A machine learning model for the prediction of peripherally inserted central catheter-related venous thrombosis among high-risk adult patients

A machine learning model for the prediction of peripherally inserted central catheter-related venous thrombosis among high-risk adult patients | 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 A machine learning model for the prediction of peripherally inserted central catheter-related venous thrombosis among high-risk adult patients Songmei Cao, Shuhua Wang, Yi Meng Fan, Bo Cheng, Liqun Zhu, Li Li, and 3 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-2559468/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: The impact of PICC-related thrombosis is worth paying attention to, and it is important to predict the risk factors for thrombosis in patients with PICC catheterization， accurate scientific assessment tools are critical forpredictingand preventing thrombosis in patients with PICCs.The main objective is to develop and validate a machine learning model for predicting the risk of peripherally inserted central catheter-related venous thrombosis. Methods: Overall, 626 patients undergoing peripherally inserted central catheter placement from January 2016 to October 2020 were enrolled. The variables included patient demographic characteristics, clinical condition, laboratory examinations, treatment, and catheter-related factors. Support vector machine and genetic algorithm were used to develop and optimize the model, respectively. SHapley Additive exPlanations was used to interpret the model. Results: The model obtained an average area under the receiver operating characteristic curve of 0.95. The SHapley Additive exPlanations summary plot was used to illustrate the effects of the top 20 features from support vector machine. This study provides a visual way to illustrate the impact of input features on the result prediction. Conclusions: The machine learning model developed based on genetic algorithm shows good predictive ability in patients with a high risk of thrombosis-related peripherally inserted central catheter. Trial registration: retrospectively registered. Machine learning Catheterization Peripheral Venous thromboembolism Support vector machine Prediction model. Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Background The Infusion Nurses Society (INS) has published the latest standard of practice for intravenous care and stated that clinical medical staff should identify the risk factors for catheter-related thrombosis before catheterization, adopt thrombotic prevention strategies during catheterization, and dynamically evaluate the suspicious symptoms of thrombus after catheterization [ 9 ], which underscores the important role of catheter-related thrombosis risk assessment. The impact of PICC-related thrombosis is worth paying attention to, and it is important to predict the risk factors for thrombosis in patients with PICC catheterization.The INS pointed out various venous thrombosis risk factors after PICC implantation [ 9 ]. Previous studies have identified PICC-related thrombosis risk factors related to complications of PICC implantation, including individual factors of patients, PICC catheters and catheter operators. Due to the numerous risk factors associated with PICC-related thrombosis, accurate scientific assessment tools are critical for predicting and preventing thrombosis in patients with PICCs. Predictive models would be helpful to estimate the risk of PICC-related thrombosis. In 2007, Seeley established a model that formed a scoring system for predicting upper extremity deep vein thrombosis in hospitalized PICC catheterization patients, consisting of five items, which had a positive predictive value of only 27.3% [ 10 ], and the list of risk factors lacked specific factors for PICC catheterization. Some scholars collected many risk factors for PICC-related thrombosis and confirmed five risk factors related to PICC thrombosis with logic regression analysis. Based on the results, the Michigan risk score for predicting PICC-associated thrombosis was established, which takes into account the specific factors associated with PICCs, such as the number of PICC lumens [ 2 ]. Chinese scholars constructed a Cox proportional hazards model to determine the risk factors and risk assessment for PICC-related deep vein thrombosis in elderly inpatients. They considered many risk factors for PICC-related thrombosis, analysed the data with traditional statistical methods, and then obtained a model including several indicators. Although the above models simplify the predictive indicators, the influencing factors of PICC-related venous thrombosis are multifactorial and complex, and the interaction effects of these factors do not have a linear relationship. Traditional statistical methods are incapable of fully mining the data. As a vital part of artificial intelligence, machine learning (ML) has great advantages, especially in mining complex data, and can better serve clinical nursing [ 11 ]. ML is already being applied in the field of clinical nursing [ 12 ]. One study established a PICC-related vein thrombosis risk assessment model for hospitalized patients with cancer based on ML [ 13 ]. Compared to the model established by Seeley, their model combined the least absolute shrinkage and selection operator (LASSO) and random forest (RF) techniques and achieved better performance in terms of the area under the receiver operating characteristic (ROC) curve (AUC). However, Liu did not use algorithms to optimize the model results; moreover, the risk factors for PICC-related thrombosis in individual patients was unable to be explained and visualized based on the model. The learning algorithm has always been the core technology for ML to learn information from data, and a good optimization algorithm can not only greatly improve the learning speed but also enhance the convergence speed and effect of the algorithm. Models based on ML can obtain better performance results by using the optimization of the algorithm. Predictive models based on ML do not require assumptions about input variables and their relationship to produce an output because they use computer programs that observe and learn from large amounts of data, so models based on such algorithms are often referred to as ‘black boxes’[ 14 ]. The interpretability and visualization of predictive ML models is an important challenge to overcome for nurses to accept and use the models. SHapley Additive exPlanations (SHAP) is a unified framework for interpreting the predictions of any ML model [ 15 ]. SHAP values can be computed either for a single prediction (individualized) or an entire dataset to explain a model’s overall behaviour (global) [ 16 ]. In recent years, despite a rapid growth of machine learning applications has been observed in healthcare, interdisciplinary subject of nursing and algorithms is still relatively low. Methods Objective The present study was designed to build a more robust support vector machine (SVM) model based on ML and optimize the convergence effect of the model by using genetic algorithm (GA), which facilitates the accurate prediction of the risk of thrombosis associated with PICC catheterization by the SVM model. In addition, we provide a visual legend of the SHAP values, which can be used to interpret our prediction model from the global level to the local level. Design This retrospective cohort study followed the transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD) statement [ 17 ]for reporting. We hypothesized that data extracted from the electronic medical records system and the medical records department could be used to significantly predict thrombosis in adult with PICC. For our analysis, we selected SVM predictive modelling method and SHAP interpret model, based on our study objectives. Patient We retrospectively reviewed the medical records of 626 patients who underwent PICC placement at the Affiliated Hospital of Jiangsu University, which is a teaching hospital with more than 1500 beds, from January 2016 to October 2020. Data from patients older than 18 years were extracted; patients with blood coagulation dysfunction and the catheter was removed because of the chemotherapy intolerance were excluded. Data Collection Our research team previously summarized the evidence of PICC-related risk factors[ 18 ]. The risk factors covered five broad categories: demographic characteristics, clinical condition, laboratory examinations, treatment, and catheter-related factors. After two rounds of expert consultation and team discussion, the variables to be collected were determined, and the measurement standards were unified. All data were collected by manual review from the electronic medical records system and the medical records department of our institution. The data were collected by 2 graduate students under the guidance of venous therapists from November 2020 to January 2021. For general information and tumor factors, the patient's medical history at admission was directly consulted; for laboratory examination results and catheter factors, examination and nursing records within 3 days before and after catheterization were obtained; for treatment factors, the patient's medical advice and disease course records within 3 months during catheterization were tracked. The PICC-related deep vein thrombosis diagnostic criteria were as follows: the presence of noncompressibility of a venous segment (CUS) or the absence of a colour or Doppler signal within the lumen of the vein (visible intraluminal thrombus) by ultrasound examination [ 19 ]. If thrombosis occurred in the vein of the PICC, it was considered to be PICC-related thrombosis. Ethical consideration Prior to the commencement of this study, we obtained ethical approval from the local research and ethics committee (SWYXLL20200121-9), and the requirement to obtain informed consent was waived because the research involved no more than minimal risk to patients. Data analysis All statistical analysis of this study was completed by Python software (version 3.7.4). Continuous variables were expressed as means ± SD for normally distributed variables and the comparisons were performed using t-test. Data that did not conform to a normal distribution are presented as median (interquartile range, IQR) and the Mann-Whitney U-test was used for group comparisons. The counting data were expressed as (%) and Chi-square test was used for comparison between groups. P < 0.05 indicated the statistical significance. SVM and random forest algorithm are implemented based on ‘sklearn’. SMOTE algorithm based on ‘imblearn’. Genetic algorithm is based on library ‘geatpy’ algorithm. SHAP is implemented based on the ‘shap’ library, which is the Python library that the main model implementation of this study relies on. Predictive Model Development Among all the variables, the overall rate of missing data was 1.16%. The missing data were processed by the RF algorithm, and then the data were standardized. In the collected dataset, the rate of PICC-related thrombosis was 26.03%, suggesting an imbalanced ratio between the two class labels. Therefore, we balanced the samples by employing the synthetic minority oversampling technique (SMOTE) algorithm to prevent overfitting in our study. The data were randomly divided, with 70% used for training and 30% used for validation. We used the SVM algorithm, which is a supervised learning model, to develop predictive models and used the associated learning algorithm to analyse data in the classification and regression analysis. Given a set of training instances, each of which is labelled as belonging to one or the other of two categories, SVM creates a model that assigns the new instance to one of the two categories, making it a nonprobabilistic binary nonlinear classifier. A SVM model is a representation of the examples as points in a new prediction space, mapped so that the examples of the separate categories are divided by a clear gap that is as wide as possible [ 20 ]. However, when the data distribution is poor or the data sample size is too large, it will be linearly inseparable. In SVM, the kernel function can be used to map the points to higher dimensions, making them separable. We used SVM with a radial basis function (RBF) kernel function to build the model. To improve the prediction effect of the model, the model was optimized using GA to maximize the AUC value. The flow diagram of the method is shown in Fig. 1 (Model building and validation procedure). First,missing values were imputed using random forest, data imbalance was handled with Synthetic Minority Over-sampling Technique (SMOTE).Secondly,support vector machine (SVM) is used to build prediction models,the genetic algorithm was used for the optimization of the model.Finally ,the remaining thirty percent of data were used for validation. Model Interpretation To break the black box theory of ML, we used SHAP values, which could transform the original linear SVM model into the sum effect of all characteristic attributes to explain the whole model. Through the SHAP values, the prediction results for each patient were presented as much as possible, and the impact of each feature on the prediction of a particular attribute was measured by the SHAP value. Evaluation Metrics Combining the features of patient data, use mainstream indexes of ML model effect to evaluate the performance of the constructed model including the confusion matrix, classification precision, accuracy, recall, F-measure (F 1 ), AUC and the Matthew’s correlation coefficient (MCC). \(MCC=\frac{{TP \times TN - FP \times FN}}{{\sqrt {(TP+FP)(TP+FN)(TN+FP)(TN+FN)} }}\) \({F_1}=2 \cdot \frac{{precision \cdot recall}}{{precision+recall}}\) \(precision=\frac{{TP}}{{TP+FP}}\) \(recall=\frac{{TP}}{{TP+FN}}\) Validity, Reliablility, And Rigour The personnel participating in data screening, medical record review and database construction have been trained uniformly. The ultrasound results were determined by the imaging technologist, who was unaware of this research. The patient’s ID does not involve patient names and other personal privacy, thus effectively protecting patient information security. In order to ensure the consistency of the model, all algorithms were operated by one person. During the construction process, the team determined the algorithm after many discussions. The research team includes clinical specialists, intravenous nurses, nursing graduate students, mathematical modeling experts and computer science professors. Results User Statistics A total of 626 patients undergoing PICC placement were selected. There were 438 people in the training set and 188 people in the test set. For the training and test sets, 114 and 49 people had thrombosis, respectively. Thirty variables were collected for each patient, including baseline characteristics, tumour factors, treatment factors, catheterization factors, and laboratory results. The demographics and variables are listed in Table 1 . For most variables, the differences between the training set and the test set were nonsignificant. Table 1 Patient characteristics and perioperative variables Variable All Training set Test set t/z/χ2 P value Patient population , n 626 438 188 Thrombosis , n 163 114 49 Demographic date Age(years) 58.00 ± 12.00 58.00 ± 11.00 58.00 ± 13.00 0.54 0.58 Man, n (%) 273(43.61) 193(44.06) 80(42.55) 0.02 0.89 BMI(kg/m 2 ) 22.74 ± 3.60 22.66 ± 3.61 22.91 ± 3.56 0.77 0.43 Smoke, n (%) 82(13.09) 59(13.47) 23(12.23) 1.04 0.31 Bed rest, n (%) 52(8.30) 33(7.53) 19(10.10) 0.07 0.79 Acute Infection, n (%) 151(24.12) 101(23.05) 50(26.89) 4.99 0.03* History of major surgery, n (%) 436(69.64) 307(70.09) 129(68.62) 0.16 0.69 History of any VTE event, n (%) 12(1.91) 4(0.91) 8(4.25) 2.00 0.16 Number of underlying diseases,n(%) 0 370(59.11) 260(59.36) 110(58.51) 0.01 0.92 1 177(28.27) 124(28.31) 53(28.19) 0.02 0.86 2 64(10.22) 42(9.59) 22(11.70) 4.32 0.05 3 10(1.60) 9(2.06) 1(1.07) 3.81 0.10 4 5(0.80) 3(0.68) 2(0.53) - - Laboratory WBC(*10 9/ L) 6.10(3.50–6.80) 6.10(3.50–6.60) 6.20(3.70–7.60) -2.40 0.02 NEUT(*10 9 /L) 63.00 ± 17.71 62.55 ± 18.10 64.06 ± 16.78 0.98 0.32 HGB(g/L) 112.34 ± 23.74 112.46 ± 23.24 112.04 ± 24.95 0.20 0.83 PLT(*10 9 /L) 180.14 ± 87.72 175.33 ± 88.03 191.34 ± 86.17 2.09 0.03* CRP(mg/L) 12.03(0-6.57) 11.50(0-5.40) 13.06(0-7.25) -0.27 0.79 PT(s) 11.35 ± 1.59 11.36 ± 1.67 11.32 ± 1.37 0.29 0.77 APTT (s) 26.18 ± 3.98 26.29 ± 3.99 25.93 ± 3.97 6.84 0.99 INR 1.00(0.91–1.02) 0.98(0.91–1.02) 1.03(0.92–1.02) -0.33 0.74 FBG(g/L) 3.65 ± 1.49 3.64 ± 1.60 3.70 ± 1.19 0.49 0.61 D-D(mg/L) 1.52(0.41–1.52) 1.35(0.40–1.41) 1.90(0.45–1.82) -1.54 0.12 Cancer related , n (%) High risk 272(43.45) 192(43.83) 80(42.55) 0 0.95 Tumor metastasis 344(54.95) 240(54.79) 104(55.32) 0.09 0.77 High risk of metastasis 88(14.05) 68(15.52) 20(10.63) 0.81 0.37 Primary tumor location Head and neck 44(7.03) 32(7.31) 12(6.38) 0.09 0.79 Upper diaphragm 231(36.90) 148(33.79) 83(44.15) Lower diaphragm 216(34.50) 163(37.21) 53(28.19) Blood system 135(21.57) 95(21.69) 40(21.28) Table 1 Patient characteristics and perioperative variables (Continued) Variable All Training set Test set t/z/χ2 P value Therapy related, n (%) Radiotherapy 160(25.55) 115(26.26) 45(23.94) 0.03 0.88 Targeted agents 150(23.96) 102(23.28) 48(25.53) 0.15 0.73 Medication history Chemotherapy 504(80.51) 351(80.14) 153(81.39) 0.09 0.77 VEGF 66(10.54) 48(10.96) 18(9.57) Ligand-inhibiting agents TKI 14(2.24) 9(2.05) 5(2.66) Hormone replacement therapy 1(0.16) 1(0.23) 0(0) Tamoxifen/raloxifence 7(1.12) 5(1.14) 2(1.06) Others 34(5.43) 24(5.48) 10(5.32) Catheter insertion , n (%) Proper PICC tip positioning 517(82.58) 368(84.02) 149(79.25) 2.37 0.12 PICC placement times 1 612(97.76) 428(97.72) 184(97.87) 0.05 0.48 ≥ 2 14(2.24) 10(2.28) 4(2.12) Arm selected for insertion Right 515(82.27) 369(84.25) 146(77.66) 0.33 0.56 Left 111(17.73) 69(15.75) 42(22.34) Vein selected for insertion Bassilic vein 552(88.18) 389(88.81) 163(86.70) 0.10 0.76 Brachial vein 60(9.58) 38(8.68) 22(11.70) Cephalic vein 9(1.44) 8(1.83) 1(1.07) Median cubital vein 5(0.80) 3(0.68) 2(0.53) Continuous variables were expressed as means ± SD for normally distributed variables and data that did not conform to a normal distribution are presented as median (interquartile range, IQR), whereas the counting data were expressed as (%). Number of underlying diseases: included the total diseases of diabetes, hypertension, coronary heart disease, infarcts, nephrotic syndrome, inflammatory bowel disease, liver disease, syphilis. Abbreviations: BMI body mass index, VTE Venous thromboembolism, WBC white cell count, NEUT neutrophils, HGB hemoglobin, PLT platelets, CRP C reactive protein, PT Prothrombin Time, APTT activated partial thromboplastin time, INR International normalized ratio, FBG fibrinogen concentration, D-D D-dimer, VEGF the vascular endothelial growth factor, TKI tyrosine kinase inhibitors. * P < 0.05 Predictive Performance And Optimization Process We constructed an SVM model with all the variables as input variables to predict the occurrence of PICC-related venous thrombosis, and the AUC is presented in Fig. 2(ROC curve analysis of the SVM models).Table 2 presents the confusion matrix for the ML model. All indexes of the SVM model on the training set are 1, which indicates that the model correctly predicted almost 100% of the data. On the test set, the AUC value of the ML model was 0.95, MCC was 0.91 and F 1 was 0.95. These results indicated that the model could sufficiently predict which PICC-catheterized patients were at high risk of thrombosis. Table 2. Confusion Matrix for Machine Learning Training data Statistical analysis Predicted true F 1 1 1 0 MCC 1 1 326 0 AUC 1 0 0 322 Precision 1 Recall 1 Accuracy 1 Test data Statistical analysis Predicted true F 1 0.95 1 0 MCC 0.91 1 133 8 AUC 0.95 0 5 132 Precision 0.96 Recall 0.94 Accuracy 0.95 MCC: Matthew’s correlation coefficient AUC: area under the receiver operating characteristic (ROC) curve F 1 : F-measure In our study, GA was used and showed that the individual optimal objective function value tended to be stable when the contemporary number reached 7 and reached the optimal value when the algebra reached 13. The average objective function value of the population tended to be approximately 0.9 when the algebra reached 18, and the optimal algebra was 27, as shown in Fig. 3 (Fitness of the genetic algorithm)which indicates that GA can converge quickly and stably. Model Interpretation And Visual Presentation To determine the features that have the greatest influence on the prediction model, the SHAP summary diagram of the trained model was drawn, as shown in Fig. 4 (SHAP summary plot of the top 20 features of the SVM model)with the top 20 features of the prediction model. This figure describes the relationships between the high and low values of elements in the training dataset and the SHAP values. According to the prediction model, the higher the characteristic SHAP value is, the higher the risk of PICC-related thrombosis. The SHAP correlation diagram in Fig. 5 (Explanation of the prediction results for specific instances) can also be used to understand how each variable affects the output of the SVM prediction model for individual patients, where ‘outvalue’ represents the predicted value of the patient and the ‘base value’ represents the mean value of thrombosis risk. A and B show two such examples. Red indicates that the contribution of this feature is positive, blue indicates that the contribution of this feature is negative, and length indicates the influence. For example, in A, a low risk value was correctly predicted. For this patient, the major contribution in red was BMI, while the major contribution in blue was the activated partial prothrombin time. This patient is a non-thrombosis patient. The model also predicted the risk of thrombosis in a patient who actually experienced thrombosis. As shown in B, the predicted value of 0.74 is higher than the mean value of 0.4739. For this patient, the overall risk of thrombosis was high. Discussion The safety and quality of care of patients with PICC-related thrombosis are still significant challenges to the quality control of the medical system. PICC-related thrombosis assessment and prediction remains a clinical concern. Regardless of patient factors and catheter factors, the risk factors leading to the occurrence of PICC-related thrombosis are complex and variable. Therefore, we sought to develop a more complex algorithm model that takes complex clinical settings into account to predict the outcome of patient complications, aiming to accurately predict the occurrence of thrombosis in individuals. Identifying thrombosis risk factors is the first step in thrombosis assessment. Before developing the model, we finalized the collected variables through two rounds of expert letters and team discussions based on evidence. The selection of these variables followed a scientifically rigorous approach, and 30 variables, including patient, tumour, iatrogenic and catheter factors, were ultimately included. Although smoking and radiation therapy were not shown in the evidence summary, they may increase the risk of blood clots [ 21 ], so we included these two variables in our study. In recent years, with the rapid development of biotechnology and the continuous optimization of oncology treatment regimens, an increasing number of patients are choosing targeted therapy at the early stage of cancer diagnosis. A meta-analysis on the use of bevacizumab for cardiovascular adverse events reported that therapy increases the risk of thrombosis [ 22 ]. At present, with the renewal of molecular targeted drugs, an increasing number of targeted drugs are being used in clinical practice. Experts suggest that targeted drug therapy should be considered. Due to the particularity of the PICC catheterization position, tumour tissue hyperplasia may oppress blood vessels, which affects the blood supply of the PICC catheterization vein. Therefore, in our study, the diaphragm muscle was used as an anatomical marker to classify the tumour types of patients, as shown in Table 2 . Based on the above variables, this study built an SVM model to predict the risk of PICC-related thrombosis in patients. We constructed an SVM model based on ML, rather than traditional mathematical models, which can only answer linear regression questions. In ML, overfitting will degrade the prediction performance of the model [ 23 ] and usually occurs when the model is too complex (i.e., contains too many parameters). In the process of building a model, appropriate measures should be taken to prevent overfitting. The SVM model itself uses a kernel function to map high-dimensional data to low-dimensional data and then divides the data based on a linear classifier, which allows SVM to have a very good fitting effect. Moreover, it can effectively solve ML problems with small sample sizes, solve nonlinear problems and has a strong generalization ability [ 24 ]. The appropriate algorithm should be selected for different data types, and SVM can effectively process various types of data. Our model included a mixture of continuous data and discrete data, and the data were not linearly separable, so SVM was suitable for the data types used in this study. Increasingly, medical research has shown great interest in the use of ML for building predictive models [ 25 ]. Some scholars have used SVM to establish prediction models and obtained a good prediction effect [ 26 ]. However, for classification learning algorithms, the AUC is a better index to evaluate ML models [ 27 ]. Theoretically, the closer the AUC is to 1, the better the model effect will be. The AUC of the prediction model was 1, which means that all the predictions were totally accurate. However, it is almost impossible to achieve a model with 100% accurate prediction, but achieving the best effect as possible is ideal. As a random global optimization algorithm, GA is probably one of the most widely known biological heuristic algorithms that can be used to design reliable and robust models for clinical and research purposes. GA, which was first proposed by Professor Holland in 1975, was originally developed by drawing on some phenomena in evolutionary biology, such as heredity, mutation, natural selection and hybridization. As a kind of nondeterministic quasi-natural algorithm, this algorithm provides a new idea for the optimization of complex systems. To optimize the effect of ML models, the hyperparameter often needs to be set manually; however, the space of the hyperparameter is very large, so the selection of the hyperparameter is more of an empirical project. GA can effectively search for and determine the hyperparameter in the parameter space so that the model effect can approach the optimal performance more quickly and stably. In our study, GA was used to solve the optimal parameter solution of SVM with RBF as the kernel function. The AUC of the model in the training set was used as the fitness evaluation function of the whole population, and the parameters gamma and C were abstracts of chromosomes. Through natural selection and mutation to generate new life populations, the optimal model was trained and then used in the test set to verify the predictive ability of the model. A variety of thrombosis risk factors in complex and real clinical situations were included in this optimized model to try to find an optimal decision boundary to make the most reasonable classification judgement for patients. Furthermore, the advantage of our study is that SHAP values were used to uncover the black box of ML. In the overall explanation of the model by SHAP, the 20 most important features to explain the model were given and arranged in sequence. Major surgery ranked first, which indicates that a history of major surgery may be the most important factor for thrombosis in patients with PICC-catheterized cancer. Evans’s study revealed that surgery longer than 1 hour is an important factor for thrombosis in patients with catheterization [ 28 ].In addition, we exploratorily included the two variables of targeted therapy and radiotherapy in our study, and the significance of these indicators was also reflected in the model. This finding suggests that nurses should pay attention to the use of targeted drugs and radiotherapy in the process of PICC catheterization in patients. In the model, the site of the primary tumour was an important predictor of thrombosis, indicating that the exploratory use of the diaphragm as an anatomical site to identify the risk of PICC-related thrombosis in patients is desirable. Among the predictive variables, tumour metastasis was also an important variable. Tumour metastasis is an important manifestation of active cancer that greatly increases the risk of thrombosis in patients. Age and medical comorbidities were also found to be important features in the model, with the risk of thrombosis (for both the first episode and recurrence) increasing exponentially with age [ 29 ]. The reasons may include medical comorbidities, decreased mobility, and possible age-related changes in clotting. As early as 2007, American scholars Seeley and Chopra constructed a logistic regression model to form a PICC-related thrombosis scoring system. A higher risk of PICC-related thrombosis was associated with a higher score [ 2 , 10 ]. This study does not present a variety of complex mathematical equations but provides a visual way to explain the model. Model visualization can support clinicians and nurses in making decisions and making treatment recommendations for patients [ 30 ]. The significant influence of each feature on the relevant prediction for each particular patient can be seen from the SHAP value, and the risk of thromboembolism for each patient is evident when compared to the baseline value. It is also important to understand how these features interact when making predictions [ 31 ]. Nurses' understanding and mastery of the SHAP value diagram is conducive to the correct evaluation and clinical decision making. Future work will focus on connecting the model results with the hospital information system by enabling the information system to actively fetch patient information to input into the model. The results can be displayed as SHAP values in the information system, and nursing managers can identify the risk according to the results of patients to facilitate the allocation of nursing work. Limitations First, the data were collected from one centre, and the model was not verified by external data. In the future, data from other centres can be collected to popularize and verify this model. Second, there are many studies that clearly show that some variables, such as hyperhomocysteinemia and thrombophilia (e.g., factor V Leiden, protein C deficiency, protein S deficiency), are risk factors for thrombosis. Due to objective reasons, some inspection indicators are not included in routine examinations in China, and the model did not include some variables with effects on thrombogenesis. Further collection of variables may change some characteristics included in the model. Conclusions Our study used an ML method to construct an SVM forecasting model, and the parameters of the model were optimized by GA. The developed model achieved excellent results in the validation dataset and can guide nurses in the precise prediction of the risk of PICC-related thrombosis in patients. The SHAP values can be used to explain the influence of 20 important variables of the model, and a legend of the SHAP values of a single patient can be drawn. Personalized legends can help nurses quickly determine the risk of PICC-related thrombosis and the most important factors affecting the patient’s risk; thus, they could offer care plans and measures for each patient to ensure the orderly development of patient treatment and care. Nurses play an important role in observing disease changes. The risk factors discussed herein should be considered when nurses perform PICC placement. This study provides a predictive PICC-related thrombosis model based on ML to help nurses identify patients at high risk of PICC-related thrombosis. With the development of information technology, it is generally accepted that good algorithms should be used to fully mine data. We should keep pace with the times in the era of big data and provide good quality care for patients in the clinic. Abbreviations abbreviation full name PICC Peripherally Inserted Central Cather SVM Support Vector Machine MCC Matthew’s Correlation ROC Receiver Operating AUC Area Under the Curve SHAP SHapley Additive LASSO Least Absolute Shrinkage and Selection Operator RF Random Forest SMOTE Synthetic Minority Oversampling Technique RBF Radial Basis Function Declarations The institutional and licensing committee approving the experiments. -All methods were carried out in accordance with relevant guidelines and regulations. -Informed consent was obtained from all subjects and their legal guardian(s). (This study was approved by the Ethics Committee of Jiangsu University Hospital under the ethics number SWYXLL20200121-9. This study is a retrospective study, conducted in accordance with the principles of the Declaration of Helsinki, only collects clinical data of patients, does not damage the psychological and physiological operation or behavior of patients, numbering patients does not involve the personal privacy such as patient name, so as to effectively protect the security of patient information.) Ethics approval and consent to participate The study was approved by the Research Ethics Board of the Affiliated Hospital of Jiangsu University, on January 01, 2020. Ethical Approval number: SWYXLL20200121-9. 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 no conflicts of interest Funding This study was funded by Management Innovation Research Project of Jiangsu Hospital Association (JSYGY-3-2019-360) and Zhenjiang Soft Science Research Project (RK2019029). Authors' contributions Songmei Cao ，Shuhua Wang ：Collection of topic selection and thesis writing; Bo Cheng，Li Li : Statistics and analysis; Liqun Zhu，Aiping Li，Hong Zhu: Organize and participate in expert correspondence and discussion; Cao Songmei: Review and finalize the papers to be published; Yimeng Fan，Yiqing Liang: Participated in data collection and paper revision All the authors reviewed the manuscript. Acknowledgments Not applicable. References Hoshal VJ. Total intravenous nutrition with peripherally inserted silicone elastomer central venous catheters. Arch Surg. 1975;110(5):644–6. https://doi.org/10.1001/archsurg.1975.01360110190032 . Chopra V, et al. The Michigan Risk Score to predict peripherally inserted central catheter-associated thrombosis. J Thromb Haemost. 2017;15(10):1951–62. https://doi.org/10.1111/ijlh.12426 . Loughran SC, Borzatta M. Peripherally inserted central catheters: a report of 2506 catheter days. JPEN J Parenter Enteral Nutr. 1995;19(2):133–6. https://doi.org/10.1177/0148607195019002133 . Ng PK, et al. Peripherally inserted central catheters in general medicine. Mayo Clin Proc. 1997;72(3):225–33. http://doi.org/10.4065/72.3.225 . Johansson E, et al. Advantages and disadvantages of peripherally inserted central venous catheters (PICC) compared to other central venous lines: a systematic review of the literature. Acta Oncol. 2013;52(5):886–92. https://doi.org/10.3109/0284186X.2013.773072 . Taxbro K, et al. Clinical impact of peripherally inserted central catheters vs implanted port catheters in patients with cancer: an open-label, randomised, two-centre trial. Br J Anaesth. 2019;122(6):734–41. https://doi.org/10.1016/j.bja.2019.01.038 . Fallouh N, et al. Peripherally Inserted Central Catheter-associated Deep Vein Thrombosis: A Narrative Review. Am J Med. 2015;128(7):722–38. https://doi.org/10.1016/j.amjmed.2015.01.027 . Burns KE, McLaren A. Catheter-related right atrial thrombus and pulmonary embolism: a case report and systematic review of the literature. Can Respir J. 2009;16(5):163–5. https://doi.org/10.1155/2009/751507 . Hinck SM. Implementing the Infusion Therapy Standards of Practice. Home Healthc Now. 2021;39(5):295. http://doi:10.1097/NHH.0000000000001002 . Seeley MA, Santiago M, Shott S. Prediction tool for thrombi associated with peripherally inserted central catheters. J Infus Nurs. 2007;30(5):280–6. https://doi.org/10.1097/01.NAN.0000292570.62763.3f . Beam AL, Kohane IS. Translating Artificial Intelligence Into Clinical Care. JAMA. 2016;316(22):2368–9. https://doi.org/10.1001/jama.2016.17217 . An R, et al. Machine learning-based patient classification system for adult patients in intensive care units: A cross-sectional study. J Nurs Manag. 2021;29(6):1752–62. https://doi.org/10.1111/jonm.13284 . Liu S, et al. Machine learning approaches for risk assessment of peripherally inserted Central catheter-related vein thrombosis in hospitalized patients with cancer. Int J Med Inform. 2019;129:175–83. https://doi.org/10.1016/j.ijmedinf.2019.06.001 . Nensa F, Demircioglu A, Rischpler C. Artificial Intelligence in Nuclear Medicine. J Nucl Med. 2019;60(Suppl 2):29. https://doi.org/10.2967/jnumed.118.220590 . S-37S. Lundberg SM, Lee S. Consistent feature attribution for tree ensembles. 2017. arXiv preprint arXiv:1705.07874. Lundberg SM, Erion GG, Lee S. Consistent individualized feature attribution for tree ensembles. 2018. arXiv preprint arXiv:1802.03888. Collins GS, et al. Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD): the TRIPOD statement. BMJ. 2015;350. g7594.https://doi.org/10.1016/j.eururo.2014.11.025 . Zhu LZRC. Evidence summary for risk assessment of PICC-related venous thrombosis. Chin J Nurs. 2017;52(10):1179–85. https://doi.org/10.37610.issn.0254–1769.2017.10.005. Bates SM, et al. Diagnosis of DVT: Antithrombotic Therapy and Prevention of Thrombosis, 9th ed: American College of Chest Physicians Evidence-Based Clinical Practice Guidelines. Chest. 2012;141(2). https://doi.org/10.1378/chest.11-2299 . Suppl): p. e351S-e418S. D'Orazio M, et al. Deciphering Cancer Cell Behavior From Motility and Shape Features: Peer Prediction and Dynamic Selection to Support Cancer Diagnosis and Therapy. Front Oncol. 2020;10:580698. .https://doi.org/10.3389/fonc.2020.580698 . Ageno W, et al. Cardiovascular risk factors and venous thromboembolism: a meta-analysis. Circulation. 2008;117(1):93–102. https://doi.org/10.1161/CIRCULATIONAHA.107.709204 . Totzeck M, Mincu RI, Rassaf T. Cardiovascular Adverse Events in Patients With Cancer Treated With Bevacizumab: A Meta-Analysis of More Than 20 000 Patients. J Am Heart Assoc. 2017;6(8). https://doi.org/10.1161/JAHA.117.006278 . Deo RC. Machine Learning in Medicine. Circulation. 2015;132(20):1920–30. https://doi.org/10.1161/CIRCULATIONAHA.115.001593 . Jin C, Wang L. Dimensionality dependent PAC-Bayes margin bound. 2012: Lake Tahoe, NV, United states. p. 1034–1042. Handelman GS, et al. eDoctor: machine learning and the future of medicine. J Intern Med. 2018;284(6):603–19. https://doi.org/10.1111/joim.12822 . Ferroni P, et al. Risk Assessment for Venous Thromboembolism in Chemotherapy-Treated Ambulatory Cancer Patients. Med Decis Making. 2017;37(2):234–42. https://doi.org/10.1177/0272989X16662654 . Huang J, Ling CX, IEEE TRANSACTIONS ON KNOWLEDGE AND DATA ENGINEERING. Using AUC and accuracy in evaluating learning algorithms., 2005. 17(3): p.299–310. https://doi.org/10.1109/TKDE.2005.50 Evans RS, et al. Risk of symptomatic DVT associated with peripherally inserted central catheters. Chest. 2010;138(4):803–10. https://doi.org/10.1378/chest.10-0154 . Gregson J, et al. Cardiovascular Risk Factors Associated With Venous Thromboembolism. JAMA Cardiol. 2019;4(2):163–73. https://doi.org/10.1001/jamacardio.2018.4537 . Sun W, et al. Development and validation of two aspiration prediction models in patients receiving nasogastric feeding. J Nurs Manag. 2020;28(6):1372–80. https://doi.org/10.1111/jonm.13093 . Li X, et al. A Time-Phased Machine Learning Model for Real-Time Prediction of Sepsis in Critical Care. Crit Care Med. 2020;48(10). p. e884-e888.https://doi.org/10.1097/CCM.0000000000004494 . Additional Declarations No competing interests reported. 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-2559468","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":174747586,"identity":"a17fa139-2da5-458d-a643-01d0d9dad39e","order_by":0,"name":"Songmei Cao","email":"","orcid":"","institution":"The Affiliated Hospital of Jiangsu University","correspondingAuthor":false,"prefix":"","firstName":"Songmei","middleName":"","lastName":"Cao","suffix":""},{"id":174747587,"identity":"2cce7105-e198-4b5b-a78c-5519f908f2f9","order_by":1,"name":"Shuhua Wang","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA+0lEQVRIie3RMUvEMBTA8ZRAp4SuF/RDvKNQkTu8r5JwUJdwi0vHwEHcdO19CydxTAl4S7Rr4RbBSejQbhYc7K2C6Y2C+Q1vCO8/PIJQEPxVkUJ8nLefXbEkSaJOTrBhpcvPWWlOTmJ+RrVdguL+bdi/2I/hyW5g7wCIqwkgE3W99CRuky+oszfgJPBZcSAXWGG2e/w9yYzM0khb8WAkGHAHcqlMjKkvqdtsPhyTup0roV8JGD6RNDJ9o8ekydeo0mY6WTVthqm+Frvm/TlSbk1YWW29t7B7mfaDXoi7Wmj8VVytkmRbdb0nGcWzHw/jN03A3dRGEATBP/cNFuFf3pulu6MAAAAASUVORK5CYII=","orcid":"","institution":"Jiangsu college of nursing","correspondingAuthor":true,"prefix":"","firstName":"Shuhua","middleName":"","lastName":"Wang","suffix":""},{"id":174747588,"identity":"8574260d-3994-4e47-ad1a-f1243f7f720b","order_by":2,"name":"Yi Meng Fan","email":"","orcid":"","institution":"The Affiliated Hospital of Jiangsu University","correspondingAuthor":false,"prefix":"","firstName":"Yi","middleName":"Meng","lastName":"Fan","suffix":""},{"id":174747589,"identity":"56e41bc7-4cf9-4400-a76a-051e726959db","order_by":3,"name":"Bo Cheng","email":"","orcid":"","institution":"Jiangsu University","correspondingAuthor":false,"prefix":"","firstName":"Bo","middleName":"","lastName":"Cheng","suffix":""},{"id":174747590,"identity":"18428813-56a1-4a26-83cf-6ce0c36037a9","order_by":4,"name":"Liqun Zhu","email":"","orcid":"","institution":"The Affiliated Hospital of Jiangsu University","correspondingAuthor":false,"prefix":"","firstName":"Liqun","middleName":"","lastName":"Zhu","suffix":""},{"id":174747591,"identity":"854a47e3-9a30-40b5-8ec5-25375c2190f7","order_by":5,"name":"Li Li","email":"","orcid":"","institution":"Jiangsu University","correspondingAuthor":false,"prefix":"","firstName":"Li","middleName":"","lastName":"Li","suffix":""},{"id":174747592,"identity":"4da90434-59af-462c-a37c-f77beffac2f2","order_by":6,"name":"YiQing Liang","email":"","orcid":"","institution":"The Affiliated Hospital of Jiangsu University","correspondingAuthor":false,"prefix":"","firstName":"YiQing","middleName":"","lastName":"Liang","suffix":""},{"id":174747593,"identity":"04b49523-f1b2-4120-8b2c-2eeb996f2ac1","order_by":7,"name":"Aiping Li","email":"","orcid":"","institution":"Taixing People's Hospital","correspondingAuthor":false,"prefix":"","firstName":"Aiping","middleName":"","lastName":"Li","suffix":""},{"id":174747594,"identity":"46d2bad7-1340-42b2-a64c-7df2d205053d","order_by":8,"name":"Hong Zhu","email":"","orcid":"","institution":"Suzhou Ninth People’s Hospital","correspondingAuthor":false,"prefix":"","firstName":"Hong","middleName":"","lastName":"Zhu","suffix":""}],"badges":[],"createdAt":"2023-02-07 09:29:28","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-2559468/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-2559468/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":32941899,"identity":"100ca5f7-ec68-492c-b352-235249fd4e57","added_by":"auto","created_at":"2023-02-14 21:38:17","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":108520,"visible":true,"origin":"","legend":"\u003cp\u003eModel building and validation procedure\u003c/p\u003e","description":"","filename":"Figure1.Modelbuildingandvalidationprocedure..png","url":"https://assets-eu.researchsquare.com/files/rs-2559468/v1/ed9cb6be2f2283f031a9d98b.png"},{"id":32942628,"identity":"6755f14f-b0bc-4ce6-96ff-6e4a1c3e1cb9","added_by":"auto","created_at":"2023-02-14 21:46:17","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":51602,"visible":true,"origin":"","legend":"\u003cp\u003eROC curve analysis of the SVM models\u003c/p\u003e","description":"","filename":"Figure2.ROCcurveanalysisoftheSVMmodels..png","url":"https://assets-eu.researchsquare.com/files/rs-2559468/v1/8751e51d77aeffcdad38a202.png"},{"id":32941896,"identity":"1e94c323-2360-40b9-ae58-5c7d77ea029c","added_by":"auto","created_at":"2023-02-14 21:38:17","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":58876,"visible":true,"origin":"","legend":"\u003cp\u003eFitness of the genetic algorithm\u003c/p\u003e","description":"","filename":"Figure3.Fitnessofthegeneticalgorithm..png","url":"https://assets-eu.researchsquare.com/files/rs-2559468/v1/71dc56080ea7f9e14d5bbc65.png"},{"id":32943110,"identity":"522010cd-93ae-499e-8ccc-30eac5f55f86","added_by":"auto","created_at":"2023-02-14 21:54:17","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":106269,"visible":true,"origin":"","legend":"\u003cp\u003eSHAP summary plot of the top 20 features of the SVM model\u003c/p\u003e","description":"","filename":"Figure4.SHAPsummaryplotofthetop20featuresoftheSVMmodel..png","url":"https://assets-eu.researchsquare.com/files/rs-2559468/v1/05f5b8829850d883f7ae705f.png"},{"id":32941898,"identity":"4189bc2f-a652-49e2-b346-130085b69863","added_by":"auto","created_at":"2023-02-14 21:38:17","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":100525,"visible":true,"origin":"","legend":"\u003cp\u003eExplanation of the prediction results for specific instances\u003c/p\u003e","description":"","filename":"Figure5.Explanationofthepredictionresultsforspecificinstances..png","url":"https://assets-eu.researchsquare.com/files/rs-2559468/v1/6aae741351175f4fe140cbee.png"},{"id":33628391,"identity":"663ac56f-2be0-4cad-a5d8-b64b7b85bcd8","added_by":"auto","created_at":"2023-03-01 14:59:42","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":842248,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-2559468/v1/5224eda6-0261-4e16-bd43-ce6f915f8100.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"A machine learning model for the prediction of peripherally inserted central catheter-related venous thrombosis among high-risk adult patients","fulltext":[{"header":"Background","content":"\u003cp\u003eThe Infusion Nurses Society (INS) has published the latest standard of practice for intravenous care and stated that clinical medical staff should identify the risk factors for catheter-related thrombosis before catheterization, adopt thrombotic prevention strategies during catheterization, and dynamically evaluate the suspicious symptoms of thrombus after catheterization [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e], which underscores the important role of catheter-related thrombosis risk assessment. The impact of PICC-related thrombosis is worth paying attention to, and it is important to predict the risk factors for thrombosis in patients with PICC catheterization.The INS pointed out various venous thrombosis risk factors after PICC implantation [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]. Previous studies have identified PICC-related thrombosis risk factors related to complications of PICC implantation, including individual factors of patients, PICC catheters and catheter operators. Due to the numerous risk factors associated with PICC-related thrombosis, accurate scientific assessment tools are critical for predicting and preventing thrombosis in patients with PICCs.\u003c/p\u003e \u003cp\u003ePredictive models would be helpful to estimate the risk of PICC-related thrombosis. In 2007, Seeley established a model that formed a scoring system for predicting upper extremity deep vein thrombosis in hospitalized PICC catheterization patients, consisting of five items, which had a positive predictive value of only 27.3% [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e], and the list of risk factors lacked specific factors for PICC catheterization. Some scholars collected many risk factors for PICC-related thrombosis and confirmed five risk factors related to PICC thrombosis with logic regression analysis. Based on the results, the Michigan risk score for predicting PICC-associated thrombosis was established, which takes into account the specific factors associated with PICCs, such as the number of PICC lumens [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. Chinese scholars constructed a Cox proportional hazards model to determine the risk factors and risk assessment for PICC-related deep vein thrombosis in elderly inpatients. They considered many risk factors for PICC-related thrombosis, analysed the data with traditional statistical methods, and then obtained a model including several indicators. Although the above models simplify the predictive indicators, the influencing factors of PICC-related venous thrombosis are multifactorial and complex, and the interaction effects of these factors do not have a linear relationship. Traditional statistical methods are incapable of fully mining the data.\u003c/p\u003e \u003cp\u003eAs a vital part of artificial intelligence, machine learning (ML) has great advantages, especially in mining complex data, and can better serve clinical nursing [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e]. ML is already being applied in the field of clinical nursing [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e]. One study established a PICC-related vein thrombosis risk assessment model for hospitalized patients with cancer based on ML [\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e]. Compared to the model established by Seeley, their model combined the least absolute shrinkage and selection operator (LASSO) and random forest (RF) techniques and achieved better performance in terms of the area under the receiver operating characteristic (ROC) curve (AUC). However, Liu did not use algorithms to optimize the model results; moreover, the risk factors for PICC-related thrombosis in individual patients was unable to be explained and visualized based on the model.\u003c/p\u003e \u003cp\u003eThe learning algorithm has always been the core technology for ML to learn information from data, and a good optimization algorithm can not only greatly improve the learning speed but also enhance the convergence speed and effect of the algorithm. Models based on ML can obtain better performance results by using the optimization of the algorithm.\u003c/p\u003e \u003cp\u003ePredictive models based on ML do not require assumptions about input variables and their relationship to produce an output because they use computer programs that observe and learn from large amounts of data, so models based on such algorithms are often referred to as \u0026lsquo;black boxes\u0026rsquo;[\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]. The interpretability and visualization of predictive ML models is an important challenge to overcome for nurses to accept and use the models. SHapley Additive exPlanations (SHAP) is a unified framework for interpreting the predictions of any ML model [\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]. SHAP values can be computed either for a single prediction (individualized) or an entire dataset to explain a model\u0026rsquo;s overall behaviour (global) [\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e]. In recent years, despite a rapid growth of machine learning applications has been observed in healthcare, interdisciplinary subject of nursing and algorithms is still relatively low.\u003c/p\u003e"},{"header":"Methods","content":"\u003ch2\u003eObjective\u003c/h2\u003e\n\u003cp\u003eThe present study was designed to build a more robust support vector machine (SVM) model based on ML and optimize the convergence effect of the model by using genetic algorithm (GA), which facilitates the accurate prediction of the risk of thrombosis associated with PICC catheterization by the SVM model. In addition, we provide a visual legend of the SHAP values, which can be used to interpret our prediction model from the global level to the local level.\u003c/p\u003e\n\u003ch3\u003eDesign\u003c/h3\u003e\n\u003cp\u003eThis retrospective cohort study followed the transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD) statement [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]for reporting. We hypothesized that data extracted from the electronic medical records system and the medical records department could be used to significantly predict thrombosis in adult with PICC. For our analysis, we selected SVM predictive modelling method and SHAP interpret model, based on our study objectives.\u003c/p\u003e\n\u003ch3\u003ePatient\u003c/h3\u003e\n\u003cp\u003e We retrospectively reviewed the medical records of 626 patients who underwent PICC placement at the Affiliated Hospital of Jiangsu University, which is a teaching hospital with more than 1500 beds, from January 2016 to October 2020. Data from patients older than 18 years were extracted; patients with blood coagulation dysfunction and the catheter was removed because of the chemotherapy intolerance were excluded.\u003c/p\u003e\n\u003ch3\u003eData Collection\u003c/h3\u003e\n\u003cp\u003eOur research team previously summarized the evidence of PICC-related risk factors[\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]. The risk factors covered five broad categories: demographic characteristics, clinical condition, laboratory examinations, treatment, and catheter-related factors. After two rounds of expert consultation and team discussion, the variables to be collected were determined, and the measurement standards were unified. All data were collected by manual review from the electronic medical records system and the medical records department of our institution. The data were collected by 2 graduate students under the guidance of venous therapists from November 2020 to January 2021. For general information and tumor factors, the patient's medical history at admission was directly consulted; for laboratory examination results and catheter factors, examination and nursing records within 3 days before and after catheterization were obtained; for treatment factors, the patient's medical advice and disease course records within 3 months during catheterization were tracked. The PICC-related deep vein thrombosis diagnostic criteria were as follows: the presence of noncompressibility of a venous segment (CUS) or the absence of a colour or Doppler signal within the lumen of the vein (visible intraluminal thrombus) by ultrasound examination [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]. If thrombosis occurred in the vein of the PICC, it was considered to be PICC-related thrombosis.\u003c/p\u003e \u003cdiv id=\"Sec7\" class=\"Section2\"\u003e \u003ch2\u003eEthical consideration\u003c/h2\u003e \u003cp\u003ePrior to the commencement of this study, we obtained ethical approval from the local research and ethics committee (SWYXLL20200121-9), and the requirement to obtain informed consent was waived because the research involved no more than minimal risk to patients.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003eData analysis\u003c/h2\u003e \u003cp\u003eAll statistical analysis of this study was completed by Python software (version 3.7.4). Continuous variables were expressed as means\u0026thinsp;\u0026plusmn;\u0026thinsp;SD for normally distributed variables and the comparisons were performed using t-test. Data that did not conform to a normal distribution are presented as median (interquartile range, IQR) and the Mann-Whitney U-test was used for group comparisons. The counting data were expressed as (%) and Chi-square test was used for comparison between groups. \u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.05 indicated the statistical significance. SVM and random forest algorithm are implemented based on \u0026lsquo;sklearn\u0026rsquo;. SMOTE algorithm based on \u0026lsquo;imblearn\u0026rsquo;. Genetic algorithm is based on library \u0026lsquo;geatpy\u0026rsquo; algorithm. SHAP is implemented based on the \u0026lsquo;shap\u0026rsquo; library, which is the Python library that the main model implementation of this study relies on.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003ePredictive Model Development\u003c/h3\u003e\n\u003cp\u003eAmong all the variables, the overall rate of missing data was 1.16%. The missing data were processed by the RF algorithm, and then the data were standardized. In the collected dataset, the rate of PICC-related thrombosis was 26.03%, suggesting an imbalanced ratio between the two class labels. Therefore, we balanced the samples by employing the synthetic minority oversampling technique (SMOTE) algorithm to prevent overfitting in our study. The data were randomly divided, with 70% used for training and 30% used for validation. We used the SVM algorithm, which is a supervised learning model, to develop predictive models and used the associated learning algorithm to analyse data in the classification and regression analysis. Given a set of training instances, each of which is labelled as belonging to one or the other of two categories, SVM creates a model that assigns the new instance to one of the two categories, making it a nonprobabilistic binary nonlinear classifier. A SVM model is a representation of the examples as points in a new prediction space, mapped so that the examples of the separate categories are divided by a clear gap that is as wide as possible [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]. However, when the data distribution is poor or the data sample size is too large, it will be linearly inseparable. In SVM, the kernel function can be used to map the points to higher dimensions, making them separable. We used SVM with a radial basis function (RBF) kernel function to build the model. To improve the prediction effect of the model, the model was optimized using GA to maximize the AUC value. The flow diagram of the method is shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e(Model building and validation procedure).\u003c/p\u003e \u003cp\u003eFirst,missing values were imputed using random forest, data imbalance was handled with Synthetic Minority Over-sampling Technique (SMOTE).Secondly,support vector machine (SVM) is used to build prediction models,the genetic algorithm was used for the optimization of the model.Finally ,the remaining thirty percent of data were used for validation.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e\n\u003ch3\u003eModel Interpretation\u003c/h3\u003e\n\u003cp\u003eTo break the black box theory of ML, we used SHAP values, which could transform the original linear SVM model into the sum effect of all characteristic attributes to explain the whole model. Through the SHAP values, the prediction results for each patient were presented as much as possible, and the impact of each feature on the prediction of a particular attribute was measured by the SHAP value.\u003c/p\u003e\n\u003ch3\u003eEvaluation Metrics\u003c/h3\u003e\n\u003cp\u003eCombining the features of patient data, use mainstream indexes of ML model effect to evaluate the performance of the constructed model including the confusion matrix, classification precision, accuracy, recall, F-measure (F\u003csub\u003e1\u003c/sub\u003e), AUC and the Matthew\u0026rsquo;s correlation coefficient (MCC).\u003c/p\u003e \u003cp\u003e \u003cspan class=\"InlineEquation\"\u003e \u003cspan class=\"mathinline\"\u003e\\(MCC=\\frac{{TP \\times TN - FP \\times FN}}{{\\sqrt {(TP+FP)(TP+FN)(TN+FP)(TN+FN)} }}\\)\u003c/span\u003e \u003c/span\u003e \u003c/p\u003e \u003cp\u003e \u003cspan class=\"InlineEquation\"\u003e \u003cspan class=\"mathinline\"\u003e\\({F_1}=2 \\cdot \\frac{{precision \\cdot recall}}{{precision+recall}}\\)\u003c/span\u003e \u003c/span\u003e \u003c/p\u003e \u003cp\u003e \u003cspan class=\"InlineEquation\"\u003e \u003cspan class=\"mathinline\"\u003e\\(precision=\\frac{{TP}}{{TP+FP}}\\)\u003c/span\u003e \u003c/span\u003e \u003c/p\u003e \u003cp\u003e \u003cspan class=\"InlineEquation\"\u003e \u003cspan class=\"mathinline\"\u003e\\(recall=\\frac{{TP}}{{TP+FN}}\\)\u003c/span\u003e \u003c/span\u003e \u003c/p\u003e\n\u003ch3\u003eValidity, Reliablility, And Rigour\u003c/h3\u003e\n\u003cp\u003eThe personnel participating in data screening, medical record review and database construction have been trained uniformly. The ultrasound results were determined by the imaging technologist, who was unaware of this research. The patient\u0026rsquo;s ID does not involve patient names and other personal privacy, thus effectively protecting patient information security. In order to ensure the consistency of the model, all algorithms were operated by one person. During the construction process, the team determined the algorithm after many discussions. The research team includes clinical specialists, intravenous nurses, nursing graduate students, mathematical modeling experts and computer science professors.\u003c/p\u003e"},{"header":"Results","content":"\u003ch3\u003eUser Statistics\u003c/h3\u003e\n\u003cp\u003eA total of 626 patients undergoing PICC placement were selected. There were 438 people in the training set and 188 people in the test set. For the training and test sets, 114 and 49 people had thrombosis, respectively. Thirty variables were collected for each patient, including baseline characteristics, tumour factors, treatment factors, catheterization factors, and laboratory results. The demographics and variables are listed in Table\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003e. For most variables, the differences between the training set and the test set were nonsignificant.\u003c/p\u003e\n\u003cdiv class=\"gridtable\"\u003e\n\u003cdiv class=\"colspec\" align=\"left\"\u003e\u0026nbsp;\u003c/div\u003e\n\u003ctable id=\"Tab1\" border=\"1\"\u003e\u003ccaption\u003e\n\u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e\n\u003cdiv class=\"CaptionContent\"\u003e\n\u003cp\u003ePatient characteristics and perioperative variables\u003c/p\u003e\n\u003c/div\u003e\n\u003c/caption\u003e\n\u003cthead\u003e\n\u003ctr\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003eVariable\u003c/p\u003e\n\u003c/th\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003eAll\u003c/p\u003e\n\u003c/th\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003eTraining set\u003c/p\u003e\n\u003c/th\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003eTest set\u003c/p\u003e\n\u003c/th\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003e\u003cem\u003et/z/\u0026chi;2\u003c/em\u003e\u003c/p\u003e\n\u003c/th\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003e\u003cem\u003eP\u003c/em\u003e value\u003c/p\u003e\n\u003c/th\u003e\n\u003c/tr\u003e\n\u003c/thead\u003e\n\u003ctbody\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u003cstrong\u003ePatient population\u003c/strong\u003e,\u003cspan class=\"BoldItalic\"\u003en\u003c/span\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e626\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e438\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e188\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u003cstrong\u003eThrombosis\u003c/strong\u003e,\u003cspan class=\"BoldItalic\"\u003en\u003c/span\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e163\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e114\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e49\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u003cstrong\u003eDemographic date\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eAge(years)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e58.00\u0026thinsp;\u0026plusmn;\u0026thinsp;12.00\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e58.00\u0026thinsp;\u0026plusmn;\u0026thinsp;11.00\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e58.00\u0026thinsp;\u0026plusmn;\u0026thinsp;13.00\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.54\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.58\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eMan,\u003cem\u003en\u003c/em\u003e(%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e273(43.61)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e193(44.06)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e80(42.55)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.02\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.89\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eBMI(kg/m\u003csup\u003e2\u003c/sup\u003e)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e22.74\u0026thinsp;\u0026plusmn;\u0026thinsp;3.60\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e22.66\u0026thinsp;\u0026plusmn;\u0026thinsp;3.61\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e22.91\u0026thinsp;\u0026plusmn;\u0026thinsp;3.56\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.77\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.43\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eSmoke,\u003cem\u003en\u003c/em\u003e(%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e82(13.09)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e59(13.47)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e23(12.23)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e1.04\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.31\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eBed rest,\u003cem\u003en\u003c/em\u003e(%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e52(8.30)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e33(7.53)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e19(10.10)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.07\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.79\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eAcute Infection,\u003cem\u003en\u003c/em\u003e(%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e151(24.12)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e101(23.05)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e50(26.89)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e4.99\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.03*\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eHistory of major surgery,\u003cem\u003en\u003c/em\u003e(%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e436(69.64)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e307(70.09)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e129(68.62)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.16\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.69\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eHistory of any VTE event,\u003cem\u003en\u003c/em\u003e(%)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e12(1.91)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e4(0.91)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e8(4.25)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e2.00\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.16\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u003cem\u003eNumber of underlying diseases,n(%)\u003c/em\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e370(59.11)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e260(59.36)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e110(58.51)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.01\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.92\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e1\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e177(28.27)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e124(28.31)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e53(28.19)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.02\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.86\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e2\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e64(10.22)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e42(9.59)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e22(11.70)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e4.32\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.05\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e3\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e10(1.60)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e9(2.06)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e1(1.07)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e3.81\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.10\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e4\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e5(0.80)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e3(0.68)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e2(0.53)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e-\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e-\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u003cstrong\u003eLaboratory\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eWBC(*10\u003csup\u003e9/\u003c/sup\u003eL)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e6.10(3.50\u0026ndash;6.80)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e6.10(3.50\u0026ndash;6.60)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e6.20(3.70\u0026ndash;7.60)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e-2.40\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.02\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eNEUT(*10\u003csup\u003e9\u003c/sup\u003e/L)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e63.00\u0026thinsp;\u0026plusmn;\u0026thinsp;17.71\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e62.55\u0026thinsp;\u0026plusmn;\u0026thinsp;18.10\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e64.06\u0026thinsp;\u0026plusmn;\u0026thinsp;16.78\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.98\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.32\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eHGB(g/L)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e112.34\u0026thinsp;\u0026plusmn;\u0026thinsp;23.74\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e112.46\u0026thinsp;\u0026plusmn;\u0026thinsp;23.24\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e112.04\u0026thinsp;\u0026plusmn;\u0026thinsp;24.95\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.20\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.83\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003ePLT(*10\u003csup\u003e9\u003c/sup\u003e/L)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e180.14\u0026thinsp;\u0026plusmn;\u0026thinsp;87.72\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e175.33\u0026thinsp;\u0026plusmn;\u0026thinsp;88.03\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e191.34\u0026thinsp;\u0026plusmn;\u0026thinsp;86.17\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e2.09\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.03*\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eCRP(mg/L)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e12.03(0-6.57)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e11.50(0-5.40)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e13.06(0-7.25)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e-0.27\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.79\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003ePT(s)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e11.35\u0026thinsp;\u0026plusmn;\u0026thinsp;1.59\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e11.36\u0026thinsp;\u0026plusmn;\u0026thinsp;1.67\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e11.32\u0026thinsp;\u0026plusmn;\u0026thinsp;1.37\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.29\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.77\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eAPTT (s)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e26.18\u0026thinsp;\u0026plusmn;\u0026thinsp;3.98\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e26.29\u0026thinsp;\u0026plusmn;\u0026thinsp;3.99\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e25.93\u0026thinsp;\u0026plusmn;\u0026thinsp;3.97\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e6.84\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.99\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eINR\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e1.00(0.91\u0026ndash;1.02)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.98(0.91\u0026ndash;1.02)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e1.03(0.92\u0026ndash;1.02)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e-0.33\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.74\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eFBG(g/L)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e3.65\u0026thinsp;\u0026plusmn;\u0026thinsp;1.49\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e3.64\u0026thinsp;\u0026plusmn;\u0026thinsp;1.60\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e3.70\u0026thinsp;\u0026plusmn;\u0026thinsp;1.19\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.49\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.61\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eD-D(mg/L)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e1.52(0.41\u0026ndash;1.52)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e1.35(0.40\u0026ndash;1.41)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e1.90(0.45\u0026ndash;1.82)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e-1.54\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.12\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u003cstrong\u003eCancer related\u003c/strong\u003e ,\u003cspan class=\"BoldItalic\"\u003en\u003c/span\u003e\u003cstrong\u003e(%)\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eHigh risk\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e272(43.45)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e192(43.83)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e80(42.55)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.95\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eTumor metastasis\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e344(54.95)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e240(54.79)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e104(55.32)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.09\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.77\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eHigh risk of metastasis\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e88(14.05)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e68(15.52)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e20(10.63)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.81\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.37\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u003cem\u003ePrimary tumor location\u003c/em\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eHead and neck\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e44(7.03)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e32(7.31)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e12(6.38)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd rowspan=\"4\" align=\"left\"\u003e\n\u003cp\u003e0.09\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd rowspan=\"4\" align=\"left\"\u003e\n\u003cp\u003e0.79\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eUpper diaphragm\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e231(36.90)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e148(33.79)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e83(44.15)\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eLower diaphragm\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e216(34.50)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e163(37.21)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e53(28.19)\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eBlood system\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e135(21.57)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e95(21.69)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e40(21.28)\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003c/tbody\u003e\n\u003c/table\u003e\n\u003c/div\u003e\n\u003cdiv class=\"gridtable\"\u003e\n\u003cdiv class=\"colspec\" align=\"left\"\u003e\u0026nbsp;\u003c/div\u003e\n\u003cdiv class=\"colspec\" align=\"char\"\u003e\u0026nbsp;\u003c/div\u003e\n\u003ctable id=\"Tab2\" border=\"1\"\u003e\u003ccaption\u003e\n\u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e\n\u003cdiv class=\"CaptionContent\"\u003e\n\u003cp\u003ePatient characteristics and perioperative variables (Continued)\u003c/p\u003e\n\u003c/div\u003e\n\u003c/caption\u003e\n\u003cthead\u003e\n\u003ctr\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003eVariable\u003c/p\u003e\n\u003c/th\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003eAll\u003c/p\u003e\n\u003c/th\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003eTraining set\u003c/p\u003e\n\u003c/th\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003eTest set\u003c/p\u003e\n\u003c/th\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003e\u003cem\u003et/z/\u0026chi;2\u003c/em\u003e\u003c/p\u003e\n\u003c/th\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003e\u003cem\u003eP\u003c/em\u003e value\u003c/p\u003e\n\u003c/th\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003eTherapy related,\u003cem\u003en\u003c/em\u003e(%)\u003c/p\u003e\n\u003c/th\u003e\n\u003cth align=\"left\"\u003e\u0026nbsp;\u003c/th\u003e\n\u003cth align=\"left\"\u003e\u0026nbsp;\u003c/th\u003e\n\u003cth align=\"left\"\u003e\u0026nbsp;\u003c/th\u003e\n\u003cth align=\"left\"\u003e\u0026nbsp;\u003c/th\u003e\n\u003cth align=\"left\"\u003e\u0026nbsp;\u003c/th\u003e\n\u003c/tr\u003e\n\u003c/thead\u003e\n\u003ctbody\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eRadiotherapy\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e160(25.55)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e115(26.26)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e45(23.94)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.03\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.88\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eTargeted agents\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e150(23.96)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e102(23.28)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e48(25.53)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.15\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.73\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u003cem\u003eMedication history\u003c/em\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eChemotherapy\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e504(80.51)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e351(80.14)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e153(81.39)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd rowspan=\"6\" align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.09\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd rowspan=\"6\" align=\"left\"\u003e\n\u003cp\u003e0.77\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eVEGF\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e66(10.54)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e48(10.96)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e18(9.57)\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eLigand-inhibiting agents TKI\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e14(2.24)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e9(2.05)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e5(2.66)\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eHormone replacement therapy\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e1(0.16)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e1(0.23)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0(0)\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eTamoxifen/raloxifence\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e7(1.12)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e5(1.14)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e2(1.06)\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eOthers\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e34(5.43)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e24(5.48)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e10(5.32)\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u003cstrong\u003eCatheter insertion\u003c/strong\u003e,\u003cspan class=\"BoldItalic\"\u003en\u003c/span\u003e\u003cstrong\u003e(%)\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eProper PICC tip positioning\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e517(82.58)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e368(84.02)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e149(79.25)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e2.37\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.12\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u003cem\u003ePICC placement times\u003c/em\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e1\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e612(97.76)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e428(97.72)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e184(97.87)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd rowspan=\"2\" align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.05\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd rowspan=\"2\" align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.48\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u0026ge;\u0026thinsp;2\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e14(2.24)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e10(2.28)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e4(2.12)\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u003cem\u003eArm selected for insertion\u003c/em\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eRight\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e515(82.27)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e369(84.25)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e146(77.66)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd rowspan=\"2\" align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.33\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd rowspan=\"2\" align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.56\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eLeft\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e111(17.73)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e69(15.75)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e42(22.34)\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e\u003cem\u003eVein selected for insertion\u003c/em\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eBassilic vein\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e552(88.18)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e389(88.81)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e163(86.70)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd rowspan=\"4\" align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.10\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd rowspan=\"4\" align=\"char\" char=\".\"\u003e\n\u003cp\u003e0.76\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eBrachial vein\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e60(9.58)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e38(8.68)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e22(11.70)\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eCephalic vein\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e9(1.44)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e8(1.83)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e1(1.07)\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eMedian cubital vein\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e5(0.80)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"char\" char=\".\"\u003e\n\u003cp\u003e3(0.68)\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e2(0.53)\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003c/tbody\u003e\n\u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003eContinuous variables were expressed as means\u0026thinsp;\u0026plusmn;\u0026thinsp;SD for normally distributed variables and data that did not conform to a normal distribution are presented as median (interquartile range, IQR), whereas the counting data were expressed as (%). Number of underlying diseases: included the total diseases of diabetes, hypertension, coronary heart disease, infarcts, nephrotic syndrome, inflammatory bowel disease, liver disease, syphilis. Abbreviations: \u003cem\u003eBMI\u003c/em\u003e body mass index, \u003cem\u003eVTE\u003c/em\u003e Venous thromboembolism, \u003cem\u003eWBC\u003c/em\u003e white cell count, \u003cem\u003eNEUT\u003c/em\u003e neutrophils, HGB hemoglobin, \u003cem\u003ePLT\u003c/em\u003e platelets, \u003cem\u003eCRP\u003c/em\u003e C reactive protein, \u003cem\u003ePT\u003c/em\u003e Prothrombin Time, \u003cem\u003eAPTT\u003c/em\u003e activated partial thromboplastin time, \u003cem\u003eINR\u003c/em\u003e International normalized ratio, \u003cem\u003eFBG\u003c/em\u003e fibrinogen concentration, \u003cem\u003eD-D\u003c/em\u003e D-dimer, \u003cem\u003eVEGF\u003c/em\u003e the vascular endothelial growth factor, \u003cem\u003eTKI\u003c/em\u003e tyrosine kinase inhibitors.\u003c/p\u003e\n\u003cp\u003e*\u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.05\u003c/p\u003e\n\u003ch3\u003ePredictive Performance And Optimization Process\u003c/h3\u003e\n\u003cp\u003eWe constructed an SVM model with all the variables as input variables to predict the occurrence of PICC-related venous thrombosis, and the AUC is presented in Fig.\u0026nbsp;2(ROC curve analysis of the SVM models).Table\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003e presents the confusion matrix for the ML model. All indexes of the SVM model on the training set are 1, which indicates that the model correctly predicted almost 100% of the data. On the test set, the AUC value of the ML model was 0.95, MCC was 0.91 and F\u003csub\u003e1\u003c/sub\u003e was 0.95. These results indicated that the model could sufficiently predict which PICC-catheterized patients were at high risk of thrombosis.\u003c/p\u003e\n\u003cdiv class=\"gridtable\"\u003e\n\u003cdiv class=\"colspec\" align=\"left\"\u003e\u0026nbsp;\u003c/div\u003e\n\u003cdiv class=\"colspec\" align=\"left\"\u003e\u0026nbsp;\u003c/div\u003e\n\u003cdiv class=\"colspec\" align=\"left\"\u003e\u0026nbsp;Table 2. Confusion Matrix for Machine Learning\n\u003ctable border=\"1\"\u003e\n\u003ctbody\u003e\n\u003ctr\u003e\n\u003ctd colspan=\"5\" width=\"546\"\u003e\n\u003cp\u003e\u003cstrong\u003eTraining data\u003c/strong\u003e\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp; \u003cstrong\u003eStatistical analysis\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"103\"\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd colspan=\"2\" width=\"221\"\u003e\n\u003cp\u003ePredicted true\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"111\"\u003e\n\u003cp\u003eF\u003csub\u003e1\u003c/sub\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"111\"\u003e\n\u003cp\u003e1\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"103\"\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"111\"\u003e\n\u003cp\u003e1\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"111\"\u003e\n\u003cp\u003e0\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"111\"\u003e\n\u003cp\u003eMCC\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"111\"\u003e\n\u003cp\u003e1\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"103\"\u003e\n\u003cp\u003e1\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"111\"\u003e\n\u003cp\u003e326\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"111\"\u003e\n\u003cp\u003e0\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"111\"\u003e\n\u003cp\u003eAUC\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"111\"\u003e\n\u003cp\u003e1\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"103\"\u003e\n\u003cp\u003e0\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"111\"\u003e\n\u003cp\u003e0\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"111\"\u003e\n\u003cp\u003e322\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"111\"\u003e\n\u003cp\u003ePrecision\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"111\"\u003e\n\u003cp\u003e1\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"103\"\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"111\"\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"111\"\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"111\"\u003e\n\u003cp\u003eRecall\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"111\"\u003e\n\u003cp\u003e1\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"103\"\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"111\"\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"111\"\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"111\"\u003e\n\u003cp\u003eAccuracy\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"111\"\u003e\n\u003cp\u003e1\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd colspan=\"5\" width=\"546\"\u003e\n\u003cp\u003e\u003cstrong\u003eTest data\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp;\u0026nbsp; Statistical analysis\u003c/strong\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"103\"\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd colspan=\"2\" width=\"221\"\u003e\n\u003cp\u003ePredicted true\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"111\"\u003e\n\u003cp\u003eF\u003csub\u003e1\u003c/sub\u003e\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"111\"\u003e\n\u003cp\u003e0.95\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"103\"\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"111\"\u003e\n\u003cp\u003e1\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"111\"\u003e\n\u003cp\u003e0\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"111\"\u003e\n\u003cp\u003eMCC\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"111\"\u003e\n\u003cp\u003e0.91\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"103\"\u003e\n\u003cp\u003e1\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"111\"\u003e\n\u003cp\u003e133\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"111\"\u003e\n\u003cp\u003e8\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"111\"\u003e\n\u003cp\u003eAUC\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"111\"\u003e\n\u003cp\u003e0.95\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"103\"\u003e\n\u003cp\u003e0\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"111\"\u003e\n\u003cp\u003e5\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"111\"\u003e\n\u003cp\u003e132\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"111\"\u003e\n\u003cp\u003ePrecision\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"111\"\u003e\n\u003cp\u003e0.96\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"103\"\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"111\"\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"111\"\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"111\"\u003e\n\u003cp\u003eRecall\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"111\"\u003e\n\u003cp\u003e0.94\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd width=\"103\"\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"111\"\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"111\"\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"111\"\u003e\n\u003cp\u003eAccuracy\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd width=\"111\"\u003e\n\u003cp\u003e0.95\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eMCC: Matthew\u0026rsquo;s correlation coefficient\u003c/p\u003e\n\u003cp\u003eAUC: area under the receiver operating characteristic (ROC) curve\u003c/p\u003e\n\u003cp\u003eF\u003csub\u003e1\u003c/sub\u003e: F-measure\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv class=\"colspec\" align=\"left\"\u003eIn our study, GA was used and showed that the individual optimal objective function value tended to be stable when the contemporary number reached 7 and reached the optimal value when the algebra reached 13. The average objective function value of the population tended to be approximately 0.9 when the algebra reached 18, and the optimal algebra was 27, as shown in Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003e(Fitness of the genetic algorithm)which indicates that GA can converge quickly and stably.\u003c/div\u003e\n\u003c/div\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003ch3\u003eModel Interpretation And Visual Presentation\u003c/h3\u003e\n\u003cp\u003eTo determine the features that have the greatest influence on the prediction model, the SHAP summary diagram of the trained model was drawn, as shown in Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e4\u003c/span\u003e(SHAP summary plot of the top 20 features of the SVM model)with the top 20 features of the prediction model. This figure describes the relationships between the high and low values of elements in the training dataset and the SHAP values. According to the prediction model, the higher the characteristic SHAP value is, the higher the risk of PICC-related thrombosis. The SHAP correlation diagram in Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e5\u003c/span\u003e(Explanation of the prediction results for specific instances) can also be used to understand how each variable affects the output of the SVM prediction model for individual patients, where \u0026lsquo;outvalue\u0026rsquo; represents the predicted value of the patient and the \u0026lsquo;base value\u0026rsquo; represents the mean value of thrombosis risk. A and B show two such examples. Red indicates that the contribution of this feature is positive, blue indicates that the contribution of this feature is negative, and length indicates the influence. For example, in A, a low risk value was correctly predicted. For this patient, the major contribution in red was BMI, while the major contribution in blue was the activated partial prothrombin time. This patient is a non-thrombosis patient. The model also predicted the risk of thrombosis in a patient who actually experienced thrombosis. As shown in B, the predicted value of 0.74 is higher than the mean value of 0.4739. For this patient, the overall risk of thrombosis was high.\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eThe safety and quality of care of patients with PICC-related thrombosis are still significant challenges to the quality control of the medical system. PICC-related thrombosis assessment and prediction remains a clinical concern. Regardless of patient factors and catheter factors, the risk factors leading to the occurrence of PICC-related thrombosis are complex and variable. Therefore, we sought to develop a more complex algorithm model that takes complex clinical settings into account to predict the outcome of patient complications, aiming to accurately predict the occurrence of thrombosis in individuals.\u003c/p\u003e \u003cp\u003eIdentifying thrombosis risk factors is the first step in thrombosis assessment. Before developing the model, we finalized the collected variables through two rounds of expert letters and team discussions based on evidence. The selection of these variables followed a scientifically rigorous approach, and 30 variables, including patient, tumour, iatrogenic and catheter factors, were ultimately included. Although smoking and radiation therapy were not shown in the evidence summary, they may increase the risk of blood clots [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e], so we included these two variables in our study. In recent years, with the rapid development of biotechnology and the continuous optimization of oncology treatment regimens, an increasing number of patients are choosing targeted therapy at the early stage of cancer diagnosis. A meta-analysis on the use of bevacizumab for cardiovascular adverse events reported that therapy increases the risk of thrombosis [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]. At present, with the renewal of molecular targeted drugs, an increasing number of targeted drugs are being used in clinical practice. Experts suggest that targeted drug therapy should be considered. Due to the particularity of the PICC catheterization position, tumour tissue hyperplasia may oppress blood vessels, which affects the blood supply of the PICC catheterization vein. Therefore, in our study, the diaphragm muscle was used as an anatomical marker to classify the tumour types of patients, as shown in Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e2\u003c/span\u003e. Based on the above variables, this study built an SVM model to predict the risk of PICC-related thrombosis in patients.\u003c/p\u003e \u003cp\u003eWe constructed an SVM model based on ML, rather than traditional mathematical models, which can only answer linear regression questions. In ML, overfitting will degrade the prediction performance of the model [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e] and usually occurs when the model is too complex (i.e., contains too many parameters). In the process of building a model, appropriate measures should be taken to prevent overfitting. The SVM model itself uses a kernel function to map high-dimensional data to low-dimensional data and then divides the data based on a linear classifier, which allows SVM to have a very good fitting effect. Moreover, it can effectively solve ML problems with small sample sizes, solve nonlinear problems and has a strong generalization ability [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e]. The appropriate algorithm should be selected for different data types, and SVM can effectively process various types of data. Our model included a mixture of continuous data and discrete data, and the data were not linearly separable, so SVM was suitable for the data types used in this study.\u003c/p\u003e \u003cp\u003eIncreasingly, medical research has shown great interest in the use of ML for building predictive models [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e]. Some scholars have used SVM to establish prediction models and obtained a good prediction effect [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e]. However, for classification learning algorithms, the AUC is a better index to evaluate ML models [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e]. Theoretically, the closer the AUC is to 1, the better the model effect will be. The AUC of the prediction model was 1, which means that all the predictions were totally accurate. However, it is almost impossible to achieve a model with 100% accurate prediction, but achieving the best effect as possible is ideal. As a random global optimization algorithm, GA is probably one of the most widely known biological heuristic algorithms that can be used to design reliable and robust models for clinical and research purposes.\u003c/p\u003e \u003cp\u003eGA, which was first proposed by Professor Holland in 1975, was originally developed by drawing on some phenomena in evolutionary biology, such as heredity, mutation, natural selection and hybridization. As a kind of nondeterministic quasi-natural algorithm, this algorithm provides a new idea for the optimization of complex systems. To optimize the effect of ML models, the hyperparameter often needs to be set manually; however, the space of the hyperparameter is very large, so the selection of the hyperparameter is more of an empirical project. GA can effectively search for and determine the hyperparameter in the parameter space so that the model effect can approach the optimal performance more quickly and stably. In our study, GA was used to solve the optimal parameter solution of SVM with RBF as the kernel function. The AUC of the model in the training set was used as the fitness evaluation function of the whole population, and the parameters gamma and C were abstracts of chromosomes. Through natural selection and mutation to generate new life populations, the optimal model was trained and then used in the test set to verify the predictive ability of the model. A variety of thrombosis risk factors in complex and real clinical situations were included in this optimized model to try to find an optimal decision boundary to make the most reasonable classification judgement for patients.\u003c/p\u003e \u003cp\u003eFurthermore, the advantage of our study is that SHAP values were used to uncover the black box of ML. In the overall explanation of the model by SHAP, the 20 most important features to explain the model were given and arranged in sequence. Major surgery ranked first, which indicates that a history of major surgery may be the most important factor for thrombosis in patients with PICC-catheterized cancer. Evans\u0026rsquo;s study revealed that surgery longer than 1 hour is an important factor for thrombosis in patients with catheterization [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e].In addition, we exploratorily included the two variables of targeted therapy and radiotherapy in our study, and the significance of these indicators was also reflected in the model. This finding suggests that nurses should pay attention to the use of targeted drugs and radiotherapy in the process of PICC catheterization in patients. In the model, the site of the primary tumour was an important predictor of thrombosis, indicating that the exploratory use of the diaphragm as an anatomical site to identify the risk of PICC-related thrombosis in patients is desirable. Among the predictive variables, tumour metastasis was also an important variable. Tumour metastasis is an important manifestation of active cancer that greatly increases the risk of thrombosis in patients. Age and medical comorbidities were also found to be important features in the model, with the risk of thrombosis (for both the first episode and recurrence) increasing exponentially with age [\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e]. The reasons may include medical comorbidities, decreased mobility, and possible age-related changes in clotting.\u003c/p\u003e \u003cp\u003eAs early as 2007, American scholars Seeley and Chopra constructed a logistic regression model to form a PICC-related thrombosis scoring system. A higher risk of PICC-related thrombosis was associated with a higher score [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e, \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]. This study does not present a variety of complex mathematical equations but provides a visual way to explain the model. Model visualization can support clinicians and nurses in making decisions and making treatment recommendations for patients [\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e]. The significant influence of each feature on the relevant prediction for each particular patient can be seen from the SHAP value, and the risk of thromboembolism for each patient is evident when compared to the baseline value. It is also important to understand how these features interact when making predictions [\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e]. Nurses' understanding and mastery of the SHAP value diagram is conducive to the correct evaluation and clinical decision making. Future work will focus on connecting the model results with the hospital information system by enabling the information system to actively fetch patient information to input into the model. The results can be displayed as SHAP values in the information system, and nursing managers can identify the risk according to the results of patients to facilitate the allocation of nursing work.\u003c/p\u003e"},{"header":"Limitations","content":"\u003cp\u003eFirst, the data were collected from one centre, and the model was not verified by external data. In the future, data from other centres can be collected to popularize and verify this model. Second, there are many studies that clearly show that some variables, such as hyperhomocysteinemia and thrombophilia (e.g., factor V Leiden, protein C deficiency, protein S deficiency), are risk factors for thrombosis. Due to objective reasons, some inspection indicators are not included in routine examinations in China, and the model did not include some variables with effects on thrombogenesis. Further collection of variables may change some characteristics included in the model.\u003c/p\u003e"},{"header":"Conclusions","content":"\u003cp\u003eOur study used an ML method to construct an SVM forecasting model, and the parameters of the model were optimized by GA. The developed model achieved excellent results in the validation dataset and can guide nurses in the precise prediction of the risk of PICC-related thrombosis in patients. The SHAP values can be used to explain the influence of 20 important variables of the model, and a legend of the SHAP values of a single patient can be drawn. Personalized legends can help nurses quickly determine the risk of PICC-related thrombosis and the most important factors affecting the patient\u0026rsquo;s risk; thus, they could offer care plans and measures for each patient to ensure the orderly development of patient treatment and care.\u003c/p\u003e \u003cp\u003eNurses play an important role in observing disease changes. The risk factors discussed herein should be considered when nurses perform PICC placement. This study provides a predictive PICC-related thrombosis model based on ML to help nurses identify patients at high risk of PICC-related thrombosis. With the development of information technology, it is generally accepted that good algorithms should be used to fully mine data. We should keep pace with the times in the era of big data and provide good quality care for patients in the clinic.\u003c/p\u003e"},{"header":"Abbreviations","content":"\u003ctable border=\"1\" cellpadding=\"0\" cellspacing=\"0\" width=\"615\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" width=\"30.73170731707317%\"\u003e\n \u003cp\u003e\u003cstrong\u003eabbreviation\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" width=\"69.26829268292683%\"\u003e\n \u003cp\u003e\u003cstrong\u003efull name\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" width=\"30.73170731707317%\"\u003e\n \u003cp\u003ePICC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" width=\"69.26829268292683%\"\u003e\n \u003cp\u003ePeripherally Inserted Central Cather\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" width=\"30.73170731707317%\"\u003e\n \u003cp\u003eSVM\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" width=\"69.26829268292683%\"\u003e\n \u003cp\u003eSupport Vector Machine\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" width=\"30.73170731707317%\"\u003e\n \u003cp\u003eMCC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" width=\"69.26829268292683%\"\u003e\n \u003cp\u003eMatthew\u0026rsquo;s Correlation\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" width=\"30.73170731707317%\"\u003e\n \u003cp\u003eROC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" width=\"69.26829268292683%\"\u003e\n \u003cp\u003eReceiver Operating\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" width=\"30.73170731707317%\"\u003e\n \u003cp\u003eAUC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" width=\"69.26829268292683%\"\u003e\n \u003cp\u003eArea Under the Curve\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" width=\"30.73170731707317%\"\u003e\n \u003cp\u003eSHAP\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" width=\"69.26829268292683%\"\u003e\n \u003cp\u003eSHapley Additive\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" width=\"30.73170731707317%\"\u003e\n \u003cp\u003eLASSO\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" width=\"69.26829268292683%\"\u003e\n \u003cp\u003eLeast Absolute Shrinkage\u003c/p\u003e\n \u003cp\u003eand Selection Operator\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" width=\"30.73170731707317%\"\u003e\n \u003cp\u003eRF\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" width=\"69.26829268292683%\"\u003e\n \u003cp\u003eRandom Forest\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" width=\"30.73170731707317%\"\u003e\n \u003cp\u003eSMOTE\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" width=\"69.26829268292683%\"\u003e\n \u003cp\u003eSynthetic Minority\u003c/p\u003e\n \u003cp\u003eOversampling Technique\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" width=\"30.73170731707317%\"\u003e\n \u003cp\u003eRBF\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" width=\"69.26829268292683%\"\u003e\n \u003cp\u003eRadial Basis Function\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e"},{"header":"Declarations","content":"\u003cp\u003eThe institutional and licensing committee approving the experiments.\u003c/p\u003e\n\u003cp\u003e-All methods were carried out in accordance with relevant guidelines and regulations.\u003cbr\u003e\u0026nbsp;-Informed consent was obtained from all subjects and their legal guardian(s).\u003c/p\u003e\n\u003cp\u003e(This study was approved by the Ethics Committee of Jiangsu University Hospital under the ethics number SWYXLL20200121-9. This study is a retrospective study, conducted in accordance with the principles of the Declaration of Helsinki, only collects clinical data of patients, does not damage the psychological and physiological operation or behavior of patients, numbering patients does not involve the personal privacy such as patient name, so as to effectively protect the security of patient information.)\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEthics approval and consent to participate\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe study was approved by the Research Ethics Board of the Affiliated Hospital of Jiangsu University, on January 01, 2020. Ethical Approval number: SWYXLL20200121-9.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent for publication\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAvailability of data and materials\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAll data generated or analysed during this study are included in this published article [and its supplementary information files].\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting interests\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare no conflicts of interest\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study was funded by Management Innovation Research Project of Jiangsu Hospital Association (JSYGY-3-2019-360) and Zhenjiang Soft Science Research Project (RK2019029).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthors\u0026apos; contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eSongmei Cao\u0026nbsp;，Shuhua Wang\u0026nbsp;：Collection of topic selection and thesis writing; Bo Cheng，Li Li : Statistics and analysis; Liqun Zhu，Aiping Li，Hong Zhu: Organize and participate in expert correspondence and discussion; Cao Songmei: Review and finalize the papers to be published; Yimeng Fan，Yiqing Liang: Participated in data collection and paper revision All the authors reviewed the manuscript.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgments\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eHoshal VJ. Total intravenous nutrition with peripherally inserted silicone elastomer central venous catheters. Arch Surg. 1975;110(5):644\u0026ndash;6. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1001/archsurg.1975.01360110190032\u003c/span\u003e\u003cspan address=\"10.1001/archsurg.1975.01360110190032\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChopra V, et al. The Michigan Risk Score to predict peripherally inserted central catheter-associated thrombosis. J Thromb Haemost. 2017;15(10):1951\u0026ndash;62. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1111/ijlh.12426\u003c/span\u003e\u003cspan address=\"10.1111/ijlh.12426\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLoughran SC, Borzatta M. Peripherally inserted central catheters: a report of 2506 catheter days. JPEN J Parenter Enteral Nutr. 1995;19(2):133\u0026ndash;6. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1177/0148607195019002133\u003c/span\u003e\u003cspan address=\"10.1177/0148607195019002133\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNg PK, et al. Peripherally inserted central catheters in general medicine. Mayo Clin Proc. 1997;72(3):225\u0026ndash;33. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://doi.org/10.4065/72.3.225\u003c/span\u003e\u003cspan address=\"10.4065/72.3.225\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJohansson E, et al. Advantages and disadvantages of peripherally inserted central venous catheters (PICC) compared to other central venous lines: a systematic review of the literature. Acta Oncol. 2013;52(5):886\u0026ndash;92. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3109/0284186X.2013.773072\u003c/span\u003e\u003cspan address=\"10.3109/0284186X.2013.773072\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTaxbro K, et al. Clinical impact of peripherally inserted central catheters vs implanted port catheters in patients with cancer: an open-label, randomised, two-centre trial. Br J Anaesth. 2019;122(6):734\u0026ndash;41. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.bja.2019.01.038\u003c/span\u003e\u003cspan address=\"10.1016/j.bja.2019.01.038\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFallouh N, et al. Peripherally Inserted Central Catheter-associated Deep Vein Thrombosis: A Narrative Review. Am J Med. 2015;128(7):722\u0026ndash;38. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.amjmed.2015.01.027\u003c/span\u003e\u003cspan address=\"10.1016/j.amjmed.2015.01.027\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBurns KE, McLaren A. Catheter-related right atrial thrombus and pulmonary embolism: a case report and systematic review of the literature. Can Respir J. 2009;16(5):163\u0026ndash;5. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1155/2009/751507\u003c/span\u003e\u003cspan address=\"10.1155/2009/751507\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHinck SM. Implementing the Infusion Therapy Standards of Practice. Home Healthc Now. 2021;39(5):295. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://doi:10.1097/NHH.0000000000001002\u003c/span\u003e\u003cspan address=\"http://doi:10.1097/NHH.0000000000001002\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSeeley MA, Santiago M, Shott S. Prediction tool for thrombi associated with peripherally inserted central catheters. J Infus Nurs. 2007;30(5):280\u0026ndash;6. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1097/01.NAN.0000292570.62763.3f\u003c/span\u003e\u003cspan address=\"10.1097/01.NAN.0000292570.62763.3f\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBeam AL, Kohane IS. Translating Artificial Intelligence Into Clinical Care. JAMA. 2016;316(22):2368\u0026ndash;9. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1001/jama.2016.17217\u003c/span\u003e\u003cspan address=\"10.1001/jama.2016.17217\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAn R, et al. Machine learning-based patient classification system for adult patients in intensive care units: A cross-sectional study. J Nurs Manag. 2021;29(6):1752\u0026ndash;62. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1111/jonm.13284\u003c/span\u003e\u003cspan address=\"10.1111/jonm.13284\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu S, et al. Machine learning approaches for risk assessment of peripherally inserted Central catheter-related vein thrombosis in hospitalized patients with cancer. Int J Med Inform. 2019;129:175\u0026ndash;83. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.ijmedinf.2019.06.001\u003c/span\u003e\u003cspan address=\"10.1016/j.ijmedinf.2019.06.001\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNensa F, Demircioglu A, Rischpler C. Artificial Intelligence in Nuclear Medicine. J Nucl Med. 2019;60(Suppl 2):29. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.2967/jnumed.118.220590\u003c/span\u003e\u003cspan address=\"10.2967/jnumed.118.220590\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. S-37S.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLundberg SM, Lee S. Consistent feature attribution for tree ensembles. 2017. \u003cem\u003earXiv preprint arXiv:1705.07874.\u003c/em\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLundberg SM, Erion GG, Lee S. Consistent individualized feature attribution for tree ensembles. 2018. \u003cem\u003earXiv preprint arXiv:1802.03888.\u003c/em\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCollins GS, et al. Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD): the TRIPOD statement. BMJ. 2015;350. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003eg7594.https://doi.org/10.1016/j.eururo.2014.11.025\u003c/span\u003e\u003cspan address=\"g7594.10.1016/j.eururo.2014.11.025\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhu LZRC. Evidence summary for risk assessment of PICC-related venous thrombosis. Chin J Nurs. 2017;52(10):1179\u0026ndash;85. https://doi.org/10.37610.issn.0254\u0026ndash;1769.2017.10.005.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBates SM, et al. Diagnosis of DVT: Antithrombotic Therapy and Prevention of Thrombosis, 9th ed: American College of Chest Physicians Evidence-Based Clinical Practice Guidelines. Chest. 2012;141(2). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1378/chest.11-2299\u003c/span\u003e\u003cspan address=\"10.1378/chest.11-2299\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. Suppl): p. e351S-e418S.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eD'Orazio M, et al. Deciphering Cancer Cell Behavior From Motility and Shape Features: Peer Prediction and Dynamic Selection to Support Cancer Diagnosis and Therapy. Front Oncol. 2020;10:580698. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e.https://doi.org/10.3389/fonc.2020.580698\u003c/span\u003e\u003cspan address=\".10.3389/fonc.2020.580698\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAgeno W, et al. Cardiovascular risk factors and venous thromboembolism: a meta-analysis. Circulation. 2008;117(1):93\u0026ndash;102. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1161/CIRCULATIONAHA.107.709204\u003c/span\u003e\u003cspan address=\"10.1161/CIRCULATIONAHA.107.709204\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTotzeck M, Mincu RI, Rassaf T. Cardiovascular Adverse Events in Patients With Cancer Treated With Bevacizumab: A Meta-Analysis of More Than 20 000 Patients. J Am Heart Assoc. 2017;6(8). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1161/JAHA.117.006278\u003c/span\u003e\u003cspan address=\"10.1161/JAHA.117.006278\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDeo RC. Machine Learning in Medicine. Circulation. 2015;132(20):1920\u0026ndash;30. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1161/CIRCULATIONAHA.115.001593\u003c/span\u003e\u003cspan address=\"10.1161/CIRCULATIONAHA.115.001593\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJin C, Wang L. Dimensionality dependent PAC-Bayes margin bound. 2012: Lake Tahoe, NV, United states. p.\u0026nbsp;1034\u0026ndash;1042.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHandelman GS, et al. eDoctor: machine learning and the future of medicine. J Intern Med. 2018;284(6):603\u0026ndash;19. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1111/joim.12822\u003c/span\u003e\u003cspan address=\"10.1111/joim.12822\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFerroni P, et al. Risk Assessment for Venous Thromboembolism in Chemotherapy-Treated Ambulatory Cancer Patients. Med Decis Making. 2017;37(2):234\u0026ndash;42. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1177/0272989X16662654\u003c/span\u003e\u003cspan address=\"10.1177/0272989X16662654\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHuang J, Ling CX, IEEE TRANSACTIONS ON KNOWLEDGE AND DATA ENGINEERING. Using AUC and accuracy in evaluating learning algorithms., 2005. 17(3): p.299\u0026ndash;310.\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1109/TKDE.2005.50\u003c/span\u003e\u003cspan address=\"10.1109/TKDE.2005.50\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eEvans RS, et al. Risk of symptomatic DVT associated with peripherally inserted central catheters. Chest. 2010;138(4):803\u0026ndash;10. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1378/chest.10-0154\u003c/span\u003e\u003cspan address=\"10.1378/chest.10-0154\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGregson J, et al. Cardiovascular Risk Factors Associated With Venous Thromboembolism. JAMA Cardiol. 2019;4(2):163\u0026ndash;73. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1001/jamacardio.2018.4537\u003c/span\u003e\u003cspan address=\"10.1001/jamacardio.2018.4537\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSun W, et al. Development and validation of two aspiration prediction models in patients receiving nasogastric feeding. J Nurs Manag. 2020;28(6):1372\u0026ndash;80. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1111/jonm.13093\u003c/span\u003e\u003cspan address=\"10.1111/jonm.13093\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLi X, et al. A Time-Phased Machine Learning Model for Real-Time Prediction of Sepsis in Critical Care. Crit Care Med. 2020;48(10). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ep. e884-e888.https://doi.org/10.1097/CCM.0000000000004494\u003c/span\u003e\u003cspan address=\"p. e884-e888.10.1097/CCM.0000000000004494\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\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":"Machine learning, Catheterization, Peripheral, Venous thromboembolism, Support vector machine, Prediction model.","lastPublishedDoi":"10.21203/rs.3.rs-2559468/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-2559468/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003e\u003cstrong\u003eBackground:\u003c/strong\u003e The impact of PICC-related thrombosis is worth paying attention to, and it is important to predict the risk factors for thrombosis in patients with PICC catheterization， accurate scientific assessment tools are critical forpredictingand preventing thrombosis in patients with PICCs.The main objective is to develop and validate a machine learning model for predicting the risk of peripherally inserted central catheter-related venous thrombosis.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMethods: \u003c/strong\u003eOverall, 626 patients undergoing peripherally inserted central catheter placement from January 2016 to October 2020 were enrolled. The variables included patient demographic characteristics, clinical condition, laboratory examinations, treatment, and catheter-related factors. Support vector machine and genetic algorithm were used to develop and optimize the model, respectively. SHapley Additive exPlanations was used to interpret the model.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eResults: \u003c/strong\u003eThe model obtained an average area under the receiver operating characteristic curve of 0.95. The SHapley Additive exPlanations summary plot was used to illustrate the effects of the top 20 features from support vector machine. This study provides a visual way to illustrate the impact of input features on the result prediction.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConclusions:\u003c/strong\u003e The machine learning model developed based on genetic algorithm shows good predictive ability in patients with a high risk of thrombosis-related peripherally inserted central catheter.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTrial registration:\u003c/strong\u003eretrospectively registered.\u003c/p\u003e","manuscriptTitle":"A machine learning model for the prediction of peripherally inserted central catheter-related venous thrombosis among high-risk adult patients","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2023-02-14 21:38:12","doi":"10.21203/rs.3.rs-2559468/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":"0f2bf18a-0439-4917-9b1c-57aa2e5a3e51","owner":[],"postedDate":"February 14th, 2023","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2023-03-01T14:59:29+00:00","versionOfRecord":[],"versionCreatedAt":"2023-02-14 21:38:12","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-2559468","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-2559468","identity":"rs-2559468","version":["v1"]},"buildId":"_2-kVJe1T_tPrBINL-cwx","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}