A new study reports that machine learning techniques may help detect ankylosing spondylitis (AS) early.
“It’s great to see the cutting-edge role machine learning can play in the early detection of patients with conditions such as AS…while still in its early stages, machine learning is clearly a way for researchers and clinicians. Kieran Walshe, Director of Health and Care Research Wales, said in a press release:
the study, “Predicting Ankylosing Spondylitis Diagnosis Using Primary Care Health Records – A Machine Learning Approachwas published in a magazine pro swan.
Possible, but more work needed to predict who is at risk of disease
Machine learning can help with early detection and diagnosis of diseases that are critical to proper care, allowing doctors to “refer patients more effectively and efficiently,” says the National Center for Population Health and Welfare Research in Wales. said Jonathan Kennedy, lead author of
“But machine learning is in the early stages of implementation. To develop it, we need more detailed data to improve predictions and clinical utility,” said Kennedy.
Because symptoms of AS, such as back pain, are common to many conditions, it is often difficult to identify who is most at risk for AS. Diagnosing disease can be a complex process, generally requiring a combination of imaging studies, laboratory tests, and other clinical evaluations. Nonetheless, early and accurate diagnosis is critical to facilitate appropriate disease treatment.
“On average, it takes an AS patient eight years from onset of symptoms to being diagnosed and treated. Machine learning may provide a useful tool to shorten this delay.” said study co-author Ernest Choi of Cardiff University.
A team of scientists in the UK used machine learning analysis of medical records to identify patterns that could help detect AS early.
“New machine learning methods may be able to identify patterns and clusters of terms/data such as diagnoses, procedures, and medications. It is observed more frequently in humans with AS,” the researchers wrote.
Using information from the Welsh National Health Database, scientists identified 543 men and 250 women aged between 15 and 35 who were diagnosed with AS. Some of these data were used to build machine learning models and the rest were used to test the models. We analyzed male and female patients separately, as AS can vary significantly by gender.
Results broadly indicated that factors predictive of AS included a history of pain and treatment with analgesics, spine x-rays, or uveitis (inflammation of the eye). , but women tended to be older at onset of symptoms compared to men with back pain who were taking several analgesics.
A previous diagnosis of other muscular and/or skeletal disorders, especially at a young age, was also predictive of future AS risk. They point out that the numbers are high,” suggesting that “the path for a woman to be diagnosed with AS is more complicated than the multiple tests, diagnoses, and referrals.” to men with AS.
Once the machine learning model was built, researchers used statistical tests to assess how accurately it could identify AS. The results generally showed good accuracy in the initial analysis, but the accuracy of the model declined significantly when applied to data from the general population.
Based on the known general population prevalence of AS, researchers estimated the ideal positive predictive value for AS in the general population to be 1.44% for men and 0.51% for women. The model used had a positive predictive value of 0.33% for men and 0.25% for women. Of note, positive predictive value is the likelihood that someone who has an abnormal test actually has the condition they are testing for.
“These characteristics indicate that the female model is closer to the ‘perfect’ model than the male model,” the researchers wrote. “This means that the female model is more accurate than the male model in identifying AS patients.”
Researchers note that these machine learning models could likely be improved with more detailed patient data.
“This study shows that machine learning may help identify people with AS and better understand their diagnostic journey through the health care system, but predictive and clinical utility remains to be seen. We need more detailed data to improve,” they concluded.
