About Me

James G. Wrightson PhD

I am a health data and meta scientist with over 15 years of experience in research and teaching. My current research applies machine learning, natural language processing, and generative AI to large population-level health datasets, focusing on measuring health system use and designing and testing digital health tools. I have over 10 years of experience teaching undergraduate and postgraduate courses in research methods, statistical analysis, and data science, and have supervised PhD, MSc, and undergraduate research projects. I have strong programming skills in R and Python and have taught courses on using R for clinical research. I am currently the Special Features Editor for open science at the Journal of Orthopaedic & Sports Physical

Papers · Google Scholar · ORCID

For a copy of my CV, please get in touch.

Recent Work

Comparing Methods to Identify which Adult Emergency Department Visits Might be Avoided: A Retrospective Analysis of the MIMIC-IV-ED Database

We compared three ways to identify potentially avoidable emergency department visits in 201,106 visits from MIMIC-IV-ED. The methods often flagged different visits: fewer than 2% of all visits were flagged by all three.

Read the medRxiv preprint · 14 November 2025

Open the full-size UpSet plot. Diagnosis alone flagged the largest group (44,119 visits); 3,170 visits were flagged by all three methods. A second panel shows cumulative totals.

Figure 1, Wrightson et al. (CC BY-NC 4.0)

GPT for RCTs? Using AI to determine adherence to clinical trial reporting guidelines

Can a large language model tell whether a clinical trial was reported properly? In a study of 113 sports medicine trial reports, GPT-4 Turbo reached 90% accuracy on the held-out text test set.

Read the paper in BMJ Open

GPT-4 Turbo achieved 90% accuracy on the held-out text test set, with a 95% confidence interval of 85% to 94%.