Statistics and Data Science Seminar: Dr Theresa Smith (University of Bath)

Title: Statistical Inference from Real-World Health Data: From Early Detection to Treatment Effectiveness 
Speaker: Dr Theresa Smith (University of Bath)
Location: Harrison 170

Abstract: Large-scale health records provide opportunities to improve disease detection and treatment decisions, but they also introduce substantial statistical challenges. In this talk, I present two applied case studies, drawn from recent work in an inflammatory disease called psoriatic arthritis, that illustrate how careful statistical modelling of complex observational health data can inform clinical decision-making.
 
The first study addresses early disease detection using routinely collected primary-care records. We develop predictive models that balance accuracy with interpretability, focusing on transparent covariate selection, stable estimation in high dimensions, and clinically meaningful interpretation of model outputs.
 
The second study examines treatment effectiveness using a national longitudinal cohort, analysing outcomes across up to ten lines of therapy. We use generalized estimating equations to model binary clinical response while accounting for repeated measures, patient heterogeneity, and confounding. We explore how estimated treatment effects evolve over successive treatment lines, providing insight into therapeutic sequencing under real-world conditions.
 
We conclude with a discussion of future work to better account for the selectively sampled nature of clinical data, and the implications of these challenges for inference and decision-making in observational health studies.