In a study presented at the annual meeting of the American College of Rheumatology, UK researchers showcase a study in which artificial intelligence was successfully used to identify ACPA positivity and CRP of greater than 12.3 mg/L to identify rheumatoid arthritis patients more likely to respond to sarilumab.
"Why do this study? We have lots of treatment options in rheumatoid arthritis, but there is no definitive guidance in what treatment to use in individual patients. So, the purpose of this study is using modern computational technique---machine learning (or, artificial intelligence)---to try to come up with clinical features and biomarkers that will help us select the right treatment for the right patient. So, we use one clinical trial in this study to generate the algorithm to identify the patient characteristics and biomarkers that will predict response to sarilumab and then we test these rules in other clinical trials to validate these observations. What we found was that patients who are anti-CCP antibody positive and has a CRP value of more than 12.3, they are more likely to respond to sarilumab than what we call rule-negative patients. We used the Monarch study in which we compared the efficacy of sarilumab with the TNF-inhibitor adalimumab. We showed that this rule is particularly true for patients who are treated with sarilumab and less so with adalimumab. So the rule is specific for predicting response to sarilumab.
"So, the clinical implication of this finding is that we know at least in the clinical trials dataset about one-third of patients will fulfill this rule. And, if patients fulfill this rule, the chances of them responding from sarilumab treatment in terms of achieving ACR70 response will increase from 7 percent to 34 percent. So a five-fold increase in patients with really large response.
"We know that in fact this rule only increased a response rate for sarilumab treatment patients from 12 to 60 percent. So, in terms of clinical patients, it would suggest that patients who are ANCA positive and have high CRP level of more than 12.3, then they are far more likely to respond to sarilumab rather than a TNF inhibitor.
"So we generate this data from clinical trial and what we want to do is apply it in routine clinical setting. First, to validate that this observation is true in routine clinical settings. But, I think, one of the potential benefits is that if this is indeed the case, then it will reduce the chance of patients not responding to a biologic treatment. And, target patients who are most likely to benefit from one class of biologic agents---in this case, sarilumab treatment.
"So I think the other benefit of this study is to demonstrate the potential use of machine learning, or artificial intelligence, in the future for helping us identify clinical characteristics and biomarkers that help us to predict response to treatment."
ABSTRACT NUMBER: 2006. Identification of a Rule to Predict Response to Sarilumab in Patients with Rheumatoid Arthritis Using Machine Learning and Clinical Trial Data. Date: Monday, November 9, 2020