Interpretable machine learning models for predicting tocilizumab response in rheumatoid arthritis using clinical data

Interpretable machine learning models for predicting tocilizumab response in rheumatoid arthritis using clinical data | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Interpretable machine learning models for predicting tocilizumab response in rheumatoid arthritis using clinical data Mengsi Ma, Xinya Chen, Xue Wu, Chen Chen, Zhuoyu Li, Qianxi Xu, and 7 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6420083/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Introduction: Approximately 30% of patients with rheumatoid arthritis (RA) do not respond to tocilizumab (TCZ). This study aimed to predict TCZ response using machine learning models trained on clinical data. Methods Baseline and follow-up data from patients with RA treated with TCZ were collected. Seven different machine learning models (logistic regression, K-nearest neighbor algorithm, random forest, support vector machine, decision tree, gradient boosting, and boosting algorithm) were trained to predict the treatment response of patients with RA across different imaging stages after three to six months of therapy. The area under the receiver operating curve (AUC) was the main performance evaluation feature to screen the best model, and the relative importance of each variable in the model was ranked using the Shapley Additive Explanation (SHAP) value. Results A total of 245 RA patients treated with TCZ were included. The logistic regression model demonstrated superior prediction performance across imaging stages, achieving AUC scores of 0.78 for patients without imaging changes, 0.73 for stage I, and 0.82 for stages II, III, and IV. Key features influencing predictions varied by imaging stage. The most important features were D-dimer concentration for patients without imaging changes, DAS28 grade for patients in stage I, and the physician's overall disease score (EGA) for patients in stages II, III, and IV, with SHAP values of 0.5180, 0.6661, and 1.3978, respectively. Discussion This study demonstrates the potential of machine learning to predict TCZ treatment outcomes and identify stage-specific features in patients with RA. Biological sciences/Immunology/Immunological disorders/Autoimmune diseases Health sciences/Medical research/Biomarkers Biological sciences/Computational biology and bioinformatics/Classification and taxonomy Biological sciences/Computational biology and bioinformatics/Data mining Biological sciences/Computational biology and bioinformatics/Data processing Biological sciences/Computational biology and bioinformatics/Machine learning Biological sciences/Computational biology and bioinformatics/Predictive medicine Health sciences/Diseases/Immunological disorders Tocilizumab Rheumatoid Arthritis Machine Learning Imaging Staging Clinical Data Figures Figure 1 Figure 2 1 Introduction Rheumatoid arthritis (RA) is a chronic systemic autoimmune disease mainly affecting the joints. The global incidence ranges from approximately 0.25–1%, leading to progressive disability, increased mortality, and high socioeconomic costs [ 1 , 2 ] . The first-line treatment for RA remains traditional disease-modifying antirheumatic drugs (DMARDs),however, most patients exhibit significant drug intolerance and only a moderate treatment response rate [ 3 ] . Advances in understanding the pathogenesis of RA have led to the development of biological agents targeting multiple pathogenic pathways. However, these drugs are expensive and are prone to drug resistance [ 4 ] . Tocilizumab, an IL-6 receptor antagonist, is a first-line biological agent used to treat RA. By inhibiting IL-6 signaling, tocilizumab reduces abnormal immune responses in RA and reduces joint destruction [ 5 , 6 ] . Although long-term clinical applications have demonstrated their efficacy and safety, approximately 30% of patients still respond poorly [ 7 ] . Early and effective treatment can improve the cost-effectiveness of RA management [ 8 ] . However, there is currently a lack of biomarkers that can predict the response of patients with RA to tocilizumab treatment. Some baseline clinical characteristics show a certain association with treatment response but are insufficient to independently predict efficacy. Machine learning, a subfield of artificial intelligence, offers promising solutions by identifying and extracting relevant features from large disease datasets and performing repeated classifications of new samples using trained models. Even with limited data, small-sample learning enables machine learning models to effectively perform learning and generalization [ 9 ] . Therefore, machine learning has been widely used in oncology for diagnosis, treatment, and prognosis research. However, its application in the treatment of rheumatic and immune diseases remains underdeveloped [ 10 ] . Although machine learning has made significant progress over the past few decades, issues surrounding the transparency and interpretability of models have emerged as key topics that need to be addressed. Clinicians often struggle to understand the internal working principles of the model, making it difficult for them to fully trust algorithm results. Explainable artificial intelligence (XAI) is a set of methods designed to enhance the transparency of model results. One widely used XAI approach is Shapley Additive Explanations (SHAP), which is based on cooperative game theory. SHAP highlights the importance of clinical indicators in predicting various diseases or patient outcomes by showing the contribution of each feature to the output [ 11 ] . It is compatible with any machine-learning model and is currently the most widely used XAI technology. Based on baseline clinical indicators of patients with RA, this study constructed a prediction model to evaluate the response of patients to tocilizumab treatment at different imaging stages using multiple machine-learning methods. The study identified key feature variables associated with efficacy prediction and used SHAP to interpret the model results. This approach aims to assist clinicians in making accurate treatment decisions and optimizing personalized treatment plans, thereby reducing unnecessary drug use and improving treatment outcomes. 2 Materials and methods 2.1 Study participants Patients with RA who visited the Department of Rheumatology and Immunology of Xinjiang Uygur Autonomous Region People's Hospital between January 2018 and December 2024 and received tocilizumab for the first time were included in this study. The inclusion criteria were: (1) age ≥ 18 years; (2) diagnosis of RA according to the 2010 ACR/EULAR classification criteria (12); (3) initiation of intravenous tocilizumab at a dose of 8 mg/kg once every four weeks for a duration of 3 to 6 months. The exclusion criteria were: (1) concurrent treatment with other biological agents in addition to tocilizumab, (2) presence of other autoimmune diseases, (3) active viral or bacterial infections, (4) malignant tumors, (5) severe cardiovascular or cerebrovascular diseases, (6) severe hepatic or renal insufficiency, and (7) severe allergy to tocilizumab. This study was approved by the Ethics Committee of Xinjiang Uygur Autonomous Region People’s Hospital (Approval No. XJPH-2023-001, Date: January 1, 2023). Written informed consent was obtained from all participants. 2.2 Data collection Demographic data such as patient age, sex, disease duration, and body mass index (BMI) were collected. Clinical data such as C-reactive protein (CRP) levels, erythrocyte sedimentation rate (ESR), tender joint count (TJC), and swollen joint count (SJC) were collected before and after treatment. Disease activity was assessed using the DAS28 score, and baseline laboratory tests, including blood counts, lymphocyte counts, and tumor markers, were also recorded. 2.2.1 Imaging staging All pre-treatment hand or foot radiographs were conducted through consensus reading by two independent specialists (a rheumatologist and a musculoskeletal radiologist) blinded to clinical data, using the Steinbrocker imaging staging system [ 13 ] . Stage I: visible osteoporosis without bone destruction; Stage II: visible osteoporosis with mild subchondral bone destruction, but no joint deformity; Stage III: visible osteoporosis, bone destruction, and joint deformity; and Stage IV: fibrous or bony ankylosis of the joints. Patients whose radiographs did not show any of these changes were classified as stage 0. 2.2.2 Disease activity assessment The DAS28 score was calculated using the following formula: $$\:\text{D}\text{A}\text{S}28\:=\:\:\left[0.56\:\times\:\:\text{s}\text{q}\text{r}\text{t}\:\left(\text{T}28\right)+\:0.28\:\times\:\:\text{s}\text{q}\text{r}\text{t}\:\left(\text{S}\text{W}28\right)+\:0.70\:\times\:\:\text{L}\text{n}\:\left(\text{E}\text{S}\text{R}\right)\right]\times\:\:1.08\:+\:0.16$$ where T28 represents the number of tender joints, SW28 represents the number of swollen joints, sqrt represents the square root, and Ln represents the natural logarithm. According to the T2T treatment goal for RA, a DAS28 score < 2.6 indicates that the RA has been in remission, while scores between 2.6 and 3.2 signify low disease activity [ 14 ] . Therefore, a DAS28 score of < 3.2 during the 3 to 6 months of tocilizumab treatment was considered effective and regarded as the "effective group,” and a DAS28 score of < 3.2 was classified as ineffective and regarded as the "ineffective group". 2.2.3 Clinical data preprocessing To prepare the clinical data for analysis, all missing data in the dataset were removed, and all features were converted into numerical formats compatible with the machine-learning algorithm [ 15 ] . Values that could not be converted were marked as missing and subsequently imputed using the median method to preserve the distribution characteristics of the data [ 16 ] . Features from clinical data had varying numerical ranges, with some values spanning between zero and one, while others extended into hundreds or even thousands. Features with larger values could dominate the model training and overshadow features with smaller values that are equally important. Therefore, we used StandardScaler to standardize features such that the mean was 0 and the standard deviation was 1. This ensured that all features were on the same scale, allowing the model to treat each feature equally, thereby improving the efficiency and accuracy of model training. 2.3 Selection of features related to efficacy prediction To reduce the risk of overfitting and improve the generalization ability of the model, the recursive feature elimination (RFE) algorithm was used to select key features related to efficacy prediction for training models [ 17 , 18 ] . First, a base model (such as a support vector machine or decision tree) was used to train the data, and then the contribution of each feature to the model performance was evaluated. Poorly performing features were iteratively eliminated, and the feature set was continuously optimized in multiple iterations until a predetermined number of features were reached. Seven machine learning algorithms were used to construct the efficacy prediction model. For patients with different imaging stages, seven machine-learning algorithms were used. These algorithms included logistic regression (LR) [ 19 ] , K nearest neighbor algorithm (KNN) [ 20 ] , random forest (RF) [ 21 ] , support vector machine (SVM) [ 22 ] , decision tree (DT) [ 23 ] , gradient boosting algorithm[GB] [ 24 ] , and boosting algorithm [ 25 ] , all of which were used to construct the short-term efficacy prediction model of tocilizumab. To avoid overfitting, the dataset was divided into a 7:3 ratio, in which 70% of the sample data were allocated to the training set. These data were used to build and train a machine-learning model, allowing it to continuously adjust its own parameters and optimize its prediction ability by learning the relationship between the features in the training set and the target variable. The remaining 30% of the sample data constituted the test dataset. The test set remained independent during the model training process and was only used to evaluate the generalization ability and performance of the model on unknown data after model training was completed. The performance of each model on the training set was evaluated using 10-fold cross-validation (StratifiedKFold) [ 26 ] . Variables such as accuracy, precision, sensitivity, specificity, and F1-Score were calculated to evaluate the model performance. The experimental artificial curve was drawn, and the area under the curve (AUC) was calculated to further evaluate the model performance [ 27 ] . 2.4 Model interpretability To visualize the model, Shapley interpretation was performed for each efficacy model. The SHAP values were positive or negative, indicating the impact on efficacy prediction. A higher absolute Shapley value denotes a greater influence on the prediction results [ 28 ] . 2.5 Statistical analysis Statistical analyses were performed using SPSS27.0 software. The Kolmogorov-Smirnov (K-S) test was used to assess data normality. Normally distributed quantitative data were expressed as mean ± standard deviation, and comparisons between the two groups were conducted using paired t-tests when variances were homogenous. Otherwise, the nonparametric ANOVA variance test was used. For non-normally distributed quantitative data, values were expressed as medians (25th quantile, 75th quantile), and the Mann-Whitney U test was used for comparison between the two groups. Count data were expressed as rate (%), and the chi-square test was used to compare between the two groups. ANOVA was used to compare multiple groups of normally distributed quantitative data, while the Kruskal-Wallis H test was employed to compare non-normally distributed quantitative and categorical data. A p-value < 0.05 was considered statistically significant. 3 Result 3.1 Patient demographics and characteristics There were no significant differences in baseline age, sex, or BMI between the two groups (Table 1). However, the ineffective treatment group had significantly higher baseline levels of ESR, TJC, SJC, DAS28-ESR, DAS28-CRP, the clinical disease activity index, and the simplified disease activity index compared to the effective group, indicating higher disease activity at baseline. In addition, a comparison of common tumor marker data revealed that the squamous cell carcinoma antigen (SCC) concentration was significantly higher in the effective treatment group compared to the ineffective treatment group (P < 0.05). Table 1 Baseline demographic and clinical characteristics of patients Effective group(n = 173) Invalid group(n = 72) p-value Gender,n(%) 0.734 male 19(11%) 9(12.5%) female 154(89.0%) 63(87.5%) Age,years 54.00(45.00,61.00) 56.00(46.25,61.00) 0.730 BMI,Kg/m2 23.67(21.10,26.92) 23.67(21.46,26.39) 0.709 Duration,months 72(13.50,177.00) 48.00(24.00,141.00) 0.227 Morning stiffness,minutes 60.00(20.00,60.00) 60.00(30.00,60.00) 0.383 RF positive,n(%) 122(70.5%) 52(72.2%) 0.789 Radiological changes,n(%) 0.222 Stage 0 42(24.3%) 13(18.1%) Stage Ⅰ 86(49.7%) 44(61.1%) Stage Ⅱ 30(17.3%) 6(8.3%) Stage Ⅲ 8(4.6%) 5(6.9%) Stage Ⅳ 7(4.0%) 4(5.6%) TJC,n(%) 2.00(1.00,10.00) 4.00(2.00,12.00) 0.010* SJC,n(%) 4.00(0.00,4.00) 6.00(0.00,7.75) 0.010* ESR,mm/h 21.50(9.00,42.75) 35.00(12.00,57.00) 0.009* CRP,mg/L 10.70(2.90,21.80) 11.05(4.63,33.62) 0.215 DAS28(ESR) 4.34(2.55,5.35) 4.88(3.11,6.44) 0.011* DAS28(CRP) 3.81(2.35,4.97) 4.42(3.08,5.65) 0.010* CDAI 13.55(4.03,23.8) 18.10(7.28,31.00) 0.031* SDAI 15.74(5.03,26.18) 20.37(9.06,33.60) 0.027* PGA 38.00(14.25,49.00) 43.00(19.50,54.50) 0.148 EGA 30.00(9.00,45.75) 37.00(16.00,47.00) 0.273 WBC,cells/109 5.59(4.23,6.64) 5.52(4.39,6.97) 0.698 RBC,cells/1012 4.10(3.80,4.39) 4.12(3.68,4.46) 0.540 HGB,g/L 115.00(103.00,124.75) 110.00(96.00,123.00) 0.192 PLT,cells/109 267.50(205.00,342.00) 255.50(210.00,343.00) 0.858 N,cell/µL 3.74(2.61,4.84) 3.75(2.71,5.03) 0.820 L,cell/µL 1.59(1.25,2.08) 1.52(1.15,1.83) 0.052 IL-6,pg/ml 12.30(2.85,38.95) 14.10(3.05,50.40) 0.287 D-dimer,mg/L 0.91(0.46,2.40) 1.01(0.65,2.90) 0.121 CD4cnt,cell/µL 699.20(474.23,870.90) 699.20(496.50,871.95) 0.861 NKT,cell/µL 74.50(37.25,98.75) 66.50(43.00,83.73) 0.777 RF,IU/ml 61.05(20.00,227.00) 83.20(20.00,288.50) 0.535 CCP,U/ml 530.85(64.23,2392.93) 811.50(184.15,2557.60) 0.365 ANA,n(%) 100.00(0.00,320.00) 100.00(0.00,100.00) 0.713 dsDNA,U/ml 10.00(10.00,10.00) 10.00(10.00,10.00) 0.797 AFP,ng/mL 2.08(2.00,2.68) 2.07(2.00,2.92) 0.994 CEA,ng/mL 1.73(1.73,2.26) 1.73(1.72,2.15) 0.825 CA125,U/mL 14.10(9.90,18.93) 17.30(10.25,19.55) 0.153 CA153,U/mL 11.70(7.00,14.70) 10.95(7.65,14.30) 0.949 CA199,U/mL 3.43(2.06,8.62) 3.42(2.06,8.26) 0.895 CYFRA21-1,µg/L 1.70(1.10,1.70) 1.70(1.10,1.70) 0.952 NSE,µg/L 10.80(10.80,10.80) 10.80(10.75,10.80) 0.396 ProGRP,pg/L 28.80(28.80,28.80) 28.80(28.80,28.80) 0.635 SCC,ng/L 0.76(0.60,0.90) 0.73(0.50,0.76) 0.008* Combination Therapy MTX,n(%) 164(94.8%) 67(93.1%) 0.593 LEF,n(%) 30(17.3%) 16(22.2%) 0.373 HCQ,n(%) 133(76.9%) 53(73.6%) 0.586 ASAP,n(%) 10(5.8%) 3(4.2%) 0.623 *p<0.05 is statistically significant. TJC number of tender joints, SJC number of swollen joints, CDAI clinical disease activity index, SDAI simplified disease activity index, PGA patient's overall disease score, EGA doctor's overall disease score, N neutrophil count, L lymphocyte count, SCC squamous cell carcinoma associated antigen, MTX methotrexate, LEF leflunomide, HCQ hydroxychloroquine, ASPA sulfasalazine. No statistically significant difference in treatment efficacy was observed among the five imaging stages. The number of patients who responded to treatment in each group was approximately 2–5 times higher than the number of patients who did not respond (Supplementary Table 1). Among these groups, baseline DAS28-CRP, disease duration, TJC, SJC, ESR, clinical disease activity index, simplified disease activity index, patient's overall disease rating, physician's overall disease rating, PLT, D-dimer, AFP, and CA199 showed statistical differences across the imaging stages (Supplementary Table 2). 3.2 Results of variable selection For patients with different imaging stages, the RFE method was used to construct the efficacy prediction model and perform feature selection. The ten most important variables were selected and the results are presented in Table 2. These variables were subsequently used to train the model. Table 2 Result of the variable selection process Radiological changes Feature 0 stage Nation、Duration、Morning stiffness、Cardiovascular history、Lymphocyte、IL-6、D-dimer、CD4 h/ T cell count、dsDNA、SCC Ⅰ stage Hand arthritis、Joint involvement、Duration of synovitis、DAS28-CRP、DAS28 grade、IL-6、NKT、NSE、SCC、Botanicals Ⅱ、Ⅲ and Ⅳ stage Gender、Duration of synovitis、Acute phase reactants、TJC、ESR、DAS28-ESR、DAS28-CRP、DAS28 grade、EGA、NSAIDs 3.3 Model evaluation As shown in Table 3, Logistic Regression achieved the highest AUC in all three training sets, indicating superior performance to the other models in distinguishing between positive and negative samples, with an AUC range of 0.73 to 0.82. Figure 1compares various machine learning models in terms of performance indicators, presenting the comprehensive performance of the model via a radar chart. Each axis of the radar chart represents a specific performance indicator. As shown in the figure, the SVM model exhibited higher precision, specificity, and accuracy, though its AUC was slightly lower than those of Logistic Regression. Table 3 AUC of the models for the response prediction LR RF GB KNN SVM DT AB 0 stage 0.78 0.66 0.62 0.66 0.64 0.62 0.70 Ⅰ stage 0.73 0.56 0.59 0.59 0.72 0.50 0.55 Ⅱ、Ⅲ and Ⅳstage 0.82 0.58 0.64 0.65 0.75 0.49 0.66 3.4 Model interpretability Using interpretable machine learning, the SHAP value for each variable was calculated. Figure 2shows the top ten most influential factors and their effect sizes. As shown in Table 2, SCC without radiographic changes, DAS28 grade of stage I, and EGA of stages II, III, and IV were the most important clinical features, with average SHAP values of 0.6577, 0.8644, and 1.3978, respectively. As previously mentioned, the higher SHAP value corresponds to a greater average contribution to the model prediction. 4 Discussion In this study, we used seven machine-learning algorithms to predict the efficacy response of patients after 3–6 months to tocilizumab treatment. The model with the best classification efficiency was selected for patients at different imaging stages. In addition, SHAP (an XAI framework) was used to quantify the impact of each variable on the model’s prediction. Compared with the other six models, Logistic Regression achieved the highest AUC value of 0.82 across the three patient validation datasets, despite lower precision and accuracy. Logistic Regression demonstrated high classification efficiency and effectively managed diverse clinical data types [ 29 ] . For small-sample clinical datasets, Logistic Regression can effectively minimize bias by reasonably selecting variables and ensuring robust model fitting with limited data, thus accurately reflecting real-world clinical situations [ 30 ] . Johansson et al. previously applied a deep learning algorithm to process and analyze clinical data of patients from randomized controlled trials, constructing a prediction model for disease remission after 24 weeks of tocilizumab treatment. Their validation set yielded an AUC of 0.70 (95% CI 0.64, 0.77). However, they did not explore the factors influencing efficacy prediction or analyze their role in the model [ 31 ] , limiting insights into the functional processes of the model. This study addresses this gap by analyzing baseline data from patients at different imaging stages and constructing a prediction model with a higher AUC. In addition, to enhance interpretability, we used SHAP values to assess the importance of clinical indicators on the model prediction results. The SHAP value of each variable revealed its influence on the model results. Specifically, the D-dimer level (for patients without imaging changes), DAS28 grade for stage I, and the EGA for stages II, III, and IV had the greatest impact. For patients with imaging changes, D-dimer level was the most critical clinical indicator in the efficacy prediction model for stages 0 and I, a novel finding in RA research. Compared to traumatic arthritis and osteoarthritis, RA synovial pathology exhibits enhanced coagulation/anticoagulation and fibrin formation/degradation processes. The enzymes, cytokines, and fibrin degradation products that maintain inflammation can be used as potential targets for therapeutic interventions, as fibrinogen and D-dimer are expressed in the synovial lining and interstitium [ 32 , 33 ] . Baseline DAS28 grade was the most important predictor of tocilizumab efficacy in patients with stage I RA, aligning with previous studies showing that higher baseline DAS28 correlated with better treatment responses [ 34 ] . Similarly, Forsblad-d’Elia et al. demonstrated that elevated DAS-28 scores predict favorable EULAR response [ 35 ] . For patients with stage II or above, the physician's overall disease score is an important clinical feature that proved more significant than objective indicators like baseline CRP and ESR in previous studies. EGA assesses baseline disease activity from a clinical perspective, highlighting its relevance in predicting outcomes. Other important clinical indicators such as sex, BMI, ESR, neutrophil count, IL-6 level, D-dimer level, and a history of cardiovascular disease have also been shown to be associated with treatment response in previous studies [ 36 , 37 ] . In this study, the short-term efficacy of tocilizumab in patients with different imaging stages ranged from 60–70%, consistent with previous clinical trials [ 38 ] . Baseline indicators of disease activity in the ineffective group were notably higher than those in the effective group. This finding may be related to tocilizumab’s ability to rapidly normalize acute-phase reactant levels, leading to significant disease remission after short-term treatment. While radiography remains the traditional gold standard imaging modality for RA, its limitations are evident. It fails to detect early disease manifestations such as inflammatory changes in soft tissues (synovitis, tension sheath inflammation, and enthesitis) or the initial stages of bone erosion. Therefore, approximately 20% of the patients included in this study had no X-ray damage at baseline, a finding attributed to this limitation [ 39 ] . The absence of complementary imaging modalities such as joint ultrasound and nuclear magnetic resonance to fully evaluate the involvement of the patient's joints and periarticular soft tissues is a limitation of this study. In addition, the relatively small sample sizes in stages III and IV may have a certain impact on the performance of the model. This study focused on analyzing the performance of each efficacy prediction model for different imaging stages and the role of selected characteristic variables in the onset of RA. However, future research should investigate the reasons for the differences in predictive features across these models. In addition, this study did not use an external test set from other RA cohort data to confirm the performance of machine learning. 5 Conclusion In conclusion, baseline clinical data were collected from RA patients before tocilizumab treatment, and imaging staging was performed based on radiographs of the hands or feet. An efficacy prediction model was then constructed using machine learning for patients with varying degrees of joint involvement after short-term treatment with tocilizumab. The model achieved an AUC of up to 0.82 and identified predictive indicators for each imaging stage, aiding clinicians in providing more personalized treatment for RA patients with different radiographic findings. However, the limited sample size poses a risk of model overfitting and unverified results. Future research should validate these findings in a larger cohort. Declarations Funding This work was supported by National Key R&D Program of China (2022YFC3602000） Competing interests The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. Ethics approval and consent to participate This study was approved by the Ethics Committee of Xinjiang Uygur Autonomous Region People’s Hospital (Approval No. XJPH-2023-001, Date: January 1, 2023). Written informed consent was obtained from all participants prior to data collection. The study was conducted in accordance with the ethical standards of the institutional and/or national research committee and with the 1964 Helsinki Declaration and its later amendments. Consent for publication All participants provided informed consent for the anonymized use of their clinical data in research publications. No personally identifiable information is included in this manuscript. Availability of data and materials The datasets generated during this study are not publicly available due to patient privacy protections but may be accessed upon reasonable request to the corresponding author, subject to institutional and ethical approvals. Code availability Not applicable. Authors’ contributions Author MM and XC: Conceptualization, Data Curation, Investigation, Writing – Original Draft. Author XW: Validation, Formal Analysis, Investigation, Writing – Review & Editing. Author CC: Conceptualization, Methodology, Writing – Review & Editing. Author ZL, QX, JC, and ZZ: Validation, Investigation, Writing – Review & Editing. Author ZG, XF, XL, and WQ: Investigation, Writing – Review & Editing. Author LW: Funding Acquisition, Supervision, Writing – Review & Editing, Data Curation. All authors contributed to the article and approved the submitted version. The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. Conditions of publication Not applicable. References Di Matteo A, Bathon JM, Emery P. Rheumatoid arthritis. Lancet . 2023;402(10416):2019-2033. https://doi.org/10.1016/S0140-6736(23)01234-5. Finckh A, Gilbert B, Hodkinson B, Bae SC, Thomas R, Deane KD, et al. Global epidemiology of rheumatoid arthritis. Nat Rev Rheumatol. 2022;18(10):591-602. https://doi.org/10.1038/s41584-022-00845-7. Suarez-Almazor, ME, Belseck, E, Shea, B, Wells, G, Tugwell, P. Methotrexate for rheumatoid arthritis. Cochrane Database Syst Rev. 2000-01-01; (2): CD000957. Aceves-Ávila, FJ, Hernández Vásquez, JR, Sicsick, S, Olguín Ortega, ML, Ramos Sánchez, MA, Urenda Quezada, A, et al. Not the same, but is it the same? Cycling of biologic agents in rheumatoid arthritis. Experience in the Instituto Mexicano del Seguro Social. Reumatol Clin (Engl Ed). 2022-01-01; 18 (6): 361-367. Wang, J, Platt, A, Upmanyu, R, Germer, S, Lei, G, Rabe, C, et al. IL-6 pathway-driven investigation of response to IL-6 receptor inhibition in rheumatoid arthritis. BMJ Open. 2013-08-19; 3 (8): e003199. Cronstein, BN. Interleukin-6--a key mediator of systemic and local symptoms in rheumatoid arthritis. Bull NYU Hosp Jt Dis. 2007-01-01; 65 Suppl 1 S11-5. Mima, T, Nishimoto, N. Clinical value of blocking IL-6 receptor. CURR OPIN RHEUMATOL. 2009-05-01; 21 (3): 224-30. Sullivan, SD, Alfonso-Cristancho, R, Carlson, J, Mallya, U, Ringold, S. Economic consequences of sequencing biologics in rheumatoid arthritis: a systematic review. J MED ECON. 2013-01-01; 16 (3): 391-6. Rashidi, HH, Tran, N, Albahra, S, Dang, LT. Machine learning in health care and laboratory medicine: General overview of supervised learning and Auto-ML. INT J LAB HEMATOL. 2021-07-01; 43 Suppl 1 15-22. Danieli, MG, Brunetto, S, Gammeri, L, Palmeri, D, Claudi, I, Shoenfeld, Y, et al. Machine learning application in autoimmune diseases: State of art and future prospectives. AUTOIMMUN REV. 2024-02-01; 23 (2): 103496. Ali, S, Akhlaq, F, Imran, AS, Kastrati, Z, Daudpota, SM, Moosa, M. The enlightening role of explainable artificial intelligence in medical & healthcare domains: A systematic literature review. COMPUT BIOL MED. 2023-11-01; 166 107555. Aletaha, D, Neogi, T, Silman, AJ, Funovits, J, Felson, DT, Bingham, CO, et al. 2010 rheumatoid arthritis classification criteria: an American College of Rheumatology/European League Against Rheumatism collaborative initiative. ANN RHEUM DIS. 2010-09-01; 69 (9): 1580-8. Kaye JJ, Fuchs HA, Moseley JW, Nance EP, Callahan LF, Pincus T. Problems with the Steinbrocker staging system for radiographic assessment of the rheumatoid hand and wrist. *Invest Radiol*. 1990;25(5):536-544. https://doi.org/10.1097/00004424-199005000-00007. Prevoo, ML, van 't Hof, MA, Kuper, HH, van Leeuwen, MA, van de Putte, LB, van Riel, PL. Modified disease activity scores that include twenty-eight-joint counts. Development and validation in a prospective longitudinal study of patients with rheumatoid arthritis. ARTHRITIS RHEUM-US. 1995-01-01; 38. Text Classification of Cornell Movie Data using Data Mining with Feature Selection Int J Eng Adv Technol. 2019-12-30; 9 (2): 2950-2955. Zhang, Z. Missing data imputation: focusing on single imputation. ANN TRANSL MED. 2016-01-01; 4 (1): 9. Zebari, R, Abdulazeez, A, Zeebaree, D, Zebari, D, Saeed, J. A Comprehensive Review of Dimensionality Reduction Techniques for Feature Selection and Feature Extraction JASTT. 2020-05-15; 1 (2): 56-70. Remeseiro, B, Bolon-Canedo, V. A review of feature selection methods in medical applications. COMPUT BIOL MED. 2019-09-01; 112 103375. Kianifard, F. Logistic Regression: A Self-Learning Text TECHNOMETRICS. 1995-02-01; 37 (1): 116-117. Altman, N. An Introduction to Kernel and Nearest-Neigh32bor Nonparametric Regression AM STAT. 1992-08-01; 46 (3): 175-185. Breiman, L. Random Forests. Machine Learning 45, 5–32 (2001). Cortes, C., Vapnik, V. Support-vector networks. Mach Learn 20, 273–297 (1995). Quinlan, J.R. Induction of decision trees. Mach Learn 1, 81–106 (1986). Friedman, Jerome H.. “Greedy function approximation: A gradient boosting machine.” Annals of Statistics 29 (2001): 1189-1232. Freund Y, Schapire RE. A decision-theoretic generalization of on-line learning and an application to boosting. J Comput Syst Sci. 1997;55(1):119-139. https://doi.org/10.1006/jcss.1997.1504. Refaeilzadeh, P., Tang, L., Liu, H. (2009). Cross-Validation. In: LIU, L., ÖZSU, M.T. (eds) Encyclopedia of Database Systems. Springer, Boston, MA. Available from:https://doi.org/10.1007/978-0-387-39940-9_565 Sokolova, M, Lapalme, G. A systematic analysis of performance measures for classification tasks INFORM PROCESS MANAG. 2009-07-01; 45 (4): 427-437. Lundberg SM, Lee SI. A unified approach to interpreting model predictions. In: Proceedings of the 31st International Conference on Neural Information Processing Systems (NeurIPS 2017). 2017:4768-4777. https://proceedings.neurips.cc/paper/2017/hash/8a20a8621978632d76c43dfd28b67767-Abstract.html. Shah, K, Patel, H, Sanghvi, D, Shah, M. A Comparative Analysis of Logistic Regression, Random Forest and KNN Models for the Text Classification Augment Hum Res. 2020-12-01; 5 (1). Bagley, SC, White, H, Golomb, BA. Logistic regression in the medical literature: standards for use and reporting, with particular attention to one medical domain. J CLIN EPIDEMIOL. 2001-10-01; 54 (10): 979-85. Johansson, FD, Collins, JE, Yau, V, Guan, H, Kim, SC, Losina, E, et al. Predicting Response to Tocilizumab Monotherapy in Rheumatoid Arthritis: A Real-world Data Analysis Using Machine Learning. J RHEUMATOL. 2021-09-01; 48 (9): 1364-1370. Ichikawa, Y, Hoshina, Y, Horiki, T, Yamada, C, Uchiyama, M, Takaya, M. Molecular markers of coagulation and fibrinolysis as indicators for the disease activity of rheumatoid arthritis Japanese Journal of Rheumatology. 1997-09-01; 7 (3): 173-181. Weinberg, JB, Pippen, AM, Greenberg, CS. Extravascular fibrin formation and dissolution in synovial tissue of patients with osteoarthritis and rheumatoid arthritis. ARTHRITIS RHEUM-US. 1991-08-01; 34 (8): 996-1005. Day, KE, Beck, LN, Heath, CH, Huang, CC, Zinn, KR, Rosenthal, EL. Identification of the optimal therapeutic antibody for fluorescent imaging of cutaneous squamous cell carcinoma. CANCER BIOL THER. 2013-03-01; 14 (3): 271-7. Forsblad-d'Elia, H, Bengtsson, K, Kristensen, LE, Jacobsson, LT. Drug adherence, response and predictors thereof for tocilizumab in patients with rheumatoid arthritis: results from the Swedish biologics register. RHEUMATOLOGY. 2015-07-01; 54 (7): 1186-93. Simons, A, Dahl, R, Espinosa-Cotton, M, Rodman, S, Rose-John, S. Abstract 5611: Inhibition of IL-6 trans-signaling in HNSCC CANCER RES. 2018-07-01; 78 (13_Supple): 5611-5611. Bykerk, VP, Ostör, AJ, Alvaro-Gracia, J, Pavelka, K, Ivorra, JA, Graninger, W, et al. Tocilizumab in patients with active rheumatoid arthritis and inadequate responses to DMARDs and/or TNF inhibitors: a large, open-label study close to clinical practice. ANN RHEUM DIS. 2012-12-01; 71 (12): 1950-4. Pers, YM, Fortunet, C, Constant, E, Lambert, J, Godfrin-Valnet, M, De Jong, A, et al. Predictors of response and remission in a large cohort of rheumatoid arthritis patients treated with tocilizumab in clinical practice. RHEUMATOLOGY. 2014-01-01; 53 (1): 76-84. Nakajima, T, Watanabe, R, Hashimoto, M, Murata, K, Murakami, K, Tanaka, M, et al. Neutrophil count reduction 1 month after initiating tocilizumab can predict clinical remission within 1 year in rheumatoid arthritis patients. RHEUMATOL INT. 2022-11-01; 42 (11): 1983-1991. Additional Declarations No competing interests reported. Supplementary Files file.docx Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-6420083","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":452878238,"identity":"2e6191ba-5262-4fe4-96f1-a36f9b3dc5c4","order_by":0,"name":"Mengsi Ma","email":"","orcid":"","institution":"People’s Hospital of Xinjiang Uygur Autonomous Region，Urumqi","correspondingAuthor":false,"prefix":"","firstName":"Mengsi","middleName":"","lastName":"Ma","suffix":""},{"id":452878239,"identity":"fb9b5f6a-bfe6-4a2a-bf26-13576462deff","order_by":1,"name":"Xinya Chen","email":"","orcid":"","institution":"College of Software, XJU, Urumqi.Xinjiang, China.","correspondingAuthor":false,"prefix":"","firstName":"Xinya","middleName":"","lastName":"Chen","suffix":""},{"id":452878240,"identity":"dc007dfd-c9f5-4a45-a792-928d52a9a043","order_by":2,"name":"Xue Wu","email":"","orcid":"","institution":"People’s Hospital of Xinjiang Uygur Autonomous Region，Urumqi","correspondingAuthor":false,"prefix":"","firstName":"Xue","middleName":"","lastName":"Wu","suffix":""},{"id":452878241,"identity":"8543189e-8d15-4bac-8efa-568916edf9c7","order_by":3,"name":"Chen Chen","email":"","orcid":"","institution":"College of Software, XJU, Urumqi.Xinjiang, China.","correspondingAuthor":false,"prefix":"","firstName":"Chen","middleName":"","lastName":"Chen","suffix":""},{"id":452878242,"identity":"74de0647-9fa6-49c7-bc8a-9f44ad685980","order_by":4,"name":"Zhuoyu Li","email":"","orcid":"","institution":"People’s Hospital of Xinjiang Uygur Autonomous Region，Urumqi","correspondingAuthor":false,"prefix":"","firstName":"Zhuoyu","middleName":"","lastName":"Li","suffix":""},{"id":452878243,"identity":"c9be7727-d13f-40d0-a4b3-e17a5b603ce8","order_by":5,"name":"Qianxi Xu","email":"","orcid":"","institution":"People’s Hospital of Xinjiang Uygur Autonomous Region，Urumqi","correspondingAuthor":false,"prefix":"","firstName":"Qianxi","middleName":"","lastName":"Xu","suffix":""},{"id":452878244,"identity":"20c42ac7-2624-433d-99ba-554fbc0f82d6","order_by":6,"name":"Ziyang Zhang","email":"","orcid":"","institution":"College of Software, XJU, Urumqi.Xinjiang, China.","correspondingAuthor":false,"prefix":"","firstName":"Ziyang","middleName":"","lastName":"Zhang","suffix":""},{"id":452878245,"identity":"63b40e60-200b-40b6-9fe4-ceb68efd0541","order_by":7,"name":"Jinchao Chen","email":"","orcid":"","institution":"College of Software, XJU, Urumqi.Xinjiang, China.","correspondingAuthor":false,"prefix":"","firstName":"Jinchao","middleName":"","lastName":"Chen","suffix":""},{"id":452878246,"identity":"2436f712-c8d3-425f-a8e2-abd4752b0fc3","order_by":8,"name":"Zhenyang Guan","email":"","orcid":"","institution":"People’s Hospital of Xinjiang Uygur Autonomous Region，Urumqi","correspondingAuthor":false,"prefix":"","firstName":"Zhenyang","middleName":"","lastName":"Guan","suffix":""},{"id":452878247,"identity":"3f4460e0-f896-4319-9428-988d83f0499c","order_by":9,"name":"Xinming Feng","email":"","orcid":"","institution":"People’s Hospital of Xinjiang Uygur Autonomous Region，Urumqi","correspondingAuthor":false,"prefix":"","firstName":"Xinming","middleName":"","lastName":"Feng","suffix":""},{"id":452878248,"identity":"e6a123dc-925b-4f11-8aa7-990c3b617d43","order_by":10,"name":"Xiangcao Liang","email":"","orcid":"","institution":"College of Software, XJU, Urumqi.Xinjiang, China.","correspondingAuthor":false,"prefix":"","firstName":"Xiangcao","middleName":"","lastName":"Liang","suffix":""},{"id":452878249,"identity":"895e57dc-bcb6-4697-8f98-e621d831b9c1","order_by":11,"name":"Wentao Qin","email":"","orcid":"","institution":"College of Software, XJU, Urumqi.Xinjiang, China.","correspondingAuthor":false,"prefix":"","firstName":"Wentao","middleName":"","lastName":"Qin","suffix":""},{"id":452878250,"identity":"f49b03d7-f5bf-411b-a868-962c72d5a50a","order_by":12,"name":"Lijun Wu","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAs0lEQVRIiWNgGAWjYFAC5gbDDz8k5NjYmw8Qq4WxwViyx8aYj+dYAvFaGHjY0hLnSeQoEKfB4EZiQ4EEz+H0NoYcBoYfFduI02JQYHE4t43h7AHGnjO3CWsxuw3UArQlt42xL4GZsY1YLTxsh9PZmHkMSNKSlsDGRqwW+/sPwYFs2MbDlnCQKL9I9hw+BopKefn5jw8++FFBhBYgYDOAsQ4QpR4ImB8Qq3IUjIJRMApGKAAALA08UVzxEBkAAAAASUVORK5CYII=","orcid":"","institution":"People’s Hospital of Xinjiang Uygur Autonomous Region，Urumqi","correspondingAuthor":true,"prefix":"","firstName":"Lijun","middleName":"","lastName":"Wu","suffix":""}],"badges":[],"createdAt":"2025-04-10 12:23:08","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6420083/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6420083/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":82355335,"identity":"eddaac7e-617d-4165-8dab-1dcbdae246ee","added_by":"auto","created_at":"2025-05-09 11:13:56","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":72261,"visible":true,"origin":"","legend":"\u003cp\u003ePerformance indicators of each model in the validation dataset across different imaging stages. (A) Comparison of AUC values for each model at different imaging stages; (B) Stage 0; (C) Stage I; (D) Stage II, III, and IV.\u003c/p\u003e","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-6420083/v1/b44e4131defe4e967eb4e0db.png"},{"id":82351601,"identity":"d7fa2a46-c8a3-4a91-bc93-9734a295c745","added_by":"auto","created_at":"2025-05-09 10:57:57","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":77589,"visible":true,"origin":"","legend":"\u003cp\u003eSHAP ranking of variables in the Logistic Regression model. (A and B) Stage 0 patients; (C and D) Stage I; (E and F) Stage II, III, and IV.\u003c/p\u003e","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-6420083/v1/e4681e3e65f3f4bc8ba97c88.png"},{"id":83438545,"identity":"6c391671-af66-4caf-8cb3-8e73ce221f33","added_by":"auto","created_at":"2025-05-26 09:02:45","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1121454,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6420083/v1/7794306e-eb15-4d0a-a6fc-dae9b75043be.pdf"},{"id":82353180,"identity":"36fba05f-bcdd-414d-a7e7-dc673b534f38","added_by":"auto","created_at":"2025-05-09 11:05:56","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"supplement","size":22544,"visible":true,"origin":"","legend":"","description":"","filename":"file.docx","url":"https://assets-eu.researchsquare.com/files/rs-6420083/v1/860e35ec6e5d211763e20c90.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Interpretable machine learning models for predicting tocilizumab response in rheumatoid arthritis using clinical data","fulltext":[{"header":"1 Introduction","content":"\u003cp\u003eRheumatoid arthritis (RA) is a chronic systemic autoimmune disease mainly affecting the joints. The global incidence ranges from approximately 0.25\u0026ndash;1%, leading to progressive disability, increased mortality, and high socioeconomic costs\u003csup\u003e[\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e, \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]\u003c/sup\u003e. The first-line treatment for RA remains traditional disease-modifying antirheumatic drugs (DMARDs),however, most patients exhibit significant drug intolerance and only a moderate treatment response rate\u003csup\u003e[\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]\u003c/sup\u003e. Advances in understanding the pathogenesis of RA have led to the development of biological agents targeting multiple pathogenic pathways. However, these drugs are expensive and are prone to drug resistance\u003csup\u003e[\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]\u003c/sup\u003e. Tocilizumab, an IL-6 receptor antagonist, is a first-line biological agent used to treat RA. By inhibiting IL-6 signaling, tocilizumab reduces abnormal immune responses in RA and reduces joint destruction\u003csup\u003e[\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e, \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]\u003c/sup\u003e. Although long-term clinical applications have demonstrated their efficacy and safety, approximately 30% of patients still respond poorly\u003csup\u003e[\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eEarly and effective treatment can improve the cost-effectiveness of RA management\u003csup\u003e[\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]\u003c/sup\u003e. However, there is currently a lack of biomarkers that can predict the response of patients with RA to tocilizumab treatment. Some baseline clinical characteristics show a certain association with treatment response but are insufficient to independently predict efficacy. Machine learning, a subfield of artificial intelligence, offers promising solutions by identifying and extracting relevant features from large disease datasets and performing repeated classifications of new samples using trained models. Even with limited data, small-sample learning enables machine learning models to effectively perform learning and generalization\u003csup\u003e[\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]\u003c/sup\u003e. Therefore, machine learning has been widely used in oncology for diagnosis, treatment, and prognosis research. However, its application in the treatment of rheumatic and immune diseases remains underdeveloped\u003csup\u003e[\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eAlthough machine learning has made significant progress over the past few decades, issues surrounding the transparency and interpretability of models have emerged as key topics that need to be addressed. Clinicians often struggle to understand the internal working principles of the model, making it difficult for them to fully trust algorithm results. Explainable artificial intelligence (XAI) is a set of methods designed to enhance the transparency of model results. One widely used XAI approach is Shapley Additive Explanations (SHAP), which is based on cooperative game theory. SHAP highlights the importance of clinical indicators in predicting various diseases or patient outcomes by showing the contribution of each feature to the output\u003csup\u003e[\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e]\u003c/sup\u003e. It is compatible with any machine-learning model and is currently the most widely used XAI technology.\u003c/p\u003e \u003cp\u003eBased on baseline clinical indicators of patients with RA, this study constructed a prediction model to evaluate the response of patients to tocilizumab treatment at different imaging stages using multiple machine-learning methods. The study identified key feature variables associated with efficacy prediction and used SHAP to interpret the model results. This approach aims to assist clinicians in making accurate treatment decisions and optimizing personalized treatment plans, thereby reducing unnecessary drug use and improving treatment outcomes.\u003c/p\u003e"},{"header":"2 Materials and methods","content":"\u003cp\u003e2.1\u0026nbsp;\u003cstrong\u003eStudy participants\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003ePatients with RA who visited the Department of Rheumatology and Immunology of Xinjiang Uygur Autonomous Region People\u0026apos;s Hospital between January 2018 and December 2024 and received tocilizumab for the first time were included in this study. The inclusion criteria were: (1) age\u0026thinsp;\u0026ge;\u0026thinsp;18 years; (2) diagnosis of RA according to the 2010 ACR/EULAR classification criteria (12); (3) initiation of intravenous tocilizumab at a dose of 8 mg/kg once every four weeks for a duration of 3 to 6 months. The exclusion criteria were: (1) concurrent treatment with other biological agents in addition to tocilizumab, (2) presence of other autoimmune diseases, (3) active viral or bacterial infections, (4) malignant tumors, (5) severe cardiovascular or cerebrovascular diseases, (6) severe hepatic or renal insufficiency, and (7) severe allergy to tocilizumab.\u003c/p\u003e\n\u003cp\u003eThis study was approved by the Ethics Committee of Xinjiang Uygur Autonomous Region People\u0026rsquo;s Hospital (Approval No. XJPH-2023-001, Date: January 1, 2023). Written informed consent was obtained from all participants.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.2 Data collection\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eDemographic data such as patient age, sex, disease duration, and body mass index (BMI) were collected. Clinical data such as C-reactive protein (CRP) levels, erythrocyte sedimentation rate (ESR), tender joint count (TJC), and swollen joint count (SJC) were collected before and after treatment. Disease activity was assessed using the DAS28 score, and baseline laboratory tests, including blood counts, lymphocyte counts, and tumor markers, were also recorded.\u003c/p\u003e\n\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e\n \u003ch2\u003e2.2.1 Imaging staging\u003c/h2\u003e\n \u003cp\u003eAll pre-treatment hand or foot radiographs were conducted through consensus reading by two independent specialists (a rheumatologist and a musculoskeletal radiologist) blinded to clinical data, using the Steinbrocker imaging staging system\u003csup\u003e[\u003cspan class=\"CitationRef\"\u003e13\u003c/span\u003e]\u003c/sup\u003e. Stage I: visible osteoporosis without bone destruction; Stage II: visible osteoporosis with mild subchondral bone destruction, but no joint deformity; Stage III: visible osteoporosis, bone destruction, and joint deformity; and Stage IV: fibrous or bony ankylosis of the joints. Patients whose radiographs did not show any of these changes were classified as stage 0.\u003c/p\u003e\n \u003cdiv id=\"Sec4\" class=\"Section3\"\u003e\n \u003ch2\u003e2.2.2 Disease activity assessment\u003c/h2\u003e\n \u003cp\u003eThe DAS28 score was calculated using the following formula:\u003c/p\u003e\n \u003cdiv id=\"Equa\" class=\"Equation\"\u003e\n \u003cdiv class=\"mathdisplay\" id=\"FileID_Equa\" name=\"EquationSource\"\u003e$$\\:\\text{D}\\text{A}\\text{S}28\\:=\\:\\:\\left[0.56\\:\\times\\:\\:\\text{s}\\text{q}\\text{r}\\text{t}\\:\\left(\\text{T}28\\right)+\\:0.28\\:\\times\\:\\:\\text{s}\\text{q}\\text{r}\\text{t}\\:\\left(\\text{S}\\text{W}28\\right)+\\:0.70\\:\\times\\:\\:\\text{L}\\text{n}\\:\\left(\\text{E}\\text{S}\\text{R}\\right)\\right]\\times\\:\\:1.08\\:+\\:0.16$$\u003c/div\u003e\n \u003c/div\u003e\n \u003cp\u003ewhere T28 represents the number of tender joints, SW28 represents the number of swollen joints, sqrt represents the square root, and Ln represents the natural logarithm. According to the T2T treatment goal for RA, a DAS28 score\u0026thinsp;\u0026lt;\u0026thinsp;2.6 indicates that the RA has been in remission, while scores between 2.6 and 3.2 signify low disease activity\u003csup\u003e[\u003cspan class=\"CitationRef\"\u003e14\u003c/span\u003e]\u003c/sup\u003e. Therefore, a DAS28 score of \u0026lt;\u0026thinsp;3.2 during the 3 to 6 months of tocilizumab treatment was considered effective and regarded as the \u0026quot;effective group,\u0026rdquo; and a DAS28 score of \u0026lt;\u0026thinsp;3.2 was classified as ineffective and regarded as the \u0026quot;ineffective group\u0026quot;.\u003c/p\u003e\n \u003c/div\u003e\n \u003cdiv id=\"Sec5\" class=\"Section3\"\u003e\n \u003ch2\u003e2.2.3 Clinical data preprocessing\u003c/h2\u003e\n \u003cp\u003eTo prepare the clinical data for analysis, all missing data in the dataset were removed, and all features were converted into numerical formats compatible with the machine-learning algorithm\u003csup\u003e[\u003cspan class=\"CitationRef\"\u003e15\u003c/span\u003e]\u003c/sup\u003e. Values that could not be converted were marked as missing and subsequently imputed using the median method to preserve the distribution characteristics of the data\u003csup\u003e[\u003cspan class=\"CitationRef\"\u003e16\u003c/span\u003e]\u003c/sup\u003e.\u003c/p\u003e\n \u003cp\u003eFeatures from clinical data had varying numerical ranges, with some values spanning between zero and one, while others extended into hundreds or even thousands. Features with larger values could dominate the model training and overshadow features with smaller values that are equally important. Therefore, we used StandardScaler to standardize features such that the mean was 0 and the standard deviation was 1. This ensured that all features were on the same scale, allowing the model to treat each feature equally, thereby improving the efficiency and accuracy of model training.\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003e2.3 Selection of features related to efficacy prediction\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003eTo reduce the risk of overfitting and improve the generalization ability of the model, the recursive feature elimination (RFE) algorithm was used to select key features related to efficacy prediction for training models\u003csup\u003e[\u003cspan class=\"CitationRef\"\u003e17\u003c/span\u003e, \u003cspan class=\"CitationRef\"\u003e18\u003c/span\u003e]\u003c/sup\u003e. First, a base model (such as a support vector machine or decision tree) was used to train the data, and then the contribution of each feature to the model performance was evaluated. Poorly performing features were iteratively eliminated, and the feature set was continuously optimized in multiple iterations until a predetermined number of features were reached. Seven machine learning algorithms were used to construct the efficacy prediction model.\u003c/p\u003e\n \u003cp\u003eFor patients with different imaging stages, seven machine-learning algorithms were used. These algorithms included logistic regression (LR)\u003csup\u003e[\u003cspan class=\"CitationRef\"\u003e19\u003c/span\u003e]\u003c/sup\u003e, K nearest neighbor algorithm (KNN)\u003csup\u003e[\u003cspan class=\"CitationRef\"\u003e20\u003c/span\u003e]\u003c/sup\u003e, random forest (RF)\u003csup\u003e[\u003cspan class=\"CitationRef\"\u003e21\u003c/span\u003e]\u003c/sup\u003e, support vector machine (SVM)\u003csup\u003e[\u003cspan class=\"CitationRef\"\u003e22\u003c/span\u003e]\u003c/sup\u003e, decision tree (DT)\u003csup\u003e[\u003cspan class=\"CitationRef\"\u003e23\u003c/span\u003e]\u003c/sup\u003e, gradient boosting algorithm[GB]\u003csup\u003e[\u003cspan class=\"CitationRef\"\u003e24\u003c/span\u003e]\u003c/sup\u003e, and boosting algorithm\u003csup\u003e[\u003cspan class=\"CitationRef\"\u003e25\u003c/span\u003e]\u003c/sup\u003e, all of which were used to construct the short-term efficacy prediction model of tocilizumab. To avoid overfitting, the dataset was divided into a 7:3 ratio, in which 70% of the sample data were allocated to the training set. These data were used to build and train a machine-learning model, allowing it to continuously adjust its own parameters and optimize its prediction ability by learning the relationship between the features in the training set and the target variable. The remaining 30% of the sample data constituted the test dataset. The test set remained independent during the model training process and was only used to evaluate the generalization ability and performance of the model on unknown data after model training was completed. The performance of each model on the training set was evaluated using 10-fold cross-validation (StratifiedKFold)\u003csup\u003e[\u003cspan class=\"CitationRef\"\u003e26\u003c/span\u003e]\u003c/sup\u003e. Variables such as accuracy, precision, sensitivity, specificity, and F1-Score were calculated to evaluate the model performance. The experimental artificial curve was drawn, and the area under the curve (AUC) was calculated to further evaluate the model performance\u003csup\u003e[\u003cspan class=\"CitationRef\"\u003e27\u003c/span\u003e]\u003c/sup\u003e.\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003e2.4 Model interpretability\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003eTo visualize the model, Shapley interpretation was performed for each efficacy model. The SHAP values were positive or negative, indicating the impact on efficacy prediction. A higher absolute Shapley value denotes a greater influence on the prediction results\u003csup\u003e[\u003cspan class=\"CitationRef\"\u003e28\u003c/span\u003e]\u003c/sup\u003e.\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003e2.5 Statistical analysis\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003eStatistical analyses were performed using SPSS27.0 software. The Kolmogorov-Smirnov (K-S) test was used to assess data normality. Normally distributed quantitative data were expressed as mean\u0026thinsp;\u0026plusmn;\u0026thinsp;standard deviation, and comparisons between the two groups were conducted using paired t-tests when variances were homogenous. Otherwise, the nonparametric ANOVA variance test was used. For non-normally distributed quantitative data, values were expressed as medians (25th quantile, 75th quantile), and the Mann-Whitney U test was used for comparison between the two groups. Count data were expressed as rate (%), and the chi-square test was used to compare between the two groups. ANOVA was used to compare multiple groups of normally distributed quantitative data, while the Kruskal-Wallis H test was employed to compare non-normally distributed quantitative and categorical data. A p-value\u0026thinsp;\u0026lt;\u0026thinsp;0.05 was considered statistically significant.\u003c/p\u003e\n \u003c/div\u003e\n\u003c/div\u003e"},{"header":"3 Result","content":"\u003cp\u003e\u003cstrong\u003e3.1 Patient demographics and characteristics\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThere were no significant differences in baseline age, sex, or BMI between the two groups (Table\u0026nbsp;1). However, the ineffective treatment group had significantly higher baseline levels of ESR, TJC, SJC, DAS28-ESR, DAS28-CRP, the clinical disease activity index, and the simplified disease activity index compared to the effective group, indicating higher disease activity at baseline. In addition, a comparison of common tumor marker data revealed that the squamous cell carcinoma antigen (SCC) concentration was significantly higher in the effective treatment group compared to the ineffective treatment group (P \u0026lt; 0.05).\u003c/p\u003e\n\u003cdiv\u003e\n \u003ctable id=\"Tab1\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv\u003eTable 1\u003c/div\u003e\n \u003cdiv\u003e\n \u003cp\u003eBaseline demographic and clinical characteristics of patients\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003ccolgroup cols=\"4\"\u003e\u003c/colgroup\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\u0026nbsp;\u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eEffective group(n = 173)\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eInvalid group(n = 72)\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003ep-value\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eGender,n(%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.734\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003emale\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e19(11%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e9(12.5%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003efemale\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e154(89.0%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e63(87.5%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eAge,years\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e54.00(45.00,61.00)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e56.00(46.25,61.00)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.730\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eBMI,Kg/m2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e23.67(21.10,26.92)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e23.67(21.46,26.39)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.709\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eDuration,months\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e72(13.50,177.00)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e48.00(24.00,141.00)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.227\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eMorning stiffness,minutes\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e60.00(20.00,60.00)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e60.00(30.00,60.00)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.383\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eRF positive,n(%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e122(70.5%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e52(72.2%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.789\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eRadiological changes,n(%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.222\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eStage 0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e42(24.3%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e13(18.1%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eStage Ⅰ\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e86(49.7%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e44(61.1%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eStage Ⅱ\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e30(17.3%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e6(8.3%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eStage Ⅲ\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e8(4.6%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e5(6.9%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eStage Ⅳ\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e7(4.0%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e4(5.6%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eTJC,n(%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e2.00(1.00,10.00)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e4.00(2.00,12.00)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.010*\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eSJC,n(%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e4.00(0.00,4.00)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e6.00(0.00,7.75)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.010*\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eESR,mm/h\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e21.50(9.00,42.75)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e35.00(12.00,57.00)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.009*\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eCRP,mg/L\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e10.70(2.90,21.80)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e11.05(4.63,33.62)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.215\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eDAS28(ESR)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e4.34(2.55,5.35)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e4.88(3.11,6.44)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.011*\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eDAS28(CRP)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e3.81(2.35,4.97)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e4.42(3.08,5.65)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.010*\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eCDAI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e13.55(4.03,23.8)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e18.10(7.28,31.00)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.031*\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eSDAI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e15.74(5.03,26.18)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e20.37(9.06,33.60)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.027*\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ePGA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e38.00(14.25,49.00)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e43.00(19.50,54.50)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.148\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eEGA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e30.00(9.00,45.75)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e37.00(16.00,47.00)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.273\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eWBC,cells/109\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e5.59(4.23,6.64)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e5.52(4.39,6.97)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.698\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eRBC,cells/1012\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e4.10(3.80,4.39)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e4.12(3.68,4.46)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.540\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eHGB,g/L\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e115.00(103.00,124.75)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e110.00(96.00,123.00)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.192\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ePLT,cells/109\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e267.50(205.00,342.00)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e255.50(210.00,343.00)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.858\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eN,cell/µL\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e3.74(2.61,4.84)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e3.75(2.71,5.03)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.820\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eL,cell/µL\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.59(1.25,2.08)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e1.52(1.15,1.83)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.052\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eIL-6,pg/ml\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e12.30(2.85,38.95)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e14.10(3.05,50.40)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.287\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eD-dimer,mg/L\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.91(0.46,2.40)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e1.01(0.65,2.90)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.121\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eCD4cnt,cell/µL\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e699.20(474.23,870.90)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e699.20(496.50,871.95)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.861\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eNKT,cell/µL\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e74.50(37.25,98.75)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e66.50(43.00,83.73)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.777\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eRF,IU/ml\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e61.05(20.00,227.00)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e83.20(20.00,288.50)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.535\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eCCP,U/ml\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e530.85(64.23,2392.93)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e811.50(184.15,2557.60)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.365\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eANA,n(%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e100.00(0.00,320.00)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e100.00(0.00,100.00)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.713\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003edsDNA,U/ml\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e10.00(10.00,10.00)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e10.00(10.00,10.00)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.797\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eAFP,ng/mL\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e2.08(2.00,2.68)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e2.07(2.00,2.92)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.994\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eCEA,ng/mL\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.73(1.73,2.26)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e1.73(1.72,2.15)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.825\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eCA125,U/mL\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e14.10(9.90,18.93)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e17.30(10.25,19.55)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.153\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eCA153,U/mL\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e11.70(7.00,14.70)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e10.95(7.65,14.30)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.949\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eCA199,U/mL\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e3.43(2.06,8.62)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e3.42(2.06,8.26)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.895\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eCYFRA21-1,µg/L\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1.70(1.10,1.70)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e1.70(1.10,1.70)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.952\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eNSE,µg/L\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e10.80(10.80,10.80)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e10.80(10.75,10.80)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.396\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eProGRP,pg/L\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e28.80(28.80,28.80)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e28.80(28.80,28.80)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.635\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eSCC,ng/L\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.76(0.60,0.90)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.73(0.50,0.76)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.008*\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eCombination Therapy\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eMTX,n(%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e164(94.8%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e67(93.1%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.593\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eLEF,n(%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e30(17.3%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e16(22.2%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.373\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eHCQ,n(%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e133(76.9%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e53(73.6%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.586\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eASAP,n(%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e10(5.8%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e3(4.2%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.623\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003ctfoot\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"4\"\u003e*p\u0026lt;0.05 is statistically significant. TJC number of tender joints, SJC number of swollen joints, CDAI clinical disease activity index, SDAI simplified disease activity index, PGA patient's overall disease score, EGA doctor's overall disease score, N neutrophil count, L lymphocyte count, SCC squamous cell carcinoma associated antigen, MTX methotrexate, LEF leflunomide, HCQ hydroxychloroquine, ASPA sulfasalazine.\u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tfoot\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003eNo statistically significant difference in treatment efficacy was observed among the five imaging stages. The number of patients who responded to treatment in each group was approximately 2–5 times higher than the number of patients who did not respond (Supplementary Table\u0026nbsp;1). Among these groups, baseline DAS28-CRP, disease duration, TJC, SJC, ESR, clinical disease activity index, simplified disease activity index, patient's overall disease rating, physician's overall disease rating, PLT, D-dimer, AFP, and CA199 showed statistical differences across the imaging stages (Supplementary Table\u0026nbsp;2).\u003c/p\u003e\n\u003cp\u003e3.2 Results of variable selection\u003c/p\u003e\n\u003cp\u003eFor patients with different imaging stages, the RFE method was used to construct the efficacy prediction model and perform feature selection. The ten most important variables were selected and the results are presented in Table\u0026nbsp;2. These variables were subsequently used to train the model.\u003c/p\u003e\n\u003cdiv\u003e\n \u003ctable id=\"Tab2\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv\u003eTable 2\u003c/div\u003e\n \u003cdiv\u003e\n \u003cp\u003eResult of the variable selection process\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003ccolgroup cols=\"2\"\u003e\u003c/colgroup\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eRadiological changes\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eFeature\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0 stage\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eNation、Duration、Morning stiffness、Cardiovascular history、Lymphocyte、IL-6、D-dimer、CD4 h/ T cell count、dsDNA、SCC\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eⅠ stage\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eHand arthritis、Joint involvement、Duration of synovitis、DAS28-CRP、DAS28 grade、IL-6、NKT、NSE、SCC、Botanicals\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eⅡ、Ⅲ and Ⅳ stage\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eGender、Duration of synovitis、Acute phase reactants、TJC、ESR、DAS28-ESR、DAS28-CRP、DAS28 grade、EGA、NSAIDs\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003e\u003cstrong\u003e3.3 Model evaluation\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAs shown in Table\u0026nbsp;3, Logistic Regression achieved the highest AUC in all three training sets, indicating superior performance to the other models in distinguishing between positive and negative samples, with an AUC range of 0.73 to 0.82. Figure\u0026nbsp;1compares various machine learning models in terms of performance indicators, presenting the comprehensive performance of the model via a radar chart. Each axis of the radar chart represents a specific performance indicator. As shown in the figure, the SVM model exhibited higher precision, specificity, and accuracy, though its AUC was slightly lower than those of Logistic Regression.\u003c/p\u003e\n\u003cdiv\u003e\n \u003ctable id=\"Tab3\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv\u003eTable 3\u003c/div\u003e\n \u003cdiv\u003e\n \u003cp\u003eAUC of the models for the response prediction\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003ccolgroup cols=\"8\"\u003e\u003c/colgroup\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\u0026nbsp;\u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eLR\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eRF\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eGB\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eKNN\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eSVM\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eDT\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eAB\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0 stage\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.78\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.66\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.62\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.66\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.64\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.62\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.70\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eⅠ stage\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.73\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.56\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.59\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.59\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.72\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.50\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.55\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eⅡ、Ⅲ and Ⅳstage\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.82\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.58\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.64\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.65\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.75\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.49\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.66\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003e\u003cstrong\u003e3.4 Model interpretability\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eUsing interpretable machine learning, the SHAP value for each variable was calculated. Figure\u0026nbsp;2shows the top ten most influential factors and their effect sizes. As shown in Table\u0026nbsp;2, SCC without radiographic changes, DAS28 grade of stage I, and EGA of stages II, III, and IV were the most important clinical features, with average SHAP values of 0.6577, 0.8644, and 1.3978, respectively. As previously mentioned, the higher SHAP value corresponds to a greater average contribution to the model prediction.\u003c/p\u003e"},{"header":"4 Discussion","content":"\u003cp\u003eIn this study, we used seven machine-learning algorithms to predict the efficacy response of patients after 3\u0026ndash;6 months to tocilizumab treatment. The model with the best classification efficiency was selected for patients at different imaging stages. In addition, SHAP (an XAI framework) was used to quantify the impact of each variable on the model\u0026rsquo;s prediction. Compared with the other six models, Logistic Regression achieved the highest AUC value of 0.82 across the three patient validation datasets, despite lower precision and accuracy. Logistic Regression demonstrated high classification efficiency and effectively managed diverse clinical data types\u003csup\u003e[\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e]\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eFor small-sample clinical datasets, Logistic Regression can effectively minimize bias by reasonably selecting variables and ensuring robust model fitting with limited data, thus accurately reflecting real-world clinical situations\u003csup\u003e[\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e]\u003c/sup\u003e. Johansson et al. previously applied a deep learning algorithm to process and analyze clinical data of patients from randomized controlled trials, constructing a prediction model for disease remission after 24 weeks of tocilizumab treatment. Their validation set yielded an AUC of 0.70 (95% CI 0.64, 0.77). However, they did not explore the factors influencing efficacy prediction or analyze their role in the model\u003csup\u003e[\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e]\u003c/sup\u003e, limiting insights into the functional processes of the model. This study addresses this gap by analyzing baseline data from patients at different imaging stages and constructing a prediction model with a higher AUC. In addition, to enhance interpretability, we used SHAP values to assess the importance of clinical indicators on the model prediction results.\u003c/p\u003e \u003cp\u003eThe SHAP value of each variable revealed its influence on the model results. Specifically, the D-dimer level (for patients without imaging changes), DAS28 grade for stage I, and the EGA for stages II, III, and IV had the greatest impact. For patients with imaging changes, D-dimer level was the most critical clinical indicator in the efficacy prediction model for stages 0 and I, a novel finding in RA research. Compared to traumatic arthritis and osteoarthritis, RA synovial pathology exhibits enhanced coagulation/anticoagulation and fibrin formation/degradation processes. The enzymes, cytokines, and fibrin degradation products that maintain inflammation can be used as potential targets for therapeutic interventions, as fibrinogen and D-dimer are expressed in the synovial lining and interstitium\u003csup\u003e[\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e, \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e]\u003c/sup\u003e. Baseline DAS28 grade was the most important predictor of tocilizumab efficacy in patients with stage I RA, aligning with previous studies showing that higher baseline DAS28 correlated with better treatment responses\u003csup\u003e[\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e]\u003c/sup\u003e. Similarly, Forsblad-d\u0026rsquo;Elia et al. demonstrated that elevated DAS-28 scores predict favorable EULAR response\u003csup\u003e[\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e]\u003c/sup\u003e. For patients with stage II or above, the physician's overall disease score is an important clinical feature that proved more significant than objective indicators like baseline CRP and ESR in previous studies. EGA assesses baseline disease activity from a clinical perspective, highlighting its relevance in predicting outcomes. Other important clinical indicators such as sex, BMI, ESR, neutrophil count, IL-6 level, D-dimer level, and a history of cardiovascular disease have also been shown to be associated with treatment response in previous studies\u003csup\u003e[\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e, \u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e]\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eIn this study, the short-term efficacy of tocilizumab in patients with different imaging stages ranged from 60\u0026ndash;70%, consistent with previous clinical trials\u003csup\u003e[\u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e]\u003c/sup\u003e. Baseline indicators of disease activity in the ineffective group were notably higher than those in the effective group. This finding may be related to tocilizumab\u0026rsquo;s ability to rapidly normalize acute-phase reactant levels, leading to significant disease remission after short-term treatment.\u003c/p\u003e \u003cp\u003eWhile radiography remains the traditional gold standard imaging modality for RA, its limitations are evident. It fails to detect early disease manifestations such as inflammatory changes in soft tissues (synovitis, tension sheath inflammation, and enthesitis) or the initial stages of bone erosion. Therefore, approximately 20% of the patients included in this study had no X-ray damage at baseline, a finding attributed to this limitation\u003csup\u003e[\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e]\u003c/sup\u003e. The absence of complementary imaging modalities such as joint ultrasound and nuclear magnetic resonance to fully evaluate the involvement of the patient's joints and periarticular soft tissues is a limitation of this study. In addition, the relatively small sample sizes in stages III and IV may have a certain impact on the performance of the model. This study focused on analyzing the performance of each efficacy prediction model for different imaging stages and the role of selected characteristic variables in the onset of RA. However, future research should investigate the reasons for the differences in predictive features across these models. In addition, this study did not use an external test set from other RA cohort data to confirm the performance of machine learning.\u003c/p\u003e"},{"header":"5 Conclusion","content":"\u003cp\u003eIn conclusion, baseline clinical data were collected from RA patients before tocilizumab treatment, and imaging staging was performed based on radiographs of the hands or feet. An efficacy prediction model was then constructed using machine learning for patients with varying degrees of joint involvement after short-term treatment with tocilizumab. The model achieved an AUC of up to 0.82 and identified predictive indicators for each imaging stage, aiding clinicians in providing more personalized treatment for RA patients with different radiographic findings. However, the limited sample size poses a risk of model overfitting and unverified results. Future research should validate these findings in a larger cohort.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis work was supported by\u0026nbsp;National Key R\u0026amp;D Program of China (2022YFC3602000）\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting interests\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare that the research was conducted in the absence of any\u003c/p\u003e\n\u003cp\u003ecommercial or financial relationships that could be construed as a potential\u003c/p\u003e\n\u003cp\u003econflict of interest.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEthics approval and consent to participate \u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study was approved by the Ethics Committee of Xinjiang Uygur Autonomous Region People\u0026rsquo;s Hospital (Approval No. XJPH-2023-001, Date: January 1, 2023). Written informed consent was obtained from all participants prior to data collection. The study was conducted in accordance with the ethical standards of the institutional and/or national research committee and with the 1964 Helsinki Declaration and its later amendments.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent for publication\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAll participants provided informed consent for the anonymized use of their clinical data in research publications. No personally identifiable information is included in this manuscript.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAvailability of data and materials\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe datasets generated during this study are not publicly available due to patient privacy protections but may be accessed upon reasonable request to the corresponding author, subject to institutional and ethical approvals. \u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCode availability\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthors\u0026rsquo; contributions\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAuthor\u0026nbsp;MM and XC: Conceptualization, Data Curation, Investigation, Writing \u0026ndash; Original Draft.\u003c/p\u003e\n\u003cp\u003eAuthor\u0026nbsp;XW: Validation, Formal Analysis, Investigation, Writing \u0026ndash; Review \u0026amp; Editing.\u003c/p\u003e\n\u003ch2\u003eAuthor\u0026nbsp;CC: Conceptualization, Methodology, Writing \u0026ndash; Review \u0026amp; Editing.\u003c/h2\u003e\n\u003cp\u003eAuthor ZL, QX, JC, and ZZ: Validation, Investigation, Writing \u0026ndash;\u0026nbsp;Review \u0026amp; Editing. \u0026nbsp;\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eAuthor ZG, XF,\u0026nbsp;XL, and WQ: Investigation, Writing \u0026ndash; Review \u0026amp; Editing.\u003c/p\u003e\n\u003ch2\u003eAuthor\u0026nbsp;LW: Funding Acquisition, Supervision, Writing \u0026ndash; Review \u0026amp; Editing, Data Curation.\u003c/h2\u003e\n\u003cp\u003eAll authors contributed to the article and approved the submitted version.\u003c/p\u003e\n\u003cp\u003eThe authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConditions of publication\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eDi Matteo A, Bathon JM, Emery P. Rheumatoid arthritis.\u003cem\u003e Lancet\u003c/em\u003e. 2023;402(10416):2019-2033. https://doi.org/10.1016/S0140-6736(23)01234-5. \u003c/li\u003e\n\u003cli\u003eFinckh A, Gilbert B, Hodkinson B, Bae SC, Thomas R, Deane KD, et al. Global epidemiology of rheumatoid arthritis. Nat Rev Rheumatol. 2022;18(10):591-602. https://doi.org/10.1038/s41584-022-00845-7.\u003c/li\u003e\n\u003cli\u003eSuarez-Almazor, ME, Belseck, E, Shea, B, Wells, G, Tugwell, P. Methotrexate for rheumatoid arthritis. Cochrane Database Syst Rev. 2000-01-01; (2): CD000957. \u003c/li\u003e\n\u003cli\u003eAceves-\u0026Aacute;vila, FJ, Hern\u0026aacute;ndez V\u0026aacute;squez, JR, Sicsick, S, Olgu\u0026iacute;n Ortega, ML, Ramos S\u0026aacute;nchez, MA, Urenda Quezada, A, et al. Not the same, but is it the same? Cycling of biologic agents in rheumatoid arthritis. Experience in the Instituto Mexicano del Seguro Social. Reumatol Clin (Engl Ed). 2022-01-01; 18 (6): 361-367. \u003c/li\u003e\n\u003cli\u003eWang, J, Platt, A, Upmanyu, R, Germer, S, Lei, G, Rabe, C, et al. IL-6 pathway-driven investigation of response to IL-6 receptor inhibition in rheumatoid arthritis. BMJ Open. 2013-08-19; 3 (8): e003199. \u003c/li\u003e\n\u003cli\u003eCronstein, BN. Interleukin-6--a key mediator of systemic and local symptoms in rheumatoid arthritis. Bull NYU Hosp Jt Dis. 2007-01-01; 65 Suppl 1 S11-5. \u003c/li\u003e\n\u003cli\u003eMima, T, Nishimoto, N. Clinical value of blocking IL-6 receptor. CURR OPIN RHEUMATOL. 2009-05-01; 21 (3): 224-30. \u003c/li\u003e\n\u003cli\u003eSullivan, SD, Alfonso-Cristancho, R, Carlson, J, Mallya, U, Ringold, S. Economic consequences of sequencing biologics in rheumatoid arthritis: a systematic review. J MED ECON. 2013-01-01; 16 (3): 391-6. \u003c/li\u003e\n\u003cli\u003eRashidi, HH, Tran, N, Albahra, S, Dang, LT. Machine learning in health care and laboratory medicine: General overview of supervised learning and Auto-ML. INT J LAB HEMATOL. 2021-07-01; 43 Suppl 1 15-22. \u003c/li\u003e\n\u003cli\u003eDanieli, MG, Brunetto, S, Gammeri, L, Palmeri, D, Claudi, I, Shoenfeld, Y, et al. Machine learning application in autoimmune diseases: State of art and future prospectives. AUTOIMMUN REV. 2024-02-01; 23 (2): 103496. \u003c/li\u003e\n\u003cli\u003eAli, S, Akhlaq, F, Imran, AS, Kastrati, Z, Daudpota, SM, Moosa, M. The enlightening role of explainable artificial intelligence in medical \u0026amp; healthcare domains: A systematic literature review. COMPUT BIOL MED. 2023-11-01; 166 107555. \u003c/li\u003e\n\u003cli\u003eAletaha, D, Neogi, T, Silman, AJ, Funovits, J, Felson, DT, Bingham, CO, et al. 2010 rheumatoid arthritis classification criteria: an American College of Rheumatology/European League Against Rheumatism collaborative initiative. ANN RHEUM DIS. 2010-09-01; 69 (9): 1580-8. \u003c/li\u003e\n\u003cli\u003eKaye JJ, Fuchs HA, Moseley JW, Nance EP, Callahan LF, Pincus T. Problems with the Steinbrocker staging system for radiographic assessment of the rheumatoid hand and wrist. *Invest Radiol*. 1990;25(5):536-544. https://doi.org/10.1097/00004424-199005000-00007.\u003c/li\u003e\n\u003cli\u003ePrevoo, ML, van \u0026apos;t Hof, MA, Kuper, HH, van Leeuwen, MA, van de Putte, LB, van Riel, PL. Modified disease activity scores that include twenty-eight-joint counts. Development and validation in a prospective longitudinal study of patients with rheumatoid arthritis. ARTHRITIS RHEUM-US. 1995-01-01; 38.\u003c/li\u003e\n\u003cli\u003eText Classification of Cornell Movie Data using Data Mining with Feature Selection Int J Eng Adv Technol. 2019-12-30; 9 (2): 2950-2955. \u003c/li\u003e\n\u003cli\u003eZhang, Z. Missing data imputation: focusing on single imputation. ANN TRANSL MED. 2016-01-01; 4 (1): 9. \u003c/li\u003e\n\u003cli\u003eZebari, R, Abdulazeez, A, Zeebaree, D, Zebari, D, Saeed, J. A Comprehensive Review of Dimensionality Reduction Techniques for Feature Selection and Feature Extraction JASTT. 2020-05-15; 1 (2): 56-70. \u003c/li\u003e\n\u003cli\u003eRemeseiro, B, Bolon-Canedo, V. A review of feature selection methods in medical applications. COMPUT BIOL MED. 2019-09-01; 112 103375. \u003c/li\u003e\n\u003cli\u003eKianifard, F. Logistic Regression: A Self-Learning Text TECHNOMETRICS. 1995-02-01; 37 (1): 116-117. \u003c/li\u003e\n\u003cli\u003eAltman, N. An Introduction to Kernel and Nearest-Neigh32bor Nonparametric Regression AM STAT. 1992-08-01; 46 (3): 175-185. \u003c/li\u003e\n\u003cli\u003eBreiman, L. Random Forests. Machine Learning 45, 5\u0026ndash;32 (2001). \u003c/li\u003e\n\u003cli\u003eCortes, C., Vapnik, V. Support-vector networks. Mach Learn 20, 273\u0026ndash;297 (1995). \u003c/li\u003e\n\u003cli\u003eQuinlan, J.R. Induction of decision trees. Mach Learn 1, 81\u0026ndash;106 (1986). \u003c/li\u003e\n\u003cli\u003eFriedman, Jerome H.. \u0026ldquo;Greedy function approximation: A gradient boosting machine.\u0026rdquo; Annals of Statistics 29 (2001): 1189-1232.\u003c/li\u003e\n\u003cli\u003eFreund Y, Schapire RE. A decision-theoretic generalization of on-line learning and an application to boosting. J Comput Syst Sci. 1997;55(1):119-139. https://doi.org/10.1006/jcss.1997.1504.\u003c/li\u003e\n\u003cli\u003eRefaeilzadeh, P., Tang, L., Liu, H. (2009). Cross-Validation. In: LIU, L., \u0026Ouml;ZSU, M.T. (eds) Encyclopedia of Database Systems. Springer, Boston, MA. Available from:https://doi.org/10.1007/978-0-387-39940-9_565\u003c/li\u003e\n\u003cli\u003eSokolova, M, Lapalme, G. A systematic analysis of performance measures for classification tasks INFORM PROCESS MANAG. 2009-07-01; 45 (4): 427-437.\u003c/li\u003e\n\u003cli\u003eLundberg SM, Lee SI. A unified approach to interpreting model predictions. In: Proceedings of the 31st International Conference on Neural Information Processing Systems (NeurIPS 2017). 2017:4768-4777. https://proceedings.neurips.cc/paper/2017/hash/8a20a8621978632d76c43dfd28b67767-Abstract.html.\u003c/li\u003e\n\u003cli\u003eShah, K, Patel, H, Sanghvi, D, Shah, M. A Comparative Analysis of Logistic Regression, Random Forest and KNN Models for the Text Classification Augment Hum Res. 2020-12-01; 5 (1).\u003c/li\u003e\n\u003cli\u003eBagley, SC, White, H, Golomb, BA. Logistic regression in the medical literature: standards for use and reporting, with particular attention to one medical domain. J CLIN EPIDEMIOL. 2001-10-01; 54 (10): 979-85. \u003c/li\u003e\n\u003cli\u003eJohansson, FD, Collins, JE, Yau, V, Guan, H, Kim, SC, Losina, E, et al. Predicting Response to Tocilizumab Monotherapy in Rheumatoid Arthritis: A Real-world Data Analysis Using Machine Learning. J RHEUMATOL. 2021-09-01; 48 (9): 1364-1370. \u003c/li\u003e\n\u003cli\u003eIchikawa, Y, Hoshina, Y, Horiki, T, Yamada, C, Uchiyama, M, Takaya, M. Molecular markers of coagulation and fibrinolysis as indicators for the disease activity of rheumatoid arthritis Japanese Journal of Rheumatology. 1997-09-01; 7 (3): 173-181.\u003c/li\u003e\n\u003cli\u003eWeinberg, JB, Pippen, AM, Greenberg, CS. Extravascular fibrin formation and dissolution in synovial tissue of patients with osteoarthritis and rheumatoid arthritis. ARTHRITIS RHEUM-US. 1991-08-01; 34 (8): 996-1005. \u003c/li\u003e\n\u003cli\u003eDay, KE, Beck, LN, Heath, CH, Huang, CC, Zinn, KR, Rosenthal, EL. Identification of the optimal therapeutic antibody for fluorescent imaging of cutaneous squamous cell carcinoma. CANCER BIOL THER. 2013-03-01; 14 (3): 271-7. \u003c/li\u003e\n\u003cli\u003eForsblad-d\u0026apos;Elia, H, Bengtsson, K, Kristensen, LE, Jacobsson, LT. Drug adherence, response and predictors thereof for tocilizumab in patients with rheumatoid arthritis: results from the Swedish biologics register. RHEUMATOLOGY. 2015-07-01; 54 (7): 1186-93. \u003c/li\u003e\n\u003cli\u003eSimons, A, Dahl, R, Espinosa-Cotton, M, Rodman, S, Rose-John, S. Abstract 5611: Inhibition of IL-6 trans-signaling in HNSCC CANCER RES. 2018-07-01; 78 (13_Supple): 5611-5611. \u003c/li\u003e\n\u003cli\u003eBykerk, VP, Ost\u0026ouml;r, AJ, Alvaro-Gracia, J, Pavelka, K, Ivorra, JA, Graninger, W, et al. Tocilizumab in patients with active rheumatoid arthritis and inadequate responses to DMARDs and/or TNF inhibitors: a large, open-label study close to clinical practice. ANN RHEUM DIS. 2012-12-01; 71 (12): 1950-4. \u003c/li\u003e\n\u003cli\u003ePers, YM, Fortunet, C, Constant, E, Lambert, J, Godfrin-Valnet, M, De Jong, A, et al. Predictors of response and remission in a large cohort of rheumatoid arthritis patients treated with tocilizumab in clinical practice. RHEUMATOLOGY. 2014-01-01; 53 (1): 76-84. \u003c/li\u003e\n\u003cli\u003eNakajima, T, Watanabe, R, Hashimoto, M, Murata, K, Murakami, K, Tanaka, M, et al. Neutrophil count reduction 1 month after initiating tocilizumab can predict clinical remission within 1 year in rheumatoid arthritis patients. RHEUMATOL INT. 2022-11-01; 42 (11): 1983-1991. \u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Tocilizumab, Rheumatoid Arthritis, Machine Learning, Imaging Staging, Clinical Data","lastPublishedDoi":"10.21203/rs.3.rs-6420083/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6420083/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eIntroduction:\u003c/h2\u003e \u003cp\u003eApproximately 30% of patients with rheumatoid arthritis (RA) do not respond to tocilizumab (TCZ). This study aimed to predict TCZ response using machine learning models trained on clinical data.\u003c/p\u003e\u003ch2\u003eMethods\u003c/h2\u003e \u003cp\u003eBaseline and follow-up data from patients with RA treated with TCZ were collected. Seven different machine learning models (logistic regression, K-nearest neighbor algorithm, random forest, support vector machine, decision tree, gradient boosting, and boosting algorithm) were trained to predict the treatment response of patients with RA across different imaging stages after three to six months of therapy. The area under the receiver operating curve (AUC) was the main performance evaluation feature to screen the best model, and the relative importance of each variable in the model was ranked using the Shapley Additive Explanation (SHAP) value.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e \u003cp\u003eA total of 245 RA patients treated with TCZ were included. The logistic regression model demonstrated superior prediction performance across imaging stages, achieving AUC scores of 0.78 for patients without imaging changes, 0.73 for stage I, and 0.82 for stages II, III, and IV. Key features influencing predictions varied by imaging stage. The most important features were D-dimer concentration for patients without imaging changes, DAS28 grade for patients in stage I, and the physician's overall disease score (EGA) for patients in stages II, III, and IV, with SHAP values of 0.5180, 0.6661, and 1.3978, respectively.\u003c/p\u003e\u003ch2\u003eDiscussion\u003c/h2\u003e \u003cp\u003eThis study demonstrates the potential of machine learning to predict TCZ treatment outcomes and identify stage-specific features in patients with RA.\u003c/p\u003e","manuscriptTitle":"Interpretable machine learning models for predicting tocilizumab response in rheumatoid arthritis using clinical data","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-05-09 10:57:52","doi":"10.21203/rs.3.rs-6420083/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"cddc7e02-ed99-456a-ac53-448032d09174","owner":[],"postedDate":"May 9th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":48163837,"name":"Biological sciences/Immunology/Immunological disorders/Autoimmune diseases"},{"id":48163838,"name":"Health sciences/Medical research/Biomarkers"},{"id":48163839,"name":"Biological sciences/Computational biology and bioinformatics/Classification and taxonomy"},{"id":48163840,"name":"Biological sciences/Computational biology and bioinformatics/Data mining"},{"id":48163841,"name":"Biological sciences/Computational biology and bioinformatics/Data processing"},{"id":48163842,"name":"Biological sciences/Computational biology and bioinformatics/Machine learning"},{"id":48163843,"name":"Biological sciences/Computational biology and bioinformatics/Predictive medicine"},{"id":48163844,"name":"Health sciences/Diseases/Immunological disorders"}],"tags":[],"updatedAt":"2025-05-30T02:23:20+00:00","versionOfRecord":[],"versionCreatedAt":"2025-05-09 10:57:52","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-6420083","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6420083","identity":"rs-6420083","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}