Blood serum to diagnose osteoarthritis – biomarkers and machine learning

Abstract

s / Osteoarthritis and Cartilage 22 (2014) S57–S489 S64 97 BLOOD SERUM TO DIAGNOSE OSTEOARTHRITIS – BIOMARKERS AND MACHINE LEARNING B.J. Heard, J.M. Rosvold, M.J. Fritzler, H. El-Gabalawy, J.P. Wiley, R.J. Krawetz. Univ. of Calgary, Calgary, AB, Canada Purpose: Although reliable diagnostics for rheumatoid arthritis (RA) exist, currently only relatively advanced osteoarthritis (OA) can be clinically diagnosed with high confidence using a combination of physical exam / functional assessment with radiographic evidence of OA related structural changes (e.g. joint space narrowing, osteophytes). The objective of this study was to develop a method for categorizing patients with OA and RA, as well as discriminating both groups from normal individuals based on a panel of inflammatory cytokines expressed in blood serum samples. Methods: Inclusion criteria required a diagnosis of mild/moderate OA performed by a sports medicine physician at the University of Calgary based on clinical symptoms for 3 months with radiographic evidence of changes associated with OA. Inclusion for RA patients required physician diagnosis (autonuclear antibody positive). RA patients were all receiving various anti-inflammatory treatments at the time serum was sampled. Normal control individuals were disqualified if any personal or family history of arthritis or autoimmune diseases were disclosed (e.g. systemic lupus erythematosus, scleroderma, polymyositis, vasculitis, spondyloarthropathies, inflammatory bowel disease, diabetesmellitus type I, and/or thyroid disease). Blood serumwas collected and a panel of 38 inflammatory proteins was quantified using Luminx Multiplexing. This panel was then used as a training set in the construction of an artificial neural network (ANN) (Matlab – ANN toolbox) to indentify patients with OA (n 1⁄4 100), RA (n 1⁄4 100), and normal individuals (n 1⁄4 100). The results for the ANN were verified by reexamining the data using a decision tree algorithm (Matlab – Treebagger Algorithm). Results: High levels of sensitivity and specificity were obtained within the ANN using all 38 cytokines (sensitivity/specificity – Norm: 100/100, OA: 100/100, RA: 100/100). These results were verified through the reinvestigation of the data set using a decision tree algorithm that returned a similar sensitivity/specificity (Norm: 100/96, OA: 100/97, RA: 95/100). Results have been summarized in Table 1, this table shows a summary of the ANN (right) training, validation, and test sets, as well as the decision tree (left) training, test, and all data sets. Data is presented as the number of patients allocated by patient type on the left, and the diagnosis (categorization by the algorithm) on the top. Conclusion: Using the panel of 38 cytokines, we were able to train an ANN to distinguish between our 3 patient groups with a high degree of sensitivity and specificity. These results were verified using a decision tree algorithm. Joint imaging techniques are improving in the diagnosis of OA, but the need still exists for a reliable non-invasive diagnostic tool to at least support an early and accurate diagnosis. Through the use of a panel of well documented inflammatory cytokines, we have developed a non-invasive test that can be used to identify patients with OA of the knee. Although there is currently no disease modifying treatment for early OA, however, the panel of cytokines that we have developed may prove useful for the development and evaluation of future