POLE: Development and Validation of a Pulmonary Embolism Prediction Model in Lung Cancer

Abstract

Background

Prediction models for cancer-associated pulmonary embolism (PE) in lung cancer patients are scarce. This study aimed to develop and validate a novel model to accurately predict PE risk in this population.

Methods

A retrospective cohort (n = 476) was used to identify PE-related risk factors and construct a predictive nomogram using Cox regression. Validation was performed in a prospective cohort (n = 140). The model's performance was compared with the Khorana score.

Results

The newly developed nomogram termed Pulmonary embOlism in Lung cancEr (POLE) included seven variables: activated partial thromboplastin time, D-dimer, serum phosphorus, carbohydrate antigen 19-9, history of lung cancer surgery, targeted therapy, and age. It demonstrated good predictive performance in the retrospective study (area under the curve [AUC]: 0.776, 95% confidence interval [CI] 0.720-0.833, P < 0.001). The AUC values at 1, 3, 6, and 9 months after lung cancer diagnosis were 0.840, 0.839, 0.799, and 0.801, respectively. In the prospective study, the AUC values for the POLE model at 1, 3, 6, and 9 months after lung cancer diagnosis were 0.859, 0.806, 0.754, and 0.746, respectively. Moreover, in both derivation and validation cohorts, there were significant differences in the probability of PE occurrence among lung cancer patients being stratified into different risk strata (all P < 0.001). The POLE model (AUC 0.762) outperformed the Khorana score (AUC 0.560).

Conclusions

The POLE model, based on seven clinical parameters, was developed and validated for predicting cancer-associated PE in lung cancer patients, demonstrating high accuracy, calibration, and stability.

Keywords

lung cancer pulmonary embolism risk factor risk assessment prediction model

Introduction

Cancer-associated venous thromboembolism (VTE) is the second leading cause of death in cancer patients.¹ Lung cancer, one of the most common malignancies worldwide,² carries a particularly high risk of VTE.³

Pulmonary embolism (PE), a potentially life-threatening form of VTE, was found⁴ in 12.4% of deceased cancer patients, with PE-related deaths accounting for 6.7% of the total population. Among patients with autopsy-confirmed PE, lung cancer patients represented the largest proportion (27.2%). A meta-analysis⁵ reported that the overall incidence of PE in lung cancer was 3.7%, ranging from 1.3% to 23.7%. First-time PE in lung cancer patients often signals severe disease and poor prognosis,⁶ highlighting the need for early risk prediction.

Existing predictive models for cancer-associated thrombosis, such as the Khorana score, have limited accuracy in lung cancer patients.⁷ Other lung cancer-specific risk scores, including the ROADMAP-CAT score⁸ and Rising-VTE/NEJ score,⁹ require further validation and are not specifically designed to predict PE. To address these gaps, we aimed to develop and validate a novel clinical scoring model to predict PE in lung cancer patients. This model is intended to enable earlier identification of high-risk individuals and timely implementation of effective prophylactic and therapeutic strategies.

Methods

Study Design

A retrospective cohort of lung cancer patients (n = 476) from a tertiary teaching hospital in Shanghai, China (January 2015–April 2021), was used to develop a predictive model for acute PE. A prospective cohort (n = 140, May 2021–May 2023) served as a validation set. All patients were followed for one year post-diagnosis. The model, designed for predicting Pulmonary embOlism in Lung cancEr, was named the POLE model. Its performance was compared with the Khorana score.¹⁰

Study Population

Inclusion criteria: (a) pathologically or cytologically confirmed primary lung cancer; (b) age >18 years; (c) absence of other concurrent primary malignancies; Exclusion criteria: (a) loss to follow-up or significant gaps in clinical data; (b) presence of chronic thromboembolic disease; (c) clinically or radiologically suspected but unconfirmed PE; (d) lung cancer patients who had previously undergone pharmacological or radiotherapeutic oncological treatment prior to presenting at our hospital.

Screening and diagnostic protocol for PE in lung cancer patients was as follows: suspected cases aged <50 years with a simplified Wells score ≥2 and D-dimer >0.5 mg/L, or cases aged ≥50 years with a simplified Wells score ≥2 and D-dimer > age × 0.01 mg/L,⁹ underwent spiral CT pulmonary angiography or ventilation/perfusion scintigraphy, interpreted independently by at least two senior radiologists. PE was diagnosed according to established imaging criteria.¹¹ All patients received oncological treatment in accordance with oncology guidelines,^12,13 prophylactic anticoagulation when indicated, and regular PE risk assessment and imaging in accordance with PE, VTE, or cancer-associated VTE guidelines.¹⁴ Final management decisions were made by attending physicians. Ethics approval (XHEC-D-2024-073) and consent procedures are described in the Declarations.

Data Collection

In the retrospective cohort, 105 PE-related variables covering patient, cancer, and treatment factors at lung cancer diagnosis were extracted from the electronic medical record system (see e-Appendix 1 for details). Oncologic treatments, VTE prophylaxis, and pulmonary vascular imaging outcomes were documented during one year of follow-up.

In the prospective cohort, general characteristics (age, sex, body mass index), variables for the POLE model and Khorana score, and pulmonary vascular imaging outcomes were collected from the same system.

Both cohorts applied identical inclusion/exclusion criteria, PE diagnostic procedures, and data extraction methods to ensure methodological consistency and minimize bias.

Statistical Analyses

Statistical analyses were performed using SPSS 26.0 and R 4.3.2. Variables with <20% missing data were imputed using the “mice” package; those with ≥20% missing data were excluded.¹⁵ Normally distributed continuous variables were expressed as mean ± standard deviation and compared with t-tests; non-normally distributed variables as median (25th percentile, P25; 75th percentile, P75) and compared with the Mann-Whitney U-test; categorical variables as counts (n) and percentages (%) and compared with chi-square or Fisher's exact tests.

In the training cohort, potential PE-related risk factors were screened by univariable Cox regression (P < 0.05), and further reduced by least absolute shrinkage and selection operator (LASSO) regression. The final predictors were analyzed using multivariable Cox regression.¹⁶ In the Cox regression analyses, continuous variables were entered as linear terms to keep the model parsimonious, with proportional hazards and a log-linear functional form prespecified. The final predictive model was selected by minimizing the Akaike information criterion (AIC),¹⁷ and a corresponding nomogram was constructed. Risk scores were calculated to stratify patients into high- and low-risk groups. Model performance was assessed with time-dependent ROC curves, area under the curve (AUC), and calibration plots; internal validation used 1000 bootstrap resamples. Time-dependent cumulative incidence curves compared PE occurrence between risk groups.

In the validation cohort, the new predictive model was compared with the Khorana score using ROC curve analysis and the DeLong test. Based on a meta-analysis,⁵ the overall incidence rate of PE in lung cancer patients is 3.7%. With a permissible error margin of 0.05, at least 55 patients needed to be included in the validation group. The actual validation cohort included 140 patients, exceeding the minimum sample size needed to evaluate the model. We acknowledge that the 5% absolute error used in the original calculation may be relatively large compared with the event rate; however, the sample size was primarily determined by the number of available patients and the need to ensure a sufficient number of events for model validation. A P value of less than 0.05 was considered statistically significant. (Additional details regarding model construction and statistical analyses are provided in e-Appendix 2)

Results

Demographics and Characteristics of Patients

According to the inclusion and exclusion criteria, 616 lung cancer patients were recruited. The process is shown in the flowchart ( Figure 1 ). The median age was 64 years, with male and female proportions of 71.1% and 28.9%, respectively. The retrospective and prospective cohorts comprised 476 and 140 patients, respectively. The demographic, clinical, and laboratory characteristics of the enrolled patients are presented in Table 1 (retrospective cohort), Table 2 (prospective cohort) and Table S1.

Table 1.

Baseline Characteristics of the Retrospective Cohort.

Variables	PE Group (n = 90)	Non-PE Group (n = 386)	P Value
Age (years)	67 (61.00, 71.25)	63 (57, 69)	0.002
Sex			0.187
Male	60 (66.70%)	284 (73.60%)
Female	30 (33.30%)	102 (26.40%)
Body Mass Index (kg/m²)	23.21 (20.94, 24.90)	22.74 (20.76, 25.02)	0.635
Performance Status Score (≤2)	90 (100%)	381 (98.7%)	0.057
Smoke			0.236
Yes	47 (52.20%)	228 (59.10%)
No	43 (47.80%)	158 (40.90%)
Comorbidity
Chronic Obstructive Pulmonary Disease	15 (16.70%)	67 (17.40%)	0.876
Cardiovascular Diseases (AF, CAD, HF)	12 (13.30%)	38 (9.80%)	0.331
Diabetes Mellitus	13 (14.40%)	52 (13.50%)	0.809
Central Venous Catheterization	3 (3.30%)	11 (2.80%)	1.000
History of Venous Thrombosis	2 (2.20%)	8 (2.10%)	1.000
Surgery	27 (30%)	190 (49.20%)	0.001
Antitumor Therapy
Antiangiogenic Therapy	6 (6.70%)	47 (12.20%)	0.135
Chemotherapy	70 (77.80%)	320 (82.90%)	0.255
Targeted Therapy	7 (7.80%)	73 (18.90%)	0.011
Immunotherapy	2 (2.20%)	18 (4.70%)	0.455
Pathology			0.008
Adenocarcinoma	50 (55.60%)	237 (61.40%)
Squamous Cell Carcinoma	12 (13.30%)	80 (20.70%)
Small Cell Carcinoma	17 (18.90%)	30 (7.80%)
Other Types	11 (12.20%)	39 (10.10%)
TNM Stage			0.033
I-IIIA	29 (32.20%)	172 (44.60%)
IIIB-IV	61 (67.80%)	214 (55.40%)
White Blood Cell (10^9/l)	6.23 (5.12, 7.69)	6.67 (5.50, 8.00)	0.057
Platelet (10^9/l)	224.50 (175.25, 270.25)	228.00 (186.00, 276.00)	0.330
Lymphocyte Percentage	20.85 (16.40, 27.63)	24.00 (17.98, 29.60)	0.026
Absolute Lymphocyte Count (10^9/l)	1.30 (1.13, 1.70)	1.54 (1.20, 1.98)	0.001
Hemoglobin (g/l)	125.50 (118, 136)	129.50 (119, 139)	0.111
Mean Platelet Volume (fl)	9.45 (8.28, 10.60)	9.00 (8.10, 10.00)	0.039
Activated Partial Thromboplastin Time (s)	30.10 (28.00, 31.65)	31.00 (29.10, 33.10)	0.002
D-dimer (mg/l)	0.66 (0.29, 1.29)	0.30 (0.14, 0.64)	<0.001
Phosphorus (mmol/l)	1.09 (0.97, 1.24)	1.14 (1.04, 1.25)	0.014
Albumin (g/l)	37.10 (34.30, 40.60)	38.25 (35.70, 41.00)	0.063
Erythrocyte Sedimentation Rate (mm/h)	25.00 (14.75, 51.75)	24.00 (12.00, 43.25)	0.198
Alpha-Hydroxybutyrate Dehydrogenase (U/L)	147 (124, 189)	132 (110, 161)	<0.001
Lactate Dehydrogenase (U/L)	217.50 (187.50, 273.25)	206 (174, 247)	0.021
Alpha-1 Microglobulin (%)	4.65 (3.78, 5.50)	4.20 (3.60, 5.30)	0.066
Carcinoembryonic Antigen (ng/ml)	5.23 (2.29, 15.10)	3.35 (1.81, 11.95)	0.017
Carbohydrate Antigen 19-9 (ng/ml)	13.50 (9.02, 57.15)	10.88 (6.74, 19.90)	<0.001
Carbohydrate Antigen 125 (ng/ml)	37.50 (19.70, 140.50)	23.92 (11.65, 61.42)	<0.001
Neuron-Specific Enolase (ng/ml)	16.66 (12.83, 25.11)	15.48 (12.75, 20.61)	0.073

Abbreviation: PE: pulmonary embolism; AF: atrial fibrillation; CAD: coronary artery disease; HF: heart failure; TNM: tumor, nodes and metastases.

Table 2.

Baseline Characteristics of the Prospective Cohort.

Variables	PE Group (n = 34)	Non-PE Group (n = 106)	P Value
Age (years)	67 (59.75, 73.25)	65 (58.00, 70.00)	0.174
Sex			0.080
Male	27 (79.40%)	67 (63.20%)
Female	7 (20.60%)	39 (36.80%)
Body Mass Index (kg/m²)	22.49 (20.93, 24.91)	23.05 (20.87, 25.15)	0.639
Surgery	7 (20.60%)	36 (34%)	0.141
Targeted Therapy	2 (5.90%)	40 (37.70%)	<0.001
Pathology			0.268
Adenocarcinoma	17 (50.00%)	71 (67%)
Squamous Cell Carcinoma	9 (26.50%)	19 (17.90%)
Small Cell Carcinoma	2 (5.90%)	6 (5.70%)
Other Types	6 (17.60%)	10 (9.40%)
White Blood Cell (10^9/l)	6.83 (5.64, 9.24)	6.39 (5.23, 7.72)	0.075
Platelet (10^9/l)	273.5 (201.25, 319.25)	223.0 (176.00, 278.25)	0.038
Absolute Lymphocyte Count (10^9/l)	1.51 (1.04, 1.66)	1.49 (1.15, 1.80)	0.473
Hemoglobin (g/l)	123.5 (111.25, 133.50)	128.5 (119.75, 141.25)	0.029
Mean Platelet Volume (fl)	9.15 (8.45, 10.03)	9.59 (8.65, 10.48)	0.313
Activated Partial Thromboplastin Time (s)	30.75 (27.73,33.35)	30.35 (29.08, 33.13)	0.505
D-dimer (mg/l)	1.28 (0.79, 2.10)	0.88 (0.45, 1.42)	0.006
Phosphorus (mmol/l)	1.09 (1.01, 1.20)	1.16 (1.09, 1.26)	0.012
Carbohydrate Antigen 19-9 (ng/ml)	11.55 (5.81, 20.50)	10.75 (6.27, 23.38)	0.769

Abbreviation: PE: pulmonary embolism.

Development of POLE Model

In the retrospective cohort, 20 potential predictors (P < 0.05) of PE were initially identified using univariable Cox regression (Table S2). LASSO regression was then applied in the training cohort for variable selection, and eight variables with non-zero coefficients were retained ( Figure 2 ). These variables were further analyzed using multivariable Cox regression to construct the POLE predictive model ( Table 3 ), which ultimately included seven variables: activated partial thromboplastin time (APTT), D-dimer, serum phosphorus, carbohydrate antigen 19-9 (CA19-9), history of lung cancer surgery, targeted therapy, and age. The risk score formula was as follows: Risk Score = level(APTT) × (−0.097) + level(D-dimer) × 0.267 + level(serum phosphorus) × (−1.103) + level(CA 19-9) × 0.001 + level(age) × 0.026 + level(history of lung cancer surgery) × 1.002 + level(targeted therapy) × 1.301. APTT, D-dimer, serum phosphorus, CA 19-9, and age were continuous variables; for “history of lung cancer surgery” and “targeted therapy”, a value of 1 was assigned for absence and 0 for presence. The POLE model was constructed based on the Cox regression equation to predict the risk of PE at 1, 3, 6, and 9 months after lung cancer diagnosis ( Figure 3 ).

Table 3.

Multivariate Cox Regression Analysis of the Clinical Parameters in the Retrospective Cohort.

Variables	HR	95%CI	Coef	P Value
Activated Partial Thromboplastin Time (s)	0.908	0.845-0.976	−0.097	0.009
D-dimer (mg/l)	1.306	1.166-1.463	0.267	<0.001
Phosphorus (mmol/l)	0.332	0.102-1.080	−1.103	0.067
Carbohydrate Antigen 19-9 (ng/ml) ^‡	1.001	1.001-1.001	0.001	<0.001
Age (years)	1.022	0.995-1.051	0.022	0.107
Surgery (no)	2.724	1.700-4.367	1.002	<0.001
Targeted Therapy (no)	3.673	1.675-8.057	1.301	0.001

Abbreviation: HR: hazard ratio; CI: confidence interval.

^‡ : 95%CI: 1.0008-1.0014

Predictive Accuracy of the new Predictive Model for PE in the Retrospective Cohort

The ROC curve analysis demonstrated good predictive accuracy for the model in the retrospective cohort (AUC: 0.776, 95% CI 0.720-0.833, P < 0.001; Figure S1A). Time-dependent ROC curves yielded AUCs of 0.840, 0.839, 0.799, and 0.801 at 1, 3, 6, and 9 months after diagnosis, respectively (Figure S1B). Internal validation with 1000 bootstrap resamples confirmed model stability (95% CI 0.717-0.828; Figure S1C). Calibration curves closely followed the diagonal line, indicating good consistency between the predicted and observed outcomes (Figure S2A-D).

The optimal risk score cutoff, determined by the Youden index, was 2.2. Patients were stratified into high- and low-risk groups, and time-dependent cumulative incidence curves showed higher PE incidence in the high-risk group (Log-rank P < 0.001; Figure 4 ).

Predictive Accuracy of the Novel Predictive Model for PE in the Prospective Cohort

The POLE model maintained good predictive accuracy in the prospective cohort (AUC: 0.762, 95% CI 0.678-0.845, P < 0.001) and outperformed the Khorana score (AUC: 0.560, 95% CI 0.443-0.676, P = 0.297; DeLong test P < 0.001; Figure 5A ). Time-dependent ROC curves demonstrated AUCs of 0.859, 0.806, 0.754, and 0.746 at 1, 3, 6, and 9 months after diagnosis, respectively ( Figure 5B ). Internal validation with 1000 bootstrap resamples showed model stability (95% CI 0.681-0.842; Figure 5C ). Calibration curves closely followed the diagonal line, indicating good agreement (Figure S3A-D). Patients were stratified into high- and low-risk groups based on a risk score cutoff of 2.2, with higher PE incidence in the high-risk group (Log-rank P < 0.001) (Figure S4).

Discussion

In this study, the POLE model was developed to predict PE in lung cancer, including seven variables: APTT, D-dimer, serum phosphorus, CA 19-9, history of lung cancer surgery, targeted therapy, and age. Although the 95% CIs for age and serum phosphorus included 1, these variables were retained based on the AIC criterion to optimize overall model performance. This approach has been widely used in predictive modeling and is supported by previous study.¹⁷ The POLE model demonstrated good predictive accuracy, consistency, and stability in both cohorts and outperformed the Khorana score.

Among the study population, 124 patients (20.1%) developed acute PE, with 90 cases (18.9%) in the retrospective cohort and 34 cases (24.3%) in the prospective cohort during a one-year follow-up. Currently, there is limited data on the annual incidence of PE in lung cancer patients. Previous literature⁵ has reported an incidence rate ranging from 1.3% to 23.7%. The incidence of PE in our study aligns with this range. Although follow-up periods were not explicitly stated in previous studies, the first six months after lung cancer diagnosis represent a high-risk period for PE.

The Khorana score, although widely used, shows limited predictive accuracy for lung cancer-specific PE.¹⁸ This limitation may be due to the diverse cancer types and broader outcomes included in its development, reducing its specificity for lung cancer-associated PE. In contrast, our study focuses solely on lung cancer patients, with PE as the only outcome, enhancing specificity. This targeted approach provides clinicians with a more precise tool for identifying high-risk lung cancer patients, aiding early detection and informed clinical decisions.

Another model¹⁹ also predicts PE risk in lung cancer using seven variables: adenocarcinoma, stage III-IV, central venous catheter, chemotherapy, serum albumin, hemoglobin, and D-dimer. Although internally and externally validated, it lacked a comparative control model and employed Logistic regression analysis, which may not fully capture the temporal relationship between lung cancer and PE risk.

Our study found that shortened APTT and elevated D-dimer levels are risk factors for PE in lung cancer patients. Previous studies also reported that shortened APTT is independently associated with increased VTE risk,^20,21 consistent with our findings.

Phosphorus is an essential mineral involved in numerous biological processes,²² primarily regulated by serum calcium concentrations and parathyroid hormone.²³ An animal study²⁴ found high-phosphorus diet promote lung cancer development and alter Protein Kinase B (AKT) signaling. In our study, serum phosphorus negatively correlated with tumor stage (r = -0.134, P = 0.001), which may indicate that higher tumor burden consumes more serum phosphorus. Our study found that the reduced serum phosphorus was a risk factor for PE development in lung cancer patients, potentially due to the heavier tumor burden associated with advanced tumor stage.²⁵ Furthermore, another study found a negative correlation between serum phosphorus and plasma levels of tissue plasminogen activator inhibitor-1 Ag (PAI-1Ag) in patients with primary hyperparathyroidism²⁶ (r = -0.453, P < 0.05), and no correlation was found between serum calcium and PAI-1, indicating a possible link between phosphorus and coagulation in lung cancer, which warrants further investigation.

CA 19-9 is a serum biomarker primarily associated with gastrointestinal cancers, particularly pancreatic cancer. In lung adenocarcinoma, serum CA 19-9 levels correlate positively with lymph node involvement and distant metastatic,²⁷ and elevated CA 19-9 levels are linked to poorer progression-free and overall survival.²⁸ Our research suggested that higher CA 19-9 levels were associated with increased PE risk in lung cancer patients. While comparable studies are limited, studies in pancreatic cancer have found a correlation between CA 19-9 and patients’ coagulation function. Elevated CA 19-9 has been associated with thrombosis in pancreatic cancer,²⁹ and its doubling time may predict VTE occurrence.³⁰ The potential correlation between CA 19-9 and circulating mucin levels may underlie its association with thrombosis.³¹ Moreover, a robust correlation has been discovered between plasma microparticle-associated tissue factor activity and CA 19-9 levels,³² as well as an increase in CA 19-9 correlating with elevated plasma thrombin concentrations.³³ These findings may help clarify how CA 19-9 contributes to the risk of PE in lung cancer patients. However, further investigation is needed to confirm these hypotheses and elucidate the mechanisms by which CA 19-9 affects the coagulation system in this population.

Our study found that targeted therapy and surgical treatment were risk factors influencing the occurrence of PE in lung cancer patients. The targeted therapy agents in our study were mainly targeted at genetic alterations such as epidermal growth factor receptor (EGFR) mutations, anaplastic lymphoma kinase (ALK) fusions, and c-ros oncogene 1 (ROS1) fusion mutations. Previous study has found that non-small cell lung cancer patients with ALK/ROS1 rearrangements are more prone to thrombosis than those with other oncogenic gene alterations.³⁴ However, we could not perform a stratified analysis due to the small sample size of patients with ALK/ROS1 rearrangements, which may limit the reliability of our results. Nonetheless, our study suggested that targeted therapy may reduce PE risk, likely due to its tumor suppression. Further large-sample studies are needed to verify whether specific gene mutations and their corresponding targeted therapies can reduce PE risk. Results on the impact of lung cancer surgery on venous thrombosis are inconsistent.^5,35 Our study suggested that surgery may reduce PE risk, possibly due to the earlier tumor stage and better health in those who underwent surgery. We found no significant effect of chemotherapy, immunotherapy, or anti-angiogenic therapy on PE development in lung cancer patients, in contrast to some previous literature.^36–38 This discrepancy may be attributed to integrated therapies in our patient cohort, which could complicate the relationship between chemotherapy and PE risk. Furthermore, our study involved a limited number of patients treated with anti-angiogenic agents and immunotherapies, potentially impacting the study's outcomes. Thus, additional large-scale, prospective studies are needed to confirm these findings and explore the relationship between anti-angiogenic therapy and PE risk in lung cancer.

Our study identified age as a risk factor for PE in lung cancer patients, with those older than 66 years being more susceptible(AUC: 0.605, sensitivity: 0.567, specificity: 0.601, 95% CI: 0.540-0.670, P = 0.002).This aligns with a previous study³⁹ that also found age over 66 years to be a potential risk factor for PE.

Although the mechanisms by which predictors such as serum phosphorus and CA 19-9 contribute to PE remain unclear, the POLE model's prospective validation and real-world data support its reliability and clinical applicability, highlighting its potential to guide risk stratification and individualized management of PE in lung cancer patients.

Limitations

The POLE model is one of the few specific models for predicting cancer-associated PE risk in lung cancer patients, offering clinical value. However, several limitations should be considered when interpreting these findings: First, the single-center design and modest sample size may limit the generalizability and application of the model. While the model is validated in our prospective cohort, future research should involve multicenter, prospective studies with larger cohorts for further validation and refinement. Second, although the retrospective and prospective cohorts were collected during different time periods, both adhered to the same inclusion and exclusion criteria, diagnostic definitions, and data collection procedures, which minimized potential bias arising from methodological differences. Nevertheless, we cannot completely exclude the influence of temporal changes in lung cancer treatment strategies or clinical practice patterns on the incidence of PE, which might have affected the model's performance in the validation cohort. In addition, early deaths and patients lost to follow-up without documented PE events were censored at the date of death or last follow-up, whichever came first, after which no further person-time was contributed. We acknowledge that this approach may introduce potential bias in estimating PE risk. Competing risk analysis, such as Fine-Gray models, could more accurately account for early deaths as competing events, and future studies will explore this approach to further validate and refine the POLE model. Third, the predictors in the POLE model were assessed at the time of lung cancer diagnosis, making it most applicable to newly diagnosed patients. Caution is needed when applying the model to patients who are not newly diagnosed. Additionally, transient risk factors for PE after diagnosis, such as trauma, infection, and new treatments, could impact the model's long-term predictive accuracy. Fourth, despite efforts to include all relevant risk factors for PE, some important factors may still be missed, which might influence its predictive accuracy. Moreover, we prespecified a linear functional form for continuous covariates to keep the model parsimonious; non-linear terms or fractional polynomials were not examined and could be explored in future work. Last, in the absence of a widely recognized model specific for evaluating PE risk in lung cancer patients, the Khorana score was selected as a control. However, it may not be an eligible control since Khorana score was designed for predicting VTE risk in all categories of cancers instead of predicting PE risk in isolated lung cancer, which may limit its specialized generalizability to the latter prediction.

Conclusions

The POLE nomogram incorporates seven predictors: APTT, D-dimer, serum phosphorus, CA 19-9, history of lung cancer surgery, targeted therapy, and age, was constructed and validated in the current study, for predicting cancer-associated PE development in lung cancer patients. It demonstrates good predictive accuracy, consistency, and stability, as well as the superiority to Khorana score. Nevertheless, the new model still needs to be validated in larger cohorts in the future.

Supplemental Material

sj-docx-1-cat-10.1177_10760296261428826 - Supplemental material for POLE: Development and Validation of a Pulmonary Embolism Prediction Model in Lung Cancer

Supplemental material, sj-docx-1-cat-10.1177_10760296261428826 for POLE: Development and Validation of a Pulmonary Embolism Prediction Model in Lung Cancer by Dongmei Wang, Wei Xiong, Xuan Huang, Fan Zhang, Fengming Xu and Fengfeng Han in Clinical and Applied Thrombosis/Hemostasis

Footnotes

List of Abbreviations

Acknowledgements

We would like to thank all the patients being studied and those who have supported this research indirectly.

ORCID iDs

Dongmei Wang

Wei Xiong

Xuan Huang

Fan Zhang

Fengming Xu

Fengfeng Han

Ethics Approval and Consent to Participate

Ethical approval to report this case was obtained from the local institutional review board (approval number XHEC-D-2024-073). Written informed consent from the participants or their next of kin in the retrospective cohort was waived due to: (1) The study involves no more than minimal risk for patients; (2) the study cannot adversely affect the rights and welfare of patients; (3) the study cannot be performed without the exemption of informed consent of patients. Written informed consent was obtained from the patient(s) in the prospective cohort for their anonymized information to be published in this article.

Consent for Publication

Not applicable.

Author Contributions

Concept and design: FH, WX. Acquisition, analysis, or interpretation of data: all authors. Drafting of the manuscript: DW. Critical review of the manuscript for important intellectual content: all authors. Statistical analysis: all authors. Administrative, technical, or material support: all authors. Supervision: FH, WX. Dongmei Wang and Wei Xiong contributed equally to the work.

Funding

The authors received no financial support for the research, authorship, and/or publication of this article.

Declaration of Conflicting Interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Data Availability and Materials

The data that support the findings of this study are available from the corresponding author upon reasonable request.

Supplemental Material

Supplemental material for this article is available online.

References

Streiff

Holmstrom

Angelini

, et al. Cancer-Associated venous thromboembolic disease, Version 2.2021, NCCN clinical practice guidelines in oncology. J Natl Compr Cancer Netw. 2021;19(10):1181-1201.

Bray

Laversanne

Sung

, et al. Global cancer statistics 2022: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin. 2024;74(3):229-263. doi:10.3322/caac.21834

Blom

Doggen

Osanto

Rosendaal

. Malignancies, prothrombotic mutations, and the risk of venous thrombosis. Jama. 2005;293(6):715-722. doi:10.1001/jama.293.6.715

Gimbel

Mulder

Bosch

FTM

, et al. Pulmonary embolism at autopsy in cancer patients. J Thromb Haemostasis. 2021;19(5):1228-1235. doi:10.1111/jth.15250

Shang

Wang

Ning

Chen

. Lung cancer and pulmonary embolism: What is the relationship? A review. J Cancer. 2018;9(17):3046-3057. doi:10.7150/jca.26008

Nouhravesh

Strange

Sindet-Pedersen

, et al. Impact of breast-, gastrointestinal-, and lung cancer on prognosis in patients with first-time pulmonary embolism: A Danish nationwide cohort study. Int J Cardiol. 2024;406:132001. doi:10.1016/j.ijcard.2024.132001

Abdel-Razeq

Sharaf

Al-Jaghbeer

, et al. COMPASS-CAT versus Khorana risk assessment model for predicting venous thromboembolic events in patients with non-small cell lung cancer on active treatment with chemotherapy and/or immunotherapy, the CK-RAM study. J Thromb Thrombolysis. 2023;56(3):447-453. doi:10.1007/s11239-023-02860-4

Syrigos

Grapsa

Sangare

, et al. Prospective assessment of clinical risk factors and biomarkers of hypercoagulability for the identification of patients with lung adenocarcinoma at risk for cancer-associated thrombosis: The observational ROADMAP-CAT study. Oncologist. 2018;23(11):1372-1381. doi:10.1634/theoncologist.2017-0530

Tsubata

Hotta

Hamai

, et al. A new risk-assessment tool for venous thromboembolism in advanced lung cancer: A prospective, observational study. J Hematol Oncol. 2022;15(1):40. doi:10.1186/s13045-022-01259-7

10.

Khorana

Kuderer

Culakova

Lyman

Francis

. Development and validation of a predictive model for chemotherapy-associated thrombosis. Blood. 2008;111(10):4902-4907. doi:10.1182/blood-2007-10-116327

11.

Konstantinides

Torbicki

Agnelli

, et al. 2014 ESC Guidelines on the diagnosis and management of acute pulmonary embolism. Kardiol Pol. 2014;72(11):997-1053. doi:10.5603/KP.2014.0211

12.

Kalemkerian

Akerley

Bogner

, et al. Small cell lung cancer. J Natl Compr Cancer Netw. 2013;11(1):78-98. doi:10.6004/jnccn.2013.0011

13.

Ettinger

Wood

Akerley

, et al. Non-small cell lung cancer, version 1.2015. J Natl Compr Cancer Netw. 2014;12(12):1738-1761. doi:10.6004/jnccn.2014.0176

14.

Lyman

Khorana

Kuderer

, et al. Venous thromboembolism prophylaxis and treatment in patients with cancer: American Society of Clinical Oncology clinical practice guideline update. J Clin Oncol. 2013;31(17):2189-2204. doi:10.1200/JCO.2013.49.1118

15.

Liang

, et al. Development and validation of a clinical risk score to predict the occurrence of critical illness in hospitalized patients with COVID-19. JAMA Intern Med. 2020;180(8):1081-1089. doi:10.1001/jamainternmed.2020.2033

16.

Liu

Huang

Yang

, et al. Nomogram for predicting overall survival in stage II-III colorectal cancer. Cancer Med. 2020;9(7):2363-2371. doi:10.1002/cam4.2896

17.

Stewart

Stephens

Hill

Whittingham

Dawson

. Model selection in occupancy models: Inference versus prediction. Ecology. 2023;104(3):e3942. doi:10.1002/ecy.3942

18.

Tsubata

Kawakado

Hamai

, et al. Identification of risk factors for venous thromboembolism and validation of the Khorana score in patients with advanced lung cancer: Based on the multicenter, prospective Rising-VTE/NEJ037 study data. Int J Clin Oncol. 2023;28(1):69-78. doi:10.1007/s10147-022-02257-y

19.

Zhu

Zhang

Gong

, et al. Derivation and external validation of a risk prediction model for pulmonary embolism in patients with lung cancer: A large retrospective cohort study. Clin Appl Thromb Hemost. 2023;29:10760296231151696. doi:10.1177/10760296231151696

20.

Zakai

Ohira

White

Folsom

Cushman

. Activated partial thromboplastin time and risk of future venous thromboembolism. Am J Med. 2008;121(3):231-238. doi:10.1016/j.amjmed.2007.10.025

21.

Tripodi

Chantarangkul

Martinelli

Bucciarelli

Mannucci

. A shortened activated partial thromboplastin time is associated with the risk of venous thromboembolism. Blood. 2004;104(12):3631-3634. doi:10.1182/blood-2004-03-1042

22.

Papadopoulou

Bountouvi

Karachaliou

. The molecular basis of calcium and phosphorus inherited metabolic disorders. Genes (Basel). 2021;12(5):734. doi:10.3390/genes12050734

23.

Ling

Zhang

. Unexpected serum phosphorus lost in an amniotic fluid embolism patient. Clin Chim Acta. 2023;538:91-93. doi:10.1016/j.cca.2022.11.015

24.

Jin

Lim

, et al. High dietary inorganic phosphate increases lung tumorigenesis and alters Akt signaling. Am J Respir Crit Care Med. 2009;179(1):59-68. doi:10.1164/rccm.200802-306OC

25.

Junjun

Pei

Ying

Kui

. Prognosis and risk factors in older patients with lung cancer and pulmonary embolism: A propensity score matching analysis. Sci Rep. 2020;10(1):1272. doi:10.1038/s41598-020-58345-4

26.

Erem

Kocak

Nuhoglu

Yilmaz

Ucuncu

. Increased plasminogen activator inhibitor-1, decreased tissue factor pathway inhibitor, and unchanged thrombin-activatable fibrinolysis inhibitor levels in patients with primary hyperparathyroidism. Eur J Endocrinol. 2009;160(5):863-868. doi:10.1530/EJE-09-0069

27.

Chen

Wang

Fang

. Analysis of tumor markers in pleural effusion and serum to verify the correlations between serum tumor markers and tumor size, TNM stage of lung adenocarcinoma. Cancer Med. 2020;9(4):1392-1399. doi:10.1002/cam4.2809

28.

Huang

Xiao

Zhou

, et al. Baseline serum tumor markers predict the survival of patients with advanced non-small cell lung cancer receiving first-line immunotherapy: A multicenter retrospective study. BMC Cancer. 2023;23(1):812. doi:10.1186/s12885-023-11312-4

29.

Woei

AJFJ

Tesselaar

Garcia Rodriguez

Romijn

Bertina

Osanto

. Tissue factor-bearing microparticles and CA19.9: Two players in pancreatic cancer-associated thrombosis? Br J Cancer. 2016;115(3):332-338. doi:10.1038/bjc.2016.170

30.

Peippo

Kurki

Seppänen

Lassila

Carpén

. CA 19-9 Doubling time in pancreatic cancer as a predictor of venous thromboembolism: A hospital database study. Acta Oncol (Stockholm, Sweden). 2020;59(2):237-241. doi:10.1080/0284186X.2019.1679881

31.

Yue

Maupin

Fallon

, et al. Enhanced discrimination of malignant from benign pancreatic disease by measuring the CA 19-9 antigen on specific protein carriers. PloS one. 2011;6(12):e29180. doi:10.1371/journal.pone.0029180

32.

Thaler

Mackman

, et al. Microparticle-associated tissue factor activity in patients with pancreatic cancer: Correlation with clinicopathological features. Eur J Clin Investig. 2013;43(3):277-285. doi:10.1111/eci.12042

33.

Mattila

Hisada

Przybyla

, et al. Levels of the cancer biomarker CA 19-9 are associated with thrombin generation in plasma from treatment-naïve pancreatic cancer patients. Thromb Res. 2021;199:21-31. doi:10.1016/j.thromres.2020.12.018

34.

Liu

Wang

, et al. High discrepancy in thrombotic events in non-small cell lung cancer patients with different genomic alterations. Transl Lung Cancer Res. 2021;10(3):1512-1524. doi:10.21037/tlcr-20-1290

35.

Howlett

Benzenine

Cottenet

Foucher

Fagnoni

Quantin

. Could venous thromboembolism and major bleeding be indicators of lung cancer mortality? A nationwide database study. BMC Cancer. 2020;20(1):461. doi:10.1186/s12885-020-06930-1

36.

Bjørnhart

Kristiansen

Asmussen

, et al. Clinical impact of venous thromboembolism in non-small cell lung cancer patients receiving immunotherapy. Thromb Res. 2023;221:164-172. doi:10.1016/j.thromres.2022.10.020

37.

Chen

Ren

, et al. Bevacizumab promotes venous thromboembolism through the induction of PAI-1 in a mouse xenograft model of human lung carcinoma. Mol Cancer. 2015;14(1):140. doi:10.1186/s12943-015-0418-x

38.

Kou

Zhou

Shi

. Enhanced procoagulant activity of platelets after chemotherapy in non-small cell lung cancer. Cancer Biol Ther. 2017;18(8):627-634. doi:10.1080/15384047.2017.1345387

39.

Shen

Huang

, et al. Prevalence and risk factors of acute pulmonary embolism in patients with lung cancer surgery. Semin Thromb Hemost. 2018;44(4):334-340. doi:10.1055/s-0037-1612625

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.97 MB