Machine Learning Model Based on Routine Blood Indicators for High-Risk Screening of Interstitial Lung Disease in Patients with Connective Tissue Disease: A Cost-Effective Triage Strategy

Machine Learning Model Based on Routine Blood Indicators for High-Risk Screening of Interstitial Lung Disease in Patients with Connective Tissue Disease: A Cost-Effective Triage Strategy | 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 Research Article Machine Learning Model Based on Routine Blood Indicators for High-Risk Screening of Interstitial Lung Disease in Patients with Connective Tissue Disease: A Cost-Effective Triage Strategy Haoran Wang, Lele Zhang, Huifang Xing, Dong Yang, Chenshen Liu, and 1 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-9173851/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 Background A machine learning model based on routine blood test indicators was constructed to predict the risk of interstitial lung disease (ILD) in patients with connective tissue disease (CTD), thereby providing a convenient tool for clinical screening of high-risk populations. Methods A total of 225 inpatients with connective tissue disease (CTD) admitted to Shanxi Provincial People's Hospital from May 2022 to May 2023 were retrospectively enrolled, including 85 cases in the CTD-ILD group and 140 cases in the pure CTD group. Clinical and laboratory data were collected, including gender, age, serum KL-6 levels, and a full set of routine blood test indicators (including derived inflammatory and immune indexes such as systemic immune-inflammation index [SII] and neutrophil-to-lymphocyte ratio [NLR]). The Adaptive Synthetic Sampling (ADASYN )technique (with three proportions: 50%, 75%, and 100%) was applied to address the issue of data imbalance, and the balanced dataset was divided into a training set and a test set at a ratio of 7:3. Five machine learning models were constructed, namely eXtreme Gradient Boosting (XGBoost), logistic regression, random forest (RF), support vector machine (SVM), and k-nearest neighbor (KNN). Hyperparameters were optimized via 10-fold cross-validation and grid search. Model performance was comprehensively evaluated from four dimensions: discriminative ability (ROC curve, PR curve), calibration (calibration plot), clinical utility (decision curve analysis [DCA]), and confusion matrix metrics (AUC, accuracy, sensitivity, etc.). The SHAP (SHapley Additive exPlanations) method was used to interpret key predictive features, and the Bootstrap method was applied to verify the stability of the models and the reliability of the results. Results Sixteen indicators, including gender, systemic immune-inflammation index (SII), white blood cell count (WBC), mean corpuscular volume (MCV) and red blood cell distribution width standard deviation (RDW-SD) exhibited statistically significant differences between the CTD-ILD group and the pure CTD group (p < 0.05). After feature screening, SII, WBC, MCV and RDW-SD were identified as the core predictive variables. The overall model performance was optimal at the 75% ADASYN sampling ratio, among which the random forest (RF) model achieved the best performance: in the validation set, the area under the curve (AUC) was 0.846, average precision (AP) was 0.896, and F1-score was 0.842. The calibration plot indicated that the model had the minimum deviation between predicted probabilities and actual risks (calibration error = 0.146), and decision curve analysis (DCA) confirmed that the model yielded net clinical benefits across the entire threshold range. SHAP analysis elucidated the action mechanisms of each core variable. Finally, 1000-time Bootstrap resampling validation showed that the RF model had a mean AUC of 0.740 ± 0.040 with a 95% confidence interval (95% CI) of [0.608, 0.848]. All performance indicators presented low coefficients of variation, demonstrating favorable stability and reliability of the model. Conclusions: The random forest (RF) model based on routine blood test indicators demonstrated favorable discriminative ability, calibration and clinical utility in the risk prediction of CTD-ILD. Although its specificity was inferior to that of serum markers such as KL-6, the model leverages the advantages of routine blood tests—high popularity, low cost and strong timeliness—and thus can serve as a convenient tool for the preliminary screening of CTD patients at high risk of ILD, and is particularly suitable for primary medical institutions and large-scale screening scenarios. Trial registration: Clinical trial number: not applicable. Connective tissue disease (CTD) interstitial lung disease (ILD) routine blood test indicators systemic immune-inflammation index (SII) mean corpuscular volume (MCV) red blood cell distribution width standard deviation (RDW-SD) white blood cell count (WBC) Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 1 | Background As the most common and severe pulmonary complication of connective tissue disease (CTD), the early diagnosis of interstitial lung disease (ILD) is crucial for improving patient prognosis. The incidence of ILD across different CTD subtypes ranges from 15% to 65%[1]. It not only severely impairs pulmonary function, presenting with progressive dyspnea, cough and reduced exercise tolerance, but also doubles the risk of death in CTD patients, with the 5-year survival rate of severe cases dropping to 50% to 70[2]. Therefore, early identification of high-risk patients with ILD among CTD cases is crucial for the timely initiation of immunosuppressive or antifibrotic interventions, delaying disease progression and improving long-term prognosis. Current clinical diagnostic methods for CTD-ILD all have certain limitations: HRCT is highly dependent on specialized equipment and costly, making it difficult to popularize in primary medical institutions; pulmonary function tests are susceptible to interference from factors such as patient compliance and concomitant airway diseases, leading to limited accuracy; invasive lung biopsy carries the risk of complications such as bleeding and infection, and has low patient acceptability, thus failing to serve as a routine screening method[3, 4]. Furthermore, the early symptoms of CTD-ILD are insidious, and most patients are diagnosed at the moderate to advanced stage, missing the optimal window for intervention[5]. Although serum KL-6 serves as a specific marker for pulmonary epithelial cell injury, with its levels positively correlated with the extent of HRCT-detected lesions and the degree of pulmonary function decline, and exhibits high sensitivity and specificity, its popularization in primary medical institutions is limited due to restrictive testing conditions[6–8]. Therefore, the development of a convenient, cost-effective and non-invasive detection tool with both early diagnostic value and wide applicability has become an urgent clinical need. Routine blood test indicators and derived inflammatory and immune indices (e.g., neutrophil-to-lymphocyte ratio [NLR], systemic immune-inflammation index [SII], monocyte-to-lymphocyte ratio [MLR]) have attracted extensive attention in the risk prediction of various inflammatory diseases and their complications due to the advantages of convenient acquisition, low detection cost and high repeatability. Existing studies have demonstrated that indicators such as NLR and SII can reflect the state of immune-inflammatory imbalance in the body, and immune-inflammatory dysregulation is one of the core pathological mechanisms underlying the onset of CTD and its complication with ILD[9]. NLR has also been shown to possess predictive value for ILD in subtypes including systemic sclerosis and primary Sjögren’s syndrome[10, 11]. Additionally, indicators such as PLR, LMR and NLR have been confirmed to have diagnostic value in rheumatoid arthritis-associated ILD[12]. However, most existing studies have only focused on the correlation between a single indicator and CTD-ILD, lacking a systematic analysis of the combined predictive value of multiple indicators. Moreover, traditional statistical methods struggle to fully explore the complex interactions among multiple indicators, resulting in limited predictive efficacy. In recent years, machine learning technology has been widely applied to the construction of disease risk prediction models due to its advantages in integrating multi-dimensional indicators and exploring complex interactions. For example, Zhang et al. used routine blood test-derived indices combined with a random forest model to predict all-cause mortality in arthritis patients with hypertension[13]; Xie et al. employed complete blood count-derived inflammatory indicators combined with LightGBM to predict pneumonia associated with acute ischemic stroke[14]. Based on this, we put forward the hypothesis: Can routine blood test indicators and their derived inflammatory and immune indices be combined with machine learning algorithms to construct a convenient, efficient and widely applicable early prediction tool for CTD-ILD, so as to provide scientific support for early clinical diagnosis? In this study, inpatients with CTD admitted to Shanxi Provincial People's Hospital from May 2022 to May 2023 were enrolled as research subjects. We collected their gender, age, and a full set of routine blood test indicators (including derived indices such as NLR, SII and MLR), and constructed five machine learning prediction models, namely eXtreme Gradient Boosting (XGBoost), Logistic Regression (LR), Random Forest Classifier (RF), Support Vector Machine (SVM) and K-Nearest Neighbors (KNN). The model performance was comprehensively evaluated from the perspectives of discriminative ability (ROC curve, PR curve), calibration (calibration plot) and clinical utility (decision curve analysis, DCA). Additionally, the SHapley Additive exPlanations (SHAP) method was used to interpret the key predictive features and their impacts on model outputs. Ultimately, the optimal prediction model was screened out, which provides a scientific tool and clinical evidence for the early and convenient assessment of the risk of ILD complication in CTD patients. 2 | Methods 2.1 | Study Population and Variable Selection In this study, we recruited inpatients with CTD, including all subtypes such as systemic lupus erythematosus (SLE), rheumatoid arthritis (RA), systemic sclerosis (SSc), and Sjögren’s syndrome (SS), who were admitted to Shanxi Provincial People's Hospital from May 2022 to May 2023. The enrolled participants were divided into two groups: the CTD-ILD group and the CTD group. Inclusion criteria included: a diagnosis of CTD based on current criteria[15–22]. The diagnosis of ILD was based on the 2013 criteria for idiopathic interstitial pneumonia (IIP) from the American Thoracic Society (ATS)/European Respiratory Society (ERS) [23]. Subjects were excluded according to the following exclusion criteria: (a) presence of other pulmonary diseases such as COPD or asthma; (b) severe heart, liver or kidney disease; (c) concomitant pulmonary tuberculosis or malignant tumors; (d) hematological diseases; (e) multiple organ failure. This study was conducted in accordance with the principles of the Declaration of Helsinki and approved by the Medical Ethics Committee of Shanxi Provincial People's Hospital (Ethical Approval No.: 2023-229). All participants provided written informed consent prior to their participation in the study. The patient characteristics of interest included gender, age, serum KL-6 (Krebs von den Lungen – 6) levels, and a full panel of routine blood test parameters: NLR (Neutrophil-Lymphocyte Ratio), SII (Systemic Immune-Inflammation Index), MLR (Monocyte-Lymphocyte Ratio), PLR (Platelet-Lymphocyte Ratio), WBC (White Blood Cell Count), NEUT% (Neutrophil Percentage), NEUT (Neutrophil Count), LYMPH% (Lymphocyte Percentage), LYMPH (Lymphocyte Count), MONO% (Monocyte Percentage), MONO (Monocyte Count), EO% (Eosinophil Percentage), EO (Eosinophil Count), BASO% (Basophil Percentage), BASO (Basophil Count), RBC (Red Blood Cell Count), HB (Hemoglobin), HCT (Hematocrit), MCV (Mean Corpuscular Volume), MCH (Mean Corpuscular Hemoglobin), MCHC (Mean Corpuscular Hemoglobin Concentration), RDW-SD (Red Cell Distribution Width-Standard Deviation), RDW-CV (Red Cell Distribution Width-Coefficient of Variation), PLT (Platelet Count), PDW (Platelet Distribution Width), PCT (Plateletcrit), and MPV (Mean Platelet Volume). Laboratory test data of all patients were collected or conducted prior to the current hospitalization. Demographic characteristics including age and gender were extracted from hospital medical records. In addition, 5 mL of fasting venous blood samples were collected from the elbow of each patient without any pre-processing. The blood samples were placed in tubes containing ethylenediaminetetraacetic acid (EDTA) anticoagulant for routine blood tests, and NLR, SII, PLR and MLR were calculated for each sample. The serum was then stored at -80 °C for further research. KL-6 was measured by latex agglutination assay using a biochemical analyzer (Beckman Coulter AU5800) with the kit provided by the manufacturer (Sekisui Medical Co., Ltd., Tokyo, Japan). Our primary outcome variable was a binary variable indicating the presence or absence of ILD in CTD patients. A heatmap was used to visualize the results of Pearson correlation analysis and examine the relationships among variables 2.2 | Handling of Imbalanced Data In the dataset of this study, there was an imbalance in the sample size between CTD patients (n=140) and CTD-ILD patients (n=85). This class distribution discrepancy may cause the model to tend to optimize the prediction accuracy of the majority class (CTD) during training, given that samples of the majority class account for a higher proportion in the loss function. Ultimately, this may lead to a significant decline in the model's ability to identify the minority class (CTD-ILD) (e.g., a low recall rate). As a high-risk population requiring early and accurate identification, an insufficient recall rate for CTD-ILD will directly undermine the clinical application value of the model (e.g., missed diagnosis of high-risk patients). To address this issue, we adopted the ADASYN (Adaptive Synthetic Sampling) method for data balancing. As an advanced oversampling technique, the core advantage of ADASYN lies in its adaptive synthesis of minority class samples: this method calculates the K-nearest neighbor (KNN) distribution density of minority class samples to generate more synthetic data for samples in sparsely distributed regions. This not only avoids the overfitting risk that may be caused by simple random oversampling (e.g., model memorization of noise due to duplicate samples) but also outperforms the traditional SMOTE method (SMOTE does not consider sample distribution density and may generate redundant samples in dense regions). ADASYN can more accurately simulate the real data distribution characteristics of the minority class (CTD-ILD) and enhance the model's ability to learn complex decision boundaries[24]. In selecting the ADASYN oversampling ratio, we comprehensively considered the degree of data imbalance and clinical requirements, and ultimately determined to test three key ratios: 50%, 75%, and 100%. The specific reasons are as follows: In the original data of this study, there were 140 samples in the majority class and 85 in the minority class, indicating a moderate imbalance. The 50% ratio (expanding the minority class to 50% of the majority class) represents mild oversampling, which retains partial characteristics of the original data distribution; the 75% ratio (expanding to 75% of the majority class) is moderate oversampling, which balances the class weights; the 100% ratio (complete balance) maximizes the weight of the minority class. These ratios cover the typical scenarios of mild-moderate-complete balance. Based on this, we implemented the ADASYN technique with oversampling ratios of 50%, 75%, and 100% (Table 1), respectively, and randomly split the balanced datasets into training and test sets at a 7:3 ratio. By comparing the optimal performance of the models and the inter-model consistency under different ratios, we selected the most appropriate ratio among the three. Table 1 Number of instances increased by ADASYN technique Percentage of ADASYN increase Class“CTD” actual 140 Class “CTD-ILD” actual 85 50% 140 130 75% 140 161 100% 140 165 2.3 | Establishment and Evaluation of Prediction Models Five machine learning algorithms were selected in this study to predict ILD complication in CTD patients, namely XGBoost, Logistic Regression, RF, SVM, and KNN. Ten-fold cross-validation (k=10) was adopted as the resampling method on the training set, with hyperparameters tuned via grid search. The validation set was used for model parameter adjustment, while the test set was applied to evaluate the overall model performance. The clinical value of the prediction models was assessed using three metrics of model quality: discrimination, calibration, and clinical usefulness. First, the discriminative ability of the models was quantified by precision-recall (PR) curve analysis. Second, calibration plots were used to evaluate model performance, assessing the calibration level and the deviation between model predictions and actual events. Third, decision curve analysis (DCA) was performed to evaluate clinical usefulness by calculating the net benefit at different threshold probabilities. In addition, confusion matrix metrics of the five models were also assessed, including average precision (AP), accuracy, sensitivity, specificity, and F1-score. 2.4 | Model Interpretation We used the SHAP method to interpret the models, which is a model interpretability approach derived from the Shapley Value in game theory. Its core objective is to quantify the contribution of each feature to the model's prediction results, thereby addressing the interpretability issue of machine learning models[25]. 2.5 | Model Validation We validated the optimal model using the Bootstrap method, a model validation approach based on resampling with replacement. Its core principle is to generate multiple bootstrap samples to simulate the distributions of different training/test sets, thereby quantitatively validating the model's generalization ability and the stability of its evaluation metrics. This method compensates for the limitations of traditional validation approaches (e.g., hold-out method, simple cross-validation) in scenarios with small sample sizes and uneven sample distributions, and is particularly suitable for robust estimation of model performance and quantification of uncertainty [26]. 2.6 | Diagnostic Performance Evaluation Method of KL-6 With interstitial lung disease (ILD) complication as the outcome event, the diagnostic efficacy of each indicator was analyzed using receiver operating characteristic (ROC) curve analysis, and the area under the curve (AUC), sensitivity, specificity, Youden's index, optimal cut-off value, and accuracy were calculated for each indicator. 2.7 | Statistical Analysis All statistical analyses were performed using R version 4.2.3 and Python version 3.11.4. 3 | Results 3.1 | Patient Characteristics A total of 225 CTD patients were enrolled in this study, among whom 85 had ILD complications.Table2 presents the comparisons of patients' gender, age, and laboratory indicators. It can be seen that gender, NLR, SII, MLR, KL-6, WBC, NEUT, NEUT%, LYMPH%, MONO%, BASO%, HCT, MCV, MCH, RDW-SD and RDW-CV showed a statistically significant difference between the two groups with p-value < 0.05. Table 2 Comparison of laboratory indicators between CTD-ILD and CTD patients. Variables ALL(n=225) CTD(n=140) CTD-ILD( n=85) p Age, Median [Q1-Q3] 57.000 [41.000;66.000] 49.500 [32.750;60.000] 63.000 [58.000;71.000] <0.001 NLR, Median [Q1-Q3] 2.612 [1.625;4.385] 2.130 [1.463;3.806] 3.404 [2.107;5.914] <0.001 SII, Median [Q1-Q3] 573.875 [320.195;1063.584] 500.263 [285.793;929.616] 684.676 [371.724;1318.792] 0.006 MLR, Median [Q1-Q3] 0.298 [0.217;0.444] 0.273 [0.211;0.385] 0.369 [0.226;0.536] 0.002 PLR, Median [Q1-Q3] 158.779 [113.245;221.569] 156.620 [120.984;212.414] 162.264 [102.721;222.581] 0.955 KL6, Median [Q1-Q3] 230.910 [156.930;586.190] 182.710 [138.832;231.675] 877.470 [431.700;1524.790] <0.001 WBC, Median [Q1-Q3] 5.830 [4.140;8.210] 5.095 [3.775;7.245] 7.340 [5.270;9.240] <0.001 NEUT, Mean (SD) 64.658 (13.484) 62.431 (13.474) 68.326 (12.748) 0.001 NEUT%, Median [Q1-Q3] 3.800 [2.310;5.540] 3.105 [2.140;4.807] 4.670 [3.280;6.770] <0.001 LYMPH, Median [Q1-Q3] 25.100 [16.900;33.500] 28.350 [19.500;35.125] 20.200 [13.400;29.000] <0.001 LYMPH%, Median [Q1-Q3] 1.350 [1.010;1.850] 1.325 [1.045;1.843] 1.380 [0.940;1.860] 0.949 MONO, Median [Q1-Q3] 7.400 [5.600;9.300] 7.550 [5.600;9.450] 7.300 [5.800;9.100] 0.692 MONO%, Median [Q1-Q3] 0.410 [0.290;0.590] 0.380 [0.270;0.520] 0.490 [0.360;0.690] <0.001 EO, Median [Q1-Q3] 1.200 [0.400;2.400] 1.200 [0.400;2.425] 1.400 [0.500;2.400] 0.849 EO%, Median [Q1-Q3] 0.060 [0.020;0.160] 0.060 [0.020;0.120] 0.090 [0.020;0.170] 0.097 BASO, Median [Q1-Q3] 0.300 [0.200;0.500] 0.300 [0.200;0.500] 0.300 [0.200;0.500] 0.747 BASO%, Median [Q1-Q3] 0.020 [0.010;0.030] 0.010 [0.010;0.030] 0.020 [0.010;0.040] 0.022 RBC, Median [Q1-Q3] 4.000 [3.630;4.310] 4.000 [3.620;4.280] 3.940 [3.670;4.430] 0.548 HB, Median [Q1-Q3] 118.000 [107.000;130.000] 116.500 [106.000;127.000] 122.000 [109.000;132.000] 0.052 HCT, Median [Q1-Q3] 0.364 [0.328;0.397] 0.361 [0.326;0.392] 0.368 [0.335;0.414] 0.026 MCV, Median [Q1-Q3] 92.300 [88.000;95.100] 90.750 [87.375;94.325] 93.500 [90.600;96.900] <0.001 MCH, Median [Q1-Q3] 29.700 [28.500;31.300] 29.450 [28.400;30.825] 30.100 [28.700;32.200] 0.032 MCHC, Median [Q1-Q3] 323.000 [315.000;331.000] 325.000 [316.000;333.000] 323.000 [315.000;330.000] 0.209 RDW-SD, Median [Q1-Q3] 46.100 [43.000;50.400] 44.400 [42.000;48.100] 48.400 [45.000;51.400] <0.001 RDW-CV, Median [Q1-Q3] 13.600 [12.900;14.900] 13.400 [12.700;14.625] 14.000 [13.200;15.100] 0.003 PLT, Median [Q1-Q3] 216.000 [171.000;280.000] 214.000 [171.750;276.500] 220.000 [168.000;291.000] 0.653 PDW, Median [Q1-Q3] 13.600 [10.700;16.000] 13.400 [10.700;16.000] 14.500 [10.075;16.025] 0.912 PCT, Median [Q1-Q3] 0.220 [0.182;0.290] 0.220 [0.184;0.280] 0.220 [0.176;0.296] 0.981 MPV, Median [Q1-Q3] 10.100 [9.300;10.900] 10.100 [9.400;10.800] 9.900 [9.300;11.025] 0.693 Sexmale1, N (%): 0.007 Female 167 (74.222%) 113 (80.714%) 54 (63.529%) Male 58 (25.778%) 27 (19.286%) 31 (36.471%) 3.2 | Feature Analysis Pearson correlation analysis was performed to examine the relationships among the aforementioned variables with statistically significant intergroup differences, and a correlation heatmap was generated accordingly (Figure 1). From the routine blood test parameters, four indicators excluding KL-6 were selected as variables for the machine learning models based on their correlation with the disease and medical professional considerations, namely WBC, SII, MCV, and RDW-SD. Table 3 shows the performance of machine learning classifiers on ADASYN-generated balanced datasets. Based on optimal model performance and inter-model consistency, the 75% ADASYN ratio achieved the best overall classification performance, ensuring minority class identification (F1-score) while retaining good discriminative ability (AUC) and overall accuracy. Table 3 Evaluation of the performance of classification models on imbalance dataset using ADASYN technique in validation set ADASYN AUC(95%CI) cutoff(95%CI) Accuracy( 95%CI) Sensitivity (95%CI) Specificity (95%CI) Positive Predictive Value (95%CI) Negative Predictive Value (95%CI) F1 Score (95%CI) Kappa(95%CI) XGBoost 50% 0.790 (0.642-0.937) 0.843(0.799-0.886) 0.684(0.629-0.739) 0.438(0.379-0.497) 0.864(0.823-0.905) 0.700(0.645-0.755) 0.679(0.623-0.735) 0.538(0.479-0.597) 0.317(0.156 -0.478) 75% 0.840 (0.697-0.982) 0.830(0.788-0.872) 0.762(0.714-0.810) 0.741(0.692-0.790) 0.800(0.755-0.845) 0.870(0.832-0.908) 0.632(0.578-0.686) 0.800(0.755-0.845) 0.510(0.337-0.683) 100% 0.739 (0.589-0.889) 0.794(0.749-0.839) 0.628(0.574-0.682) 0.565(0.509-0.621) 0.700(0.649-0.751) 0.684(0.632-0.736) 0.583(0.528-0.638) 0.619(0.564-0.674) 0.262(0.110-0.414) logistic 50% 0.824 (0.684-0.963) 0.448(0.389-0.507) 0.789(0.740-0.838) 1.000(1.000-1.000) 0.636(0.579-0.693) 0.667(0.611-0.723) 1.000(1.000-1.000) 0.800(0.752-0.848) 0.596(0.426-0.766) 75% 0.756 (0.578-0.933) 0.437(0.381-0.493) 0.762(0.714-0.810) 0.889(0.854-0.924) 0.533(0.477-0.589) 0.774(0.727-0.821) 0.727(0.677-0.777) 0.828(0.785-0.871) 0.449(0.277-0.621) 100% 0.748 (0.597-0.898) 0.503(0.447-0.559) 0.651(0.598-0.704) 0.696(0.644-0.748) 0.600(0.545-0.655) 0.667(0.614-0.720) 0.632(0.578-0.686) 0.681(0.629-0.733) 0.297(0.139-0.455) RandomForest 50% 0.908 (0.820-0.996) 0.600(0.542-0.658) 0.816(0.770-0.862) 0.688(0.633-0.743) 0.909(0.875-0.943) 0.846(0.803-0.889) 0.800(0.752-0.848) 0.759(0.708-0.810) 0.612(0.443-0.781) 75% 0.846 (0.725-0.967) 0.550(0.494-0.606) 0.786(0.740-0.832) 0.889(0.854-0.924) 0.600(0.545-0.655) 0.800(0.755-0.845) 0.750(0.701-0.799) 0.842(0.801-0.883) 0.512(0.339-0.685) 100% 0.668 (0.497-0.840) 0.600(0.545-0.655) 0.698(0.646-0.750) 0.696(0.644-0.748) 0.700(0.649-0.751) 0.727(0.677-0.777) 0.667(0.614-0.720) 0.711(0.660-0.762) 0.394(0.225-0.563) SVM 50% 0.878 (0.772-0.983) 0.624(0.566-0.682) 0.789(0.740-0.838) 0.625(0.567-0.683) 0.909(0.875-0.943) 0.833(0.789-0.877) 0.769(0.719-0.819) 0.714(0.660-0.768) 0.553(0.381-0.725) 75% 0.746 (0.552-0.939) 0.485(0.429-0.541) 0.762(0.714-0.810) 0.852(0.812-0.892) 0.600(0.545-0.655) 0.793(0.747-0.839) 0.692(0.640-0.744) 0.821(0.778-0.864) 0.466(0.293-0.639) 100% 0.698 (0.527-0.868) 0.638(0.584-0.692) 0.698(0.646-0.750) 0.609(0.554-0.664) 0.800(0.755-0.845) 0.778(0.731-0.825) 0.640(0.586-0.694) 0.683(0.631-0.735) 0.402(0.232-0.572) KNN 50% 0.835 (0.706-0.965) 0.800(0.752-0.848) 0.789(0.740-0.838) 0.688(0.633-0.743) 0.864(0.823-0.905) 0.786(0.737-0.835) 0.792(0.744-0.840) 0.733(0.680-0.786) 0.561(0.389-0.733) 75% 0.851 (0.717-0.985) 0.600(0.545-0.655) 0.762(0.714-0.810) 0.852(0.812-0.892) 0.600(0.545-0.655) 0.793(0.747-0.839) 0.692(0.640-0.744) 0.821(0.778-0.864) 0.466(0.293-0.639) 100% 0.732 (0.584-0.879) 0.800(0.755-0.845) 0.698(0.646-0.750) 0.565(0.509-0.621) 0.850(0.810-0.890) 0.812(0.765-0.859) 0.630(0.575-0.685) 0.667(0.614-0.720) 0.406(0.236-0.576) 3.3 | Model Development and Evaluation Five machine learning algorithms, namely XGBoost, Logistic Regression, Random Forest (RF), Support Vector Machine (SVM), and K-Nearest Neighbor (KNN), were selected in this study to construct prediction models. Model development and training were completed on the training set, and a comprehensive performance evaluation was carried out on the validation and test sets from three dimensions: discrimination, calibration, and clinical usefulness. Specifically, this involved plotting the ROC curves for the training and validation sets, precision-recall (PR) curves and calibration plots for the validation set, and conducting decision curve analysis (DCA). Additionally, metrics associated with the confusion matrix were calculated to quantify the model performance. After hyperparameter tuning and horizontal algorithm comparison, the average precision (AP) of all five models exceeded 0.75, confirming that all models possessed a fundamental predictive ability (Figure 2). Among them, the RF model exhibited the most outstanding performance across all core metrics: the AUC of the validation set reached 0.846, the average precision (AP) was 0.896, and the F1-score was as high as 0.842, demonstrating excellent discriminative ability and a well-balanced performance in identifying positive and negative samples. The results of the calibration plot showed that the RF model had the smallest deviation between the predicted probability and the actual risk of ILD occurrence, with a calibration error of only 0.146, which was significantly superior to other models (XGBoost: 0.149, Logistic Regression: 0.190, SVM: 0.169, KNN: 0.141), thus achieving the optimal calibration accuracy (Figure3 A). DCA analysis revealed all models provided a net clinical benefit across all threshold values, with the RF model’s net benefit curve leading in most intervals. This confirms its superior clinical utility for assisting clinicians in screening high-risk patients and developing intervention strategies (Figure3 B). ROC curve analysis showed that the AUC of the RF model reached 1.000 in the training set with the curve close to the upper left corner, indicating an extremely strong classification ability; its AUC remained at a high level of 0.846 in the validation set, which was significantly superior to other models (XGBoost: 0.840, Logistic Regression: 0.756, SVM: 0.746, KNN: 0.851), demonstrating a stable generalization ability (Figure4). Taken together, the RF model exhibited comprehensive advantages in discriminative ability (AUC: 0.846, AP: 0.896), calibration accuracy (calibration error: 0.146), balanced identification of positive and negative samples (F1-score: 0.842), and clinical decision-making value (leading net benefit in DCA), and was thus identified as the optimal prediction model in this study. 3.4 | Model Interpretation Figure 5 A presents the SHAP summary plot of the RF prediction model, which consists of four features ranked by their impact on ILD complication in CTD patients. A higher SHAP value of a feature indicates a higher risk of ILD complication in CTD patients. Red represents high feature values, purple denotes feature values close to the overall mean, and blue indicates low feature values. Figure 5 B presents a bar plot of the mean absolute SHAP values for each feature, which quantifies the average impact of each feature on the model output. A longer blue bar indicates a greater average impact of the corresponding feature on predicting ILD occurrence in CTD patients. Figure 6 provides an example for predicting the risk of ILD complication in CTD patients. In this case, the RF model predicted a risk of 0.90 for ILD complication (reference value: 1.00). 3.5 | Model Validation To further verify the stability and result reliability of the optimal RF model in predicting the risk of ILD in CTD patients, the Bootstrap resampling method (sampling ratio: 75%) was adopted for repeated validation, with a total of 1000 resampled samples generated and corresponding ROC curves constructed. The analysis results showed that the AUC values of the 1000 Bootstrap resamplings ranged from 0.608 to 0.848, with a mean AUC of 0.740±0.040 (Figure 7), indicating that the model had a small fluctuation in predictive performance across different sample subsets and its core discriminative ability exhibited good robustness. Combined with the detailed statistical indicators from the Bootstrap validation (Table 4), all key performance parameters of the model exhibited a low degree of variation: the coefficient of variation (CV) for AUROC was 0.054 and that for the F1-score was 0.050, indicating that the model had strong stability in both its comprehensive classification performance and balanced ability to identify positive and negative samples. The mean accuracy (ACC) was 0.687±0.039 with a 95% confidence interval (CI) of [0.604, 0.758], and the balanced accuracy (balanced ACC) was 0.677±0.040 with a 95% CI of [0.599, 0.753]. These results reflect that the model had good consistency in identifying the two outcomes (ILD complication / no ILD complication) across different sample distributions, with no obvious class bias observed. In terms of core clinical application indicators, the model had a mean recall of 0.799±0.066 with a 95% CI of [0.653, 0.918], indicating a low risk of missed diagnosis for CTD-ILD high-risk patients—a characteristic of great value in clinical screening scenarios. The positive predictive value (PPV) was 0.713±0.040 with a 95% CI of [0.636, 0.787], suggesting a high proportion of actual ILD complication among patients predicted as high-risk by the model, which helps reduce unnecessary further examinations. In contrast, the mean specificity was 0.556±0.075 with a CV of 0.134, representing the parameter with a relatively high degree of variation among all indicators. This is speculated to be associated with the distribution heterogeneity of ILD-positive cases in the sample; however, combined with the net clinical benefit revealed by the DCA analysis, this level of specificity still meets the basic requirements for clinical decision-making. The negative predictive value (NPV) was 0.485±0.039 with a 95% CI of [0.406, 0.555], indicating that a certain proportion of patients predicted as low-risk by the model still had actual ILD complications. In clinical application, a comprehensive judgment should be made by combining the specific clinical characteristics of patients to avoid missed diagnosis caused by over-reliance on the model results alone. In summary, the results of Bootstrap resampling validation demonstrated that all performance metrics of the optimal RF model possessed favorable stability and reliability. The narrow distribution of AUC values, low coefficient of variation, and reasonable confidence interval range further supported the clinical application value of this model in predicting the risk of ILD complication in CTD patients, and its predictive results can provide a robust evidence-based basis for clinicians to formulate screening strategies and intervention plans. Table 4 Random Forest Model with Bootstrap Validation (1000 Resamples) - Core Performance Metrics Metric Mean Std Ci low Ci high CV AUROC 0.740 0.040 0.660 0.816 0.054 ACC 0.687 0.039 0.604 0.758 0.057 balanced_ACC 0.677 0.040 0.599 0.753 0.059 recall 0.799 0.066 0.653 0.918 0.082 specificity 0.556 0.075 0.405 0.714 0.134 F1 0.732 0.037 0.653 0.800 0.050 PPV 0.713 0.040 0.636 0.787 0.056 NPV 0.485 0.039 0.406 0.555 0.079 3.6 | Diagnostic Performance of KL-6 Table 5 and Figure 8 show the diagnostic performance of the KL-6 index for high-risk screening of 225 CTD-ILD patients in this study: the AUC value reached 0.927 (95% confidence interval: 0.890–0.958), with a sensitivity of 0.776, a specificity of 0.957 and a Youden's index of 0.734. When 406.320 was used as the optimal cut-off value, the overall diagnostic accuracy of KL-6 for this disease reached 0.884 (95% confidence interval: 0.820–0.936). Table 5 Diagnostic performance statistics of KL-6 and 4 model variables for disease screening (N=225), including key parameters such as AUC value, sensitivity, specificity, and optimal threshold for each indicator. N AUC value (95% CI) Sensitivity (95% CI) Specificity (95% CI) Youden Index (95% CI) Optimal Cut-off Value (95% CI) Accuracy (95% CI) KL-6 225 0.927(0.890-0.958) 0.776(0.710-0.921) 0.957(0.772-0.989) 0.734(0.638-0.837) 406.320(233.817-432.680) 0.884(0.820-0.936) WBC 225 0.687(0.618-0.752) 0.671(0.438-0.924) 0.643(0.425-0.857) 0.313(0.251-0.441) 6.050(4.650-7.856) 0.667(0.589-0.747) SII 225 0.609(0.529-0.668) 0.612(0.279-1.000) 0.593(0.130-0.917) 0.205(0.127-0.323) 580.748(174.721-1531.001) 0.601(0.440-0.696) RDW-SD 225 0.683(0.620-0.748) 0.776(0.493-0.872) 0.550(0.448-0.800) 0.326(0.232-0.448) 45.000(43.938-48.300) 0.666(0.589-0.727) MCV 225 0.640(0.556-0.705) 0.765(0.313-0.851) 0.493(0.413-0.906) 0.258(0.167-0.390) 90.600(90.200-96.200) 0.621(0.546-0.691) 4 | Discussion In this study, a machine learning model was constructed and validated based on routine blood test indicators, aiming to provide a convenient tool for predicting the risk of ILD complication in CTD patients. Through the comparison of clinical data from 225 CTD patients, SII, WBC, MCV and RDW-SD were finally identified as the core variables for CTD-ILD prediction after feature screening. Five prediction models were established based on the four core variables, among which the Random Forest (RF) model exhibited the most prominent performance, with an AUC of 0.846, an average precision (AP) of 0.896, a sensitivity of 0.889, a specificity of 0.600, an F1-score of 0.842 in the validation set and a calibration error of only 0.146; Decision Curve Analysis (DCA) confirmed that it yielded a net clinical benefit across the entire threshold range. SHAP analysis revealed the mechanism of action of each core variable. Finally, after 1000 times of Bootstrap resampling validation, the RF model had a mean AUC of 0.740 ± 0.040 with a 95% confidence interval of [0.608, 0.848], and all performance metrics showed low coefficients of variation, demonstrating favorable stability and reliability. In the current diagnosis of CTD-ILD, various detection methods have obvious limitations: HRCT involves expensive equipment and ionizing radiation, has low popularization at the primary care level, and is not suitable for large-scale screening; pulmonary function tests are susceptible to interference from patient compliance, leading to limited accuracy; invasive lung biopsy carries the risk of complications and has low patient acceptability. None of these methods can be used as a routine screening tool[ 3 , 4 ]. As a specific marker of type Ⅱ alveolar epithelial cells, KL-6 can directly reflect the degree of alveolar epithelial injury and is a clinically recognized indicator for definitive diagnosis and disease assessment[ 8 , 27 ]. In the high-risk screening of 225 CTD-ILD patients in this study, KL-6 exhibited prominent advantages in accurate diagnosis with an AUC of 0.927, a specificity of 0.957 and an accuracy of 0.884, yet it is limited by the reliance on specific detection equipment, low availability at the primary care level and relatively high testing costs. Previous diagnostic studies on routine blood test indicators have mostly focused on the predictive value of individual markers. Several studies have shown that patients with CTD-ILD often present with an elevated total white blood cell count (WBC) or an increased neutrophil ratio (NEUT%)[ 28 , 29 ]. Other studies have found that the RDW level in patients with CTD-ILD is typically higher than that in patients with CTD alone, and a higher RDW value is associated with poorer pulmonary function and lower overall survival, suggesting that it can serve as a potential predictive marker for disease severity and prognosis[ 28 , 30 ]. Other studies have also developed RF prediction models based on serum markers such as KL-6 and SP-D, which achieved an 85% prediction accuracy on unseen test data[ 31 ]. However, they are also difficult to apply for large-scale screening due to high testing costs and low availability at the primary care level. Compared with the aforementioned existing studies, the prediction model in this study was developed based on routine blood test indicators, which not only offers core advantages of convenient detection, low cost, and high accessibility at the primary care level. Although the AUC (0.846) and specificity (0.600) of the model in the validation set are lower than those of KL-6 (AUC = 0.927, specificity = 0.957), it exhibits a superior sensitivity in comparison (0.889 for the model vs. 0.776 for KL-6), thus demonstrating more prominent efficacy in reducing the risk of missed diagnosis. Additionally, this model integrates the complex interactive effects of four core indicators (SII, WBC, MCV, RDW-SD), outputs individualized risk probabilities, and supports clinical stratified screening, making it more suitable for large-scale preliminary screening and scenarios with limited medical resources. The four core indicators of the model established in this study cover the key pathophysiological links of CTD-ILD from different dimensions and have clear detection significance, yet the diagnostic efficacy of each individual indicator is relatively limited, making it difficult to accurately distinguish between CTD-ILD patients and those with CTD alone. This is consistent with the clinical pathological mechanism: the pathogenesis of CTD-ILD results from the combined action of multiple factors such as immune-inflammatory imbalance, tissue injury, and hematopoietic dysfunction. A single indicator can only reflect abnormalities in one of these links and cannot fully capture the complex pathological characteristics of the disease, thus having limited predictive value when applied alone. Given the limitations of individual indicators, this study comprehensively integrated the four core indicators using machine learning algorithms, and the constructed RF model achieved a significant improvement in diagnostic efficacy. Leveraging the advantages of the RF algorithm, the model fully explores the complex interactive effects among the indicators and realizes multi-dimensional integration of the systemic inflammatory level reflected by WBC, the immune-inflammatory imbalance state represented by SII, the hematopoietic function changes associated with MCV, and the red blood cell heterogeneity embodied by RDW-SD. SHAP analysis revealed that the mean absolute SHAP value of RDW-SD ranked first among the four core variables, making it the key feature driving model prediction. In the study sample, the median RDW-SD in the CTD-ILD group (48.400 fL) was significantly higher than that in the CTD-only group (44.400 fL), with a statistically significant intergroup difference (p < 0.001). This result is consistent with the conclusion of previous studies that abnormally elevated RDW is associated with the development of CTD-ILD 32 . From a pathophysiological perspective, patients with ILD exhibit diffuse inflammatory and fibrotic lesions in the pulmonary interstitium, which directly impairs the integrity of the alveolar-capillary barrier, reduces gas exchange efficiency, and induces a chronic, occult state of mild hypoxia. This persistent hypoxia stimulates alterations in the bone marrow hematopoietic microenvironment, leading to heterogeneous maturation during erythropoiesis and thereby increasing red blood cell size heterogeneity. Additionally, the abnormally activated immune-inflammatory response in patients with CTD releases large amounts of reactive oxygen species, inflammatory cytokines and other substances, which directly damage the lipid and protein structures of the red blood cell membrane, impair the morphological stability of red blood cells, and further exacerbate the variation in red blood cell volume distribution[ 30 , 32 ]. Furthermore, an elevation in RDW-SD may also indirectly reflect disease activity and the status of pulmonary function impairment. Existing studies have shown that the RDW level in CTD-ILD patients is negatively correlated with pulmonary function indices (e.g., forced vital capacity, diffusing capacity), and a higher RDW-SD indicates a more severe degree of pulmonary interstitial fibrosis and a more obvious gas exchange disorder 32 . This characteristic enables RDW-SD to serve not only as a risk prediction indicator for CTD-ILD but also as a reference for assessing disease severity, further highlighting its application value in clinical practice. As a classic indicator reflecting the activity of systemic inflammation, the median white blood cell count (WBC) in the CTD-ILD group (7.340×10⁹/L) was higher than that in the CTD-only group (5.095×10⁹/L). Activated leukocytes (especially neutrophils and monocytes) accumulate in the pulmonary interstitium, release inflammatory mediators, and participate in fibroblast activation, thereby driving the progression of pulmonary fibrosis[ 33 – 35 ]. he systemic immune-inflammation index (SII) integrates platelet, neutrophil and lymphocyte counts, and can comprehensively reflect the degree of immune-inflammatory imbalance[ 36 ]. In this study, the median SII in the CTD-ILD group (684.676) was significantly higher than that in the CTD-only group (500.263), indicating that the enhanced inflammation mediated by neutrophils and lymphocyte suppression jointly promote the development of ILD. Neutrophil-released elastase damages alveolar epithelial cells, while lymphocyte reduction impairs the ability to regulate inflammation[ 37 , 38 ]. Mean corpuscular volume (MCV) reflects the average size of red blood cells. In this study, the median MCV in the CTD-ILD group (93.500 fL) was higher than that in the CTD-only group (90.750 fL), which may be associated with erythropoietic disorders caused by chronic inflammation or metabolic disturbances of vitamin B₁₂ and folic acid. Additionally, studies have suggested an association between MCV and hypoxic status in patients with pulmonary fibrosis, enabling it to indirectly reflect pulmonary function impairment[ 32 , 39 ]. This model features a clear and easy-to-implement application pathway in real clinical settings, which can be seamlessly integrated into the routine diagnosis and treatment process for CTD patients, and is particularly suitable for primary medical institutions and large-scale screening needs. In clinical practice, for all CTD patients attending the clinic, after routine blood tests (already widely used as a basic examination), data on the four core indicators (WBC, SII, MCV, RDW-SD) can be directly extracted and input into this RF model to quickly generate individualized probability of ILD onset. For high-risk patients, it is recommended to prioritize further confirmatory examinations, such as serum KL-6 testing combined with HRCT imaging assessment, to clarify the presence of ILD and the scope of lesions. Meanwhile, immunosuppressive or anti-fibrotic intervention regimens should be formulated as early as possible, and routine blood tests and pulmonary function tests should be rechecked every 3–6 months to dynamically monitor changes in risk. For moderate-risk patients, the primary disease treatment can be strengthened first to control inflammatory activity; routine blood tests should be rechecked every 6–12 months and the model risk should be reassessed, and confirmatory examinations should be initiated promptly if the risk escalates. For low-risk patients, routine follow-up for the primary disease is sufficient, and model reassessment should be conducted during annual rechecks to avoid missed diagnosis of occult ILD. This "routine testing + model preliminary screening + stratified confirmation" model not only avoids imposing additional testing burdens on patients (relying on existing routine blood test data) but also greatly improves the screening efficiency for high-risk populations of ILD and reduces the waste of resources such as unnecessary HRCT and KL-6 tests. At the same time, its high sensitivity (0.889) effectively reduces the risk of missed diagnosis, ensures that the opportunity for early intervention is not lost, and achieves an optimal balance between clinical benefits and medical costs. When constructing this model, patient basic information other than gender and age, as well as patients’ chief complaints and physical examination findings, were not included as variables, because the model focuses on application scenarios in primary medical institutions and large-scale screening. The core advantage of the model design is its rapid operation relying on routine blood test indicators with high popularity and objective detection, avoiding the operational burden, time cost and labor consumption caused by additional collection of complex clinical information, which is in line with the efficiency-first demand of screening scenarios. Core routine blood test indicators such as SII, WBC, MCV and RDW-SD can fully cover the pathophysiological mechanisms related to the onset of CTD-ILD, such as immune-inflammatory imbalance and tissue injury, enabling efficient and convenient risk prediction without relying on other basic information, chief complaints or physical examination findings. This study also has several limitations. First, it adopted a single-center retrospective design with a relatively limited sample size, which may lead to selection bias. Additionally, the concentrated distribution of patient regions, ethnicities, and disease subtypes in a single center reduces the external generalizability of the model. Second, there is a lack of multicenter external validation. Given the variations in medical standards, detection equipment, and patient population characteristics across different centers, the application effectiveness of the model in other institutions has not been verified, which restricts its nationwide promotion. Third, stratified analysis by CTD disease subtypes was not performed. The pathogenesis and risk factors for ILD comorbidity may differ among subtypes such as systemic lupus erythematosus, rheumatoid arthritis, and systemic sclerosis, and the predictive efficacy of the model for patients with a single subtype needs further verification. Fourth, the model has a relatively low specificity, with a mean value of 0.556 ± 0.075 and a coefficient of variation of 0.134, which may result in a certain number of false positive results. Finally, long-term follow-up data of patients were not collected, making it impossible to evaluate the long-term predictive value of the model for the onset of CTD-ILD and to verify the correlation between the model's predictive results and patient prognosis. To address the limitations of this study, future work may be carried out in the following aspects: conducting multicenter prospective studies to enroll CTD patients from medical institutions in different regions and at different levels, covering more disease subtypes and population characteristics; further expanding the sample size, performing stratified analysis by CTD disease subtypes, and constructing subtype-specific prediction models to improve the predictive efficacy for patients with a single subtype. In summary, the RF model constructed in this study based on routine blood test indicators has demonstrated favorable discriminative ability, calibration, and clinical practicality in predicting the risk of CTD-ILD. By comparing with existing clinical testing methods, it highlights its unique advantages in popularity, cost, and timeliness. It can serve as a convenient tool for the preliminary screening of high-risk populations for ILD in CTD patients, and is particularly suitable for primary medical institutions and large-scale screening scenarios. 5 | Conclusions The random forest model based on routine blood test indicators (SII, WBC, MCV, RDW-SD) constructed in this study has good discriminative ability, calibration and clinical utility, and the model is stable and reliable with high sensitivity. Relying on the advantages of routine blood test indicators such as wide popularity, low cost and strong timeliness, the model can serve as a convenient and effective preliminary screening tool for CTD patients with high risk of ILD, which is of great significance for improving the early identification rate of CTD-ILD, perfecting the clinical screening system of CTD-ILD, and guiding the rational allocation of medical resources. Although the model has certain limitations such as low specificity and single-center research design, the research results still provide important clinical evidence and scientific ideas for the early screening of CTD-ILD, and lay a foundation for the subsequent multi-center prospective research and model optimization. Abbreviations Abbreviation Full Name CTD Connective Tissue Disease ILD Interstitial Lung Disease CTD-ILD Connective Tissue Disease-Associated Interstitial Lung Disease SII Systemic Immune-Inflammation Index WBC White Blood Cell Count MCV Mean Corpuscular Volume RDW-SD Red Blood Cell Distribution Width-Standard Deviation NLR Neutrophil-Lymphocyte Ratio MLR Monocyte-Lymphocyte Ratio PLR Platelet-Lymphocyte Ratio KL-6 Krebs von den Lungen – 6 ADASYN Adaptive Synthetic Sampling RF Random Forest XGBoost eXtreme Gradient Boosting SVM Support Vector Machine KNN k-Nearest Neighbor AUC Area Under the Curve AP Average Precision DCA Decision Curve Analysis SHAP SHapley Additive exPlanations HRCT High-Resolution Computed Tomography COPD Chronic Obstructive Pulmonary Disease SLE Systemic Lupus Erythematosus RA Rheumatoid Arthritis SSc Systemic Sclerosis SS Sjögren’s Syndrome IIP Idiopathic Interstitial Pneumonia ATS American Thoracic Society ERS European Respiratory Society EDTA Ethylenediaminetetraacetic Acid CV Coefficient of Variation CI Confidence Interval ACC Accuracy PPV Positive Predictive Value NPV Negative Predictive Value TIMPs Tissue Inhibitors of Matrix Metalloproteinases MMPs Matrix Metalloproteinases TGF-β Transforming Growth Factor-β PDGF Platelet-Derived Growth Factor IFN-γ Interferon-γ IL Interleukin TNF-α Tumor Necrosis Factor-α HIF-1α Hypoxia-Inducible Factor-1α RBC Red Blood Cell Count HB Hemoglobin HCT Hematocrit MCH Mean Corpuscular Hemoglobin MCHC Mean Corpuscular Hemoglobin Concentration RDW-CV Red Blood Cell Distribution Width-Coefficient of Variation PLT Platelet Count PDW Platelet Distribution Width PCT Plateletcrit MPV Mean Platelet Volume NEUT Neutrophil Count NEUT% Neutrophil Percentage LYMPH Lymphocyte Count LYMPH% Lymphocyte Percentage MONO Monocyte Count MONO% Monocyte Percentage EO Eosinophil Count EO% Eosinophil Percentage BASO Basophil Count BASO% Basophil Percentage Declarations Ethics approval and consent to participate The Medical Ethics Committee of Shanxi provincial people’s hospital approved the study (2023-229). Consent for publication Not applicable Availability of data and materials All data generated or analysed during this study are included in this published article [and its supplementary information files] Competing interests The authors declare that they have no competing interests Funding The authors received no funding for this work. Authors' contributions Wang Haoran and Zhang Lele are co-first authors. Wang Haoran was responsible for designing the research plan, performing most of the experiments, collecting and analyzing experimental data, and drafting the initial manuscript. Zhang Lele assisted in experimental operation, data collation and manuscript revision. Xing Huifang, Yang Dong and Liu Chenshen participated in part of the experimental operation, sample collection and data verification. As the corresponding author, Liang Hongping was responsible for supervising the entire research process, providing guidance on experimental design and data analysis, critically revising the important academic content of the manuscript, and reviewing and finalizing the manuscript for submission Acknowledgements Not applicable References Joy GM, Arbiv OA, Wong CK, Lok SD, Adderley NA, Dobosz KM, et al. Prevalence, imaging patterns and risk factors of interstitial lung disease in connective tissue disease: a systematic review and meta-analysis. Eur Respir Rev. 2023;32:220210. https://doi.org/10.1183/16000617.0210-2022. Jaeger VK, Wirz EG, Allanore Y, Rossbach P, Riemekasten G, Hachulla E, et al. Incidences and Risk Factors of Organ Manifestations in the Early Course of Systemic Sclerosis: A Longitudinal EUSTAR Study. PLoS ONE. 2016;11:e0163894. https://doi.org/10.1371/journal.pone.0163894. Herzog EL, Mathur A, Tager AM, Feghali-Bostwick C, Schneider F, Varga J. Review: Interstitial Lung Disease Associated With Systemic Sclerosis and Idiopathic Pulmonary Fibrosis: How Similar and Distinct? Arthritis & Rheumatology. 2014;66:1967–78. https://doi.org/10.1002/art.38702. Frauenfelder T, Winklehner A, Nguyen TDL, Dobrota R, Baumueller S, Maurer B, et al. Screening for interstitial lung disease in systemic sclerosis: performance of high-resolution CT with limited number of slices: a prospective study. Annals of the Rheumatic Diseases. 2014;73:2069–73. https://doi.org/10.1136/annrheumdis-2014-205637. Hernandez-Gonzalez F, Prieto-González S, Brito-Zeron P, Cuerpo S, Sanchez M, Ramirez J, et al. Impact of a systematic evaluation of connective tissue disease on diagnosis approach in patients with interstitial lung diseases. Medicine. 2020;99:e18589. https://doi.org/10.1097/MD.0000000000018589. Yamakawa H, Hagiwara E, Kitamura H, Yamanaka Y, Ikeda S, Sekine A, et al. Serum KL-6 and surfactant protein-D as monitoring and predictive markers of interstitial lung disease in patients with systemic sclerosis and mixed connective tissue disease. J Thorac Dis. 2017;9:362–71. https://doi.org/10.21037/jtd.2017.02.48. Lee JS, Lee EY, Ha Y-J, Kang EH, Lee YJ, Song YW. Serum KL-6 levels reflect the severity of interstitial lung disease associated with connective tissue disease. Arthritis Res Ther. 2019;21:58. https://doi.org/10.1186/s13075-019-1835-9. Xing H, Liang H. The clinical value of KL-6 for predicting the occurrence and severity of connective tissue disease-associated interstitial lung disease is not affected by CTD type or treatment. PeerJ. 2024;12:e17792. https://doi.org/10.7717/peerj.17792. Ruta VM, Man AM, Alexescu TG, Motoc NS, Tarmure S, Ungur RA, et al. Neutrophil-to-Lymphocyte Ratio and Systemic Immune-Inflammation Index—Biomarkers in Interstitial Lung Disease. Medicina. 2020;56:381. https://doi.org/10.3390/medicina56080381. Zhang K, Li W, He C, He X, Hou J. The ratio of neutrophil to lymphocyte predicts interstitial lung disease and its prognosis in patients with primary Sjögren’s syndrome: a retrospective analysis. Ann Palliat Med. 2021;10:6493–501. https://doi.org/10.21037/apm-21-1043. Ati̇Lla N, Yildirim Çeti̇N G, Balkarli A. Association of neutrophil/lymphocyte ratio with the degree ofinterstitial lung disease in systemic sclerosis. Turk J Med Sci. 2016;46:1871–4. https://doi.org/10.3906/sag-1601-87. Chen Q, Chen D, Xu X, Liu Y, Yin T, Li D. Platelet/Lymphocyte, Lymphocyte/Monocyte, and Neutrophil/Lymphocyte Ratios as Biomarkers in Patients with Rheumatoid Arthritis and Rheumatoid Arthritis-Associated Interstitial Lung Disease. Med Sci Monit. 2019;25:6474–81. https://doi.org/10.12659/MSM.916583. K Z, X M, X Z, G Q, C Z. Machine learning based association between inflammation indicators (NLR, PLR, NPAR, SII, SIRI, and AISI) and all-cause mortality in arthritis patients with hypertension: NHANES 1999-2018. Frontiers in public health. 2025;13. https://doi.org/10.3389/fpubh.2025.1559603. Xie M, Liu Z, Dai F, Cao Z, Wang X. Predicting Stroke-Associated Pneumonia in Acute Ischemic Stroke: A Machine Learning Model Development and Validation Study with CBC-Derived Inflammatory Indices. Int J Gen Med. 2025;18:3117–28. https://doi.org/10.2147/IJGM.S524450. Aggarwal R, Rider LG, Ruperto N, Bayat N, Erman B, Feldman BM, et al. 2016 American College of Rheumatology/European League Against Rheumatism criteria for minimal, moderate, and major clinical response in adult dermatomyositis and polymyositis: An International Myositis Assessment and Clinical Studies Group/Paediatric Rheumatology International Trials Organisation Collaborative Initiative. Annals of the Rheumatic Diseases. 2017;76:792–801. https://doi.org/10.1136/annrheumdis-2017-211400. Alarcón-Segovia D, Cardiel MH. Comparison between 3 diagnostic criteria for mixed connective tissue disease. Study of 593 patients. J Rheumatol. 1989;16:328–34. Aletaha D, Neogi T, Silman AJ, Funovits J, Felson DT, Bingham III CO, et al. 2010 Rheumatoid arthritis classification criteria: An American College of Rheumatology/European League Against Rheumatism collaborative initiative. Arthritis & Rheumatism. 2010;62:2569–81. https://doi.org/10.1002/art.27584. Aringer M, Costenbader K, Daikh D, Brinks R, Mosca M, Ramsey-Goldman R, et al. 2019 European League Against Rheumatism/American College of Rheumatology classification criteria for systemic lupus erythematosus. Ann Rheum Dis. 2019;78:1151–9. https://doi.org/10.1136/annrheumdis-2018-214819. Leavitt RY, Fauci AS, Bloch DA, Michel BA, Hunder GG, Arend WP, et al. The American College of Rheumatology 1990 criteria for the classification of Wegener’s granulomatosis. Arthritis Rheum. 1990;33:1101–7. https://doi.org/10.1002/art.1780330807. Mosca M, Neri R, Bombardieri S. Undifferentiated connective tissue diseases (UCTD): a review of the literature and a proposal for preliminary classification criteria. Clin Exp Rheumatol. 1999;17:615–20. Shiboski SC, Shiboski CH, Criswell LA, Baer AN, Challacombe S, Lanfranchi H, et al. American College of Rheumatology classification criteria for Sjögren’s syndrome: a data-driven, expert consensus approach in the Sjögren’s International Collaborative Clinical Alliance cohort. Arthritis Care Res (Hoboken). 2012;64:475–87. https://doi.org/10.1002/acr.21591. van den Hoogen F, Khanna D, Fransen J, Johnson SR, Baron M, Tyndall A, et al. 2013 classification criteria for systemic sclerosis: an American college of rheumatology/European league against rheumatism collaborative initiative. Ann Rheum Dis. 2013;72:1747–55. https://doi.org/10.1136/annrheumdis-2013-204424. Travis WD, Costabel U, Hansell DM, King TE, Lynch DA, Nicholson AG, et al. An official American Thoracic Society/European Respiratory Society statement: Update of the international multidisciplinary classification of the idiopathic interstitial pneumonias. Am J Respir Crit Care Med. 2013;188:733–48. https://doi.org/10.1164/rccm.201308-1483ST. Haibo He, Yang Bai, Garcia EA, Shutao Li. ADASYN: Adaptive synthetic sampling approach for imbalanced learning. In: 2008 IEEE International Joint Conference on Neural Networks (IEEE World Congress on Computational Intelligence). Hong Kong, China: IEEE; 2008. p. 1322–8. https://doi.org/10.1109/IJCNN.2008.4633969. Chen H, Lundberg SM, Lee S-I. Explaining a series of models by propagating Shapley values. Nat Commun. 2022;13:4512. https://doi.org/10.1038/s41467-022-31384-3. Henderson AR. The bootstrap: A technique for data-driven statistics. Using computer-intensive analyses to explore experimental data. Clinica Chimica Acta. 2005;359:1–26. https://doi.org/10.1016/j.cccn.2005.04.002. El-Beheidy R, Domouky AM, Zidan H, Amer YA. Serum KL-6 as predictive and prognostic marker of interstitial lung disease in childhood connective tissue diseases: a pilot study. Reumatismo. 2021;73. https://doi.org/10.4081/reumatismo.2021.1399. Shi S, Chen L, Gui X, Chen L, Qiu X, Yu M, et al. Association of Red Blood Cell Distribution Width Levels with Connective Tissue Disease-Associated Interstitial Lung Disease (CTD-ILD). Disease Markers. 2021;2021:1–7. https://doi.org/10.1155/2021/5536360. Yang HJ, Dong ZX, Wang Y, Li JB, Liu YY. Diagnostic value of IL-27 and KL-6 in the diagnosis of connective tissue disease-associated interstitial lung disease[J]. Chin J Int Respir , 2020, 40 :911-916. Liu C, Yang J, Lu Z. Study on the Red Blood Cell Distribution Width in Connective Tissue Disease Associated with Interstitial Lung Disease. Biomed Res Int. 2020;2020:8130213. https://doi.org/10.1155/2020/8130213. Györfi A-H, Filla T, Dickel N, Möller F, Li Y-N, Bergmann C, et al. Performance of serum biomarkers reflective of different pathogenic processes in systemic sclerosis-associated interstitial lung disease. Rheumatology. 2024;63:962–9. https://doi.org/10.1093/rheumatology/kead332. Luo C, Shi J, Zhang J, Lin Y, Pan Y, Zhang J, et al. Survival and early outcomes following lung transplantation for interstitial lung disease associated with non-scleroderma connective tissue disease: a national cohort study. Clin Exp Rheumatol. 2025;43:477–85. https://doi.org/10.55563/clinexprheumatol/tjnyz5. Ahmed S, Handa R. Management of Connective Tissue Disease-related Interstitial Lung Disease. Curr Pulmonol Rep. 2022;11:86–98. https://doi.org/10.1007/s13665-022-00290-w. Wang J, Li K, Hao D, Li X, Zhu Y, Yu H, et al. Pulmonary fibrosis: pathogenesis and therapeutic strategies. MedComm (2020). 2024;5:e744. https://doi.org/10.1002/mco2.744. Ding L, Yang J, Zhang C, Zhang X, Gao P. Neutrophils Modulate Fibrogenesis in Chronic Pulmonary Diseases. Front Med (Lausanne). 2021;8:616200. https://doi.org/10.3389/fmed.2021.616200. Zhang Y, Chen B, Wang L, Wang R, Yang X. Systemic immune-inflammation index is a promising noninvasive marker to predict survival of lung cancer. Medicine (Baltimore). 2019;98:e13788. https://doi.org/10.1097/MD.0000000000013788. Rzepka-Wrona P, Miądlikowska E, Skoczyński S, Barczyk A, Piotrowski W. Patterns of lung fibrosis in patients with interstitial pneumonia with autoimmune features and connective tissue diseases-associated interstitial lung disease-a narrative review. Ann Palliat Med. 2022;11:2110–30. https://doi.org/10.21037/apm-21-3974. Ishiwari M, Kono Y, Togashi Y, Kobayashi K, Kikuchi R, Kogami M, et al. Prognosis of Connective Tissue Disease Related Interstitial Lung Disease after Initiation of Long-Term Oxygen Therapy: Comparison with Idiopathic Pulmonary Fibrosis. OJRD. 2024;14:111–21. https://doi.org/10.4236/ojrd.2024.144011. Ma C, Meng K, Shi S, Zhao T, Chen S, Zhou X, et al. Clinical significance of interleukin-6, total bilirubin, CD3 + CD4 + T cells counts in the acute exacerbation of connective tissue disease-associated interstitial lung disease: a cross-sectional study. Eur J Med Res. 2023;28:393. https://doi.org/10.1186/s40001-023-01384-0. Additional Declarations No competing interests reported. Supplementary Files Rawdataset.xlsx ExplanationofVariableSelectioninthePredictionModel.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-9173851","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":613824343,"identity":"060f3c4c-58fb-41e3-bec7-b608f67f04e0","order_by":0,"name":"Haoran Wang","email":"","orcid":"","institution":"Shanxi Provincial People's Hospital","correspondingAuthor":false,"prefix":"","firstName":"Haoran","middleName":"","lastName":"Wang","suffix":""},{"id":613824344,"identity":"b04fb524-613a-4f68-a288-16ac4a5feb88","order_by":1,"name":"Lele Zhang","email":"","orcid":"","institution":"Shanxi Provincial People's Hospital","correspondingAuthor":false,"prefix":"","firstName":"Lele","middleName":"","lastName":"Zhang","suffix":""},{"id":613824345,"identity":"30147b79-4c53-4557-b865-597b959cec78","order_by":2,"name":"Huifang Xing","email":"","orcid":"","institution":"Shanxi Provincial People's Hospital","correspondingAuthor":false,"prefix":"","firstName":"Huifang","middleName":"","lastName":"Xing","suffix":""},{"id":613824347,"identity":"be553379-4670-4416-8a38-cfbdccfd3c8c","order_by":3,"name":"Dong Yang","email":"","orcid":"","institution":"Shanxi Provincial People's Hospital","correspondingAuthor":false,"prefix":"","firstName":"Dong","middleName":"","lastName":"Yang","suffix":""},{"id":613824348,"identity":"7c3967ea-31d4-4c1c-a165-7c613ff63269","order_by":4,"name":"Chenshen Liu","email":"","orcid":"","institution":"Shanxi Provincial People's Hospital","correspondingAuthor":false,"prefix":"","firstName":"Chenshen","middleName":"","lastName":"Liu","suffix":""},{"id":613824349,"identity":"8c6d53a1-6a29-4eac-8057-37e2317b0499","order_by":5,"name":"Hongping Liang","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAzUlEQVRIiWNgGAWjYBADZgYe5gMHPvwgTQtb4sGZPSTZw8NjfJiDjQiF8jPSn3262XaY3eDMmQ+HGXgY5PnFDuDXAlRpPDu37TCzwdneDYcLLBgMZ85OIKCFvYeZGazlPO+GwzN4GBIMbhPQIt/M/hiqhefBYR42IrQwHG8whmg528NAnBaQX5hzzqUzS545ZgAMZAnCfgGG2GPmnDLrZL4zyY8/fPhhI88vTchhIMDI1pyscADMlCBCORj8qbOTbyBW8SgYBaNgFIw4AABGAEfV/I5kZQAAAABJRU5ErkJggg==","orcid":"","institution":"Shanxi Provincial People's Hospital","correspondingAuthor":true,"prefix":"","firstName":"Hongping","middleName":"","lastName":"Liang","suffix":""}],"badges":[],"createdAt":"2026-03-20 02:08:24","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-9173851/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-9173851/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":105786386,"identity":"a687f23b-b5ad-448a-9b50-9c688a1bd727","added_by":"auto","created_at":"2026-03-31 06:45:33","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":234986,"visible":true,"origin":"","legend":"\u003cp\u003eCorrelation between factors\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-9173851/v1/ab10cd50e0c39ae72f17ed04.png"},{"id":105904563,"identity":"fab18612-1012-44b9-87b6-3b428002e792","added_by":"auto","created_at":"2026-04-01 10:09:32","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":248224,"visible":true,"origin":"","legend":"\u003cp\u003eEvaluation of the prediction model for CTD Patients Complicated with ILD, the average precision recall curves, indicating the trade off between precision and recall.\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-9173851/v1/505cfe283b32dd5a94cb2e08.png"},{"id":105786411,"identity":"2add0c73-b426-4223-9645-29c258b7f967","added_by":"auto","created_at":"2026-03-31 06:45:33","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":306872,"visible":true,"origin":"","legend":"\u003cp\u003eExamples of calibration plots and decision curve for predicting with various models: XGBoost, logistic, RandomForest, SVM, and KNN. (A)Calibration curve (Validation). The dotted line on the graph represents the perfect match between the observed (y - axis, Fraction of positives) and predicted (x - axis, Mean predicted value) probabilities. A closer distance between a model's curve and the dotted line indicates greater calibration accuracy. (B) Validation Decision Curve. The curves show the mean net benefit against threshold probabilities based on decisions from model outputs. The curves referred to as “Treat All” and “Treat None” represent two kinds of extreme decisions\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-9173851/v1/9cf47304464e6a266e6bb7d9.png"},{"id":105786389,"identity":"9771fe02-8e7d-42cd-9314-9fd1deb2a042","added_by":"auto","created_at":"2026-03-31 06:45:33","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":350302,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003e(A)\u003c/strong\u003eROC curve of the training set. This figure shows the performance of various machine learning models in distinguishing positive and negative samples on the training data. \u003cstrong\u003e(B)\u003c/strong\u003e ROC curve of the validation set. This figure shows the generalization performance of the same models on unseen validation data.\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-9173851/v1/f82ea53426be7836d9b9abb0.png"},{"id":105904006,"identity":"ac6b63df-282f-4d61-bbf3-8aaa76fc04e8","added_by":"auto","created_at":"2026-04-01 10:00:16","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":168973,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003e(A)\u003c/strong\u003e is a SHAP scatter plot, showing the impact of different features on the model output. The horizontal axis represents the SHAP value (reflecting the degree of influence of features on the RF model output, with positive values promoting and negative values inhibiting), the vertical axis represents feature names, and the color of the points represents the level of feature values (red for high and blue for low). \u003cstrong\u003e(B)\u003c/strong\u003e is a bar chart of the mean absolute SHAP values for each feature. The horizontal axis represents the mean SHAP value (reflecting the average magnitude of the feature's impact on the model output), and the vertical axis represents feature names.\u003c/p\u003e","description":"","filename":"5.png","url":"https://assets-eu.researchsquare.com/files/rs-9173851/v1/13bcb67daeda462033b703e6.png"},{"id":105904489,"identity":"e97f065e-cf7d-42c7-ace1-a5770070aec1","added_by":"auto","created_at":"2026-04-01 10:08:58","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":51380,"visible":true,"origin":"","legend":"\u003cp\u003eSHAP force plot. The horizontal axis represents the predicted probability range, and the \"base value\" is the average predicted risk for all samples. The red segments push the predicted risk upward, while the blue segments pull the predicted risk downward.\u003c/p\u003e","description":"","filename":"6.png","url":"https://assets-eu.researchsquare.com/files/rs-9173851/v1/f63e41633602958b403b6159.png"},{"id":105904470,"identity":"7ddd18ba-f164-4363-a71e-4de3864ed39f","added_by":"auto","created_at":"2026-04-01 10:08:51","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":201170,"visible":true,"origin":"","legend":"\u003cp\u003eThe figure shows the distribution range of ROC curves from 1000 resamples and the mean ROC curve, with a mean AUC of 0.740±0.040 and a 95% confidence interval of [0.608, 0.848].\u003c/p\u003e","description":"","filename":"7.png","url":"https://assets-eu.researchsquare.com/files/rs-9173851/v1/15d8932ce7c5e4431a9df445.png"},{"id":105786393,"identity":"6c84c96e-8e06-4447-8a80-08519d22797c","added_by":"auto","created_at":"2026-03-31 06:45:33","extension":"png","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":114890,"visible":true,"origin":"","legend":"\u003cp\u003eThe receiver operating characteristic (ROC) curves of KL6, WBC, SII, RDW-SD, and MCV. The x-axis denotes the false positive rate (1-specificity), and the y-axis denotes the true positive rate (sensitivity); the dashed line represents the baseline for random guessing.\u003c/p\u003e","description":"","filename":"8.png","url":"https://assets-eu.researchsquare.com/files/rs-9173851/v1/147f53f7cba822c500962bb2.png"},{"id":107707261,"identity":"0a2f4051-38a8-4184-8378-b9fd8b17edcf","added_by":"auto","created_at":"2026-04-24 09:19:55","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2027089,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-9173851/v1/1c4a6593-db52-45b5-a6ed-a20d640e1277.pdf"},{"id":105904091,"identity":"cd64a0b6-9542-4dad-9f24-12150254a63a","added_by":"auto","created_at":"2026-04-01 10:03:58","extension":"xlsx","order_by":0,"title":"","display":"","copyAsset":false,"role":"supplement","size":64533,"visible":true,"origin":"","legend":"","description":"","filename":"Rawdataset.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-9173851/v1/602d052764e194f524fe4750.xlsx"},{"id":105904366,"identity":"839343c4-b5c5-41dd-a2c8-1faa5ba5383c","added_by":"auto","created_at":"2026-04-01 10:07:43","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":157158,"visible":true,"origin":"","legend":"","description":"","filename":"ExplanationofVariableSelectioninthePredictionModel.docx","url":"https://assets-eu.researchsquare.com/files/rs-9173851/v1/dc39c7f0c41dd899b5770d08.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Machine Learning Model Based on Routine Blood Indicators for High-Risk Screening of Interstitial Lung Disease in Patients with Connective Tissue Disease: A Cost-Effective Triage Strategy","fulltext":[{"header":"1 | Background","content":"\u003cp\u003eAs the most common and severe pulmonary complication of connective tissue disease (CTD), the early diagnosis of interstitial lung disease (ILD) is crucial for improving patient prognosis. The incidence of ILD across different CTD subtypes ranges from 15% to 65%[1]. It not only severely impairs pulmonary function, presenting with progressive dyspnea, cough and reduced exercise tolerance, but also doubles the risk of death in CTD patients, with the 5-year survival rate of severe cases dropping to 50% to 70[2]. Therefore, early identification of high-risk patients with ILD among CTD cases is crucial for the timely initiation of immunosuppressive or antifibrotic interventions, delaying disease progression and improving long-term prognosis.\u003c/p\u003e\n\u003cp\u003eCurrent clinical diagnostic methods for CTD-ILD all have certain limitations: HRCT is highly dependent on specialized equipment and costly, making it difficult to popularize in primary medical institutions; pulmonary function tests are susceptible to interference from factors such as patient compliance and concomitant airway diseases, leading to limited accuracy; invasive lung biopsy carries the risk of complications such as bleeding and infection, and has low patient acceptability, thus failing to serve as a routine screening method[3, 4]. Furthermore, the early symptoms of CTD-ILD are insidious, and most patients are diagnosed at the moderate to advanced stage, missing the optimal window for intervention[5]. Although serum KL-6 serves as a specific marker for pulmonary epithelial cell injury, with its levels positively correlated with the extent of HRCT-detected lesions and the degree of pulmonary function decline, and exhibits high sensitivity and specificity, its popularization in primary medical institutions is limited due to restrictive testing conditions[6\u0026ndash;8]. Therefore, the development of a convenient, cost-effective and non-invasive detection tool with both early diagnostic value and wide applicability has become an urgent clinical need.\u003c/p\u003e\n\u003cp\u003eRoutine blood test indicators and derived inflammatory and immune indices (e.g., neutrophil-to-lymphocyte ratio [NLR], systemic immune-inflammation index [SII], monocyte-to-lymphocyte ratio [MLR]) have attracted extensive attention in the risk prediction of various inflammatory diseases and their complications due to the advantages of convenient acquisition, low detection cost and high repeatability.\u003c/p\u003e\n\u003cp\u003eExisting studies have demonstrated that indicators such as NLR and SII can reflect the state of immune-inflammatory imbalance in the body, and immune-inflammatory dysregulation is one of the core pathological mechanisms underlying the onset of CTD and its complication with ILD[9]. NLR has also been shown to possess predictive value for ILD in subtypes including systemic sclerosis and primary Sj\u0026ouml;gren\u0026rsquo;s syndrome[10, 11].\u0026nbsp;Additionally, indicators such as PLR, LMR and NLR have been confirmed to have diagnostic value in rheumatoid arthritis-associated ILD[12]. However, most existing studies have only focused on the correlation between a single indicator and CTD-ILD, lacking a systematic analysis of the combined predictive value of multiple indicators. Moreover, traditional statistical methods struggle to fully explore the complex interactions among multiple indicators, resulting in limited predictive efficacy. In recent years, machine learning technology has been widely applied to the construction of disease risk prediction models due to its advantages in integrating multi-dimensional indicators and exploring complex interactions. For example, Zhang et al. used routine blood test-derived indices combined with a random forest model to predict all-cause mortality in arthritis patients with hypertension[13]; Xie et al. employed complete blood count-derived inflammatory indicators combined with LightGBM to predict pneumonia associated with acute ischemic stroke[14]. Based on this, we put forward the hypothesis: Can routine blood test indicators and their derived inflammatory and immune indices be combined with machine learning algorithms to construct a convenient, efficient and widely applicable early prediction tool for CTD-ILD, so as to provide scientific support for early clinical diagnosis?\u003c/p\u003e\n\u003cp\u003eIn this study, inpatients with CTD admitted to Shanxi Provincial People\u0026apos;s Hospital from May 2022 to May 2023 were enrolled as research subjects. We collected their gender, age, and a full set of routine blood test indicators (including derived indices such as NLR, SII and MLR), and constructed five machine learning prediction models, namely eXtreme Gradient Boosting (XGBoost), Logistic Regression (LR), Random Forest Classifier (RF), Support Vector Machine (SVM) and K-Nearest Neighbors (KNN). The model performance was comprehensively evaluated from the perspectives of discriminative ability (ROC curve, PR curve), calibration (calibration plot) and clinical utility (decision curve analysis, DCA). Additionally, the SHapley Additive exPlanations (SHAP) method was used to interpret the key predictive features and their impacts on model outputs. Ultimately, the optimal prediction model was screened out, which provides a scientific tool and clinical evidence for the early and convenient assessment of the risk of ILD complication in CTD patients.\u003c/p\u003e"},{"header":"2 | Methods","content":"\u003cp\u003e2.1 | Study Population and Variable Selection\u003c/p\u003e\n\u003cp\u003eIn this study, we recruited inpatients with CTD, including all subtypes such as systemic lupus erythematosus (SLE), rheumatoid arthritis (RA), systemic sclerosis (SSc), and Sj\u0026ouml;gren\u0026rsquo;s syndrome (SS), who were admitted to Shanxi Provincial People\u0026apos;s Hospital from May 2022 to May 2023. The enrolled participants were divided into two groups: the CTD-ILD group and the CTD group.\u003c/p\u003e\n\u003cp\u003eInclusion criteria included: a diagnosis of CTD based on current criteria[15\u0026ndash;22]. The diagnosis of ILD was based on the 2013 criteria for idiopathic interstitial pneumonia (IIP) from the American Thoracic Society (ATS)/European Respiratory Society (ERS) [23]. Subjects were excluded according to the following exclusion criteria: (a) presence of other pulmonary diseases such as COPD or asthma; (b) severe heart, liver or kidney disease; (c) concomitant pulmonary tuberculosis or malignant tumors; (d) hematological diseases; (e) multiple organ failure.\u003c/p\u003e\n\u003cp\u003eThis study was conducted in accordance with the principles of the Declaration of Helsinki and approved by the Medical Ethics Committee of Shanxi Provincial People\u0026apos;s Hospital (Ethical Approval No.: 2023-229). All participants provided written informed consent prior to their participation in the study.\u003c/p\u003e\n\u003cp\u003eThe patient characteristics of interest included gender, age, serum KL-6 (Krebs von den Lungen \u0026ndash; 6) levels, and a full panel of routine blood test parameters: NLR (Neutrophil-Lymphocyte Ratio), SII (Systemic Immune-Inflammation Index), MLR (Monocyte-Lymphocyte Ratio), PLR (Platelet-Lymphocyte Ratio), WBC (White Blood Cell Count), NEUT% (Neutrophil Percentage), NEUT (Neutrophil Count), LYMPH% (Lymphocyte Percentage), LYMPH (Lymphocyte Count), MONO% (Monocyte Percentage), MONO (Monocyte Count), EO% (Eosinophil Percentage), EO (Eosinophil Count), BASO% (Basophil Percentage), BASO (Basophil Count), RBC (Red Blood Cell Count), HB (Hemoglobin), HCT (Hematocrit), MCV (Mean Corpuscular Volume), MCH (Mean Corpuscular Hemoglobin), MCHC (Mean Corpuscular Hemoglobin Concentration), RDW-SD (Red Cell Distribution Width-Standard Deviation), RDW-CV (Red Cell Distribution Width-Coefficient of Variation), PLT (Platelet Count), PDW (Platelet Distribution Width), PCT (Plateletcrit), and MPV (Mean Platelet Volume).\u003c/p\u003e\n\u003cp\u003eLaboratory test data of all patients were collected or conducted prior to the current hospitalization. Demographic characteristics including age and gender were extracted from hospital medical records. In addition, 5 mL of fasting venous blood samples were collected from the elbow of each patient without any pre-processing. The blood samples were placed in tubes containing ethylenediaminetetraacetic acid (EDTA) anticoagulant for routine blood tests, and NLR, SII, PLR and MLR were calculated for each sample. The serum was then stored at -80 \u0026deg;C for further research. KL-6 was measured by latex agglutination assay using a biochemical analyzer (Beckman Coulter AU5800) with the kit provided by the manufacturer (Sekisui Medical Co., Ltd., Tokyo, Japan).\u003c/p\u003e\n\u003cp\u003eOur primary outcome variable was a binary variable indicating the presence or absence of ILD in CTD patients. A heatmap was used to visualize the results of Pearson correlation analysis and examine the relationships among variables\u003c/p\u003e\n\u003cp\u003e2.2 | Handling of Imbalanced Data\u003c/p\u003e\n\u003cp\u003eIn the dataset of this study, there was an imbalance in the sample size between CTD patients (n=140) and CTD-ILD patients (n=85). This class distribution discrepancy may cause the model to tend to optimize the prediction accuracy of the majority class (CTD) during training, given that samples of the majority class account for a higher proportion in the loss function. Ultimately, this may lead to a significant decline in the model\u0026apos;s ability to identify the minority class (CTD-ILD) (e.g., a low recall rate). As a high-risk population requiring early and accurate identification, an insufficient recall rate for CTD-ILD will directly undermine the clinical application value of the model (e.g., missed diagnosis of high-risk patients).\u003c/p\u003e\n\u003cp\u003eTo address this issue, we adopted the ADASYN (Adaptive Synthetic Sampling) method for data balancing. As an advanced oversampling technique, the core advantage of ADASYN lies in its adaptive synthesis of minority class samples: this method calculates the K-nearest neighbor (KNN) distribution density of minority class samples to generate more synthetic data for samples in sparsely distributed regions. This not only avoids the overfitting risk that may be caused by simple random oversampling (e.g., model memorization of noise due to duplicate samples) but also outperforms the traditional SMOTE method (SMOTE does not consider sample distribution density and may generate redundant samples in dense regions). ADASYN can more accurately simulate the real data distribution characteristics of the minority class (CTD-ILD) and enhance the model\u0026apos;s ability to learn complex decision boundaries[24].\u003c/p\u003e\n\u003cp\u003eIn selecting the ADASYN oversampling ratio, we comprehensively considered the degree of data imbalance and clinical requirements, and ultimately determined to test three key ratios: 50%, 75%, and 100%. The specific reasons are as follows: In the original data of this study, there were 140 samples in the majority class and 85 in the minority class, indicating a moderate imbalance. The 50% ratio (expanding the minority class to 50% of the majority class) represents mild oversampling, which retains partial characteristics of the original data distribution; the 75% ratio (expanding to 75% of the majority class) is moderate oversampling, which balances the class weights; the 100% ratio (complete balance) maximizes the weight of the minority class. These ratios cover the typical scenarios of mild-moderate-complete balance.\u003c/p\u003e\n\u003cp\u003eBased on this, we implemented the ADASYN technique with oversampling ratios of 50%, 75%, and 100% (Table 1), respectively, and randomly split the balanced datasets into training and test sets at a 7:3 ratio. By comparing the optimal performance of the models and the inter-model consistency under different ratios, we selected the most appropriate ratio among the three.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 1\u003c/strong\u003e Number of instances increased by ADASYN technique\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 192px;\"\u003e\n \u003cp\u003e\u003cstrong\u003ePercentage of\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003eADASYN increase\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 192px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eClass\u0026ldquo;CTD\u0026rdquo;\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003eactual 140\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 192px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eClass \u0026ldquo;CTD-ILD\u0026rdquo;\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003eactual 85\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 192px;\"\u003e\n \u003cp\u003e50%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 192px;\"\u003e\n \u003cp\u003e140\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 192px;\"\u003e\n \u003cp\u003e130\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 192px;\"\u003e\n \u003cp\u003e75%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 192px;\"\u003e\n \u003cp\u003e140\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 192px;\"\u003e\n \u003cp\u003e161\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 192px;\"\u003e\n \u003cp\u003e100%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 192px;\"\u003e\n \u003cp\u003e140\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 192px;\"\u003e\n \u003cp\u003e165\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e2.3 | Establishment and Evaluation of Prediction Models\u003c/p\u003e\n\u003cp\u003eFive machine learning algorithms were selected in this study to predict ILD complication in CTD patients, namely XGBoost, Logistic Regression, RF, SVM, and KNN. Ten-fold cross-validation (k=10) was adopted as the resampling method on the training set, with hyperparameters tuned via grid search. The validation set was used for model parameter adjustment, while the test set was applied to evaluate the overall model performance. The clinical value of the prediction models was assessed using three metrics of model quality: discrimination, calibration, and clinical usefulness. First, the discriminative ability of the models was quantified by precision-recall (PR) curve analysis. Second, calibration plots were used to evaluate model performance, assessing the calibration level and the deviation between model predictions and actual events. Third, decision curve analysis (DCA) was performed to evaluate clinical usefulness by calculating the net benefit at different threshold probabilities. In addition, confusion matrix metrics of the five models were also assessed, including average precision (AP), accuracy, sensitivity, specificity, and F1-score.\u003c/p\u003e\n\u003cp\u003e2.4 | Model Interpretation\u003c/p\u003e\n\u003cp\u003eWe used the SHAP method to interpret the models, which is a model interpretability approach derived from the Shapley Value in game theory. Its core objective is to quantify the contribution of each feature to the model\u0026apos;s prediction results, thereby addressing the interpretability issue of machine learning models[25].\u003c/p\u003e\n\u003cp\u003e2.5 | Model Validation\u003c/p\u003e\n\u003cp\u003eWe validated the optimal model using the Bootstrap method, a model validation approach based on resampling with replacement. Its core principle is to generate multiple bootstrap samples to simulate the distributions of different training/test sets, thereby quantitatively validating the model\u0026apos;s generalization ability and the stability of its evaluation metrics. This method compensates for the limitations of traditional validation approaches (e.g., hold-out method, simple cross-validation) in scenarios with small sample sizes and uneven sample distributions, and is particularly suitable for robust estimation of model performance and quantification of uncertainty [26].\u003c/p\u003e\n\u003cp\u003e2.6 | Diagnostic Performance Evaluation Method of KL-6\u003c/p\u003e\n\u003cp\u003eWith interstitial lung disease (ILD) complication as the outcome event, the diagnostic efficacy of each indicator was analyzed using receiver operating characteristic (ROC) curve analysis, and the area under the curve (AUC), sensitivity, specificity, Youden\u0026apos;s index, optimal cut-off value, and accuracy were calculated for each indicator.\u003c/p\u003e\n\u003cp\u003e2.7 | Statistical Analysis\u003c/p\u003e\n\u003cp\u003eAll statistical analyses were performed using R version 4.2.3 and Python version 3.11.4.\u003c/p\u003e"},{"header":"3 | Results","content":"\u003cp\u003e3.1 | Patient Characteristics\u003c/p\u003e\n\u003cp\u003eA total of 225 CTD patients were enrolled in this study, among whom 85 had ILD complications.Table2 presents the comparisons of patients\u0026apos; gender, age, and laboratory indicators. It can be seen that gender, NLR, SII, MLR, KL-6, WBC, NEUT, NEUT%, LYMPH%, MONO%, BASO%, HCT, MCV, MCH, RDW-SD and RDW-CV showed a statistically significant difference between the two groups with p-value \u0026lt; 0.05.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 2\u003c/strong\u003e Comparison of laboratory indicators between CTD-ILD and CTD patients.\u003c/p\u003e\n\u003ctable style=\"width: 100%;\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e\u003cstrong\u003eVariables\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e\u003cstrong\u003eALL(n=225)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" colspan=\"2\"\u003e\n \u003cp\u003e\u003cstrong\u003eCTD(n=140)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" colspan=\"2\"\u003e\n \u003cp\u003e\u003cstrong\u003eCTD-ILD( n=85)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" colspan=\"2\"\u003e\n \u003cp\u003ep\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eAge, Median [Q1-Q3]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e57.000 [41.000;66.000]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e49.500 [32.750;60.000]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e63.000 [58.000;71.000]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eNLR, Median [Q1-Q3]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e2.612 [1.625;4.385]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e2.130 [1.463;3.806]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e3.404 [2.107;5.914]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eSII, Median [Q1-Q3]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e573.875 [320.195;1063.584]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e500.263 [285.793;929.616]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e684.676 [371.724;1318.792]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e0.006\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eMLR, Median [Q1-Q3]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.298 [0.217;0.444]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e0.273 [0.211;0.385]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e0.369 [0.226;0.536]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e0.002\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003ePLR, Median [Q1-Q3]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e158.779 [113.245;221.569]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e156.620 [120.984;212.414]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e162.264 [102.721;222.581]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e0.955\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eKL6, Median [Q1-Q3]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e230.910 [156.930;586.190]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e182.710 [138.832;231.675]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e877.470 [431.700;1524.790]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eWBC, Median [Q1-Q3]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e5.830 [4.140;8.210]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e5.095 [3.775;7.245]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e7.340 [5.270;9.240]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eNEUT, Mean (SD)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e64.658 (13.484)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e62.431 (13.474)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e68.326 (12.748)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eNEUT%, Median [Q1-Q3]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e3.800 [2.310;5.540]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e3.105 [2.140;4.807]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e4.670 [3.280;6.770]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eLYMPH, Median [Q1-Q3]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e25.100 [16.900;33.500]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e28.350 [19.500;35.125]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e20.200 [13.400;29.000]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eLYMPH%, Median [Q1-Q3]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e1.350 [1.010;1.850]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e1.325 [1.045;1.843]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e1.380 [0.940;1.860]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e0.949\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eMONO, Median [Q1-Q3]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e7.400 [5.600;9.300]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e7.550 [5.600;9.450]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e7.300 [5.800;9.100]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e0.692\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eMONO%, Median [Q1-Q3]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.410 [0.290;0.590]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e0.380 [0.270;0.520]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e0.490 [0.360;0.690]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eEO, Median [Q1-Q3]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e1.200 [0.400;2.400]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e1.200 [0.400;2.425]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e1.400 [0.500;2.400]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e0.849\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eEO%, Median [Q1-Q3]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.060 [0.020;0.160]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e0.060 [0.020;0.120]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e0.090 [0.020;0.170]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e0.097\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eBASO, Median [Q1-Q3]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.300 [0.200;0.500]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e0.300 [0.200;0.500]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e0.300 [0.200;0.500]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e0.747\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eBASO%, Median [Q1-Q3]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.020 [0.010;0.030]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e0.010 [0.010;0.030]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e0.020 [0.010;0.040]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e0.022\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eRBC, Median [Q1-Q3]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e4.000 [3.630;4.310]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e4.000 [3.620;4.280]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e3.940 [3.670;4.430]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e0.548\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eHB, Median [Q1-Q3]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e118.000 [107.000;130.000]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e116.500 [106.000;127.000]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e122.000 [109.000;132.000]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e0.052\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eHCT, Median [Q1-Q3]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.364 [0.328;0.397]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e0.361 [0.326;0.392]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e0.368 [0.335;0.414]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e0.026\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eMCV, Median [Q1-Q3]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e92.300 [88.000;95.100]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e90.750 [87.375;94.325]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e93.500 [90.600;96.900]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eMCH, Median [Q1-Q3]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e29.700 [28.500;31.300]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e29.450 [28.400;30.825]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e30.100 [28.700;32.200]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e0.032\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eMCHC, Median [Q1-Q3]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e323.000 [315.000;331.000]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e325.000 [316.000;333.000]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e323.000 [315.000;330.000]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e0.209\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eRDW-SD, Median [Q1-Q3]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e46.100 [43.000;50.400]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e44.400 [42.000;48.100]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e48.400 [45.000;51.400]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eRDW-CV, Median [Q1-Q3]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e13.600 [12.900;14.900]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e13.400 [12.700;14.625]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e14.000 [13.200;15.100]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e0.003\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003ePLT, Median [Q1-Q3]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e216.000 [171.000;280.000]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e214.000 [171.750;276.500]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e220.000 [168.000;291.000]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e0.653\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003ePDW, Median [Q1-Q3]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e13.600 [10.700;16.000]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e13.400 [10.700;16.000]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e14.500 [10.075;16.025]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e0.912\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003ePCT, Median [Q1-Q3]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.220 [0.182;0.290]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e0.220 [0.184;0.280]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e0.220 [0.176;0.296]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e0.981\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eMPV, Median [Q1-Q3]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e10.100 [9.300;10.900]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e10.100 [9.400;10.800]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e9.900 [9.300;11.025]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e0.693\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eSexmale1, N (%):\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e0.007\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eFemale\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e167 (74.222%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e113 (80.714%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e54 (63.529%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eMale\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e58 (25.778%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e27 (19.286%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\n \u003cp\u003e31 (36.471%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e3.2 | Feature Analysis\u003c/p\u003e\n\u003cp\u003ePearson correlation analysis was performed to examine the relationships among the aforementioned variables with statistically significant intergroup differences, and a correlation heatmap was generated accordingly (Figure 1). From the routine blood test parameters, four indicators excluding KL-6 were selected as variables for the machine learning models based on their correlation with the disease and medical professional considerations, namely WBC, SII, MCV, and RDW-SD.\u003c/p\u003e\n\u003cp\u003eTable 3 shows the performance of machine learning classifiers on ADASYN-generated balanced datasets. Based on optimal model performance and inter-model consistency, the 75% ADASYN ratio achieved the best overall classification performance, ensuring minority class identification (F1-score) while retaining good discriminative ability (AUC) and overall accuracy.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 3\u003c/strong\u003e Evaluation of the performance of classification models on imbalance dataset using ADASYN technique in validation set\u003c/p\u003e\n\u003ctable style=\"width: 99%;\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e\u003cstrong\u003eADASYN\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e\u003cstrong\u003eAUC(95%CI)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e\u003cstrong\u003ecutoff(95%CI)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003eAccuracy(\u003cstrong\u003e95%CI)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003eSensitivity\u003cstrong\u003e(95%CI)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003eSpecificity\u003cstrong\u003e(95%CI)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003ePositive Predictive Value\u003cstrong\u003e(95%CI)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003eNegative Predictive Value \u003cstrong\u003e(95%CI)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003eF1 Score\u003cstrong\u003e(95%CI)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e\u003cstrong\u003eKappa(95%CI)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"3\"\u003e\n \u003cp\u003e\u003cstrong\u003eXGBoost\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e50%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.790 (0.642-0.937)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.843(0.799-0.886)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.684(0.629-0.739)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.438(0.379-0.497)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.864(0.823-0.905)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.700(0.645-0.755)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.679(0.623-0.735)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.538(0.479-0.597)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.317(0.156 -0.478)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e75%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.840 (0.697-0.982)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.830(0.788-0.872)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.762(0.714-0.810)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.741(0.692-0.790)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.800(0.755-0.845)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.870(0.832-0.908)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.632(0.578-0.686)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.800(0.755-0.845)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.510(0.337-0.683)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e100%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.739 (0.589-0.889)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.794(0.749-0.839)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.628(0.574-0.682)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.565(0.509-0.621)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.700(0.649-0.751)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.684(0.632-0.736)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.583(0.528-0.638)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.619(0.564-0.674)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.262(0.110-0.414)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"3\"\u003e\n \u003cp\u003e\u003cstrong\u003elogistic\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e50%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.824 (0.684-0.963)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.448(0.389-0.507)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.789(0.740-0.838)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e1.000(1.000-1.000)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.636(0.579-0.693)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.667(0.611-0.723)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e1.000(1.000-1.000)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.800(0.752-0.848)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.596(0.426-0.766)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e75%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.756 (0.578-0.933)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.437(0.381-0.493)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.762(0.714-0.810)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.889(0.854-0.924)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.533(0.477-0.589)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.774(0.727-0.821)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.727(0.677-0.777)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.828(0.785-0.871)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.449(0.277-0.621)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e100%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.748 (0.597-0.898)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.503(0.447-0.559)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.651(0.598-0.704)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.696(0.644-0.748)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.600(0.545-0.655)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.667(0.614-0.720)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.632(0.578-0.686)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.681(0.629-0.733)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.297(0.139-0.455)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"3\"\u003e\n \u003cp\u003e\u003cstrong\u003eRandomForest\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e50%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.908 (0.820-0.996)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.600(0.542-0.658)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.816(0.770-0.862)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.688(0.633-0.743)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.909(0.875-0.943)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.846(0.803-0.889)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.800(0.752-0.848)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.759(0.708-0.810)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.612(0.443-0.781)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e75%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.846 (0.725-0.967)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.550(0.494-0.606)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.786(0.740-0.832)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.889(0.854-0.924)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.600(0.545-0.655)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.800(0.755-0.845)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.750(0.701-0.799)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.842(0.801-0.883)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.512(0.339-0.685)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e100%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.668 (0.497-0.840)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.600(0.545-0.655)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.698(0.646-0.750)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.696(0.644-0.748)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.700(0.649-0.751)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.727(0.677-0.777)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.667(0.614-0.720)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.711(0.660-0.762)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.394(0.225-0.563)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"3\"\u003e\n \u003cp\u003e\u003cstrong\u003eSVM\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e50%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.878 (0.772-0.983)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.624(0.566-0.682)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.789(0.740-0.838)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.625(0.567-0.683)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.909(0.875-0.943)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.833(0.789-0.877)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.769(0.719-0.819)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.714(0.660-0.768)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.553(0.381-0.725)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e75%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.746 (0.552-0.939)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.485(0.429-0.541)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.762(0.714-0.810)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.852(0.812-0.892)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.600(0.545-0.655)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.793(0.747-0.839)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.692(0.640-0.744)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.821(0.778-0.864)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.466(0.293-0.639)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e100%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.698 (0.527-0.868)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.638(0.584-0.692)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.698(0.646-0.750)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.609(0.554-0.664)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.800(0.755-0.845)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.778(0.731-0.825)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.640(0.586-0.694)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.683(0.631-0.735)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.402(0.232-0.572)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"3\"\u003e\n \u003cp\u003e\u003cstrong\u003eKNN\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e50%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.835 (0.706-0.965)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.800(0.752-0.848)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.789(0.740-0.838)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.688(0.633-0.743)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.864(0.823-0.905)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.786(0.737-0.835)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.792(0.744-0.840)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.733(0.680-0.786)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.561(0.389-0.733)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e75%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.851 (0.717-0.985)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.600(0.545-0.655)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.762(0.714-0.810)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.852(0.812-0.892)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.600(0.545-0.655)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.793(0.747-0.839)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.692(0.640-0.744)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.821(0.778-0.864)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.466(0.293-0.639)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e100%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.732 (0.584-0.879)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.800(0.755-0.845)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.698(0.646-0.750)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.565(0.509-0.621)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.850(0.810-0.890)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.812(0.765-0.859)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.630(0.575-0.685)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.667(0.614-0.720)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e0.406(0.236-0.576)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n\u003cp\u003e3.3 | Model Development and Evaluation\u003c/p\u003e\n\u003cp\u003eFive machine learning algorithms, namely XGBoost, Logistic Regression, Random Forest (RF), Support Vector Machine (SVM), and K-Nearest Neighbor (KNN), were selected in this study to construct prediction models. Model development and training were completed on the training set, and a comprehensive performance evaluation was carried out on the validation and test sets from three dimensions: discrimination, calibration, and clinical usefulness. Specifically, this involved plotting the ROC curves for the training and validation sets, precision-recall (PR) curves and calibration plots for the validation set, and conducting decision curve analysis (DCA). Additionally, metrics associated with the confusion matrix were calculated to quantify the model performance.\u003c/p\u003e\n\u003cp\u003eAfter hyperparameter tuning and horizontal algorithm comparison, the average precision (AP) of all five models exceeded 0.75, confirming that all models possessed a fundamental predictive ability (Figure 2). Among them, the RF model exhibited the most outstanding performance across all core metrics: the AUC of the validation set reached 0.846, the average precision (AP) was 0.896, and the F1-score was as high as 0.842, demonstrating excellent discriminative ability and a well-balanced performance in identifying positive and negative samples. The results of the calibration plot showed that the RF model had the smallest deviation between the predicted probability and the actual risk of ILD occurrence, with a calibration error of only 0.146, which was significantly superior to other models (XGBoost: 0.149, Logistic Regression: 0.190, SVM: 0.169, KNN: 0.141), thus achieving the optimal calibration accuracy (Figure3 A).\u003c/p\u003e\n\u003cp\u003eDCA analysis revealed all models provided a net clinical benefit across all threshold values, with the RF model\u0026rsquo;s net benefit curve leading in most intervals. This confirms its superior clinical utility for assisting clinicians in screening high-risk patients and developing intervention strategies (Figure3 B).\u003c/p\u003e\n\u003cp\u003eROC curve analysis showed that the AUC of the RF model reached 1.000 in the training set with the curve close to the upper left corner, indicating an extremely strong classification ability; its AUC remained at a high level of 0.846 in the validation set, which was significantly superior to other models (XGBoost: 0.840, Logistic Regression: 0.756, SVM: 0.746, KNN: 0.851), demonstrating a stable generalization ability (Figure4).\u003c/p\u003e\n\u003cp\u003eTaken together, the RF model exhibited comprehensive advantages in discriminative ability (AUC: 0.846, AP: 0.896), calibration accuracy (calibration error: 0.146), balanced identification of positive and negative samples (F1-score: 0.842), and clinical decision-making value (leading net benefit in DCA), and was thus identified as the optimal prediction model in this study.\u003cv:shape id=\"_x0000_i1029\" type=\"#_x0000_t75\"\u003e\u0026nbsp;\u003cv:imagedata src=\"file:///C%3A/Users/btr8097/AppData/Local/Packages/oice_16_974fa576_32c1d314_17d4/AC/Temp/msohtmlclip1/01/clip_image006.png\" o:title=\"\"\u003e\u0026nbsp;\u003c/v:imagedata\u003e\u0026nbsp;\u003c/v:shape\u003e\n\u003c/p\u003e\n\u003cp\u003e3.4 | Model Interpretation\u003c/p\u003e\n\u003cp\u003eFigure 5 A presents the SHAP summary plot of the RF prediction model, which consists of four features ranked by their impact on ILD complication in CTD patients. A higher SHAP value of a feature indicates a higher risk of ILD complication in CTD patients. Red represents high feature values, purple denotes feature values close to the overall mean, and blue indicates low feature values.\u003c/p\u003e\n\u003cp\u003eFigure 5 B presents a bar plot of the mean absolute SHAP values for each feature, which quantifies the average impact of each feature on the model output. A longer blue bar indicates a greater average impact of the corresponding feature on predicting ILD occurrence in CTD patients.\u003c/p\u003e\n\u003cp\u003eFigure 6 provides an example for predicting the risk of ILD complication in CTD patients. In this case, the RF model predicted a risk of 0.90 for ILD complication (reference value: 1.00).\u003c/p\u003e\n\u003cp\u003e3.5 | Model Validation\u003c/p\u003e\n\u003cp\u003eTo further verify the stability and result reliability of the optimal RF model in predicting the risk of ILD in CTD patients, the Bootstrap resampling method (sampling ratio: 75%) was adopted for repeated validation, with a total of 1000 resampled samples generated and corresponding ROC curves constructed. The analysis results showed that the AUC values of the 1000 Bootstrap resamplings ranged from 0.608 to 0.848, with a mean AUC of 0.740\u0026plusmn;0.040 (Figure 7), indicating that the model had a small fluctuation in predictive performance across different sample subsets and its core discriminative ability exhibited good robustness.\u003c/p\u003e\n\u003cp\u003eCombined with the detailed statistical indicators from the Bootstrap validation (Table 4), all key performance parameters of the model exhibited a low degree of variation: the coefficient of variation (CV) for AUROC was 0.054 and that for the F1-score was 0.050, indicating that the model had strong stability in both its comprehensive classification performance and balanced ability to identify positive and negative samples. The mean accuracy (ACC) was 0.687\u0026plusmn;0.039 with a 95% confidence interval (CI) of [0.604, 0.758], and the balanced accuracy (balanced ACC) was 0.677\u0026plusmn;0.040 with a 95% CI of [0.599, 0.753]. These results reflect that the model had good consistency in identifying the two outcomes (ILD complication / no ILD complication) across different sample distributions, with no obvious class bias observed.\u003c/p\u003e\n\u003cp\u003eIn terms of core clinical application indicators, the model had a mean recall of 0.799\u0026plusmn;0.066 with a 95% CI of [0.653, 0.918], indicating a low risk of missed diagnosis for CTD-ILD high-risk patients\u0026mdash;a characteristic of great value in clinical screening scenarios. The positive predictive value (PPV) was 0.713\u0026plusmn;0.040 with a 95% CI of [0.636, 0.787], suggesting a high proportion of actual ILD complication among patients predicted as high-risk by the model, which helps reduce unnecessary further examinations. In contrast, the mean specificity was 0.556\u0026plusmn;0.075 with a CV of 0.134, representing the parameter with a relatively high degree of variation among all indicators. This is speculated to be associated with the distribution heterogeneity of ILD-positive cases in the sample; however, combined with the net clinical benefit revealed by the DCA analysis, this level of specificity still meets the basic requirements for clinical decision-making.\u003c/p\u003e\n\u003cp\u003eThe negative predictive value (NPV) was 0.485\u0026plusmn;0.039 with a 95% CI of [0.406, 0.555], indicating that a certain proportion of patients predicted as low-risk by the model still had actual ILD complications. In clinical application, a comprehensive judgment should be made by combining the specific clinical characteristics of patients to avoid missed diagnosis caused by over-reliance on the model results alone.\u003c/p\u003e\n\u003cp\u003eIn summary, the results of Bootstrap resampling validation demonstrated that all performance metrics of the optimal RF model possessed favorable stability and reliability. The narrow distribution of AUC values, low coefficient of variation, and reasonable confidence interval range further supported the clinical application value of this model in predicting the risk of ILD complication in CTD patients, and its predictive results can provide a robust evidence-based basis for clinicians to formulate screening strategies and intervention plans.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 4\u003c/strong\u003e Random Forest Model with Bootstrap Validation (1000 Resamples) - Core Performance Metrics\u003c/p\u003e\n\u003ctable style=\"width: 3.6e+2pt;border: none;\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e\u003cstrong\u003eMetric\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e\u003cstrong\u003eMean\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e\u003cstrong\u003eStd\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e\u003cstrong\u003eCi low\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e\u003cstrong\u003eCi high\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e\u003cstrong\u003eCV\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e\u003cstrong\u003eAUROC\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e0.740\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e0.040\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e0.660\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e0.816\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e0.054\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e\u003cstrong\u003eACC\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e0.687\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e0.039\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e0.604\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e0.758\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e0.057\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e\u003cstrong\u003ebalanced_ACC\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e0.677\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e0.040\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e0.599\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e0.753\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e0.059\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e\u003cstrong\u003erecall\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e0.799\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e0.066\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e0.653\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e0.918\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e0.082\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e\u003cstrong\u003especificity\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e0.556\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e0.075\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e0.405\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e0.714\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e0.134\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e\u003cstrong\u003eF1\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e0.732\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e0.037\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e0.653\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e0.800\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e0.050\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e\u003cstrong\u003ePPV\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e0.713\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e0.040\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e0.636\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e0.787\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e0.056\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e\u003cstrong\u003eNPV\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e0.485\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e0.039\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e0.406\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e0.555\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e0.079\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e3.6 | Diagnostic Performance of KL-6\u003c/p\u003e\n\u003cp\u003eTable 5 and Figure 8 show the diagnostic performance of the KL-6 index for high-risk screening of 225 CTD-ILD patients in this study: the AUC value reached 0.927 (95% confidence interval: 0.890\u0026ndash;0.958), with a sensitivity of 0.776, a specificity of 0.957 and a Youden\u0026apos;s index of 0.734. When 406.320 was used as the optimal cut-off value, the overall diagnostic accuracy of KL-6 for this disease reached 0.884 (95% confidence interval: 0.820\u0026ndash;0.936).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 5\u003c/strong\u003e Diagnostic performance statistics of KL-6 and 4 model variables for disease screening (N=225), including key parameters such as AUC value, sensitivity, specificity, and optimal threshold for each indicator.\u003c/p\u003e\n\u003ctable style=\"width: 4.9e+2pt;border: none;\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\" colspan=\"2\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e\u003cstrong\u003eN\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e\u003cstrong\u003eAUC value (95% CI)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e\u003cstrong\u003eSensitivity\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003e(95% CI)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e\u003cstrong\u003eSpecificity\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003e(95% CI)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e\u003cstrong\u003eYouden Index (95% CI)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e\u003cstrong\u003eOptimal Cut-off Value (95% CI)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e\u003cstrong\u003eAccuracy\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;(95% CI)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e\u003cstrong\u003eKL-6\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" colspan=\"2\"\u003e\n \u003cp\u003e225\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e0.927(0.890-0.958)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e0.776(0.710-0.921)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e0.957(0.772-0.989)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e0.734(0.638-0.837)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e406.320(233.817-432.680)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e0.884(0.820-0.936)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e\u003cstrong\u003eWBC\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" colspan=\"2\"\u003e\n \u003cp\u003e225\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e0.687(0.618-0.752)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e0.671(0.438-0.924)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e0.643(0.425-0.857)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e0.313(0.251-0.441)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e6.050(4.650-7.856)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e0.667(0.589-0.747)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e\u003cstrong\u003eSII\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" colspan=\"2\"\u003e\n \u003cp\u003e225\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e0.609(0.529-0.668)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e0.612(0.279-1.000)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e0.593(0.130-0.917)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e0.205(0.127-0.323)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e580.748(174.721-1531.001)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e0.601(0.440-0.696)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e\u003cstrong\u003eRDW-SD\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" colspan=\"2\"\u003e\n \u003cp\u003e225\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e0.683(0.620-0.748)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e0.776(0.493-0.872)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e0.550(0.448-0.800)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e0.326(0.232-0.448)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e45.000(43.938-48.300)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e0.666(0.589-0.727)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e\u003cstrong\u003eMCV\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\" colspan=\"2\"\u003e\n \u003cp\u003e225\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e0.640(0.556-0.705)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e0.765(0.313-0.851)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e0.493(0.413-0.906)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e0.258(0.167-0.390)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e90.600(90.200-96.200)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd nowrap=\"\"\u003e\n \u003cp\u003e0.621(0.546-0.691)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e"},{"header":"4 | Discussion","content":"\u003cp\u003eIn this study, a machine learning model was constructed and validated based on routine blood test indicators, aiming to provide a convenient tool for predicting the risk of ILD complication in CTD patients. Through the comparison of clinical data from 225 CTD patients, SII, WBC, MCV and RDW-SD were finally identified as the core variables for CTD-ILD prediction after feature screening. Five prediction models were established based on the four core variables, among which the Random Forest (RF) model exhibited the most prominent performance, with an AUC of 0.846, an average precision (AP) of 0.896, a sensitivity of 0.889, a specificity of 0.600, an F1-score of 0.842 in the validation set and a calibration error of only 0.146; Decision Curve Analysis (DCA) confirmed that it yielded a net clinical benefit across the entire threshold range. SHAP analysis revealed the mechanism of action of each core variable. Finally, after 1000 times of Bootstrap resampling validation, the RF model had a mean AUC of 0.740\u0026thinsp;\u0026plusmn;\u0026thinsp;0.040 with a 95% confidence interval of [0.608, 0.848], and all performance metrics showed low coefficients of variation, demonstrating favorable stability and reliability.\u003c/p\u003e \u003cp\u003eIn the current diagnosis of CTD-ILD, various detection methods have obvious limitations: HRCT involves expensive equipment and ionizing radiation, has low popularization at the primary care level, and is not suitable for large-scale screening; pulmonary function tests are susceptible to interference from patient compliance, leading to limited accuracy; invasive lung biopsy carries the risk of complications and has low patient acceptability. None of these methods can be used as a routine screening tool[\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e, \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]. As a specific marker of type Ⅱ alveolar epithelial cells, KL-6 can directly reflect the degree of alveolar epithelial injury and is a clinically recognized indicator for definitive diagnosis and disease assessment[\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e, \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e]. In the high-risk screening of 225 CTD-ILD patients in this study, KL-6 exhibited prominent advantages in accurate diagnosis with an AUC of 0.927, a specificity of 0.957 and an accuracy of 0.884, yet it is limited by the reliance on specific detection equipment, low availability at the primary care level and relatively high testing costs. Previous diagnostic studies on routine blood test indicators have mostly focused on the predictive value of individual markers. Several studies have shown that patients with CTD-ILD often present with an elevated total white blood cell count (WBC) or an increased neutrophil ratio (NEUT%)[\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e, \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e]. Other studies have found that the RDW level in patients with CTD-ILD is typically higher than that in patients with CTD alone, and a higher RDW value is associated with poorer pulmonary function and lower overall survival, suggesting that it can serve as a potential predictive marker for disease severity and prognosis[\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e, \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e]. Other studies have also developed RF prediction models based on serum markers such as KL-6 and SP-D, which achieved an 85% prediction accuracy on unseen test data[\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e]. However, they are also difficult to apply for large-scale screening due to high testing costs and low availability at the primary care level.\u003c/p\u003e \u003cp\u003eCompared with the aforementioned existing studies, the prediction model in this study was developed based on routine blood test indicators, which not only offers core advantages of convenient detection, low cost, and high accessibility at the primary care level. Although the AUC (0.846) and specificity (0.600) of the model in the validation set are lower than those of KL-6 (AUC\u0026thinsp;=\u0026thinsp;0.927, specificity\u0026thinsp;=\u0026thinsp;0.957), it exhibits a superior sensitivity in comparison (0.889 for the model vs. 0.776 for KL-6), thus demonstrating more prominent efficacy in reducing the risk of missed diagnosis. Additionally, this model integrates the complex interactive effects of four core indicators (SII, WBC, MCV, RDW-SD), outputs individualized risk probabilities, and supports clinical stratified screening, making it more suitable for large-scale preliminary screening and scenarios with limited medical resources.\u003c/p\u003e \u003cp\u003eThe four core indicators of the model established in this study cover the key pathophysiological links of CTD-ILD from different dimensions and have clear detection significance, yet the diagnostic efficacy of each individual indicator is relatively limited, making it difficult to accurately distinguish between CTD-ILD patients and those with CTD alone. This is consistent with the clinical pathological mechanism: the pathogenesis of CTD-ILD results from the combined action of multiple factors such as immune-inflammatory imbalance, tissue injury, and hematopoietic dysfunction. A single indicator can only reflect abnormalities in one of these links and cannot fully capture the complex pathological characteristics of the disease, thus having limited predictive value when applied alone. Given the limitations of individual indicators, this study comprehensively integrated the four core indicators using machine learning algorithms, and the constructed RF model achieved a significant improvement in diagnostic efficacy. Leveraging the advantages of the RF algorithm, the model fully explores the complex interactive effects among the indicators and realizes multi-dimensional integration of the systemic inflammatory level reflected by WBC, the immune-inflammatory imbalance state represented by SII, the hematopoietic function changes associated with MCV, and the red blood cell heterogeneity embodied by RDW-SD.\u003c/p\u003e \u003cp\u003eSHAP analysis revealed that the mean absolute SHAP value of RDW-SD ranked first among the four core variables, making it the key feature driving model prediction. In the study sample, the median RDW-SD in the CTD-ILD group (48.400 fL) was significantly higher than that in the CTD-only group (44.400 fL), with a statistically significant intergroup difference (p\u0026thinsp;\u0026lt;\u0026thinsp;0.001). This result is consistent with the conclusion of previous studies that abnormally elevated RDW is associated with the development of CTD-ILD \u003csup\u003e32\u003c/sup\u003e. From a pathophysiological perspective, patients with ILD exhibit diffuse inflammatory and fibrotic lesions in the pulmonary interstitium, which directly impairs the integrity of the alveolar-capillary barrier, reduces gas exchange efficiency, and induces a chronic, occult state of mild hypoxia. This persistent hypoxia stimulates alterations in the bone marrow hematopoietic microenvironment, leading to heterogeneous maturation during erythropoiesis and thereby increasing red blood cell size heterogeneity. Additionally, the abnormally activated immune-inflammatory response in patients with CTD releases large amounts of reactive oxygen species, inflammatory cytokines and other substances, which directly damage the lipid and protein structures of the red blood cell membrane, impair the morphological stability of red blood cells, and further exacerbate the variation in red blood cell volume distribution[\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e, \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e]. Furthermore, an elevation in RDW-SD may also indirectly reflect disease activity and the status of pulmonary function impairment. Existing studies have shown that the RDW level in CTD-ILD patients is negatively correlated with pulmonary function indices (e.g., forced vital capacity, diffusing capacity), and a higher RDW-SD indicates a more severe degree of pulmonary interstitial fibrosis and a more obvious gas exchange disorder\u003csup\u003e32\u003c/sup\u003e. This characteristic enables RDW-SD to serve not only as a risk prediction indicator for CTD-ILD but also as a reference for assessing disease severity, further highlighting its application value in clinical practice.\u003c/p\u003e \u003cp\u003eAs a classic indicator reflecting the activity of systemic inflammation, the median white blood cell count (WBC) in the CTD-ILD group (7.340\u0026times;10⁹/L) was higher than that in the CTD-only group (5.095\u0026times;10⁹/L). Activated leukocytes (especially neutrophils and monocytes) accumulate in the pulmonary interstitium, release inflammatory mediators, and participate in fibroblast activation, thereby driving the progression of pulmonary fibrosis[\u003cspan additionalcitationids=\"CR34\" citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e]. he systemic immune-inflammation index (SII) integrates platelet, neutrophil and lymphocyte counts, and can comprehensively reflect the degree of immune-inflammatory imbalance[\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e]. In this study, the median SII in the CTD-ILD group (684.676) was significantly higher than that in the CTD-only group (500.263), indicating that the enhanced inflammation mediated by neutrophils and lymphocyte suppression jointly promote the development of ILD. Neutrophil-released elastase damages alveolar epithelial cells, while lymphocyte reduction impairs the ability to regulate inflammation[\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e, \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e]. Mean corpuscular volume (MCV) reflects the average size of red blood cells. In this study, the median MCV in the CTD-ILD group (93.500 fL) was higher than that in the CTD-only group (90.750 fL), which may be associated with erythropoietic disorders caused by chronic inflammation or metabolic disturbances of vitamin B₁₂ and folic acid. Additionally, studies have suggested an association between MCV and hypoxic status in patients with pulmonary fibrosis, enabling it to indirectly reflect pulmonary function impairment[\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e, \u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eThis model features a clear and easy-to-implement application pathway in real clinical settings, which can be seamlessly integrated into the routine diagnosis and treatment process for CTD patients, and is particularly suitable for primary medical institutions and large-scale screening needs. In clinical practice, for all CTD patients attending the clinic, after routine blood tests (already widely used as a basic examination), data on the four core indicators (WBC, SII, MCV, RDW-SD) can be directly extracted and input into this RF model to quickly generate individualized probability of ILD onset. For high-risk patients, it is recommended to prioritize further confirmatory examinations, such as serum KL-6 testing combined with HRCT imaging assessment, to clarify the presence of ILD and the scope of lesions. Meanwhile, immunosuppressive or anti-fibrotic intervention regimens should be formulated as early as possible, and routine blood tests and pulmonary function tests should be rechecked every 3\u0026ndash;6 months to dynamically monitor changes in risk. For moderate-risk patients, the primary disease treatment can be strengthened first to control inflammatory activity; routine blood tests should be rechecked every 6\u0026ndash;12 months and the model risk should be reassessed, and confirmatory examinations should be initiated promptly if the risk escalates. For low-risk patients, routine follow-up for the primary disease is sufficient, and model reassessment should be conducted during annual rechecks to avoid missed diagnosis of occult ILD. This \"routine testing\u0026thinsp;+\u0026thinsp;model preliminary screening\u0026thinsp;+\u0026thinsp;stratified confirmation\" model not only avoids imposing additional testing burdens on patients (relying on existing routine blood test data) but also greatly improves the screening efficiency for high-risk populations of ILD and reduces the waste of resources such as unnecessary HRCT and KL-6 tests. At the same time, its high sensitivity (0.889) effectively reduces the risk of missed diagnosis, ensures that the opportunity for early intervention is not lost, and achieves an optimal balance between clinical benefits and medical costs.\u003c/p\u003e \u003cp\u003eWhen constructing this model, patient basic information other than gender and age, as well as patients\u0026rsquo; chief complaints and physical examination findings, were not included as variables, because the model focuses on application scenarios in primary medical institutions and large-scale screening. The core advantage of the model design is its rapid operation relying on routine blood test indicators with high popularity and objective detection, avoiding the operational burden, time cost and labor consumption caused by additional collection of complex clinical information, which is in line with the efficiency-first demand of screening scenarios. Core routine blood test indicators such as SII, WBC, MCV and RDW-SD can fully cover the pathophysiological mechanisms related to the onset of CTD-ILD, such as immune-inflammatory imbalance and tissue injury, enabling efficient and convenient risk prediction without relying on other basic information, chief complaints or physical examination findings.\u003c/p\u003e \u003cp\u003eThis study also has several limitations. First, it adopted a single-center retrospective design with a relatively limited sample size, which may lead to selection bias. Additionally, the concentrated distribution of patient regions, ethnicities, and disease subtypes in a single center reduces the external generalizability of the model. Second, there is a lack of multicenter external validation. Given the variations in medical standards, detection equipment, and patient population characteristics across different centers, the application effectiveness of the model in other institutions has not been verified, which restricts its nationwide promotion. Third, stratified analysis by CTD disease subtypes was not performed. The pathogenesis and risk factors for ILD comorbidity may differ among subtypes such as systemic lupus erythematosus, rheumatoid arthritis, and systemic sclerosis, and the predictive efficacy of the model for patients with a single subtype needs further verification. Fourth, the model has a relatively low specificity, with a mean value of 0.556\u0026thinsp;\u0026plusmn;\u0026thinsp;0.075 and a coefficient of variation of 0.134, which may result in a certain number of false positive results. Finally, long-term follow-up data of patients were not collected, making it impossible to evaluate the long-term predictive value of the model for the onset of CTD-ILD and to verify the correlation between the model's predictive results and patient prognosis.\u003c/p\u003e \u003cp\u003eTo address the limitations of this study, future work may be carried out in the following aspects: conducting multicenter prospective studies to enroll CTD patients from medical institutions in different regions and at different levels, covering more disease subtypes and population characteristics; further expanding the sample size, performing stratified analysis by CTD disease subtypes, and constructing subtype-specific prediction models to improve the predictive efficacy for patients with a single subtype.\u003c/p\u003e \u003cp\u003eIn summary, the RF model constructed in this study based on routine blood test indicators has demonstrated favorable discriminative ability, calibration, and clinical practicality in predicting the risk of CTD-ILD. By comparing with existing clinical testing methods, it highlights its unique advantages in popularity, cost, and timeliness. It can serve as a convenient tool for the preliminary screening of high-risk populations for ILD in CTD patients, and is particularly suitable for primary medical institutions and large-scale screening scenarios.\u003c/p\u003e"},{"header":"5 | Conclusions","content":"\u003cp\u003eThe random forest model based on routine blood test indicators (SII, WBC, MCV, RDW-SD) constructed in this study has good discriminative ability, calibration and clinical utility, and the model is stable and reliable with high sensitivity. Relying on the advantages of routine blood test indicators such as wide popularity, low cost and strong timeliness, the model can serve as a convenient and effective preliminary screening tool for CTD patients with high risk of ILD, which is of great significance for improving the early identification rate of CTD-ILD, perfecting the clinical screening system of CTD-ILD, and guiding the rational allocation of medical resources. Although the model has certain limitations such as low specificity and single-center research design, the research results still provide important clinical evidence and scientific ideas for the early screening of CTD-ILD, and lay a foundation for the subsequent multi-center prospective research and model optimization.\u003c/p\u003e\n\u003cp\u003e\u003cbr\u003e\u003c/p\u003e"},{"header":"Abbreviations","content":"\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003eAbbreviation\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003e\u003cstrong\u003eFull Name\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eCTD\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eConnective Tissue Disease\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eILD\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eInterstitial Lung Disease\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eCTD-ILD\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eConnective Tissue Disease-Associated Interstitial Lung Disease\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eSII\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eSystemic Immune-Inflammation Index\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eWBC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eWhite Blood Cell Count\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eMCV\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eMean Corpuscular Volume\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eRDW-SD\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eRed Blood Cell Distribution Width-Standard Deviation\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eNLR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eNeutrophil-Lymphocyte Ratio\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eMLR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eMonocyte-Lymphocyte Ratio\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003ePLR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003ePlatelet-Lymphocyte Ratio\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eKL-6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eKrebs von den Lungen \u0026ndash; 6\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eADASYN\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eAdaptive Synthetic Sampling\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eRF\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eRandom Forest\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eXGBoost\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eeXtreme Gradient Boosting\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eSVM\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eSupport Vector Machine\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eKNN\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003ek-Nearest Neighbor\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eAUC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eArea Under the Curve\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eAP\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eAverage Precision\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eDCA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eDecision Curve Analysis\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eSHAP\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eSHapley Additive exPlanations\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eHRCT\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eHigh-Resolution Computed Tomography\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eCOPD\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eChronic Obstructive Pulmonary Disease\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eSLE\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eSystemic Lupus Erythematosus\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eRA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eRheumatoid Arthritis\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eSSc\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eSystemic Sclerosis\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eSS\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eSj\u0026ouml;gren\u0026rsquo;s Syndrome\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eIIP\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eIdiopathic Interstitial Pneumonia\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eATS\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eAmerican Thoracic Society\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eERS\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eEuropean Respiratory Society\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eEDTA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eEthylenediaminetetraacetic Acid\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eCV\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eCoefficient of Variation\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eCI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eConfidence Interval\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eACC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eAccuracy\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003ePPV\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003ePositive Predictive Value\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eNPV\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eNegative Predictive Value\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eTIMPs\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eTissue Inhibitors of Matrix Metalloproteinases\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eMMPs\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eMatrix Metalloproteinases\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eTGF-\u0026beta;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eTransforming Growth Factor-\u0026beta;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003ePDGF\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003ePlatelet-Derived Growth Factor\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eIFN-\u0026gamma;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eInterferon-\u0026gamma;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eIL\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eInterleukin\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eTNF-\u0026alpha;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eTumor Necrosis Factor-\u0026alpha;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eHIF-1\u0026alpha;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eHypoxia-Inducible Factor-1\u0026alpha;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eRBC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eRed Blood Cell Count\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eHB\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eHemoglobin\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eHCT\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eHematocrit\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eMCH\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eMean Corpuscular Hemoglobin\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eMCHC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eMean Corpuscular Hemoglobin Concentration\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eRDW-CV\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eRed Blood Cell Distribution Width-Coefficient of Variation\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003ePLT\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003ePlatelet Count\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003ePDW\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003ePlatelet Distribution Width\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003ePCT\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003ePlateletcrit\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eMPV\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eMean Platelet Volume\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eNEUT\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eNeutrophil Count\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eNEUT%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eNeutrophil Percentage\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eLYMPH\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eLymphocyte Count\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eLYMPH%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eLymphocyte Percentage\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eMONO\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eMonocyte Count\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eMONO%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eMonocyte Percentage\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eEO\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eEosinophil Count\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eEO%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eEosinophil Percentage\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eBASO\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eBasophil Count\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003eBASO%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd\u003e\n \u003cp\u003eBasophil Percentage\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e"},{"header":"Declarations","content":"\u003cp\u003eEthics approval and consent to participate\u003c/p\u003e\n\u003cp\u003eThe Medical Ethics Committee of Shanxi provincial people\u0026rsquo;s hospital approved the study (2023-229).\u003c/p\u003e\n\u003cp\u003eConsent for publication\u003c/p\u003e\n\u003cp\u003eNot applicable\u003c/p\u003e\n\u003cp\u003eAvailability of data and materials\u003c/p\u003e\n\u003cp\u003eAll\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003edata generated or analysed during this study are included in this published article [and its supplementary information files]\u003c/p\u003e\n\u003cp\u003eCompeting interests\u003c/p\u003e\n\u003cp\u003eThe authors declare that they have no competing interests\u003c/p\u003e\n\u003cp\u003eFunding\u003c/p\u003e\n\u003cp\u003eThe authors received no funding for this work.\u003c/p\u003e\n\u003cp\u003eAuthors\u0026apos; contributions\u003c/p\u003e\n\u003cp\u003eWang Haoran and Zhang Lele are co-first authors. Wang Haoran was responsible for designing the research plan, performing most of the experiments, collecting and analyzing experimental data, and drafting the initial manuscript. Zhang Lele assisted in experimental operation, data collation and manuscript revision. Xing Huifang, Yang Dong and Liu Chenshen participated in part of the experimental operation, sample collection and data verification. As the corresponding author, Liang Hongping was responsible for supervising the entire research process, providing guidance on experimental design and data analysis, critically revising the important academic content of the manuscript, and reviewing and finalizing the manuscript for submission\u003c/p\u003e\n\u003cp\u003eAcknowledgements\u003c/p\u003e\n\u003cp\u003eNot applicable\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eJoy GM, Arbiv OA, Wong CK, Lok SD, Adderley NA, Dobosz KM, et al. Prevalence, imaging patterns and risk factors of interstitial lung disease in connective tissue disease: a systematic review and meta-analysis. Eur Respir Rev. 2023;32:220210. https://doi.org/10.1183/16000617.0210-2022.\u003c/li\u003e\n\u003cli\u003eJaeger VK, Wirz EG, Allanore Y, Rossbach P, Riemekasten G, Hachulla E, et al. Incidences and Risk Factors of Organ Manifestations in the Early Course of Systemic Sclerosis: A Longitudinal EUSTAR Study. PLoS ONE. 2016;11:e0163894. https://doi.org/10.1371/journal.pone.0163894.\u003c/li\u003e\n\u003cli\u003eHerzog EL, Mathur A, Tager AM, Feghali-Bostwick C, Schneider F, Varga J. Review: Interstitial Lung Disease Associated With Systemic Sclerosis and Idiopathic Pulmonary Fibrosis: How Similar and Distinct? Arthritis \u0026amp; Rheumatology. 2014;66:1967\u0026ndash;78. https://doi.org/10.1002/art.38702.\u003c/li\u003e\n\u003cli\u003eFrauenfelder T, Winklehner A, Nguyen TDL, Dobrota R, Baumueller S, Maurer B, et al. Screening for interstitial lung disease in systemic sclerosis: performance of high-resolution CT with limited number of slices: a prospective study. Annals of the Rheumatic Diseases. 2014;73:2069\u0026ndash;73. https://doi.org/10.1136/annrheumdis-2014-205637.\u003c/li\u003e\n\u003cli\u003eHernandez-Gonzalez F, Prieto-Gonz\u0026aacute;lez S, Brito-Zeron P, Cuerpo S, Sanchez M, Ramirez J, et al. Impact of a systematic evaluation of connective tissue disease on diagnosis approach in patients with interstitial lung diseases. Medicine. 2020;99:e18589. https://doi.org/10.1097/MD.0000000000018589.\u003c/li\u003e\n\u003cli\u003eYamakawa H, Hagiwara E, Kitamura H, Yamanaka Y, Ikeda S, Sekine A, et al. Serum KL-6 and surfactant protein-D as monitoring and predictive markers of interstitial lung disease in patients with systemic sclerosis and mixed connective tissue disease. J Thorac Dis. 2017;9:362\u0026ndash;71. https://doi.org/10.21037/jtd.2017.02.48.\u003c/li\u003e\n\u003cli\u003eLee JS, Lee EY, Ha Y-J, Kang EH, Lee YJ, Song YW. Serum KL-6 levels reflect the severity of interstitial lung disease associated with connective tissue disease. Arthritis Res Ther. 2019;21:58. https://doi.org/10.1186/s13075-019-1835-9.\u003c/li\u003e\n\u003cli\u003eXing H, Liang H. The clinical value of KL-6 for predicting the occurrence and severity of connective tissue disease-associated interstitial lung disease is not affected by CTD type or treatment. PeerJ. 2024;12:e17792. https://doi.org/10.7717/peerj.17792.\u003c/li\u003e\n\u003cli\u003eRuta VM, Man AM, Alexescu TG, Motoc NS, Tarmure S, Ungur RA, et al. Neutrophil-to-Lymphocyte Ratio and Systemic Immune-Inflammation Index\u0026mdash;Biomarkers in Interstitial Lung Disease. Medicina. 2020;56:381. https://doi.org/10.3390/medicina56080381.\u003c/li\u003e\n\u003cli\u003eZhang K, Li W, He C, He X, Hou J. The ratio of neutrophil to lymphocyte predicts interstitial lung disease and its prognosis in patients with primary Sj\u0026ouml;gren\u0026rsquo;s syndrome: a retrospective analysis. Ann Palliat Med. 2021;10:6493\u0026ndash;501. https://doi.org/10.21037/apm-21-1043.\u003c/li\u003e\n\u003cli\u003eAti̇Lla N, Yildirim \u0026Ccedil;eti̇N G, Balkarli A. Association of neutrophil/lymphocyte ratio with the degree ofinterstitial lung disease in systemic sclerosis. Turk J Med Sci. 2016;46:1871\u0026ndash;4. https://doi.org/10.3906/sag-1601-87.\u003c/li\u003e\n\u003cli\u003eChen Q, Chen D, Xu X, Liu Y, Yin T, Li D. Platelet/Lymphocyte, Lymphocyte/Monocyte, and Neutrophil/Lymphocyte Ratios as Biomarkers in Patients with Rheumatoid Arthritis and Rheumatoid Arthritis-Associated Interstitial Lung Disease. Med Sci Monit. 2019;25:6474\u0026ndash;81. https://doi.org/10.12659/MSM.916583.\u003c/li\u003e\n\u003cli\u003eK Z, X M, X Z, G Q, C Z. Machine learning based association between inflammation indicators (NLR, PLR, NPAR, SII, SIRI, and AISI) and all-cause mortality in arthritis patients with hypertension: NHANES 1999-2018. Frontiers in public health. 2025;13. https://doi.org/10.3389/fpubh.2025.1559603.\u003c/li\u003e\n\u003cli\u003eXie M, Liu Z, Dai F, Cao Z, Wang X. Predicting Stroke-Associated Pneumonia in Acute Ischemic Stroke: A Machine Learning Model Development and Validation Study with CBC-Derived Inflammatory Indices. Int J Gen Med. 2025;18:3117\u0026ndash;28. https://doi.org/10.2147/IJGM.S524450.\u003c/li\u003e\n\u003cli\u003eAggarwal R, Rider LG, Ruperto N, Bayat N, Erman B, Feldman BM, et al. 2016 American College of Rheumatology/European League Against Rheumatism criteria for minimal, moderate, and major clinical response in adult dermatomyositis and polymyositis: An International Myositis Assessment and Clinical Studies Group/Paediatric Rheumatology International Trials Organisation Collaborative Initiative. Annals of the Rheumatic Diseases. 2017;76:792\u0026ndash;801. https://doi.org/10.1136/annrheumdis-2017-211400.\u003c/li\u003e\n\u003cli\u003eAlarc\u0026oacute;n-Segovia D, Cardiel MH. Comparison between 3 diagnostic criteria for mixed connective tissue disease. Study of 593 patients. J Rheumatol. 1989;16:328\u0026ndash;34.\u003c/li\u003e\n\u003cli\u003eAletaha D, Neogi T, Silman AJ, Funovits J, Felson DT, Bingham III CO, et al. 2010 Rheumatoid arthritis classification criteria: An American College of Rheumatology/European League Against Rheumatism collaborative initiative. Arthritis \u0026amp; Rheumatism. 2010;62:2569\u0026ndash;81. https://doi.org/10.1002/art.27584.\u003c/li\u003e\n\u003cli\u003eAringer M, Costenbader K, Daikh D, Brinks R, Mosca M, Ramsey-Goldman R, et al. 2019 European League Against Rheumatism/American College of Rheumatology classification criteria for systemic lupus erythematosus. Ann Rheum Dis. 2019;78:1151\u0026ndash;9. https://doi.org/10.1136/annrheumdis-2018-214819.\u003c/li\u003e\n\u003cli\u003eLeavitt RY, Fauci AS, Bloch DA, Michel BA, Hunder GG, Arend WP, et al. The American College of Rheumatology 1990 criteria for the classification of Wegener\u0026rsquo;s granulomatosis. Arthritis Rheum. 1990;33:1101\u0026ndash;7. https://doi.org/10.1002/art.1780330807.\u003c/li\u003e\n\u003cli\u003eMosca M, Neri R, Bombardieri S. Undifferentiated connective tissue diseases (UCTD): a review of the literature and a proposal for preliminary classification criteria. Clin Exp Rheumatol. 1999;17:615\u0026ndash;20.\u003c/li\u003e\n\u003cli\u003eShiboski SC, Shiboski CH, Criswell LA, Baer AN, Challacombe S, Lanfranchi H, et al. American College of Rheumatology classification criteria for Sj\u0026ouml;gren\u0026rsquo;s syndrome: a data-driven, expert consensus approach in the Sj\u0026ouml;gren\u0026rsquo;s International Collaborative Clinical Alliance cohort. Arthritis Care Res (Hoboken). 2012;64:475\u0026ndash;87. https://doi.org/10.1002/acr.21591.\u003c/li\u003e\n\u003cli\u003evan den Hoogen F, Khanna D, Fransen J, Johnson SR, Baron M, Tyndall A, et al. 2013 classification criteria for systemic sclerosis: an American college of rheumatology/European league against rheumatism collaborative initiative. Ann Rheum Dis. 2013;72:1747\u0026ndash;55. https://doi.org/10.1136/annrheumdis-2013-204424.\u003c/li\u003e\n\u003cli\u003eTravis WD, Costabel U, Hansell DM, King TE, Lynch DA, Nicholson AG, et al. An official American Thoracic Society/European Respiratory Society statement: Update of the international multidisciplinary classification of the idiopathic interstitial pneumonias. Am J Respir Crit Care Med. 2013;188:733\u0026ndash;48. https://doi.org/10.1164/rccm.201308-1483ST.\u003c/li\u003e\n\u003cli\u003eHaibo He, Yang Bai, Garcia EA, Shutao Li. ADASYN: Adaptive synthetic sampling approach for imbalanced learning. In: 2008 IEEE International Joint Conference on Neural Networks (IEEE World Congress on Computational Intelligence). Hong Kong, China: IEEE; 2008. p. 1322\u0026ndash;8. https://doi.org/10.1109/IJCNN.2008.4633969.\u003c/li\u003e\n\u003cli\u003eChen H, Lundberg SM, Lee S-I. Explaining a series of models by propagating Shapley values. Nat Commun. 2022;13:4512. https://doi.org/10.1038/s41467-022-31384-3.\u003c/li\u003e\n\u003cli\u003eHenderson AR. The bootstrap: A technique for data-driven statistics. Using computer-intensive analyses to explore experimental data. Clinica Chimica Acta. 2005;359:1\u0026ndash;26. https://doi.org/10.1016/j.cccn.2005.04.002.\u003c/li\u003e\n\u003cli\u003eEl-Beheidy R, Domouky AM, Zidan H, Amer YA. Serum KL-6 as predictive and prognostic marker of interstitial lung disease in childhood connective tissue diseases: a pilot study. Reumatismo. 2021;73. https://doi.org/10.4081/reumatismo.2021.1399.\u003c/li\u003e\n\u003cli\u003eShi S, Chen L, Gui X, Chen L, Qiu X, Yu M, et al. Association of Red Blood Cell Distribution Width Levels with Connective Tissue Disease-Associated Interstitial Lung Disease (CTD-ILD). Disease Markers. 2021;2021:1\u0026ndash;7. https://doi.org/10.1155/2021/5536360.\u003c/li\u003e\n\u003cli\u003eYang HJ, Dong ZX, Wang Y, Li JB, Liu YY. Diagnostic value of IL-27 and KL-6 in the diagnosis of connective tissue disease-associated interstitial lung disease[J]. \u003cem\u003eChin J Int Respir\u003c/em\u003e, 2020, \u003cstrong\u003e40\u003c/strong\u003e:911-916.\u003c/li\u003e\n\u003cli\u003eLiu C, Yang J, Lu Z. Study on the Red Blood Cell Distribution Width in Connective Tissue Disease Associated with Interstitial Lung Disease. Biomed Res Int. 2020;2020:8130213. https://doi.org/10.1155/2020/8130213.\u003c/li\u003e\n\u003cli\u003eGy\u0026ouml;rfi A-H, Filla T, Dickel N, M\u0026ouml;ller F, Li Y-N, Bergmann C, et al. Performance of serum biomarkers reflective of different pathogenic processes in systemic sclerosis-associated interstitial lung disease. Rheumatology. 2024;63:962\u0026ndash;9. https://doi.org/10.1093/rheumatology/kead332.\u003c/li\u003e\n\u003cli\u003eLuo C, Shi J, Zhang J, Lin Y, Pan Y, Zhang J, et al. Survival and early outcomes following lung transplantation for interstitial lung disease associated with non-scleroderma connective tissue disease: a national cohort study. Clin Exp Rheumatol. 2025;43:477\u0026ndash;85. https://doi.org/10.55563/clinexprheumatol/tjnyz5.\u003c/li\u003e\n\u003cli\u003eAhmed S, Handa R. Management of Connective Tissue Disease-related Interstitial Lung Disease. Curr Pulmonol Rep. 2022;11:86\u0026ndash;98. https://doi.org/10.1007/s13665-022-00290-w.\u003c/li\u003e\n\u003cli\u003eWang J, Li K, Hao D, Li X, Zhu Y, Yu H, et al. Pulmonary fibrosis: pathogenesis and therapeutic strategies. MedComm (2020). 2024;5:e744. https://doi.org/10.1002/mco2.744.\u003c/li\u003e\n\u003cli\u003eDing L, Yang J, Zhang C, Zhang X, Gao P. Neutrophils Modulate Fibrogenesis in Chronic Pulmonary Diseases. Front Med (Lausanne). 2021;8:616200. https://doi.org/10.3389/fmed.2021.616200.\u003c/li\u003e\n\u003cli\u003eZhang Y, Chen B, Wang L, Wang R, Yang X. Systemic immune-inflammation index is a promising noninvasive marker to predict survival of lung cancer. Medicine (Baltimore). 2019;98:e13788. https://doi.org/10.1097/MD.0000000000013788.\u003c/li\u003e\n\u003cli\u003eRzepka-Wrona P, Miądlikowska E, Skoczyński S, Barczyk A, Piotrowski W. Patterns of lung fibrosis in patients with interstitial pneumonia with autoimmune features and connective tissue diseases-associated interstitial lung disease-a narrative review. Ann Palliat Med. 2022;11:2110\u0026ndash;30. https://doi.org/10.21037/apm-21-3974.\u003c/li\u003e\n\u003cli\u003eIshiwari M, Kono Y, Togashi Y, Kobayashi K, Kikuchi R, Kogami M, et al. Prognosis of Connective Tissue Disease Related Interstitial Lung Disease after Initiation of Long-Term Oxygen Therapy: Comparison with Idiopathic Pulmonary Fibrosis. OJRD. 2024;14:111\u0026ndash;21. https://doi.org/10.4236/ojrd.2024.144011.\u003c/li\u003e\n\u003cli\u003eMa C, Meng K, Shi S, Zhao T, Chen S, Zhou X, et al. Clinical significance of interleukin-6, total bilirubin, CD3\u0026thinsp;+\u0026thinsp;CD4\u0026thinsp;+\u0026thinsp;T cells counts in the acute exacerbation of connective tissue disease-associated interstitial lung disease: a cross-sectional study. Eur J Med Res. 2023;28:393. https://doi.org/10.1186/s40001-023-01384-0.\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":"Connective tissue disease (CTD), interstitial lung disease (ILD), routine blood test indicators, systemic immune-inflammation index (SII), mean corpuscular volume (MCV), red blood cell distribution width standard deviation (RDW-SD), white blood cell count (WBC)","lastPublishedDoi":"10.21203/rs.3.rs-9173851/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-9173851/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003e\u003cstrong\u003eBackground\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eA machine learning model based on routine blood test indicators was constructed to predict the risk of interstitial lung disease (ILD) in patients with connective tissue disease (CTD), thereby providing a convenient tool for clinical screening of high-risk populations.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMethods\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eA total of 225 inpatients with connective tissue disease (CTD) admitted to Shanxi Provincial People's Hospital from May 2022 to May 2023 were retrospectively enrolled, including 85 cases in the CTD-ILD group and 140 cases in the pure CTD group. Clinical and laboratory data were collected, including gender, age, serum KL-6 levels, and a full set of routine blood test indicators (including derived inflammatory and immune indexes such as systemic immune-inflammation index [SII] and neutrophil-to-lymphocyte ratio [NLR]). The Adaptive Synthetic Sampling (ADASYN )technique (with three proportions: 50%, 75%, and 100%) was applied to address the issue of data imbalance, and the balanced dataset was divided into a training set and a test set at a ratio of 7:3. Five machine learning models were constructed, namely eXtreme Gradient Boosting (XGBoost), logistic regression, random forest (RF), support vector machine (SVM), and k-nearest neighbor (KNN). Hyperparameters were optimized via 10-fold cross-validation and grid search. Model performance was comprehensively evaluated from four dimensions: discriminative ability (ROC curve, PR curve), calibration (calibration plot), clinical utility (decision curve analysis [DCA]), and confusion matrix metrics (AUC, accuracy, sensitivity, etc.). The SHAP (SHapley Additive exPlanations) method was used to interpret key predictive features, and the Bootstrap method was applied to verify the stability of the models and the reliability of the results.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eResults\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eSixteen indicators, including gender, systemic immune-inflammation index (SII), white blood cell count (WBC), mean corpuscular volume (MCV) and red blood cell distribution width standard deviation (RDW-SD) exhibited statistically significant differences between the CTD-ILD group and the pure CTD group (p \u0026lt; 0.05). After feature screening, SII, WBC, MCV and RDW-SD were identified as the core predictive variables. The overall model performance was optimal at the 75% ADASYN sampling ratio, among which the random forest (RF) model achieved the best performance: in the validation set, the area under the curve (AUC) was 0.846, average precision (AP) was 0.896, and F1-score was 0.842. The calibration plot indicated that the model had the minimum deviation between predicted probabilities and actual risks (calibration error = 0.146), and decision curve analysis (DCA) confirmed that the model yielded net clinical benefits across the entire threshold range. SHAP analysis elucidated the action mechanisms of each core variable. Finally, 1000-time Bootstrap resampling validation showed that the RF model had a mean AUC of 0.740 ± 0.040 with a 95% confidence interval (95% CI) of [0.608, 0.848]. All performance indicators presented low coefficients of variation, demonstrating favorable stability and reliability of the model.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConclusions:\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe random forest (RF) model based on routine blood test indicators demonstrated favorable discriminative ability, calibration and clinical utility in the risk prediction of CTD-ILD. Although its specificity was inferior to that of serum markers such as KL-6, the model leverages the advantages of routine blood tests—high popularity, low cost and strong timeliness—and thus can serve as a convenient tool for the preliminary screening of CTD patients at high risk of ILD, and is particularly suitable for primary medical institutions and large-scale screening scenarios.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTrial registration:\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eClinical trial number: not applicable.\u003c/p\u003e","manuscriptTitle":"Machine Learning Model Based on Routine Blood Indicators for High-Risk Screening of Interstitial Lung Disease in Patients with Connective Tissue Disease: A Cost-Effective Triage Strategy","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-03-31 06:45:22","doi":"10.21203/rs.3.rs-9173851/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":"6e5e4ae8-27c7-4465-bec9-92b35ffb12e9","owner":[],"postedDate":"March 31st, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2026-04-23T08:41:30+00:00","versionOfRecord":[],"versionCreatedAt":"2026-03-31 06:45:22","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-9173851","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-9173851","identity":"rs-9173851","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}