centre in Romania, comprising 9008 administrative keys from 8090 women, with index first-trimester examinations from 11 August 2010 to 31 August 2026. Record-level linkage to the source export was verified in a 500-record sample. Delivery-data availability was assessed in records that had passed a gestational-age-derived late-gestation reference date plus a 90-day lag before the analysis cutoff, with 0- and 180-day sensitivity analyses, and it was compared descriptively across four calendar epochs. We distinguished observations not generated, not measured and not documented, and we report a sensitivity analysis restricted to one key per woman.

Among 8436 outcome-mature keys, availability of examinations in both later gestational windows rose from 431/1631 (26.4%) in 2010-2014 to 514/1149 (44.7%) in 2024-2026, while documentation of delivery mode fell from 932/1631 (57.1%) to 123/1149 (10.7%), in absolute as well as relative terms. The divergence persisted under both alternative maturity lags and when each woman contributed a single key (26.0% to 44.1%, and 56.3% to 9.6%). Mean arterial pressure, a measurement made within the service, rose from 0.06% to 95.2% over the same period. Within the contracting subset with a documented mode, recorded indications became more frequent (1.7% to 42.3%), so the composition of the information that returned changed as its volume fell. Cord blood gases were an investigator-confirmed structural zero within the study pathway and the analysed export contained no field for the 5 min Apgar score, so the intended acid-base-supported endpoint could not be reconstructed. All proportions describe documentation present in the analysed export at extraction, not the occurrence of clinical events.

Evaluating prenatal prediction requires adequate ascertainment of the relevant pregnancy outcomes, and whether that condition holds is an empirical question for each service. Increasing antenatal observation did not ensure continuity of delivery information, and the composition and informational content of the outcomes that did return also changed. Documentation failure arose both from fields left empty and from fields the record did not contain. Beyond research, inadequate outcome ascertainment may limit a service's ability to audit and evaluate its own prenatal screening performance. These data do not test the biological hypothesis; they identify requirements for a valid future test.