You can read more about the ConcePTION project here (uio.no).
Why is a pregnancy algorithm needed?
Reliable identification and dating of pregnancies is essential for studying medication use and safety during pregnancy. However, pregnancy information can be distributed across birth registries, hospital records, primary care records and other healthcare data sources.
Importantly, approaches relying mainly on records generated at the end of pregnancy may miss early pregnancy losses, pregnancies with unknown outcomes and pregnancies that are still ongoing when data are extracted. This can introduce selection bias in pregnancy research.
The aim of this study was to develop and apply a Pregnancy Algorithm (PA) for identifying and dating pregnancies, enabling federated perinatal pharmacoepidemiological studies across data partners using the ConcePTION Common Data Model (CDM).
How does the ConcePTION Pregnancy Algorithm work?
The algorithm was applied to 10 electronic healthcare data sources across six European countries, including Norway.
Pregnancy-related information from birth registries, hospital records, primary care and other sources was reconciled into pregnancy episodes with estimated start and end dates and outcomes. A random forest model was used to estimate gestational age when information on pregnancy start was missing.
The algorithm was tailored to local data characteristics and iteratively verified by each data partner.
Study Results
The algorithm identified more than 2.7 million pregnancies among more than 2.2 million individuals. Across databases:
- 50–83% ended in live birth
- 1–15% in elective termination
- 4–10% in spontaneous abortion
- 2–34% had an unknown outcome
The first pregnancy record occurred at a median of 47–280 gestational days, demonstrating major differences in how early pregnancies could be detected across data sources. The random forest model estimated gestational age for 6–89% of records, with an RMSE of 17–50 days.
Results from Norway
In Norway, the algorithm identified 393,725 pregnancies:
- 70.7% ended in live birth
- 15.3% in elective termination
- 10.3% in spontaneous abortion
- 2.4% had an unknown outcome
The median gestational age at the first pregnancy record was 130 days (IQR 97–136). The random forest model had an RMSE of 28 days for records requiring imputation of both pregnancy start and end dates.
What is RMSE?
RMSE stands for Root Mean Squared Error. It measures how far a model's predictions are from the observed values, while giving greater weight to larger errors. In this study, it measures the accuracy of the random forest model in predicting gestational age.
For Norway (UOSL), an RMSE of 28 days means that the model's predicted gestational ages differed from the observed values by roughly four weeks in terms of the model's typical error magnitude, although RMSE is not literally the mean absolute difference. Lower RMSE indicates better predictive accuracy.