A model to assess the risk of peripherally inserted central venous catheter-related thrombosis in patients with breast cancer: a retrospective cohort study

Limited risk assessment tool to stratify the risk of PICC-related thrombosis (PICC-RVT) in breast cancer patients. This study developed a model to assess the risk of PICC-RVT in breast cancer patients. We conducted a retrospective cohort study of 1284 breast cancer patients receiving PICC insertion from January 1, 2015, to August 31, 2019, at a cancer specialized hospital in Hunan province, China. The entire population was divided into two groups at a ratio of 3:1 which included a derivation sample (n = 978), and a validation sample (n = 284). PICC-RVT was confirmed by ultrasonography in the presence of clinical symptoms and signs. PICC-RVT occurred in 40 (4.09%) of the derivation sample patients. Multivariable analysis identified 9 variables: chronic obstructive pulmonary disease, prior central venous catheter placement, higher level of platelets, higher level of D-dimer, lower level of activated partial thromboplastin time, menopause, no prior breast surgery, upper extremity lymphedema, and endocrine therapy. Points were assigned to each variable according to regression coefficient. The model had an area under the receiver operating characteristics curve (AUC) of 0.850 (95% CI 0.776 to 0.924), The Hosmer–Lemeshow goodness-of-fit was 5.780 (p = 0.328). At a cutoff value of 3.5, the sensitivity and specificity were 75% and 83%, respectively. Several disease-specific factors of breast cancer (e.g., menopause, endocrine therapy, and upper extremity lymphedema) play important roles in the development of PICC-RVT. Patients at higher PICC-RVT risk could be candidates for close post-insertion monitoring and interventions to prevent PICC-RVT.


Introduction
Breast cancer is the most common cancer in female patients [1]. With the development of medical service, the 5-year survival rate of breast cancer is as high as 90.9%. [2] Chemotherapy is one of the main treatment methods for breast cancer. It is very important for breast cancer treatment to ensure the safe infusion of chemotherapy drugs. Peripherally inserted central venous catheter (PICC) is a kind of catheter inserted from peripheral vein such as basilica or brachial vein to central venous, which can effectively prevent the extravasation of chemotherapy drugs, and provide a safe "life access" for breast cancer patients [3]. Despite these advantages, the application of PICC can also cause some catheter-related complications, among which PICC-related thrombosis (PICC-RVT) has been one of the most common and detrimental complications that could cause discomfort experience such as swelling and pain in the limb, and even cause pulmonary embolism which can Si-yi Peng and Tao Wei contributed equally to this work and should be considered as the joint first authors * Xu-ying Li lixuying@hnca.org.cn 1 The Early Clinical Trial Center in The Affiliated Cancer Hospital of Xiangya School of Medicine, Central South University/Hunan Cancer Hospital, Changsha, Hunan Province, China endanger patients' life [4]. Fortunately, it could be effectively prevented by pharmacological or non-pharmacological methods [5,6]. However, preventive anticoagulation therapy will increase the risk of bleeding; current guidelines do not recommend routine preventive anticoagulation therapy [7]. The latest consensus on the prevention and treatment of catheter-related thrombosis in China pointed out that preventive anticoagulant treatment was still necessary for people at high risk of thrombosis [8]. If so, how can we screen out the high-risk group of catheter-related thrombosis in clinical practice? Building predictive models may be one of the solutions. Many researchers have assessed PICC-RVT risk prediction models to guide medical staff to take measures as early as possible [9][10][11]. In recent 5 years, some prediction models have been constructed for PICC-RVT. Seeley and Chopra team from the USA reported PICC-RVT prediction model, respectively. Both of their target population was patients who had a PICC insertion, and most of them were cancer patients [9,10]. Since 2016, scholar from China began to explore the construction of predictive models for PICC-RVT. Hao's team first reported PICC-RVT risk prediction nomogram models for cancer patients [12]. However, these models have not been widely used, and their clinical effectiveness needs further exploration.
Nevertheless, it was lacking in breast cancer patients. For breast cancer patients after surgery, the affected side of the upper limb lymphatic reflux is blocked, it is not suitable for further puncture, and the main infusion procedure would be completed by the healthy side of the limb which makes it difficult for second catheter placement if the first catheter fails to infusion [13]. Besides, female's application of contraceptives, hormone replacement therapy, genetic, and environmental factors could lead to changes in vascular endothelial function, platelet activity, and fibrinolytic activity, and then result in an increased risk of thrombosis [14]. Furthermore, research showed that endocrine therapy and chemotherapy played roles in the development of venous thrombosis in breast cancer patients [15], and it may also cause catheter-related thrombosis.
At present, few research were reported on the risk factors of PICC-RVT in breast cancer patients [16,17], and the research on the specific PICC-RVT prediction tool for breast cancer patients has not been reported. It is necessary to build a PICC-RVT risk scoring system for breast cancer patients which would help guide the clinical decision-making regarding thromboprophylaxis strategies.

Study design and patients
A retrospective, cohort study was conducted in patients with breast cancer receiving PICC insertion from January

Data collection
We obtained the patients identity number from the vascular access clinic in which most of breast cancer patients had their PICC inserted. Then, we entered the hospital information system (HIS) to collect potential related factors for PICC-RVT, which consisted of four parts: (1) patient-related data: age, body mass index (BMI), ABO blood type, hypertension, diabetes, chronic obstructive pulmonary disease (COPD), smoking history, and central venous catheter (CVC) placement history; (2) laboratory test results before insertion: hemoglobin (Hb), platelet (PLT) count, fibrinogen (FIB) level, D-dimer, and activated partial thromboplastin time (APTT); (3) PICC insertion data: date of PICC insertion, insertion vein, insertion side, placement attempts, and catheter/vessel diameter; and (4) breast cancer-specific data: menopause, cancer stage, pathological type, metastasis, cancer side, lymphedema, use of anthracyclines, platinum-based drugs, trastuzumab, radiation therapy, and endocrine therapy. Trained nurses regularly checked missing data to ensure data quality. When patients reported arm pain or swelling limb, we would use ultra-sonography (Philipsiu22, Netherlands) for further diagnostic. All data were extracted by two investigators independently. To protect patient privacy and confidentiality, we just use the inpatient number as the unique identifier to extract the data needed for this study as listed above.

Primary outcome
The primary outcome was the incidence of symptomatic PICC-RVT, mainly included thrombophlebitis and deep vein thrombosis. This outcome was selected primarily because it can cause patients' discomfort, disable the catheter function, and delay the optimal timing of treatment. When patients reported arm pain or swelling limb, qualified sonographer would use ultrasonography (Philipsiu22, Netherlands) for further diagnostic.

Statistical analysis
Data analysis was performed using SPSS software (version 18, SPSS Inc, IBM, NY, USA). We randomly selected 75% of the total samples for model derivation, and the remaining 25% samples for model validation. Demographic and clinical variable were compared between the two sample groups. Then we abstracted the data of model derivation for further analysis. Demographic and clinical variable were screened for possible inclusion into the risk prediction model. The dependent variable in the analysis was the development of symptomatic PICC-RVT. Those with P value ≤ 0.10 in univariate analysis (including Fisher exact test, χ 2 test, and Mann-Whitney U test, as appropriate) were retained for further consideration. Multivariable logistic regression analysis was used in a stepwise entry process with the P-value set at < 0.05 to identify the relevant predictive factors.
From the logistic regression statistical outputs, the contribution of the individual PICC-RVT risk factors was weighted by beta-coefficients of the final model. To simplify calculations using these weights in the risk scoring system, the beta-coefficients were rounded to the nearest unit value. A summary PICC-RVT risk scoring system in breast cancer was then assigned to each patient by simply adding up transformed beta-coefficients values for each risk factor they possessed.
The predictive accuracy of the final risk scoring system was then determined by testing the specificity, sensitivity, and area under the receiver-operating characteristic (ROC) curve (AUC). We did these testing both in the case of model derivation and model validation. A predictive instrument with an AUC of ≥ 0.70 is considered to have good discrimination.

Sample characteristics
From January 1, 2015, to August 31, 2019, a total of 1375 PICCs were inserted in 1375 patients with breast cancer at the vascular access clinic. The final sample size included in this study was 1262 after exclusion of 113 cases due to the absence of catheter removal. All of the 1262 patients were female, with a median age of 47 years (interquartile range [IQR] 42-53 years). Two hundred and ninety-nine patients (23.69%) presented one or more comorbidities, including diabetes, hypertension, and COPD. Ninety-three percent of the patients (n = 1177) had an invasive ductal carcinoma, and 89.1% of patients (n = 1124) had breast surgery. Among them, 98.3% of the patients (n = 1105) had axillary lymph node surgical resection. All catheters in this study were made of silicon, three-way valve type, with single lumen.
According to the pre-set allocation ratio, 978 people were allocated in the derivation group and 284 people were arranged in the validation group. There was no statistical difference between the two groups in demographic data, PICC insertion data, medical history data, breast cancer-specific data, laboratory examination data, etc. Detailed data can be found in Table 1.

Characteristics of PICC-RVT in breast cancer patients
PICC-RVT developed in 4.09% (40/978) of the validation sample and in 3.52% (10/284) of the deviation sample; the overall incidence rate was 3.96% (50/1262). Among 50 cases of catheter-related thrombosis, 62% (n = 31) occurred in a single site, among which 38.7% (n = 12) were located in basilic vein alone, and 19.4% (n = 6) in axillary veins, 12.9% (n = 4) in internal jugular vein, and 12.9% (n = 4) in the subclavian vein. The median indwelling time of PICC was 122 days (IQR: 81-145 days). The shortest indwelling time of PICC was 1 day, and the longest was 396 days. The indwelling time for PICC-RVT ranged from 3 to 380 days, with median indwelling time 119.5 days (IQR: 50.5 ~ 172.5 days). The peak incidence of PICC-RVT occurred after post-insertion day 180 (range, 3 to 380 days). The detailed incidence of PICC-related thrombosis in different indwelling periods is shown in Table 2.

Univariate and multivariate analyses for PICC-RVT
The bivariable analysis comparing the risk factors in those with and without PICC-RVT in the derivation sample is displayed in Table 3. Univariate analysis screened out 14 variables related to PICC-RVT with P value ≤ 0.10, which included COPD, CVC placement history, Hb, PLT, D-dimer, FIB, APTT, menopause, tumor metastasis, breast surgery, upper limb lymphedema, anthracycline chemotherapy, platinum chemotherapy, and endocrine therapy.
After the initial univariate screening, the potential predictive factors associated with PICC-RVT were retained for further analysis using multivariable logistic regression analysis in a stepwise entry process with the P-value set at < 0.05 to identify the relevant predictive factors. There were nine final variables retained in the model that were considered significant PICC-RVT predictive factors, including CVC placement history, COPD, PLT, D-dimer, APTT, menopause, breast cancer surgery, upper limb lymphedema, and endocrine therapy (P < 0.05) ( Table 4). These variables were selected for the development of the prediction rule. In addition, the time of PICC retention could not be determined until PICC removal; therefore, it was not included in the risk prediction rule.

Derivation of the prediction rule
Based on the results of the multivariable model, a final prediction rule was developed by adding predictor score if it was present. Score values were the regression betacoefficients rounded to the nearest unit value for the specific parameter. An individual patients' score was calculated in the following manner: if a parameter was present, the corresponding score value was multiplied by 1, if not by 0. Adding these values' results were the patients' total score. Table 5 shows the parameters and the corresponding weight.
The final product was a scoring system with sum between 0 and 15, a higher score was associated with an increased risk for PICC-RVT in breast cancer. The model was continued with a ROC curve analysis and a test of the area under the curve (AUC). The findings indicated that the AUC was acceptable at 0.850 (95%CI: 0.776, 0.924), indicating a good accuracy of the scoring system (Fig. 1). Applying this scoring system to the deviation sample and using a cut-point of 3.5 or more to identify high-risk patients, sensitivity was 75% and specificity was 83.2%. The results of the Hosmer-Lemeshow goodness-of-fit test (5.780, p = 0.328) indicated that the model had a satisfactory prediction effect requiring no further calibration.

Validation of the scoring system
Applying the PICC-RVT risk scoring system derived from deviation sample to the validation sample and using a cut-point of 3.5 or more to identify high risk patients in PICC-RVT, we find the AUC was 0.882 (95% CI: 0.781, 0.984) (Fig. 1), with sensitivity 70% and specificity 84.7%, respectively.

Discussion
The study is the first to develop a simple and validated prediction tool to assess PICC-RVT in breast cancer patients, to our knowledge. Among the 1262 breast cancer patients included in this study, 50 developed symptomatic PICC-RVT; the incidence was 3.96%. This finding is consistent with the findings of other studies on the incidence of PICC-RVT in breast cancer patients [16][17][18]. Most studies about catheter-related thrombosis in breast cancer patients in recent years are retrospective studies, and there may be a certain degree of positive events missed, meaning that the incidence might be underestimated. In fact, we did underestimate its incidence. Research results from a prospective, multicenter cohort of breast cancer patients undergoing insertion of a CVC for chemotherapy showed that the catheter-related thrombosis cumulative probability was 9.6% at 3 months and 11.5% at 6 months. This showed a higher incidence than that in all retrospective studies [19]. Despite the inherent  limitations of this retrospective study design, the present study provided important reference for PICC-RVT risk prediction in breast cancer patients. The results of this study showed that the incidence of thrombosis in different time periods is different, showing a trend of first increasing, then decreasing, and then increasing. In PICC with an indwelling time of more than half a year, the incidence of catheter-related thrombosis reached 12.22%, which was far higher than the average incidence (3.96%) obtained in this study. It is suggested that the indwelling time of the catheter should not exceed 180 days. When the indwelling time of the catheter exceeds 180 days, it is necessary to strengthen the inspection of such patients. It should be noted that this study did not explore the relationship between indwelling time and thrombosis under the premise of controlling other variables. We just analyzed the incidence of PICC-RVT from different indwelling time interval; further research is needed in this topic.
The risk factors of catheter-related thrombosis can be divided into three aspects: patient related, catheter related, and operation and treatment related [3]. Patient-related factors are generally more difficult to control than the other two aspects; therefore, patient-related risk factors are mainly used to predict high-risk groups of thrombosis, which in turn prompts clinical attention to such groups of people. For breast cancer patients, former research indicated that invasive lobular carcinoma was more prone to catheter-related thrombosis [19], but it was shown no difference in our study. This may have the following two reasons: firstly, it may be because of the difference in the extent of catheter-related thrombosis (CRT). Outcome variable in former research included symptomatic and asymptomatic CRT, and the incidence of all CRT was higher than that in our study. Positive CRT case in invasive lobular carcinoma was enough for a significance level in the results of statistical analysis. Secondly, it may be related to the relatively small proportion of invasive lobular carcinoma in breast cancer in Chinese population [20]. Among the catheter-related factors, the diameter   [3,21]. Former research showed that the patients with a ratio of catheter to vein diameter above 0.45 had a higher risk of CRT [22]. In our study, we did not find any difference in the occurrence of CRT; this may because the cut-off value is not applicable to our data or to Chinese population. Some studies also indicated that the ABO blood type played roles in the development of CRT [23,24], but the results are currently controversial in research [25]. Research group from Australia revealed that patients undergoing PICC insertion with a blood group B appear to have a higher risk of venous thrombosis, while research from China showed that the risk of PICC-RVT with non-O blood type is significantly higher than that in patients with blood type O. In our study, we did not find any difference of CRT incidence between ABO groups; our data can provide a reference to the ongoing systematic review of this topic.
In this study, the specific characteristic of breast cancer was also included in the final model. The variables identified as being important predictors for PICC-RVT in breast cancer patients included COPD, history of CVC, PLT level, APTT level, D-dimer level, menopause, breast surgery, upper limb lymphedema, and endocrine therapy, all of which were objectively measurable, and the last four variables were the specific characteristic for breast cancer. The results showed that patients presenting with menopause were more likely to develop venous thrombosis; this may be attributed to menopausal hormone therapy [26]. The risk of PICC-RVT in breast cancer patients without breast surgery was three times than that in patients with breast surgery. This may because the tumor in the body activates the coagulation chain and weakens the fibrinolysis system when the breast tumor is not surgically removed, which would further lead to thrombosis development [27]. Most of breast cancer patients will have axillary lymph node dissection on the affected side, and a few people will have opposite lymph node dissection, which would cut off the pathway of lymphatic reflux and cause lymphedema [28]. This study found that the risk of PICC-RVT in breast cancer patients with lymphedema was 28 times than that without lymphedema. In case of lymphedema, a large amount of lymph leakage will cause blood concentration. Swelling of the limb will compress the blood vessels and slow down the blood flow. Besides, the limited mobility of the limbs with lymphedema increases the risk of thrombosis [29,30]. Meanwhile, lymphedema patients are prone to infection and inflammation [31]. It has been reported that the more serious the infection, the higher the risk thrombosis [32]. Fortunately, lymphoedema is becoming rare as lymph node sentinel technique has emerged and surgical axillary dissection has improved. It should be noted that for patients with breast cancer, the "non-affected side" for insertion should always be chosen and down written on the informed consent patient's form [33]. In this study, we found that the risk of PICC-related thrombosis in patients with endocrinotherapy was more than 4 times than that in patients without endocrinotherapy, which may explain the significantly increased risk of venous thrombosis [34,35]. Endocrine therapy is one of the main means of systemic treatment for breast cancer, and it is also the treatment method that making breast cancer specific to other tumors [36].
We developed the scoring system with score ≥ 3.5 indicated high risk of PICC-RVT in breast cancer patients. The results show that the AUC in both samples was greater than 0.8, and the sensitivity and specificity were more than 70%, and the Hosmer-Lemeshow goodness-of-fit test showed that there was no significant difference between the prediction results of PICC-RVT of the scoring systems and the actual results; the scoring system has acceptable prediction accuracy. The scoring system developed in the current study allows the corporation of breast cancer-specific factors that will identify patients at high risk and ensure they are receiving appropriate prophylaxis. The scoring system is applicable to breast cancer patients, which is the most common cancer in females. Besides, it can discriminate between high-and low-risk patients. The risk threshold for medical decision-making can also be shifted up or down, depending on a patients' or clinician's risk tolerance. The primary use of this tool as we see it is to modify current antithrombosis prophylaxis to avoid unnecessary thrombosis, identify patients who require additional education regarding PICC-RVT management, and to monitor current antiemetic protocols.
Despite the potential benefits of the PICC-RVT risk model in breast cancer patients, several limitations in the current study need to be acknowledged. Even though we have validated the risk scoring system, the fact that it was carried out in the same institution may lead to a certain degree of bias in the selection of patients. The model should undergo external prospective validation on a new sample of patients. In addition, as we applied a retrospective design to develop the model, it is generally accepted that this kind of research design could lead to underreporting the incidence of PICC-RVT, so the actual rate of PICC-RVT may be higher than what we reported.
Despite the above limitations, this study has important implications for future research and clinical practice. Patients who are stratified into high risk for PICC-RVT would benefit from frequent PICC-RVT screening and implementation of prevention strategies. Importantly, this PICC-RVT prediction rule in patients with breast cancer provides a method to stratify at-risk patients before insertion for such interventions to ultimately reduce PICC-RVT occurrence and cost of PICC-RVT.

Conclusion
In this study, we developed a PICC-RVT risk prediction scoring system in breast cancer patients. The disease-specific factors of breast cancer, such as endocrine therapy, lymphedema of upper extremity play important roles in the development of PICC-RVT. The clinical application of the scoring system will be an important source of individual patient risk information for the oncology clinician and may enhance patient care by optimizing the use of the anticoagulation in a proactive manner.