Ankylosing spondylitis (AS) is the second most common type of inflammatory arthritis, often affecting teens and young adults. Symptoms of AS include back pain, stiffness, inflammation of the joints (arthritis), inflammation where tendons attach to bones (enthesitis), and fatigue. Over time, these conditions can lead to spinal fusion, severely affecting quality of life, especially in young people.
Unfortunately, the diagnosis of AS is slow, taking up to 10 years after symptom onset, and usually requires x-rays. The slow progression of symptoms and lack of access to definitive testing contribute to these delays.
However, early detection of this condition can halt the degenerative process and maintain a good quality of life for those affected, potentially making a big difference.
Read more: Unexplained back pain? It could be ankylosing spondylitis
In our study, we explored the possibility of using regularly collected medical data from GPs and hospitals combined with advanced machine learning techniques to identify AS at an earlier stage. Machine learning uses algorithms to analyze sample data to enable predictions and decisions without explicit programming.
We analyzed the data separately for men and women, and the results could change the way GPs detect and diagnose AS.
valuable tool
Anonymous data from Swansea University Medical School’s National Data Repository were used to conduct the study. Patients with AS were identified and matched with people without diagnostic records.
Analysis of this data suggests that factors such as low back pain, uveitis (inflammation of the middle layer of the eye), and non-steroidal anti-inflammatory drug use under the age of 20 are factors associated with an increased risk of developing the disease. It turns out. AS in men.
In contrast, our model revealed that women were more likely than men to experience AS symptoms later in life and were more likely to be dependent on multiple analgesics. This probably indicates that the condition is more likely to be misdiagnosed in females.
Machine learning is a valuable tool for profiling and understanding the characteristics of people who are likely to develop AS. It performs well on test datasets with artificially high prevalence.
However, when applied to the general population of GPs and hospitals where AS is uncommon, even the best models achieve only a low positive predictive value of 1.4%. (This is the probability that a person really has AS after a positive test result.)
Therefore, multiple models may need to be used over time to narrow the population and improve this predictive value, thus expediting the diagnosis of AS.
Zhao Jormami/Shutterstock
Recognize challenges
Machine learning technology has great potential to improve patient care. However, it is also important to recognize the challenges associated with using these techniques effectively.
These models rely on diverse and comprehensive high-quality data to produce reliable and accurate results. However, privacy concerns, data sensitivity, and lack of standardization can limit medical data. Therefore, these limitations can compromise model accuracy and reliability.
It’s important to recognize that machine learning on this topic is still in its early stages. Further development of this requires the collection of more detailed data to improve predictive rates and clinical utility.
Read more: From ‘crazy’ provocateurs to IBM’s failed AI superproject: The controversial story of how data transformed healthcare
However, our study shows great potential for machine learning to help identify patients with AS and better understand the diagnostic process through the healthcare system.
We know that early detection and diagnosis of AS is critical to ensuring the best outcome for patients. We believe machine learning can help with this. It also has the potential to empower general practitioners to find and refer patients more effectively and efficiently.
