Machine learning used to improve outcome for arthritic kids
It's a sad fact that children can develop arthritis, and while some end up going into remission, the disease becomes much worse in others. A new machine-learning technique is reportedly able to predict which kids will fall into which category, allowing for their treatment to be tailored accordingly.
Scientists from the University of Toronto started out with a collection of clinical data relating to 640 arthritic children, which was gathered between 2005 and 2010. That dataset included each patient's symptoms, along with their outcomes. Machine learning algorithms were then used to sift through all of that information, looking for recurrent patterns.
What it discovered was that most of the children could be sorted into one of several groups, depending on the area of their body in which the joint pain was present – those areas included the pelvic region, fingers, wrists, toes, knees and ankles. Some of the patients, however, didn't fit neatly into one group, as their joint pain wasn't localized. It was these children who took longer to go into remission, ultimately faring worse than the others.
It is now hoped that as soon as a patient is identified as belonging to that last non-group, doctors can set about administering fairly potent medications, perhaps improving the outcome. On the other hand, if it's determined that a child is likely to soon enter remission anyway, the use of medication can be minimized – this will both reduce costs, and spare the patient from unnecessarily enduring side effects.
"Knowing which children will benefit from which treatment at which time is really the cornerstone of personalized medicine and the question doctors and families want answered when children are first diagnosed," says U Toronto's Prof. Rae Yeung.
A paper on the research, which also involved Prof. Quaid Morris and recently-graduated student Simon Eng, was published this week in the journal PLOS Medicine.