Mother-infant linked UK electronic birth cohorts representing 17.5 million births harmonised to the OMOP common data model

This paper describes the successful harmonization of five diverse UK electronic birth cohorts, encompassing over 17.5 million births, into the OMOP Common Data Model to create a standardized, federated resource that enables large-scale, reproducible maternal and child health research across England, Scotland, and Wales without sharing individual-level data.

The Gist

Imagine you are trying to solve a massive puzzle about the health of mothers and babies across the entire United Kingdom. You have five different teams, each holding a huge pile of puzzle pieces. The problem? Each team's pieces are different shapes, different colors, and have different pictures on the back. One team uses square pieces, another uses triangles, and they all speak slightly different "languages" to describe the same thing (like a baby's birth weight or a mother's blood pressure).

If you tried to mix them all together, it would be a mess. You couldn't see the big picture.

This paper is about a project called MIREDA (Mother and Infant Research Electronic Data Analysis) that decided to fix this. They built a giant, universal "adapter" that turns all those different puzzle pieces into the exact same shape and color, so they can finally fit together perfectly.

Here is how they did it, explained simply:

1. The Goal: One Big Picture

The researchers wanted to study over 17.5 million births across England, Scotland, and Wales. That's a lot of babies! But because the data came from different hospitals and regions, it was scattered and messy. They wanted to combine these records to answer big questions: Why do some babies arrive early? How does a mother's health affect her child years later? Does the place you live change your birth outcomes?

2. The Solution: The "Universal Translator" (OMOP)

To make the data talk to each other, they used a standard blueprint called the OMOP Common Data Model. Think of this like a universal power adapter.

• Before: A British plug (UK hospital data) couldn't fit into a French socket (a Scottish hospital database).
• After: The MIREDA team built an adapter that turns every British plug into a standard shape that fits everywhere.

Now, a "smoking during pregnancy" record from Wales looks exactly the same as a "smoking during pregnancy" record from London. They speak the same language.

3. The Tricky Part: The Mother-Baby Connection

Most computer databases are designed to track one person at a time. But pregnancy is special because it involves two people (mother and baby) who are deeply connected for a short time, and then the baby becomes their own person.

Imagine a database that only knows about "Alice" and "Bob," but doesn't know they are mother and son.

• The Problem: Standard databases don't have a built-in "Family Tree" button.
• The Fix: The team created a special "linking card" (called a fact_relationship table). It's like a name tag that says, "Alice is the mother of Bob, and they were together during this specific pregnancy." This allows researchers to follow the mother's story and the baby's story separately, but also see how they are connected.

4. The Process: The "Factory"

The team didn't move the actual data (which would be unsafe and illegal). Instead, they sent the "instructions" (the adapter rules) to each hospital's secure computer room (called a Trusted Research Environment).

• The Factory: Inside each secure room, a machine (software) took the messy local data, ran it through the "Universal Translator," and turned it into the clean, standard OMOP format.
• The Result: The raw data never left the hospital. Only the standardized, anonymized results were shared. This is like sending a recipe to a chef in a locked kitchen; the chef cooks the meal and sends you the taste, but you never see the ingredients or the kitchen.

5. The Challenges (The "Glitches")

Even with a great plan, there were bumps in the road:

• Missing Addresses: The standard model only remembers your current address. But for babies, where you lived during pregnancy matters. They had to build a special "time-travel notebook" (a helper table) to keep track of where families lived in the past.
• Lost Details: Sometimes, the standard model is too simple. For example, if a mother said "I don't know" about her ethnicity, the standard model might just say "Missing." The team had to find clever ways to keep those "I don't know" answers so they didn't lose important nuance.
• Drug Names: One hospital might call a medicine "Paracetamol," another "Acetaminophen," and a third uses a code like "12345." They had to create a massive dictionary to translate all these names into one standard name.

6. Why This Matters

Now, researchers can run a single computer program that asks a question (e.g., "How many babies were born early in 2020?") and it instantly runs that question across all 17.5 million records in England, Scotland, and Wales.

• No more waiting: They don't have to wait years to get permission to share data.
• Better answers: They can spot rare diseases or rare side effects because they have such a huge crowd of people to study.
• Fairness: They can compare if a baby born in a poor area has different outcomes than one in a rich area, across the whole country.

The Bottom Line

This paper is a blueprint for how to turn a chaotic pile of different medical records into one giant, organized library. It's like taking five different languages and creating a single, perfect dictionary so that doctors and scientists from all over the UK can finally work together to make mothers and babies healthier.

In short: They built a universal translator for baby data, allowing the UK to finally "speak" as one voice to improve healthcare for everyone.

Technical Summary

1. Problem Statement

Despite the availability of large-scale electronic health records (EHRs) for maternal and child health research in the UK, significant barriers prevent cross-cohort and cross-national analysis:

• Data Heterogeneity: Existing UK birth cohorts (e.g., Born in Wales, Born in Bradford) differ in study design, coding systems (SNOMED, Read, ICD-10, OPCS-4), and variable definitions.
• Model Limitations: The Observational Medical Outcomes Partnership (OMOP) Common Data Model (CDM) is designed around individual patients. It lacks native structures to explicitly represent the complex relationships between mothers, infants, and pregnancy episodes, which are central to perinatal research.
• Data Silos: Previous attempts to link mother-infant data often relied on non-standard extensions (reducing interoperability) or attributed pregnancy data solely to the mother (preventing longitudinal infant tracking).
• Privacy Constraints: Research requires secure, federated analysis where raw data cannot leave local Trusted Research Environments (TREs).

2. Methodology

The Mother and Infant Research Electronic Data Analysis (MIREDA) partnership developed a reproducible pipeline to harmonize five major UK cohorts into the OMOP CDM (version 5.4).

A. Data Sources

The harmonized resource integrates five cohorts covering England, Scotland, and Wales:

1. Born in Wales (BiW): Linked health, education, census, and social care data (SAIL Databank).
2. Born in Scotland (BiS): Maternity EHR (TRAK) linked to national administrative outcomes.
3. Born in Bradford (BiB): Maternity records linked to education and social care.
4. Born in South London (eLIXIR): Neonatal/maternity records (BadgerNet/EPIC) linked to primary care and mental health data.
5. CPRD (Gold/Aurum): Primary care data linked to Hospital Episode Statistics and the CPRD Pregnancy Register.

B. The ETL (Extract, Transform, Load) Pipeline

The transformation was conducted locally within each cohort's TRE using the OHDSI toolset:

• Profiling: WhiteRabbit was used to scan source databases and generate metadata reports.
• Mapping: Carrot Mapper was used to map source variables to OMOP standard concepts (SNOMED, RxNorm, ICD-10).
  • Challenge: UK drug codes (Read/DM+D) were pre-processed to map to RxNorm.
  • Challenge: UK ethnicity classifications were mapped to standard OMOP "race" concepts for broad categories, while detailed ethnicity was preserved in the observation domain to avoid data loss.
• Transformation: Carrot Transform executed JSON-based rules to load data into the OMOP schema.

C. Key Technical Innovations (Solutions to Challenges)

1. Mother-Infant Linkage: Instead of non-standard extensions, the authors utilized the standard fact_relationship table.
   • Linked maternal and infant person_ids using specific relationship_concept_ids (e.g., 'mother' = 4248584, 'child' = 4285883).
   • Represented pregnancy episodes as condition_occurrence records (concept ID 4299535) with start/end dates derived from LMP and delivery.
2. Location History: Since OMOP CDM v5.4 only stores the current address, a non-standard helper table (location_history) was implemented (aligned with CDM v6.0 conventions) to support longitudinal analysis of time-varying demographic variables.
3. Many-to-One Mapping: A coordinated strategy ensured consistent domain assignment for variables like blood pressure and smoking status across all sites.

3. Key Contributions

• Scale: Creation of a harmonized dataset representing 17.5 million live births and over 11.4 million mothers.
• Standardization: Successful mapping of heterogeneous UK data sources into a single, interoperable OMOP CDM structure without compromising the ability to link mothers and infants.
• Federated Capability: The pipeline enables "write-once, run-anywhere" analysis. Researchers can develop code once and execute it across all cohorts within their respective TREs, sharing only aggregate results, not individual-level data.
• Open Resources: All mapping rules, transformation scripts, and documentation are publicly available in the MIREDA GitHub repository, allowing for extension to other cohorts.

4. Results and Data Characteristics

The paper provides detailed descriptive statistics (Tables 2–5) of the harmonized data:

• Birth Outcomes:
  • Singletons: 99% of births.
  • Gestational Age: 82.2% term, 5.4% preterm, 3.3% post-term.
  • Delivery Mode: 20.6% Caesarean section (C-section), with significant regional variation (e.g., 39.7% in South London vs. 19.9% in Wales).
• Maternal Demographics:
  • Mean maternal age: 29.4 years.
  • Ethnicity: Significant variation in missing data rates across cohorts (e.g., 65.5% missing in Born in Wales vs. 10.6% in CPRD Aurum), highlighting the need for continued harmonization work.
  • Socioeconomic Status: The Index of Multiple Deprivation (IMD) shows distinct regional profiles (e.g., Born in Bradford has 70.1% in the most deprived quintile).
• Clinical Conditions:
  • Gestational diabetes: 3.5% overall.
  • Pre-eclampsia: 1.5% overall.
  • Smoking at booking: 2.5% overall (with high variation in recording completeness).

5. Significance and Limitations

Significance:

• Rare Event Analysis: The massive sample size (17.5M) allows for the study of rare exposures and adverse outcomes that smaller cohorts cannot support.
• Policy Evaluation: Enables cross-national comparisons of healthcare policies (e.g., C-section rates) and their impact on population health.
• Trial Emulation: Supports the emulation of randomized controlled trials using real-world data at a national scale.
• Interoperability: Demonstrates that complex mother-infant relationships can be modeled within the standard OMOP CDM, setting a precedent for international perinatal research networks.

Limitations:

• Data Loss in Standardization: Mapping to standard vocabularies can result in granularity loss (e.g., specific UK ethnicity sub-categories mapped to broad "race" categories).
• Absence of Negative Data: OMOP CDM does not natively represent the absence of a condition. A lack of a recorded diagnosis may mean the condition is absent or simply unrecorded.
• Non-Health Data: Linked non-health data (e.g., education, social care) are not standardized within the OMOP CDM framework.
• Access Governance: While the data is harmonized, researchers must still apply separately to each TRE, which can be time-consuming.
• Geographic Gaps: Northern Ireland data was not included due to system changes, though future integration is planned.

Conclusion

This work establishes a foundational infrastructure for large-scale, federated maternal and child health research in the UK. By solving the specific challenge of mother-infant linkage within the OMOP CDM, the MIREDA partnership has created a reusable, scalable resource that facilitates reproducible, cross-cohort studies while maintaining strict data privacy and governance standards.