Development and validation of a nomogram to assess postoperative venous thromboembolism risk in patients with stage IA non‐small cell lung cancer

Abstract Background Venous thromboembolism (VTE) is a common postoperative complication in patients with lung cancer that seriously affects prognosis and quality of life. At present, the detection rate of patients with early‐stage lung cancer is increasing, but there are few studies on the risk factors for postoperative venous thromboembolism (VTE) in patients with stage IA non‐small cell lung cancer (NSCLC). This study aimed to establish a nomogram for predicting the probability of postoperative VTE risk in patients with stage IA NSCLC. Methods The clinical data of 452 patients with stage IA NSCLC from January 2017 to January 2022 in our center were retrospectively analyzed and randomly divided into a training set and a validation set at a ratio of 7:3. Independent risk factors were identified by univariate and multivariate logistic regression analyses, and a nomogram was established based on the results and internally validated. The predictive power of the nomogram was evaluated by receiver operating characteristic curve (ROC), calibration curve, and decision curve analysis (DCA). Results The nomogram prediction model included three risk factors: age, preoperative D‐dimer, and intermuscular vein dilatation. The areas under the ROC curve of this predictive model were 0.832 (95% CI: 0.732–0.924) and 0.791 (95% CI: 0.668–0.930) in the training and validation sets, respectively, showing good discriminative power. In addition, the probability of postoperative VTE occurrence predicted by the nomogram was consistent with the actual occurrence probability. In the decision curve, the nomogram model had a better net clinical benefit at a threshold probability of 5%–90%. Conclusion This study is the first to develop a nomogram for predicting the risk of postoperative VTE in patients with stage IA NSCLC; this nomogram can accurately and intuitively evaluate the probability of VTE in these patients and help clinicians make decisions on prevention and treatment.


| INTRODUCTION
Venous thromboembolism (VTE), which mainly includes deep venous thrombosis (DVT) and pulmonary thromboembolism (PE), is one of the most common complications of malignant tumors. Compared with the general population, cancer patients have a significantly increased risk of VTE, 1,2 especially lung cancer, which currently has the highest incidence of VTE. 3,4 In addition, surgical resection undoubtedly increases the probability of VTE in patients during lung cancer treatment. The occurrence of postoperative VTE events affects the treatment of the primary disease and significantly increases the risk of death of patients. 5,6 According to previous studies, the incidence of VTE after lung cancer surgery is as high as 8.1%-23.1%. [7][8][9] Therefore, we need to pay attention to the early prevention and diagnosis of thrombosis in patients after surgery. At present, to identify the high-risk groups of VTE as early as possible, a variety of VTE risk assessment models (RAM) have been proposed, including Caprini RAM, 10 Padua RAM, 11 Khorana RAM, 12 Rogers RAM, 13 etc. Different risk assessment models are used for different populations. Among them, the modified Caprini RAM 10 is widely used in thoracic surgery and has a good stratification effect. However, with the widespread application of low-dose, high-resolution spiral CT in the screening of lung cancer and novel coronavirus pneumonia in recent years, 14 the detection rate of early-stage lung cancer has increased significantly, resulting in a great change in the characteristics of the lung cancer population in current thoracic surgery. Whether the modified Caprini risk assessment model is equally applicable to patients with early-stage lung cancer is unclear. In our center, we previously conducted a preliminary study on VTE incidence after surgery for stage IA NSCLC and reported that the number of lymph nodes removed was an independent risk factor. 15 However, in the era of precision medicine, it is essential to identify additional high-risk factors for VTE in early-stage lung cancer patients, especially in patients with stage IA lung cancer, and it is urgent to establish an accurate postoperative VTE prediction model.
A nomogram is an intuitive and visual prediction tool that can accurately predict each patient's risk probability of clinical events. [16][17][18] A nomogram can make the prediction model more targeted by adding useful potential biomarkers, thereby achieving precise and personalized treatment. Nomograms have been applied to studies of various malignancies and have proven to be a reliable tool for predicting cancer prognosis. 19,20 In addition, its accuracy has been confirmed in predicting postoperative VTE of gynecological tumors, spinal metastatic tumors, and breast cancer. 17,21,22 However, there is currently a lack of accurate predictive models for postoperative VTE in stage IA NSCLC. Therefore, this study attempts to establish a nomogram graph model to predict the risk probability of postoperative VTE in patients with stage IA NSCLC. Through this model, clinicians can accurately identify patients with a high risk of postoperative VTE and can vividly show the probability of postoperative VTE to patients.

| Patient enrollment
We retrospectively analyzed 452 patients with stage pIA lung cancer who underwent lung resection at our center from January 2017 to January 2022. Inclusion criteria: (1) Primary NSCLC with pathological stage IA; (2) no VTE before surgery; and (3) complete clinical data. Exclusion criteria: (1) Patient refused surgical treatment; (2) benign lesions or nonstage IA NSCLC; (3) VTE before surgery; (4) perioperative use of anticoagulant drugs; (5) presence of hematological diseases; and (6) no lower extremity vascular ultrasound examination was performed after operation, or other data were missing. All patients were pathologically staged according to the eighth edition of lung cancer staging published by the American Joint Committee on Cancer/International Union Against Cancer. 23 The follow-up period ended when the patient was discharged.

| Outcome and variables
The outcome variable in this study was the occurrence of VTE events before discharge in patients with stage IA NSCLC after surgery. All patients in our study underwent lower extremity vascular ultrasound examinations by two professional sonographers before and after the operation to determine whether there was DVT. Computed tomography pulmonary angiography (CTPA) should be performed to determine whether PE exists if the patient has K E Y W O R D S non-small cell lung cancer, stage IA, venous thromboembolism, nomogram, caprini risk assessment model typical symptoms of PE (chest pain, hemoptysis, dyspnea, or persistent hypoxemia), a high Caprini score (≥9), or a newly diagnosed postoperative DVT.
The following data were recorded through the electronic medical record system: basic information included age, sex, length of hospital stay, body mass index (BMI); surgical information included surgical approach, the extent of resection, operation time, bleeding, and the number of lymph nodes removed; coagulation function indicators included platelet (PLT), activated partial thromboplastin time (APTT), prothrombin time (PT), and D-dimer levels preoperatively and postoperatively; preoperative pulmonary function indicators included forced exhalation in the first, second volume (FEV1), forced vital capacity (FVC), and maximum ventilation volume (MVV); imaging information included nodule morphology, nodule location, and intermuscular vein dilation (IVD); and pathological information included pathological diameter, pathological type, and T stage.

| Nomogram construction and validation
All enrolled patients were randomly divided into a training set (n = 318) and a validation set (n = 134) at a ratio of 7:3 by computer. This ratio can ensure the maximum utilization of samples and make the validation set have a sufficient sample size. Previous studies have confirmed this ratio. 21,24,25 Univariate and multivariate logistic regression analyses were used in the training set to determine the independent risk factors for postoperative VTE, and finally, a nomogram was constructed based on the factors screened out by the multivariate logistic regression results. The area under the curve (AUC) and C-index were used to discriminate the accuracy of the model for distinguishing between VTE and non-VTE patients. The calibration curve was applied to describe the consistency between the predicted probability and the actual probability. The closer the two lines are, the closer the predicted incidence rate is to the actual incidence rate, indicating that the nomogram prediction model has a better consistency. Finally, decision curve analysis (DCA) was used to evaluate the net benefit of patients and clinical utility by quantifying the net benefit at different threshold probabilities in the validation set.

| Statistical analyses
All continuous variables were compared between groups using t-test and are described as the mean ± standard deviation. All categorical variables were compared between groups using the chi-square test and are described as proportions. Receiver operating characteristic (ROC) curves were plotted by the "pROC" R package. Nomograms were constructed using the "rms" R package, and calibration plots were drawn using the "rms" R package using 1000 bootstrap resampling and obtained C-index. Clinical decision curves were drawn using the "rmda" package. Statistical analysis was performed using IBM SPSS Statistics 26.0 and R 4.0.3 (version 4.0.3; http://www.Rproj ect.org), and a two-sided p value <0.05 was considered significant.

| Patient characteristics
A total of 452 patients were finally included, including 179 males and 273 females, with an average age of 57.86 ± 10.15 years. Forty (8.9%) patients had postoperative VTE events, all of which were DVTs. Patients were randomly divided into 318 patients (70%) in the training set and 134 patients (30%) in the validation set at a ratio of 7:3. The screening process is shown in the flowchart ( Figure 1). Except for the extent of resection (p = 0.027), pathological diameter (p = 0.015), and pathological type (p = 0.033), the other characteristics of the two groups were comparable, and there was no significant difference in distribution. The demographic and clinicopathological characteristics of the two groups are shown in Table 1.

| Nomogram construction
A nomogram was constructed from the results of multivariate logistic regression analysis of the training set ( Figure 2). The nomogram showed that age had the greatest influence on postoperative VTE, followed by preoperative D-dimer and intermuscular vein dilatation. Each parameter of each factor in these variables is assigned a score on the model, and a cumulative total risk score is finally calculated. Then, drawing a straight line down can intuitively estimate the probability of postoperative VTE for each patient.

| Calibration and validation of the nomogram
Receiver operating characteristic (ROC) curves were plotted in the training and validation sets. The AUC values of the modified Caprini RAM were 0.728 (95% CI: 0.639-0.816) and 0.598 (95% CI: 0.440-0.755). The results are shown in Figure 3A and B. The AUC values of the nomogram in the training set and validation set were 0.832 (95% CI: 0.732-0.924) and 0.791 (95% CI: 0.668-0.930), respectively, which were higher than those of the modified Caprini RAM. It is suggested that the nomogram prediction model has better discriminative power, and the results are shown in Figure 3C and D. In addition, the calibration curve was applied to evaluate the agreement between the predicted probability and the true probability of the nomogram prediction model. The results are shown in Figure 4A   fitting consistency with the real curve. The C-index values are 0.832 and 0.790, respectively, indicating that the model has significant discrimination.

| Clinical use
The DCA of the nomogram in the validation set is shown in Figure 5. In the decision curve, the nomogram for predicting VTE showed more benefit than all or none if the patient's threshold probability was between 5% and 90%.

| DISCUSSION
Surgery remains the mainstay of treatment for NSCLC. VTE is a common complication after surgery in patients with NSCLC, and its occurrence significantly increases patient mortality. [3][4][5][6] The number of patients with stage IA NSCLC undergoing thoracic surgery has increased rapidly in recent years. 26,27 However, there are few studies on the incidence and risk factors for postoperative VTE in these patients. It is necessary to better understand the risk factors for postoperative VTE in such patients. In addition, establishing a good VTE prediction model can help clinicians accurately identify high-risk patients with VTE promptly. This study attempted to establish a reliable nomogram of 452 patients to predict postoperative VTE in stage IA NSCLC to predict the risk probability of thrombosis in these patients accurately. The nomogram prediction model established in this study included age, preoperative D-dimer, and intermuscular vein dilation. It has accurate prediction power, with a C-index of 0.832, and has been internally verified. The results also confirmed that the model has a good performance of distinction. A nomogram is a method to accurately predict the probability of occurrence of each individual outcome event by integrating multiple predictors. Previous studies have confirmed that nomograms can accurately predict the risk of VTE in patients. 17,21,22 Wang et al. 17 used age, D-dimer, body mass index (BMI), and a surgical approach to establish a nomogram to accurately predict the probability of postoperative VTE in patients with gynecological malignancies, and the C-index was 0.721. Similarly, Zhang et al. 21 and Li et al. 22 used a nomogram to establish a predictive model for postoperative VTE of spinal metastatic tumors and postoperative VTE of breast cancer, respectively, which can achieve individualized and accurate prediction.
It is well known that age is an important influencing factor for VTE, and the results of our study show that age is still an independent risk factor for stage IA NSCLC. The risk of postoperative VTE increased by 1.095 times for each additional 10 years of age, consistent with previous lung cancer-related studies. [28][29][30] Among them, Wang et al. 30 showed that the incidence of postoperative VTE in patients older than 60 years was significantly higher than that in patients younger than 60 years (17.1% vs. 6.3%). In addition, in patients undergoing surgery for gynecological tumors or breast cancer, advanced age is also a high-risk factor for postoperative VTE. 17,21,22 D-dimer is a degradation product of fibrin and is a sensitive indicator for judging fibrinolytic status and coagulation function. Several studies 28,[30][31][32] have confirmed that D-dimer is an independent risk predictor of VTE. The results of this study are highly consistent with those of previous studies, 9,22,31,32 all of which suggested that pretreatment D-dimer elevation significantly increased the risk of VTE in patients. Tian et al. showed that the AUC value of preoperative D-dimer in VTE after lung resection was 0.70 and suggested that patients with elevated preoperative D-dimer should undergo VTE prophylaxis. 9 In addition, Li et al.'s study 22 confirmed that with an increase in the preoperative D-dimer value to varying degrees, the risk of VTE in breast cancer patients after surgery increased by 0.831-4.036 times. A previous prospective, the singlecenter study also confirmed that increased D-dimer in patients before chemotherapy was significantly associated with a higher risk of VTE after chemotherapy. 32 The results of our study suggest that preoperative D-dimer elevation is an independent risk factor for stage IA NSCLC. Therefore, paying attention to the level of D-dimer before treatment can help clinicians identify some patients with a prothrombotic state in time.
Calf intermuscular vein dilatation is also an independent factor for postoperative VTE in patients with stage IA NSCLC. Most of the patients with VTE events in this study had calf intermuscular vein thrombosis, a specific type of DVT. 33 Therefore, calf intermuscular vein dilatation was considered a risk predictor for research analysis. The calf intermuscular venous plexus has many branches, a small diameter, a thin wall, few venous valves, slow blood flow, and the absence of deep fascia and other hard tissues around it; thus, it is also prone to thrombosis. Recent studies related to lung cancer have shown that the incidence of VTE in patients with varicose veins of the lower extremities is approximately five times that of patients without varicose veins. 34 In addition, a previous study on risk factor analysis of postoperative VTE in patients with lung cancer showed that calf intermuscular vein dilatation was an independent risk factor for postoperative VTE after lung cancer, which was consistent with the results of our study. 35 This study shows that the modified Caprini RAM has certain limitations when evaluating postoperative VTE in patients with stage IA NSCLC, and the nomogram has excellent discriminative ability. When a nomogram predicts the probability of postoperative VTE in patients with stage IA NSCLC, it is not only convenient to apply but also has accurate and personalized results. In both datasets of this study, the AUC values of the nomogram were higher than those of the modified Caprini RAM, which had better predictive power. At present, the modified CapriniRAM is the most widely used in the field of thoracic surgery. It evaluates VTE risk in patients in the form of scores and has good risk stratification. A previous prospective study 7 confirmed that when lung resection patients were assessed by Caprini RAM, the incidence of VTE in the low-risk group, intermediate-risk group, and high-risk group was 0%, 12.3%, and 40.0%, respectively, with a good stratification effect and was validated in a recent multicenter study. 36  Bold values were all P < 0.05, indicating that the factor was statistically significant, and then these factors were included in the multivariate analysis. allows for the addition of potential biomarkers to achieve precise individualized predictions. Our study has the following limitations. First, singlecenter, retrospective studies reduce the generalizability  (A, B). The x-axis shows the model's predicted probability, and the y-axis shows the actual probability. of the model. Second, although the validation set in the study was randomly assigned, a large amount of multicenter data is still needed for external validation. Finally, patients after discharge were not considered in this study, which may underestimate the incidence of postoperative VTE in patients with stage IA NSCLC.

| CONCLUSION
In summary, our study retrospectively analyzed 452 patients with stage IA NSCLC, established a nomogram based on three independent risk factors to predict the risk probability of postoperative VTE in patients with stage IA NSCLC and validated the model, confirming its good prediction performance. We believe that the application of this model will allow clinicians to accurately assess the risk of postoperative VTE in patients with stage IA NSCLC, thereby providing patients with appropriate individualized prevention and treatment strategies.

AUTHOR CONTRIBUTIONS
Yongsheng Cai and Honghong Dong contributed to manuscript writing, acquisition of data and analysis, and interpretation of data; Jinbai Miao and Qirui Chen contributed to conception and design, revising it critically for important intellectual content. Xinyang Li and Yi Liu have been involved in figure preparation and acquisition of data. Bin Hu and Hui Li revises it critically for the manuscript; All authors gave final approval of the version to be published.