Comparison of baseline clinical characteristics between lung cancer patients with or without VTE

This study enrolled 158 participants that met the inclusion criteria. Thrombotic events were reported in 27 (17.1%) patients (VTE group) within six months after the diagnosis of lung cancer. The baseline clinical characteristics of the VTE group and the non-VTE group patients are summarized in Table 1. Overall, 76.6% (n = 121) of the study participants were male. There were no significant differences in the gender distribution between the two groups. The median age of patients diagnosed with thrombosis was 71.41 years, which was comparatively higher than the median age of patients in the non-thrombosis group (64.36 years). Furthermore, 29.6% (n = 8) of the patients in the VTE group showed a BMI ≥ 25 kg/m². Therefore, the median BMI of the VTE group was significantly higher than the BMI of patients in the non-VTE group. Lung cancer patients with advanced stage or distant metastases were more likely to develop VTE than those with localized lesions (92.6% vs. 67.9%, P = 0.009; Table 1). Furthermore, the number of patients with ECOG PS scores ≥ 2 were significantly higher in the VTE group compared to the non-VTE group (81.5% vs. 21.7%, p < 0.001). Besides, history of VTE was an important risk factor of lung cancer-associated thrombosis. The number of patients with history of VTE were significantly higher in the VTE group compared to the non-VTE group (37.0% vs. 4.6%, p < 0.001). Moreover, we did not observe significant differences in the incidence rates of thrombotic events between patients with non-small cell lung cancer (NSCLC) and those with small cell lung cancer (SCLC). However, the incidence rates of thrombotic events were significantly higher in the lung cancer patients with hypertension compared to those without hypertension (63.0% vs. 36.6%) in our study.

Table 1 Characteristics between the VTE group and the non-VTE group Identification of serum biomarkers associated with VTE

We then analyzed the relationship between thrombotic events and the coagulation-and thrombus-related biomarkers to identify the biomarkers associated with early diagnosis of thrombosis. Firstly, the Kolmogorov-Smirnov test results showed that all the continuous variables analyzed did not conform to normal distribution. Therefore, they were represented as median (interquartile range). The Mann-Whitney U test was used to analyze the differences in these variables between the VTE and the non-VTE groups. The serum levels of FDP, D-dimer, TM, TAT, PIC, and t-PAIC were significantly higher in the VTE group compared with the non-VTE group (P < 0.05) (Table 2).

Table 2 Comparison between the VTE group and the non-VTE group.(M [Q1,Q3]). Receiver Operating Curve (ROC) analysis of the diagnostic efficacy of the VTE-related biomarkers in the cohorts of lung cancer patients

The area under the curve (AUC) values of FDP, D-dimer, TM, TAT, PIC, and t-PAIC for diagnosing VTE were 0.869, 0.860, 0.817, 0.802, 0.745, and 0.685, respectively. The cut-off values for FDP, D-dimer, TM, TAT, PIC, and t-PAIC were 2.84 mg/L, 0.77 µg/mL, 9.75 TU/ml, 2.25 ng/ml, 0.80 µg/ml, and 7.35 ng/ml, respectively(Table 3; Fig. 1). The sensitivity values of FDP, D-dimer, TM, TAT, PIC, and t-PAIC for the diagnosis of VTE were 96.3%, 92.6%, 88.9%, 88.9%, 92.6% and 77.8%, respectively. The specificity values of FDP, D-dimer, TM, TAT, PIC, and t-PAIC for the diagnosis of VTE were 45.7%, 40.3%, 40.3%, 48.1%, 60.5%, and 45.0%, respectively. Furthermore, we found that the combined diagnostic efficacy of the six biomarkers was higher than the diagnostic efficacy of the individual biomarkers (AUC value of 0.912) (Fig. 2)

Table 3 Diagnostic efficiency of each biomarker in lung cancer-associated VTE. Fig. 1

Receiver operating characteristic (ROC) curve analysis of each biomarker in diagnosis of VTE.

Fig. 2

Receiver operating characteristic (ROC) curve analysis of combined 6 biomarkers in diagnosis of VTE.

Selection of the valuable variables for building new model by multivariable identification of VTE-associated risk factors using logistic multiple regression analysis

We analyzed the results of the Chi-square test and the sample sizes, and identified the following factors as candidated risk factor for further screening: clinical stage, hypertension, BMI ≥ 25 Kg/m², ECOG PS score ≥ 2, history of VTE, FDP ≥ 2.84 mg/L, D-dimer ≥ 0.77 µg/g/ml, TM ≥ 9.75 TU/ml, TAT ≥ 2.25 ng/ml, PIC ≥ 0.80 µg/ml, and tPAI-C ≥ 7.35 ng/ml. Subsequently, the collinearity of these factors was analyzed. The results showed multicollinearity between numerous clinical factors with FDP and D-dimer (Table 4). The high correlation between the coagulation-related biomarkers and the thrombus-related biomarkers reduced the accuracy of the risk prediction model. Therefore, we excluded the aforementioned non-compliance indicators, such as clinical stage, hypertension, BMI ≥ 25 Kg/m², and PIC. The remaining indicators were used as covariates and multivariate logistic regression analysis was performed with the occurrence of VTE as the dependent variable. The results showed that TM ≥ 9.75 TU/ml (OR = 1.616, 95%CI:1.219–20.763, P = 0.026), TAT ≥ 2.25 ng/ml(OR = 2.480, 95%CI:2.401–59.386, P = 0.002), t-PAIC (OR = 1.578, 95%CI:1.301–18.055, P = 0.019), history of VTE (OR = 2.071, 95%CI:1.630-38.617, P = 0.010), and ECOG PS score ≥ 2 (OR = 2.208, 95%CI:2.536–32.616, P = 0.001) were independent risk factors for the lung cancer-associated thrombosis (Table 5).

Table 4 Results of multicollinearity analysis Table 5 Risk factors for thrombosis in lung cancer patients by multivariate analysis Construction and validation of a new risk prediction model for lung cancer-associated thrombosis

A new risk prediction model was constructed by incorporating the independent risk factors. The data was imported into the R software [11] and the rms package [12] was used to create a nomogram. Each predictor in the nomogram was assigned a score on the “Points” axis. The sum of scores for all the variables were assigned to the “Total Points” axis. The total points corresponded to the predicted probability of VTE associated with lung cancer(Fig. 3).

Fig. 3

Nomogram of the new risk prediction model. (Abbreviation:PS: ECOG PS score ≥ 2; HIST: History of VTE; TM: TM ≥ 9.75TU/ml; TAT: TAT ≥ 2.25ng/ml; t-PAIC: t-PAIC ≥ 7.35ng/mL)

The Omnibus test of model coefficients showed that the new VTE risk prediction model was significant (χ²=69.377, P < 0.001). The Hosmer and Lemeshow test indicated high goodness of fit test for the new VTE risk prediction model (P = 0.662 > 0.05). The C-statistic value for the discrimination ability of the new risk prediction model was 0.889 (95% CI: 0.829–0.933). The C-statistic value for the new VTE risk prediction model was 0.862 after internal validation using bootstrap resampling. The calibration plot showed good agreement between the calibration curve and the ideal curve, thereby suggesting good agreement between the predicted incidence rate of the model and the actual incidence rate (Fig. 4).

Fig. 4

Calibration curve of the new model

Comparison of diagnostic efficacy between the new risk prediction model and the other two

Next, we compared the diagnostic efficacies of the new VTE prediction model with the KRS and the PRS. The − 2loglikelihood ratio of the new prediction model was 75.126, more effective than the KRS (139.674) and the PRS (96.188)(Table 6). The C-statistic values for the new VTE risk prediction model, the KRS, and the PRS were 0.889, 0.533, and 0.751 respectively (Fig. 5). This validated the higher diagnostic efficacy of the new VTE prediction model. Furthermore, we verified the clinical applicability of the new risk prediction model. Clinical decision curve analysis can visualize the clinical benefit of the model at different thresholds, and the results show that the new risk prediction model has a higher net benefit rate than the KRS and the PRS(Fig. 6).

Fig. 5

Comparison of C-statistic values. (Abbreviation: NEW RPM:new risk prediction model,Padua score:Padua risk score)

Fig. 6

Comparison of Clinical decision curve analysis. (Abbreviation: NEW:new risk prediction model,KRS:Khorana risk score,Padua:Padua risk score)

Table 6 Comparison of C- statistics between the new RPM and the other two