PhD Studentship: Improving outcomes of childbirth and addressing disparity between ethnic groups with Causal Artificial Intelligence (AI) – PhD Public Health

In the UK, about one in seven newborns each year are unwell enough to need care in a neonatal unit. Among so many mothers, there is wide variation in the factors that may contribute to poor outcomes for babies. These factors can include the mother’s health before pregnancy, the number and type of tests and checks during pregnancy, and the course of labour and any interventions during childbirth. The complexity and inconsistency of the data make it difficult to identify the true causes of poor outcomes.

An emerging field in health data science is Causal Artificial Intelligence (AI). In traditional analysis it is possible to find statistical links between factors but not to identify which factors cause changes in others. Existing AI methods can learn to recognise combinations of several factors and accurately calculate an outcome for each patient, but the rules used are too complex to be interpreted by humans. The novel aspect of Causal AI is that the model is designed to clearly explain the logic of why each outcome is predicted. This in- built cause-and-effect setup will prove important for patient and clinician trust in the model.

2. Problem or issue to be investigated.

Infant outcomes in pregnancy vary markedly by ethnic group. In England, infant mortality rates are twice as high among Black infants (6.8 per 1000 live births) and around one and a half times as high among Asian or Asian British infants (5.4 per 1000 live births) than for White infants (3.2 per 1000 live births).

3. Research questions/aims and objectives.

Research question: What factors, present before or during pregnancy and childbirth, does Causal AI identify as influencing infant outcome? Which are modifiable? Which modifiable factors could reduce ethnic disparities in infant outcomes?

Aim: This PhD will develop and apply advanced Causal AI methods to a maternity dataset covering the Liverpool area to identify modifiable care factors that could help reduce the pronounced ethnic inequalities in infant outcomes during pregnancy and birth.

Objectives:

1. Build a causal model to develop a data-driven understanding of the direct and indirect causes of poor infant outcomes.
2. Find the strength of the causal effect of each factor and so identify which factors have the potential to make the biggest changes to outcomes.
3. Validate the results with expert knowledge from obstetricians.
4. Identify the modifiable factors which are actionable of follow-up. These will have an optimal balance of having a significant effect on outcomes and being achievable for the relevant patients.

4. Proposed methodology and methods.

The data available covers over 30,000 pregnancies in the Liverpool area and has details on over 200 aspects of each, including the mother’s pre-pregnancy health, the birth, and the immediate neonatal care. The principal indicators of outcome are: the number of days receiving assistance with breathing through ventilation in neonatal care (including days intubated); and infant death.

Causal AI is a key emerging area that the Engineering and Physical Sciences Research Council (EPSRC) has identified as needing development in the UK, and is supporting in application to health through the CHAI (Causality in Healthcare AI) network, of which the University of Exeter is a member

Advert information