Session Type: Abstract Submissions (ACR)
Treatment using biologics is widely used for rheumatoid arthritis (RA) in these days. Prediction of the clinical response to biologics prior to the administration is important mainly because of their price, efficacy and adverse events. Several methods for predicting the response to biologics have been reported, however, it is hard to know if they are applicable for the real-world patients. The aim of the present study was to generate a novel method for predicting the clinical response to infliximab, using machine-learning algorithm with only clinical data obtained before treatment in real-world RA patients.
We obtained about 40 components of the clinical examinations from 141 patients with RA (Group-1) before infliximab (IFX) treatment at the Saitama Medical University Hospital and Jichi Medical University Hospital. Patients fulfilling the 1987 revised ACR classification criteria were assessed for overall disease activity using DAS28-CRP before IFX treatment and after 12 weeks and were divided into two groups (responder group and non-responder group) according to EULAR response criteria. These data were used for training. To determine the best machine-learning algorithm and the best clinical parameter set, 50 algorithms in the WEKA software package, which consisted of a collection of machine-learning algorithms for data mining tasks, were compared using the training data. As a criterion for selecting combinations of algorithm and parameter set, the accuracy of prediction was required over 95%. However, the predictive accuracies were lower, mainly in the 70-80%. Therefore, we developed the technique to weight each clinical parameter in the training data, and could generate the high-versatility prediction score over 90 %. Next, the selected combination methods were applied to other clinical data which were obtained from 38 patients with RA (Group-2) before IFX treatment at St.Luke’s International Hospital and Hamamatsu Medical University Hospital. Finally, the best prediction method was selected from the combinations of algorithm and parameter set.
The combination of Multilayer Perception algorithm (neural network) and 9 clinical parameters shows the best accuracy performance, compared to the others. This prediction method could completely reproduce (100% accuracy) the result of training data (Group-1). This method was applied to the clinical data of other hospitals (Group-2) and could predict the good or moderate response to IFX with 92% accuracy. The positive prediction value of this method was 93.5%, while the negative value was 85.7%.
We have developed a novel method for predicting the clinical response to IFX, using Multilayer Perceptron algorithm with 9 clinical parameters of RA patients before treatment. This method has predicted the clinical response of IFX on different groups of RA patients with 92% accuracy. We believe that our method for predicting the response to IFX in real-world RA patients has advantages over the other methods in several points including easy usability, cost-effectiveness and accuracy.
Mitsubishi Tanabe Pharma.,
« Back to 2013 ACR/ARHP Annual Meeting
ACR Meeting Abstracts - https://acrabstracts.org/abstract/a-novel-method-predicting-good-response-using-only-background-clinical-data-in-ra-patients-treated-with-infliximab/