This repo hosts Ethan de Villiers' script for working with the StartRight Dataset on Beverley Shields' and Angus Jones' Classification models research at the University of Exeter (Diabetes team).
Background and Aim: Approximately 30% of those developing classical type-1 diabetes (T1D) in adults are initially diagnosed and treated as having type-2 diabetes, without insulin. We aimed to determine whether previously developed diabetes classification models (the Exeter T1DT2D Calculator https://www.diabetesgenes.org/t1dt2d-prediction-model/) are able to identify those with initially non-insulin treated diabetes who will progress to insulin within 3 years of diagnosis.
Methods: We assessed performance of 3 established classification models for adult-onset T1D in identifying adults with newly diagnosed non-insulin treated diabetes who progress to insulin treatment within 3 years of diagnosis in the prospective StartRight study. Model 1 (age and BMI), model 2 (age, BMI, GAD and IA2 islet-autoantibodies, and model 3 (all features and T1D genetic risk score (T1DGRS)).
Results: See Presentation, poster or delve into the script ("Classification Models.Rmd") for further details!
Conclusion: Prediction models for identifying adult onset T1D have high performance for identifying patients initially treated without insulin who progress to insulin within 3 years of diagnosis and improve prediction over and above use of islet-antibodies alone.
Thanks to the supportive and expertise from the Diabetes team in Exeter, my research was awarded a 15 minute presentation at Diabetes UK 2023. Within this repo there is an E-poster file (P86, .pdf) and a presentation file (A44, .pptx) which showcases the presentation, and the final result of the script.
As this project has been finalised, an LTS (long term support) branch has been created to immortalise the script, version and all other relevant details for future review.
I cannot thank my supervisors, Beverley Shields and Angus Jones enough for their endless support, patience and time. Without them this fantastic project and year would not have been possible, and much less so enjoyable. I further thank Pedro "Peter" Cardoso, Katie Young, Laura Guedemann, Julieanne Knupp, Nicholas Thomas, Richard Orum, John Dennis, Harry Green, Tim McDonald, Jean Claude Katte, Kashyap Patel, Ines Barroso, TJ McKinley, Catherine Angwin and Andrew Hattersley for all of their support, constructive feedback, and meeting-time snacks.