Jayadev Chethan, Hulley Philippa, Swales Catherine, Snelling Sarah, Collins Gary, Taylor Peter, Price Andrew
Royal National Orthopaedic Hospital NHS Trust, Stanmore, UK.
Nuffield Department of Orthopaedics, Rheumatology, and Musculoskeletal Sciences, University of Oxford, Oxford, UK.
Bone Joint Res. 2020 Oct 12;9(9):623-632. doi: 10.1302/2046-3758.99.BJR-2019-0192.R1. eCollection 2020 Sep.
The lack of disease-modifying treatments for osteoarthritis (OA) is linked to a shortage of suitable biomarkers. This study combines multi-molecule synovial fluid analysis with machine learning to produce an accurate diagnostic biomarker model for end-stage knee OA (esOA).
Synovial fluid (SF) from patients with esOA, non-OA knee injury, and inflammatory knee arthritis were analyzed for 35 potential markers using immunoassays. Partial least square discriminant analysis (PLS-DA) was used to derive a biomarker model for cohort classification. The ability of the biomarker model to diagnose esOA was validated by identical wide-spectrum SF analysis of a test cohort of ten patients with esOA.
PLS-DA produced a streamlined biomarker model with excellent sensitivity (95%), specificity (98.4%), and reliability (97.4%). The eight-biomarker model produced a fingerprint for esOA comprising type IIA procollagen N-terminal propeptide (PIIANP), tissue inhibitor of metalloproteinase (TIMP)-1, a disintegrin and metalloproteinase with thrombospondin motifs 4 (ADAMTS-4), monocyte chemoattractant protein (MCP)-1, interferon-γ-inducible protein-10 (IP-10), and transforming growth factor (TGF)-β3. Receiver operating characteristic (ROC) analysis demonstrated excellent discriminatory accuracy: area under the curve (AUC) being 0.970 for esOA, 0.957 for knee injury, and 1 for inflammatory arthritis. All ten validation test patients were classified correctly as esOA (accuracy 100%; reliability 100%) by the biomarker model.
SF analysis coupled with machine learning produced a partially validated biomarker model with cohort-specific fingerprints that accurately and reliably discriminated esOA from knee injury and inflammatory arthritis with almost 100% efficacy. The presented findings and approach represent a new biomarker concept and potential diagnostic tool to stage disease in therapy trials and monitor the efficacy of such interventions.Cite this article: 2020;9(9):623-632.
骨关节炎（OA）缺乏改善病情的治疗方法与合适生物标志物的短缺有关。本研究将多分子滑液分析与机器学习相结合，以生成用于终末期膝骨关节炎（esOA）的准确诊断生物标志物模型。
使用免疫测定法对esOA患者、非OA膝关节损伤患者和炎性膝关节炎患者的滑液（SF）进行35种潜在标志物分析。采用偏最小二乘判别分析（PLS-DA）得出用于队列分类的生物标志物模型。通过对10例esOA患者的测试队列进行相同的广谱SF分析，验证生物标志物模型诊断esOA的能力。
PLS-DA生成了一个精简的生物标志物模型，具有出色的敏感性（95%）、特异性（98.4%）和可靠性（97.4%）。八生物标志物模型生成了esOA的指纹图谱，包括IIA型前胶原N端前肽（PIIANP）、金属蛋白酶组织抑制剂（TIMP）-1、含血小板反应蛋白基序的解聚素和金属蛋白酶4（ADAMTS-4）、单核细胞趋化蛋白（MCP）-1、干扰素-γ诱导蛋白-10（IP-10）和转化生长因子（TGF）-β3。受试者工作特征（ROC）分析显示出出色的鉴别准确性：esOA的曲线下面积（AUC）为0.970，膝关节损伤为0.957，炎性关节炎为1。生物标志物模型将所有10例验证测试患者正确分类为esOA（准确性100%；可靠性100%）。
滑液分析与机器学习相结合产生了一个部分验证的生物标志物模型，该模型具有队列特异性指纹图谱，能够以近100%的效能准确、可靠地将esOA与膝关节损伤和炎性关节炎区分开来。所呈现的研究结果和方法代表了一种新的生物标志物概念和潜在的诊断工具，可用于治疗试验中的疾病分期以及监测此类干预措施的疗效。引用本文：2020；9（9）：623-632。
