We are an inter-disciplinary team of UK scientists with expertise in obstetrics, women's and child health, epidemiology, climate science, inflammation, computational modelling, machine learning and artificial intelligence. Together we have a long history with existing strengths underlying preterm birth research that crosses multiple disciplines and an excellent track record of publications and awards leading research in preterm birth.

We aim to develop and validate a deep learning model to predict the risk of preterm birth and other adverse pregnancy outcomes using data from EPIC electronic health records at University College London Hospital Trust (UCLH) for a cohort of 18000 patients. We will obtain corresponding data on exposure to ambient pollution using non-identifiers for postcode (area) and date of delivery (month). The model will review the temporal sequence of events within a patient's medical history and current pregnancy, identifying significant interactions and will predict the risk of preterm birth. It will also determine the threshold and gestation at which pollution exposure has the greatest impact.

Preterm birth is the leading cause of neonatal morbidity and mortality worldwide. Children born prematurely have higher rates of cerebral palsy, sensory deficits, learning disabilities and respiratory illness. In the UK, approximately 60,000 babies are born prematurely each year. This is equivalent to 1 in 9 pregnancies in England and the numbers increase to 1 in every 7 pregnancies in London. In around 40% of cases, the cause of preterm birth are unknown. Current algorithms to predict preterm birth are limited in their ability to identify women at highest risk of delivering preterm and do not consider genetic, lifestyle and environmental circumstances within their prediction. With the rapid development of machine learning and deep learning, it is now possible to develop models which can consider a higher number of variables within their predictive algorithm, to formulate a patient specific prediction of risk. There is growing evidence that maternal exposure to air pollution during pregnancy is associated with an increased risk of preterm birth. Exposure to air pollution may be associated with poor placental function, pre-eclampsia, and poor fetal growth although there is limited data on these adverse pregnancy outcomes, all of which can lead to preterm birth. At present, many of the recent epidemiological studies in this area lack detailed and matching clinical data sets without gaps in electronic records.

This study aims to:

1. Link data on air pollution exposure with highly detailed clinical data sets extracted from patient electronic health records from University College London Hospital NHS Trust (UCLH)
2. Develop a computational model which can accurately predict the gestation at which a patient will deliver in weeks and days
3. Using the model, identify the timepoints in pregnancy that air pollution has the greatest impact on pregnancy outcomes

Inclusion Criteria:

• We aim to include data from pregnant women who delivered at University College London Hospitals from 2019 onwards after the start of the EPIC electronic patient record. The is no specified age range for this study, so as to improve inclusivity. We also aim to represent minority ethnic groups and patients with social deprivation within our dataset.

Exclusion Criteria:

• We will exclude data from patients with an incomplete duration of follow-up due to transfer of antenatal care for delivery at another trust. Patients with incomplete past obstetric history data, inaccurate estimations of gestational age (e.g. due to late booking of the pregnancy) and missing data for 'postcode of usual address' will also be excluded. Patients who are less than 18 years of age will be excluded.

Pregnant women over the age of 18 years old

We aim to include data from pregnant women who delivered at UCLH from 2019 when EPIC was launched and until the end of 2023. There is no specified upper age range for this study. To improve inclusivity, we will aim to collect information from all women booking and delivering at UCLH to ensure minority ethnic groups and patients with social deprivation or with additional pregnancy complicating disorders are included within our dataset.

Data from the EPIC electronic health records database will be anonymized at UCLH to create a secondary dataset with anonymized identifier for patient identifier, postcode (area) and delivery date (month). Raw data screened. Patients excluded according to criteria.