机器学习在类风湿性关节炎诊疗及并发症预测中的研究进展
doi: 10.7655/NYDXBNSN250252
严舒桐 , 刘琪
天津中医药大学医学技术学院,天津 300000
基金项目: 天津市教委科研计划项目(2023KJ143)
Machine learning in rheumatoid arthritis:advances in clinical diagnosis,treatment,and complication prediction
YAN Shutong , LIU Qi
College of Medical Technology,Tianjin University of Traditional Chinese Medicine,Tianjin 300000 ,China
摘要
类风湿性关节炎(rheumatoid arthritis,RA)作为一种慢性系统性自身免疫性疾病,以滑膜炎和进行性关节破坏为特征,具有高致残率与复杂并发症风险,严重威胁患者生活质量。尽管传统诊疗方案可以显著改善患者症状,但其早期诊断困难、治疗反应个体差异大及心血管疾病、间质性肺病等并发症的发生率高,仍是临床实践中的焦点问题。近年来人工智能与机器学习技术的快速发展,为突破RA诊疗瓶颈提供了全新机遇。通过深度挖掘RA多模态医学数据,如影像学、基因组学和电子健康记录等,机器学习模型在早期诊断、治疗反应分层管理及并发症风险建模,如心血管事件预警中均展现出潜在优势。但数据异质性、模型可解释性不足及临床转化障碍等问题仍制约其广泛应用。文章旨在系统梳理机器学习在RA诊疗中的最新研究进展,为RA依据机器学习技术实现精准诊疗提供理论依据与实践参考。
Abstract
Rheumatoid arthritis(RA)is a chronic systemic autoimmune disease characterized by synovitis and progressive joint destruction,leading to a high disability rate and complex complications that severely compromise patients’quality of life. Although traditional diagnosis and treatment have been shown to significantly enhance patient symptoms,challenges such as the difficulty of early diagnosis,substantial individual variability in treatment response,and the high prevalence of complications,including cardiovascular diseases and interstitial lung disease,continue to be focal points in clinical practice. Recent rapid advancements in artificial intelligence and machine learning(ML)technologies offer novel opportunities to overcome these bottlenecks in RA management. By deeply mining RA multimodal medical data,such as patient imaging,genomics,and electronic health records,ML models have shown potential advantages in early diagnosis,stratified management of treatment response,and modeling of complication risk,such as early warning of cardiovascular events. However,barriers such as data heterogeneity,limited model interpretability,and challenges in clinical translation hinder its widespread adoption. This paper systematically reviews the latest research progress in applying ML to RA diagnosis and treatment,aiming to provide a theoretical foundation and practical reference for realizing ML-driven precision medicine in RA.
类风湿性关节炎(rheumatoid arthritis,RA)是一种慢性系统性自身免疫性疾病,在过去30年里,RA及其相关并发症发病率不断上升。有研究显示,2050年患者数将达到3亿[1]。其发病机制与多种炎症细胞浸润引起的关节缺氧和免疫失调密切相关[2]。RA引起的长期慢性骨骼、关节炎症可导致疼痛、关节畸形,严重可致滑膜组织失去功能[3]。目前主要采取抗风湿药的常规疗法来改善患者病情,最常使用的药物为甲氨蝶呤与糖皮质激素,它们能够抑制干扰素等致炎因子、诱导白介素(interleukin,IL)⁃10 等抗炎因子的产生,但该类药物长期使用会出现明显的不良反应,如代谢紊乱、胃肠道溃疡和骨质疏松等,且该类药物仅能改善患者症状,难以根治 RA[4]。现阶段RA的诊断主要依据患者临床表现、实验室检查以及影像学检查,但其早期症状无特异性[5-6],常与骨关节炎等疾病混淆,现有指标也无法全面反映疾病严重程度以及伴发并发症的风险,同时由于人群异质性,不同患者对治疗药物的敏感度也存在明显差异[7-9],亟待强化诊断体系以提高临床对RA的治疗效力。
近年来人工智能(artificial intelligence,AI)逐渐兴起,机器学习(machine learning,ML)作为AI 的一个分支,是一种利用计算机系统整合已有数据进行自主学习与深度分析的技术。通过构建模型,ML 可从现有数据中发现规律,从而对新数据进行预测或决策。由于ML能够处理复杂、高维的数据,并通过变量间的关联得出规律,因此被广泛应用于医疗领域进行疾病诊断、预测疾病进展和药物研发等[10]
ML算法目前分为有监督、无监督、半监督和强化学习4种,监督学习通过标记数据进行训练,主要包括分类分析与回归分析,常见模型有支持向量机 (support vector machine,SVM)、随机森林(random forest,RF)、最小绝对收缩和选择运算符(least abso⁃ lute shrinkage and selection operator,LASSO)等;无监督学习对未标记数据进行处理,模型需要自己发现数据中的结构和模式,包括聚类、降维和特征学习;半监督学习兼具监督和无监督学习的特点,通过利用少量标记数据和大量未标记数据进行模型训练;强化学习通过与环境交互解决问题,通过不断试错进行纠正[11]。不同算法各具优势,因此要依据目的选择合适模型,当单一模型无法有效解决问题时,亦可采用混合模型达到最优效果。现阶段RA诊疗模型常依托监督学习模式,主要应用算法见表1。近年来,ML的快速发展,为RA诊疗提供了新的机遇与挑战,文章旨在系统梳理ML在RA及其并发症诊疗中的最新研究进展,为RA诊疗的智能决策提供理论依据与实践参考。
1RA诊疗常用算法汇总表
Table1Summary table of common algorithms for RA diagnosis and treatment
1 影像学ML多模态数据赋能诊疗RA
ML 结合影像学检测对 RA 的早期诊断有极高价值。未分化关节炎(undifferentiated arthritis,UA) 有进展为 RA 的可能,预测其进展仍是临床难题。基于此,Hu等[12] 对432例UA患者进行为期1年的随访,其中152例(35.2%)进展为RA,RF模型集成18个关节超声评分系统(ultrasound 18,US18)与临床基线数据,对该结果展现出极高的预测灵敏度和准确度,在此基础上,利用沙普利加性解释(Shapley additive explanations,SHAP)分析还找出关键变量,证实 US18 中 2 级关节计数、US18 总分和肿胀关节计数对早期RA的检测至关重要。另一研究团队通过风湿病门诊招募符合入组条件的143例RA患者,将其随机分为训练集和验证集,放射科医生将训练集患者的临床数据、静态和动态超声影像结果依据欧洲抗风湿病联盟⁃风湿病评估量表(European Alliance of Associations for Rheumatology ⁃ Outcome Measures in Rheumatology,EULAR⁃OMERACT)、滑膜内血流情况(0~3分)和滑膜增生肥大(0~3分)进行赋分,用以构建卷积神经网络评分模型(ResNet50)进而评估验证集 RA 的活动度,最终模型评估结果与医生诊断保持一致,提示该模型有辅助放射科医生进行 RA 诊断的潜能[13]
ML 结合影像学检测也可监测 RA 并发症。骨质疏松症是 RA 患者最常见的关节并发症,为了找寻更快捷的评估工具,Saito等[14] 尝试使用手部 X线片计算第二掌皮质指数(second metacarpal cortical index,2MCI)预测骨密度(bone mineral density, BMD),最终成功结合临床指标构建出了具有高性能与高泛化能力预测RA患者骨质疏松的ML模型。间质性肺病(interstitial lung disease,ILD)是RA的关节以外并发症,McDermott 等[15] 使用定量计算机断层扫描(quantitative computed tomography,QCT)对 RA 患者及健康对照肺部数据利用多变量线性回归分析,发现 RA 与肺部较高的间质变化相关,且合并肺气肿的 RA 患者病死率远高于单纯 RA 患者。 DAS⁃28评分是临床评价风湿活动性的传统指标[16], Zhou 等[17] 通过对 RA 患者 28 个关节评分(Disease Activity Score based on 28⁃joint count,DAS28)结合实验室检测指标等,构建的列线图模型则可用于 RA 患者并发 ILD 风险的预测,其中高龄、吸烟、DAS28 评分升高、影像学关节分期较高(Ⅱ 期和 Ⅲ 期)和甲氨蝶呤治疗等被确定为 RA⁃ILD 的独立危险因素。还有多项研究使用深度卷积神经网络(convolu⁃ tional neural network,CNN)对十几万张胸部有异常的患者X线片进行分析,建立了名为DLAD⁃10(Deep⁃ Learning Algorithm Detecting10 Common Abnormal⁃ ities)和CheXNeXt(Chest X⁃ray NeXt)的DL模型,可及时有效地辅助医生判断患者胸部的异常表现,同样有诊断RA⁃ILD的潜能[17-18]
2 分子生物学ML多模态数据赋能诊疗RA
ML 结合基因检测结果有效提高 RA 的诊断率。循环抗瓜氨酸肽抗体(anti ⁃cyclic citrullinated peptide antibody,ACPA)是 RA 诊断的特异性指标,但很多患者早期的ACPA呈现阴性,这就使得RA诊断常出现延迟[19]。有研究收集173例RA患者外周血CD4+ T细胞行基因芯片检测,以发病初期明确为 RA的111份样本作为训练集构建出12基因诊断模型,该模型对于验证集ACPA阴性的RA患者的诊断灵敏度为0.85,特异度为0.75,提示基因测序数据的多基因特征模型有利于早期诊断RA[20]。类似地,一些研究利用基因表达公开数据集(gene expression omnibus,GEO)或收集样本的RA测序数据,使用加权基因相关网络分析(weighted gene co ⁃expression network analysis,WGCNA)与LASSO、支持向量机递归特征消除(support vector machine recursive feature elimination,SVM⁃RFE)和 RF 算法等识别出滑膜中 RRM2、DLGAP5、KIF11、AKR1C3、MCEE、POLE4 和 PFKM,CD8 + T 细胞中的 GDF15、IGLC1、IGHM、 CD8A、GZMA 和 PRF1,血小板细胞关联密切的 MAPK3、ACTB、ACTG1、VAV2、PTPN6和ACTN1 [21-25],滑膜趋化因子CXCL4与CXCL7均可作为RA早期发生的标志[26]。也有研究使用单细胞测序结果结合极端梯度提升(extreme gradient boosting,XGBoost)算法识别出 RA 中 MIER1、PPP1CB、ICOS、GADD45A 和 CD3D 这 5 个上调基因与 SLFN5、PIP4K2A 和 IL6ST 这3个下调基因可以联合进行RA诊断[27]。一研究团队选取 72 例 RA、8 例 UA 和 13 例健康志愿者的外周血单个核细胞(peripheral blood mononuclear cell,PBMC)DNA样本进行甲基化测序,使用SVM算法构建UA向RA分化的DNA甲基化特征谱,进而监测UA进展[28],而Geng等[29] 则通过合并GEO数据库 RA 滑膜组织的基因测序结果,共有 233 例 RA 和 126例对照,通过构建LASSO模型找出了IGF2BP3和 YTHDC2两个m6A甲基化调节因子,相比健康者,RA 患者的IGF2BP3和YTHDC2显著升高。另一研究利用公共数据库数据集对 RA 患者外泌体差异表达 lncRNA使用SVM、KNN、RF和Logit算法进行DL,找出了DLEU2、FAM13A⁃AS1、MEG3和SNHG15 这4种具有RA代表性的基因,它们有作为RA早期诊疗靶标的潜能[30]。针对代谢综合征相关 RA(metabolic syndrome ⁃ rheumatoid arthritis,MetS ⁃ RA)患者,从 GEO 数据库中获得了 3 个 RA 数据集和 1 个 MetS 数据集,利用LASSO和RF识别出TYK2和TRAF2可作为 MetS⁃RA 的潜在标志物[31]。另外,使用 ML 整合患者临床数据可对患者的疼痛等级进行精确分类,从而帮助医生开展治疗[32]。也有研究基于图谱的基因表达模块鉴定(graph ⁃based gene expression module identification,GbGMI)算法表明滑膜成纤维细胞中815个基因与疼痛有关[33]。前述研究均表明 ML 结合基因测序数据对 RA 的精准诊断和明确分级有潜在价值。
ML 利用分子检测结果亦可辅助 RA 并发症的预判。有研究使用153例伴发ILD的RA和单纯RA 患者静脉血标本,采取LASSO、RF、偏最小二乘回归 (partial least squares regression,PLS)共同找出KL⁃6、 D⁃二聚体和CA19⁃9可作为判断RA⁃ILD的潜在血液生物标志物[34]。无监督ML通过聚类分析显示,RA 与精神/行为类或心血管类、慢性疼痛障碍类合并症相关性最高[35-36],约 15%的 RA 患者会出现抑郁症状,有研究使用LASSO结合SVM/RF算法对RA合并重度抑郁症患者的特异性差异表达基因构建模型,找出了 AURKA、BTN3A2、CXCL10、ERAP2、MARCO、 PLA2G7、EAF1、SDCBP和RNF19B作为潜在治疗靶点[37-38]。还有研究发现RA会加重动脉粥样硬化性心血管疾病,使用 RF 或递归神经网络(recurrent neural network,RNN)等算法可有效辅助判断RA患者并发心血管疾病的风险[39-40]。Feng 等[41] 使用 LASSO、RF和逻辑回归(logistic regression,LR)交叉找出低密度脂蛋白(low⁃density lipoprotein,LDL)、调节性T细胞(regulatory T cell,Treg)占比和总胆固醇 (total cholesterol,TC)水平是RA伴发心血管疾病的危险因素。LASSO 与 RF 算法表明,NFIL3、EED、 GRK2、MAP3K11、RMI1和TPST1 [42],还有9种miRNA (let⁃7c⁃5p、miR⁃30e⁃5p、miR⁃4446⁃3p、miR⁃126⁃5p、 miR⁃3168、miR⁃425⁃5p、miR⁃126⁃3p、miR⁃30a⁃5p 和 miR⁃125a⁃5p)对RA伴动脉粥样硬化具有较高的诊断价值[43]。风湿症状可能伴随恶性肿瘤,为此 Gente等[44] 使用多变量Logistic回归分析,选取了4种代谢物浓度和1种脂质比值作为模型变量,构建出曲线下面积(area under curve,AUC)为 0.995 的 RA 诊断模型,其灵敏度与特异度也较高,可用于预测包括RA在内的约2/3的风湿性肌肉骨骼疾病(rheu⁃ matic musculoskeletal disorder,RMD)患者同期患有癌症的风险。
3 细胞标志物ML多模态数据赋能诊疗RA
ML结合不同种类细胞相关标志物可进一步辅助诊断 RA。有研究报道细胞铁死亡、铜死亡等可加重RA的病理过程,通过GEO数据库中RA数据集的差异基因与844个铁死亡相关基因取交集,LASSO 分析交集基因发现 MMP13 升高与 GABARAPL1 降低可作为 RA 的标志[45]。类似地,铜死亡相关基因 SLC35A1、PRPF39、MAP4K3、TMX1 和 FAM96A 具有潜在的诊断价值,可用于预测 RA 患者的病理结局[46]。乳酸化作用在免疫细胞代谢及基因表达改变中具有重要调节作用,Fu 等[47] 采用 134 种 ML 算法组合对RA基因测序结果进行训练,找出7个核心促乳酸化相关基因,该诊断模型 AUC 可达 0.918, qRT ⁃PCR 证实 NDUFB3、NGLY1 和 SLC25A4 在 RA 中高表达。为研究乳酸代谢与RA的关系,Yang等[48] 使用类似方法找到与巨噬细胞功能相关的KCNN4 和SLC25A4以及参与树突状细胞免疫的GATA2基因,它们均有望成为RA治疗靶点。Adami等[49] 从骨代谢角度出发,使用主成分分析(principal compo⁃ nent analysis,PCA)方法发现骨中较低的 PINP 和 B⁃ALP可有效诊断RA。另外,根据血清傅里叶变换红外光谱和DL算法构建的MSResNet模型也具有良好区分强直性脊柱炎(ankylosing spondylitis,AS)、骨关节炎(osteoarthritis,OA)和RA患者的性能[50]。以上均说明ML 结合与RA 密切相关的细胞标志物亦有提高临床RA诊断能力的潜在价值。
4 ML能预测RA药物疗效
推动精确医学、提高药物疗效是治疗 RA 的关键[51]。Li 等[52] 收集了 56 例甲氨蝶呤治疗失败 (methotrexate⁃inadequate response,MTX⁃IR)并计划采用肿瘤坏死因子抑制剂(tumor necrosis factor interference,TNFi)治疗的RA患者的血浆外泌体,对其基因测序结果使用RF和LASSO回归各自产生核心基因后作交集,并最终确定has⁃circ0002715与has⁃ circ0001946为关键基因,可用于预测MTX⁃IR RA患者及作为开启TNFi治疗的标志物。并且,监测MTX 治疗前后4周内的基因变化也有助于判断患者是否适合MTX治疗[53]。同时,MTX⁃IR 还与高淋巴细胞计数有关,淋巴细胞计数高于2 000个/mL 的患者, MTX治疗更易失败[54]。Zhang等[55] 开发出新型纳米级材料运载药物至发病部位进行释放,提高了药物治疗RA的精准度,减少了药物不良反应,同时结合自适应神经元模糊推理系统(adaptive neuron fuzzy inference system,ANFIS)和XGBoost ML模型可预测该方式药物的治疗效能。还有研究围绕治疗RA的一种中药白花蛇舌草,利用网络药理学和生物信息学展开分析,最终通过LASSO、RF和SVM⁃RFE算法及分子对接确认MMP9为该药物治疗RA的核心靶点[56]。Ye等[57] 利用7年间的RA患者就诊数据及就诊时气象因素和空气污染,利用长短期记忆(long short⁃term memory,LSTM)模型有效预测RA 患者就诊情况,有利于优化医疗资源分配。
约40%的RA患者对特定药物治疗无反应,使用 ML可针对性地给予不同患者个性化的用药建议[58-59]。 Rivellese等[60] 使用利妥昔单抗以及托珠单抗治疗的 RA患者滑膜组织行RNA测序,利用10×10倍嵌套交叉验证方法结合测序结果详细找出了与利妥昔单抗相关的40个基因和与托珠单抗相关的39个基因,其预测药物有效性的准确度分别达 0.744 和 0.681。 Ukalovic等[61] 依托患者临床数据利用交叉验证方法开发5种RA生物特定疾病修饰药物(disease⁃modify⁃ ing drug,DMARD)的疗效预测模型,发现阿达木单抗(adalimumab,ADA)和依那西普(etanercept,ETN) 联用治疗效能最高。对RA患者外周血CD4+ T细胞进行RNA测序及DNA甲基化测序,利用XGBoost和 RF 算法依据差异基因和差异甲基化位点(differen⁃ tially methylated position,DMP)训练出的模型可有效预测患者对ADA和ETN的治疗反应,相比药物治疗无效者,有疗效的患者呈现基因高甲基化状态,其准确率分别达 84.7%和 88.0%[62-63]。类似地,使用 SVM算法开发出的结合临床基线数据和外周血7种 DMP 评分的预后模型可用于判断患者对来氟米特治疗是否敏感[64]。有研究整合了 RA 外周血基因的转录组图谱,并对外周血中404个上调的差异表达基因(differentially expressed gene,DEG)进行功能富集分析,使用ML中的无监督聚类分析确定3种最佳分类亚型,分别为中性粒细胞驱动、干扰素驱动和 CD8+ T 细胞驱动,有助于针对RA的个性化给药[65]。 Baxter等[66] 使用XGBoost结合5 481例患者临床监测数据训练出判断 RA 患者 5 年接受手术的概率 (AUC=0.90±0.02)以及接受关节置换还是其他手术 (AUC=0.58±0.10)的模型,以期实现对RA患者的精准治疗。
RA除可造成关节与器官损伤外,还会伴随其他多系统临床症状。RA易造成严重感染(serious infection, SI),Hetland等[67] 通过统计学结合ML技术对8 404例使用托法替布治疗的 RA 患者数据进行分析,进一步证实了年龄较大、男性、既往感染史和基线时皮质类固醇使用与SI风险密切相关,而Mehta等[68] 则是使用LASSO分析得出,相比非炎性风湿及肌肉骨骼疾病,RA患者SI风险显著增加。
5 总结与展望
随着 ML 的逐步推广,医学领域也开始不断尝试融合这种先进方法以寻找各类疾病相关的靶向基因、通路和生物标志物等,进而发现潜在诊疗靶点及治疗途径。目前已开发出大量 RA 相关的 ML 模型(表2),这些模型的构建为参与测试的患者提供了病情分类、基因差异分析、生存期预估相关的结果,有助于患者清晰了解自身状况,为患者的治疗与预后提供了新方向,且ML可以有效地将影像、血液检测和基因检测,如临床基础数据、关节评分、免疫细胞计数、特异性抗体检测、核酸标志物检测等多维度数据进行整合分析,高精准度、低错误率地迅速产生诊断建议,辅助临床医生高效诊疗RA,具有潜在应用价值(图1)。但其应用于临床实际仍存在较大困难,挑战之一是如何有效融合多模态数据,依据《中国类风湿关节炎发展报告2020》[69],RA 的诊断标准主要包括患者伴有晨僵等临床症状,手部X线检查显示骨质疏松以及血清类风湿因子水平升高等,这些数据检测手段不同,结果差异大,现有的算法进行融合分析仍存在技术困难。再者若想将所建立的模型直接应用于临床疾病诊断,则需要把所开发的算法模型嵌入现有的如影像、检验等大型仪器中,这就需要提高仪器之间的信息互传以及数据兼容性,目前这一工作难度较高,且伴随科技飞速发展,RA诊断模型的开发愈来愈全面和深入,实时数据更新和系统维护也将成为重点和难点。且 ML 训练模型需要依赖大量原始数据,但当前许多研究中使用的样本量偏少,尽管研究者们会通过采取交叉验证、特征降维等方式来对模型进行复查以获得较高准确率,但仍无法完全弥补因数据量过少所引起的模型不确定性,无法实现推广,导致投入临床使用的模型较少,更多仅是停留在实验室研究测试层面。为了提高模型的质量、通用性和可用性,后续则需要纳入更多数据并进行多中心验证。
即使基于ML诊疗RA仍存在诸多风险与挑战,但其依然具有一定的意义和价值,持续优化 ML 搭建诊疗RA的可推广、可持续性模型具有提高RA患者早期诊断率以及改善预后的潜能。如哈佛大学研究团队开发了一款临床组织病理学成像评估基础模型,已在全球多家医院验证了其对肿瘤标本的诊断准确性,预计在2~3年内推出商业化工具[70];另外北京邮电大学科研团队通过多种医学数据和综合评估微调构建的大型语言模型ClinicalGPT,在训练过程中融入广泛且多样化的现实世界数据,如病历、特定领域知识和多轮对话咨询,通过医学知识问答、医疗检查、患者咨询和病历诊断分析等进行多维度验证,其诊断能力均显著优于其他模型[71]。虽然前述模型并未特别针对 RA 诊疗,但模型构建机制对于开发 RA 诊疗模型有很高的借鉴价值。
2ML赋能诊断RA及其并发症
Table2ML empowers diagnosis of RA and its complications
(续表2)
RF:random forest;CNN:convolutional neural network;DCNN:deep convolutional neural network;ANN:artificial neural network;Bi⁃LSTM:bidi⁃ rectional long short term memory;AM:additive model;SVM:support vector machine;SVM⁃RFE:support vector machine recursive feature elimination; SVM⁃RBF:support vector machine radial basis function;LASSO:least absolute shrinkage and selection operator;LSTM:long short term memory; MSResNet:multiscale residual network;GBT:gradient boosting decision tree;SSGB:stability selection gradient boosting;LGBM:light gradient boosting machine;XGBoost:eXtreme gradient boosting;glmBoost:generalized linear model boost;GMM:gaussian mixture model;ANFIS:adaptive neuron fuzzy inference system;RC:ridge regression;ENR:elastic net regression;KNN:K nearest neighbor;PLS:partial least squares regression;CC:consensus clus⁃ ter plus;DEC:deep embedded cluster;LR:logistic regression;MA:multivariate analysis;MLR:multiple linear regression;SFS:sequential forward selection.
1ML辅助RA诊疗流程图(Office Plus绘制)
Figure1ML assisted RA diagnosis and treatment process diagram(drawn by Office Plus)
现阶段还需突破ML对于不同维度数据的整合分析能力,伴随AI,例如DeepSeek等相关大模型的开发,大幅提高了对不同数据的整合能力,并呈现数据处理过程的详细解析。同时这种诊断模型是整个医院甚至整个社会公用的,这也就需要全社会建立公认算法,并加强对医护人员的持续培训。最后,加强多中心交互实践,汇集多家医院、诊疗中心以及实验室的检测数据进行整合分析,建立可推广、普适性的诊疗模型,同期构建诊疗模型监管网络,杜绝人为篡改或擅用医疗数据,以便真正助力诊疗,服务患者。
利益冲突声明:
所有作者声明无利益冲突。
Conflict of Interests:
All authors declare no conflict of interests.
作者贡献声明:
严舒桐负责文献检索、论文初稿撰写;刘琪负责选题指导、论文检查审核修订。
Authors Contributions:
YAN Shutong was responsible for literature search,initial draft writing of paper;LIU Qi was responsible for topic selec⁃ tion guidance,paper inspection,review and revision.
1ML辅助RA诊疗流程图(Office Plus绘制)
Figure1ML assisted RA diagnosis and treatment process diagram(drawn by Office Plus)
1RA诊疗常用算法汇总表
Table1Summary table of common algorithms for RA diagnosis and treatment
2ML赋能诊断RA及其并发症
Table2ML empowers diagnosis of RA and its complications
DONG Q, WU J Y, ZHANG H G,et al. Deciphering im-munometabolic landscape in rheumatoid arthritis:integrative multiomics,explainable machine learning and experimental validation[J]. J Inflamm Res,2025,18:637-652
YANG F H, LV J Q, HUANG Y L,et al. A supramolecular assembly strategy for the treatment of rheumatoid arthritis with ultrasound-augmented inflammatory microenvironment reprograming[J]. Biomaterials,2025,316:123006
AO Y, LAN Q, YU T H,et al. Cellular senescence-associated genes in rheumatoid arthritis: Identification and functional analysis[J]. PLoS One,2025,20(1):e0317364
SAWAN S, KUMARI A, MAJIE A,et al.siRNA-based nanotherapeutic approaches for targeted delivery in rheumatoid arthritis[J]. Biomater Adv,2025,168:214120
MAJITHIA V, GERACI S A. Rheumatoid arthritis:diagnosis and management[J]. Am J Med,2007,120(11):936-939
VITTECOQ O, BREVET P, GERARD B,et al. On difficulties to define prognostic factors for clinical practice in rheumatoid arthritis[J]. RMD Open,2024,10(3):e004472
PARK S H, JEON H L, KIM S C,et al. Utilization of targeted disease-modifying anti-rheumatic drugs(DMARDs)for managing rheumatoid arthritis in the last decade:a two-country comparison[J]. Int J Clin Pharmacol Ther,2021,59(10):639-644
COX N, MALLEN C D, SCOTT I C. Pharmacological pain management in patients with rheumatoid arthritis:a narrative literature review[J]. BMC Med,2025,23(1):54
KERSCHBAUMER A, SEPRIANO A, SMOLEN J S,et al. Efficacy of pharmacological treatment in rheumatoid arthritis:a systematic literature research informing the 2019 update of the EULAR recommendations for management of rheumatoid arthritis[J]. Ann Rheum Dis,2020,79(6):744-759
TANAKA M, AKIYAMA Y, MORI K,et al. Machine learning-based analyses of contributing factors for the development of hypertension:a comparative study[J]. Clin Exp Hypertens,2025,47(1):2449613
SARKER I H. Machine learning:algorithms,real-world applications and research directions[J]. SN Comput Sci,2021,2(3):160
HU X L, LIU X M, XU Y,et al. Machine learning-based prediction model integrating ultrasound scores and clinical features for the progression to rheumatoid arthritis in patients with undifferentiated arthritis[J]. Clin Rheuma-tol,2025,44(2):649-659
HE X L, WANG M, ZHAO C Y,et al. Deep learning-based automatic scoring models for the disease activity of rheumatoid arthritis based on multimodal ultrasound images[J]. Rheumatology(Oxford),2024,63(3):866-873
SAITO R, FUJII T, MURATA K,et al. Prediction models incorporating second metacarpal cortical index for osteoporosis in rheumatoid arthritis:externally validated machine learning models developed using data from the KURAMA cohort[J]. Int J Rheum Dis,2024,27(10):e15358
MCDERMOTT G C, HAYASHI K, YOSHIDA K,et al. Rheumatoid arthritis,quantitative parenchymal lung features and mortality among smokers[J]. Rheumatology(Oxford),2025,64(1):133-142
朱巧英, 丁炎, 宣旻, 等. 类风湿性关节炎超声评分与DAS-28评分的相关性研究[J]. 南京医科大学学报(自然科学版),2019,39(1):90-93. ZHU Q Y, DING Y, XUAN W,et al. Study on the correlation between ultrasound score and DAS-28 score in rheumatoid arthritis[J]. Journal of Nanjing Medical University(Natural Sciences),2019,39(1):90-93
ZHOU W Y, YANG Y Y, SU L. Development and validation of predictive factors influencing rheumatoid arthritis associated with interstitial lung disease[J]. Clin Rheuma-tol,2025,44(2):601-613
RAJPURKAR P, IRVIN J, BALL R L,et al. Deep learning for chest radiograph diagnosis:a retrospective comparison of the CheXNeXt algorithm to practicing radiologists[J]. PLoS Med,2018,15(11):e1002686
O’NEIL L J, SPICER V, SMOLIK I,et al. Association of a serum protein signature with rheumatoid arthritis development[J]. Arthritis Rheumatol,2021,73(1):78-88
PRATT A G, SWAN D C, RICHARDSON S,et al. A CD4 T cell gene signature for early rheumatoid arthritis implicates interleukin 6-mediated STAT3 signalling,particularly in anti-citrullinated peptide antibody-negative disease[J]. Ann Rheum Dis,2012,71(8):1374-1381
WU Y K, LIU C D, LIU C,et al. Machine learning and weighted gene co-expression network analysis identify a three-gene signature to diagnose rheumatoid arthritis[J]. Front Immunol,2024,15:1387311
GUO Z Y, MA Y Y, WANG Y Q,et al. Identification and validation of metabolism-related genes signature and immune infiltration landscape of rheumatoid arthritis based on machine learning[J]. Aging(Albany NY),2023,15(9):3807-3825
WANG Q Z, LI Q Q, WANG R H,et al. Identification of CD8 + T cell-related biomarkers and immune infiltration characteristic of rheumatoid arthritis[J]. Aging(Albany NY),2024,16(2):1399-1413
ZHAO Z Y, REN J, XIE S P,et al. Identification of biomarkers associated with CD8+ T cells in rheumatoid arthritis and their pan-cancer analysis[J]. Front Immunol,2022,13:1044909
LIU Y C, JIANG H X, KANG T L,et al. Platelets-related signature based diagnostic model in rheumatoid arthritis using WGCNA and machine learning[J]. Front Immunol,2023,14:1204652
YEO L, ADLARD N, BIEHL M,et al. Expression of chemokines CXCL4 and CXCL7 by synovial macrophages defines an early stage of rheumatoid arthritis[J]. Ann Rheum Dis,2016,75(4):763-771
DING Q, XU Q Y, HONG Y N,et al. Integrated analysis of single-cell RNA-seq,bulk RNA-seq, Mendelian randomization,and eQTL reveals T cell-related nomogram model and subtype classification in rheumatoid arthritis[J]. Front Immunol,2024,15:1399856
DE LA CALLE-FABREGAT C, NIEMANTSVERDRIET E, CAÑETE J D,et al. Prediction of the progression of undifferentiated arthritis to rheumatoid arthritis using DNA methylation profiling[J]. Arthritis Rheumatol,2021,73(12):2229-2239
GENG Q S, CAO X X, FAN D P,et al. Diagnostic gene signatures and aberrant pathway activation based on m6A methylation regulators in rheumatoid arthritis[J]. Front Immunol,2022,13:1041284
ZHOU T S, YANG C L, WANG J Q,et al. Identification of serum exosomal lncRNAs and their potential regulation of characteristic genes of fibroblast-like synoviocytes in rheumatoid arthritis[J]. Int Immunopharmacol,2024,143(Pt 2):113382
HUANG Y F, YUE S K, QIAO J H,et al. Identification of diagnostic genes and drug prediction in metabolic syndrome-associated rheumatoid arthritis by integrated bioinformatics analysis,machine learning,and molecular docking[J]. Front Immunol,2024,15:1431452
LÖTSCH J, ALFREDSSON L, LAMPA J. Machine-learning-based knowledge discovery in rheumatoid arthritis-related registry data to identify predictors of persistent pain[J]. Pain,2020,161(1):114-126
BAI Z L, BARTELO N, ASLAM M,et al. Synovial fibroblast gene expression is associated with sensory nerve growth and pain in rheumatoid arthritis[J]. Sci Transl Med,2024,16(742):eadk3506
QIN Y, WANG Y L, MENG F X,et al. Identification of biomarkers by machine learning classifiers to assist diagnose rheumatoid arthritis-associated interstitial lung disease[J]. Arthritis Res Ther,2022,24(1):115
CROWSON C S, GUNDERSON T M, DAVIS J M 3rd,et al. Using unsupervised machine learning methods to cluster comorbidities in a population-based cohort of patients with rheumatoid arthritis[J]. Arthritis Care Res(Hobo-ken),2023,75(2):210-219
ENGLAND B R, YANG Y, ROUL P,et al. Identification of multimorbidity patterns in rheumatoid arthritis through machine learning[J]. Arthritis Care Res(Hoboken),2023,75(2):220-230
ZHOU T T, SUN J J, TANG L D,et al. Potential diagnostic markers and therapeutic targets for rheumatoid arthritis with comorbid depression based on bioinformatics analysis[J]. Front Immunol,2023,14:1007624
JIANG W, WANG X C, TAO D X,et al. Identification of common genetic characteristics of rheumatoid arthritis and major depressive disorder by bioinformatics analysis and machine learning[J]. Front Immunol,2023,14:1183115
TIWARI E, SHRIMANKAR D, MAINDARKAR M,et al. Artificial intelligence-based cardiovascular/stroke risk stratification in women affected by autoimmune disorders:a narrative survey[J]. Rheumatol Int,2025,45(1):14
AL-MAINI M, MAINDARKAR M, KITAS G D,et al. Artificial intelligence-based preventive,personalized and precision medicine for cardiovascular disease/stroke risk assessment in rheumatoid arthritis patients:a narrative review[J]. Rheumatol Int,2023,43(11):1965-1982
FENG M, MENG F X, JIA Y H,et al. Exploration of risk factors for cardiovascular disease in patients with rheumatoid arthritis:a retrospective study[J]. Inflammation,2024
LIU F Z, HUANG Y, LIU F H,et al. Identification of immune-related genes in diagnosing atherosclerosis with rheumatoid arthritis through bioinformatics analysis and machine learning[J]. Front Immunol,2023,14:1126647
ORMSETH M J, SOLUS J F, SHENG Q H,et al. Plasma miRNAs improve the prediction of coronary atherosclerosis in patients with rheumatoid arthritis[J]. Clin Rheuma-tol,2021,40(6):2211-2219
GENTE K, FEISST M, MARX D,et al. Altered serum metabolome as an indicator of paraneoplasia or concomitant cancer in patients with rheumatic disease[J]. Ann Rheum Dis,2024,83(8):974-983
CAI Y, DENG L C, YAO J. Analysis and identification of ferroptosis-related diagnostic markers in rheumatoid arthritis[J]. Ann Med,2024,56(1):2397572
ZHOU Y, LI X, NG L,et al. Identification of copper death-associated molecular clusters and immunological profiles in rheumatoid arthritis[J]. Front Immunol,2023,14:1103509
FU W C, WANG T B, LU Y H,et al. The role of lacty-lation in plasma cells and its impact on rheumatoid arthritis pathogenesis: Insights from single-cell RNA sequencing and machine learning[J]. Front Immunol,2024,15:1453587
YANG F, SHEN J Y, ZHAO Z M,et al. Unveiling the link between lactate metabolism and rheumatoid arthritis through integration of bioinformatics and machine learning[J]. Sci Rep,2024,14(1):9166
ADAMI G, FASSIO A, ROSSINI M,et al. Machine learning to characterize bone biomarkers profile in rheumatoid arthritis[J]. Front Immunol,2023,14:1291727
WU X, SHUAI W, CHEN C,et al. Rapid screening for autoimmune diseases using Fourier transform infrared spectroscopy and deep learning algorithms[J]. Front Immu-nol,2023,14:1328228
LIN C M A, COOLES F A H, ISAACS J D. Precision medicine:the precision gap in rheumatic disease[J]. Nat Rev Rheumatol,2022,18(12):725-733
LI G Q, CHEN H Y, SHEN J C,et al. Unveiling new therapeutic horizons in rheumatoid arthritis:an in-depth exploration of circular RNAs derived from plasma exosomes[J]. J Orthop Surg Res,2025,20(1):109
PLANT D, MACIEJEWSKI M, SMITH S,et al. Profiling of gene expression biomarkers as a classifier of methotrexate nonresponse in patients with rheumatoid arthritis[J]. Arthritis Rheumatol,2019,71(5):678-684
DUQUESNE J, BOUGET V, COURNÈDE P H,et al. Machine learning identifies a profile of inadequate responder to methotrexate in rheumatoid arthritis[J]. Rheumatology(Oxford),2023,62(7):2402-2409
ZHANG Z Y, YE M T, GE Y S,et al. Eco-friendly nanotechnology in rheumatoid arthritis: ANFIS-XGBoost enhanced layered nanomaterials[J]. Environ Res,2024,262(Pt 1):119832
JIANG J, HUANG M, ZHANG S S,et al. Identification of Hedyotis diffusa Willd-specific mRNA-miRNA-lncRNA network in rheumatoid arthritis based on network pharmacology,bioinformatics analysis,and experimental verification[J]. Sci Rep,2024,14(1):6291
YE Z, YE B J, MING Z L,et al. Forecasting rheumatoid arthritis patient arrivals by including meteorological factors and air pollutants[J]. Sci Rep,2024,14(1):17840
MENDOZA-PINTO C, SÁNCHEZ-TECUATL M, BERRA-ROMANI R,et al. Machine learning in the prediction of treatment response in rheumatoid arthritis:a systematic review[J]. Semin Arthritis Rheum,2024,68:152501
KALWEIT M, BURDEN A M, BOEDECKER J,et al. Patient groups in rheumatoid arthritis identified by deep learning respond differently to biologic or targeted synthetic DMARDs[J]. PLoS Comput Biol,2023,19(6):e1011073
RIVELLESE F, SURACE A E A, GOLDMANN K,et al. Rituximab versus tocilizumab in rheumatoid arthritis:synovial biopsy-based biomarker analysis of the phase 4 R4RA randomized trial[J]. Nat Med,2022,28(6):1256-1268
UKALOVIC D, LEEB B F, RINTELEN B,et al. Prediction of ineffectiveness of biological drugs using machine learning and explainable AI methods:data from the Austrian Biological Registry BioReg[J]. Arthritis Res Ther,2024,26(1):44
HAGEMAN I, MOL F, ATIQI S,et al. Novel DNA methy-lome biomarkers associated with adalimumab response in rheumatoid arthritis patients[J]. Front Immunol,2023,14:1303231
TAO W Y, CONCEPCION A N, VIANEN M,et al. Mul-tiomics and machine learning accurately predict clinical response to adalimumab and etanercept therapy in patients with rheumatoid arthritis[J]. Arthritis Rheumatol,2021,73(2):212-222
CHEN Y L, WANG Q, LIU H N,et al. The prognostic value of whole-genome DNA methylation in response to leflunomide in patients with rheumatoid arthritis[J]. Front Immunol,2023,14:1173187
CHANG M J, FENG Q F, HAO J W,et al. Deciphering the molecular landscape of rheumatoid arthritis offers new insights into the stratified treatment for the condition[J]. Front Immunol,2024,15:1391848
BAXTER N B, LIN C H, WALLACE B I,et al. Development of a machine learning model to predict the use of surgery in patients with rheumatoid arthritis[J]. Arthritis Care Res(Hoboken),2024,76(5):636-643
HETLAND M L, STRANGFELD A, BONFANTI G,et al. Machine learning prediction and explanatory models of serious infections in patients with rheumatoid arthritis treated with tofacitinib[J]. Arthritis Res Ther,2024,26(1):153
MEHTA B, PEDRO S, OZEN G,et al. Serious infection risk in rheumatoid arthritis compared with non-inflammatory rheumatic and musculoskeletal diseases:a US national cohort study[J]. RMD Open,2019,5(1):e000935
曾小峰, 田新平, 李梦涛. 中国类风湿关节炎发展报告-2020[M]. 沈阳: 辽宁科学技术出版社,2021. ZENG X F, TIAN X P, LI M T. Report on the development of rheumatoid arthritis in China.2020[M]. Sheng-yang:liaoning Science and Technology Publisher,2021
WANG X Y, ZHAO J H, MAROSTICA E,et al. A pathology foundation model for cancer diagnosis and prognosis prediction[J]. Nature,2024,634(8035):970-978
WANG G Y, YANG G X, DU Z X,et al. ClinicalGPT: Large language models finetuned with diverse medical data and comprehensive evaluation[EB/OL].[2023-06-16].https://arxiv.org/abs/2306.09968v1
关闭