NHS England Data Science PhD Internships

Causal Inference to Estimate the Impact of Interventions

Keywords: Causal inference, bayes, tabular

Need: Healthcare interventions, such as community interventions to deliver proactive care or reduce inequalities, promise great opportunity and result in the roll-out of a variety of small-scale programmes. However, the evidence base regarding such interventions is limited and often lacks statistical expertise in design precluding analysis through traditional randomisation frameworks.

There are many approaches which attempt to identify causal inference from counterfactual analysis (e.g. Difference-In-Difference, Effect of Treatment in the Treated, Propensity Score Matching, Potential Outcomes Framework, Granger causality, …) or build causal representations of variable relationships (e.g. Casual Bayesian Networks, Structural Equation Modelling, causal layers in deep learning, …).

However, often in healthcare the data quality and/or knowledge of the high-level causal interactions between variables is poor making causal inference difficult. This project aims explore causal methods for healthcare interventions asking:

• What are the most effective machine learning and statistical methods for causal inference in different healthcare scenarios?
• How can we integrate domain knowledge and expert input into causal AI models to improve their accuracy and reliability?
• What are the best practices for validating causal models in the healthcare settings?
• How can we address issues of data quality, missing data, and confounding variables in causal models?

This work is intended to be in partnership with North Central London ICB with a focus on their community data.

Current Knowledge/Examples & Possible Techniques/Approaches:

• An overview of ETT
• Structural Causal Model with Expert Augmented Knowledge to Estimate the Effect of Oxygen Therapy on Mortality in the ICU
• R Package - CausalImpact
• A scoping review of causal methods
• The Case for Causal AI
• Causal Machine Learning for predicting treatment outcomes
• Causal Healthcare AI

Related Previous Internship Projects: N/A

Enables Future Work: Build a foundational workflow and understanding to enable estimation of impact of community interventions from observational data where appropriate.

Outcome/Learning Objectives: A better understanding of when causal inference can be used to investigate the impact of a programme, generation of a workflow to enable the analysis of future initiatives

Datasets: HealtheIntent linked data, UK Longitudinal Linkage Collaborative

Desired skill set: When applying please highlight any experience around causal inference, direct-acyclic graph discovery, coding experience (including any coding in the open), any other data science experience you feel relevant.

---

Return to list of all available projects.

This site is open source. Improve this page.