< Back to Industry Collaborators
Funder: UCB Pharma and Health and Care research
Overview: This study explores the application of machine learning techniques to identify predictive profiles for the development of ankylosing spondylitis (AS). Using routine data from the Secure Anonymised Information Linkage databank, decision trees were developed separately for men and women based on key features associated with AS development. The models demonstrated good predictive performance within test datasets (positive predictive value 70%-80%), but challenges arise in real-world application due to the low prevalence of AS in the general population. The findings highlight the potential of machine learning in aiding early AS diagnosis and warrant further investigation into cost-effectiveness and real-world implementation.
Why SAIL was valuable:
- SAIL facilitated the linkage of diverse datasets in a secure and anonymized manner. This allowed researchers to access a wide range of healthcare information while protecting patient confidentiality.
- SAIL ensured patient confidentiality and privacy by anonymizing and securely linking individual-level data.
- The use of SAIL facilitated collaborative research efforts across different institutions and healthcare settings.
- Model development using machine learning.
Results:
- The researchers used machine learning methods to develop predictive models for identifying individuals likely to develop ankylosing spondylitis (AS) in the future
- The developed machine learning models showed promising performance within test datasets, with positive predictive values ranging from 70% to 80%. This means that the models were effective in identifying individuals who would later receive a diagnosis of AS based on their characteristics and medical history.
- Men with AS tended to present with lower back pain, uveitis, and nonsteroidal anti-inflammatory drug (NSAID) use at a younger age (under 20 years).
Women with AS typically had an older age of symptom presentation compared to men, often experiencing symptoms in the age range of 20-30 years. They also had a higher prevalence of multiple pain relief medications.
