Risk factors for venous thromboembolism in patients with metastatic bone tumors: development and validation of a machine learning-based prediction model

doi:10.3969/j.issn.1674-8115.2026.10.008

Abstract

Abstract:

Objective ·To identify independent risk factors for venous thromboembolism (VTE) in patients with metastatic bone tumors and to develop and validate a machine learning-based prediction model. Methods ·This retrospective cohort study included 439 patients with metastatic bone tumors treated at the Clinical Diagnosis and Treatment Center for Metastatic Bone Tumors, Shanghai Sixth People's Hospital, from January 2012 to December 2022 for model training and internal validation. Collected data included demographic characteristics (sex and age at diagnosis of metastatic bone tumors), primary tumor sites, Karnofsky performance status, treatment modalities (radiotherapy and surgery), spinal cord compression, admission complete blood count parameters (red blood cell count, white blood cell count, hemoglobin, and platelet count), coagulation indices (D-dimer, fibrinogen, and prothrombin time), and sites of bone metastases. Patients were divided into the VTE group (n=153) and the non-VTE group (n=286) based on VTE outcomes. Multivariate Logistic regression was used to identify risk factors. Five machine learning models—CatBoost, LightGBM, XGBoost, Gaussian process, and Logistic regression—were trained and evaluated using ten-fold cross-validation to predict VTE risk and identify the optimal model for clinical risk stratification. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, and specificity. A prospective cohort including 65 patients with metastatic bone tumors admitted between January 2023 and December 2024 was used to validate the performance of the model with the highest AUC. Shapley additive explanations (SHAP) was employed for clinical interpretation of model features. Results ·There were no significant differences in age, sex distribution, or Karnofsky performance status between the VTE group and the non-VTE group (all P>0.05). The most common primary tumors were lung cancer (40.3%), breast cancer (12.3%), and prostate cancer (7.3%). The most common sites of bone metastases were the spine (64.7%), pelvis (53.5%), ribs and clavicle (33.9%), and lower limbs (28.7%). Three independent risk factors were identified: lower limb bone metastases (OR=1.89, 95%CI 1.17‒3.07, P=0.010), elevated D-dimer levels (OR=1.06, 95%CI 1.02‒1.09, P<0.001), and prolonged prothrombin time (OR=1.09, 95%CI 1.01‒1.17, P=0.034). CatBoost demonstrated the best performance, with an AUC of 0.82 (95%CI 0.73‒0.90) in the internal test set and 0.80 (95%CI 0.67‒0.93) in the prospective validation set. SHAP analysis confirmed D-dimer as the most important predictor (mean |SHAP|=1.41). Conclusion ·Lower limb bone metastases, elevated D-dimer levels, and prolonged prothrombin time are independent risk factors for VTE in patients with metastatic bone tumors. The CatBoost-based prediction model demonstrated robust performance in prospective validation, showing potential for clinical application in predicting VTE risk in patients with metastatic bone tumors.

Key words: metastatic bone tumor, venous thromboembolism (VTE), risk factor, machine learning, CatBoost, prediction model

CLC Number:

R738.1

Ruan Shunyi, Sun Yi, Wang Qiong, Chang Yujie, Gu Yifeng, Xu Yongming, Wang Zhiyu, Zhao Hui. Risk factors for venous thromboembolism in patients with metastatic bone tumors: development and validation of a machine learning-based prediction model[J]. Journal of Shanghai Jiao Tong University (Medical Science), 2026,(10): 1383-1391.

Figures/Tables 6

TrendMD

References 26

[1]	Ashrani A A, Gullerud R E, Petterson T M, et al. Risk factors for incident venous thromboembolism in active cancer patients: a population based case-control study[J]. Thromb Res, 2016, 139: 29-37.
[2]	Khorana A A, Francis C W, Culakova E, et al. Thromboembolism is a leading cause of death in cancer patients receiving outpatient chemotherapy[J]. J Thromb Haemost, 2007, 5(3): 632-634.
[3]	Sørensen H T, Mellemkjær L, Olsen J H, et al. Prognosis of cancers associated with venous thromboembolism[J]. N Engl J Med, 2000, 343(25): 1846-1850.
[4]	Chew H K, Wun T, Harvey D J, et al. Incidence of venous thromboembolism and the impact on survival in breast cancer patients[J]. J Clin Oncol, 2007, 25(1): 70-76.
[5]	Pfaundler N, Limacher A, Stalder O, et al. Prognosis in patients with cancer-associated venous thromboembolism: comparison of the RIETE-VTE and modified Ottawa score[J]. J Thromb Haemost, 2020, 18(5): 1154-1161.
[6]	Mundy G R. Metastasis to bone: causes, consequences and therapeutic opportunities[J]. Nat Rev Cancer, 2002, 2(8): 584-593.
[7]	Brodowicz T, O'Byrne K, Manegold C. Bone matters in lung cancer[J]. Ann Oncol, 2012, 23(9): 2215-2222.
[8]	齐典文, 张国川, 扈文海, 等. 骨转移癌患者下肢深静脉血栓形成的危险因素分析[J]. 中华肿瘤杂志, 2014, 36(6): 469-472.
	Qi D W, Zhang G C, Hu W H, et al. Analysis of risk factors for deep vein thrombosis of the lower extremity in patients with bone metastase[J]. Chinese Journal of Oncology, 2014, 36(6): 469-472.
[9]	Mioc M L, Prejbeanu R, Vermesan D, et al. Deep vein thrombosis following the treatment of lower limb pathologic bone fractures: a comparative study[J]. BMC Musculoskelet Disord, 2018, 19(1): 213.
[10]	Walker M S, Miller P J E, Namjoshi M, et al. Relationship between incidence of fracture and health-related quality-of-life in metastatic breast cancer patients with bone metastases[J]. J Med Econ, 2013, 16(1): 179-189.
[11]	Garcia-Garcia V, Barca-Hernando M, Lopez-Ruz S, et al. Relevance of recurrent venous thromboembolism according to location of metastasis in patients with cancer-associated thrombosis. A cohort study[J]. Thromb Res, 2025, 245: 109228.
[12]	Khorana A A, DeSancho M T, Liebman H, et al. Prediction and prevention of cancer-associated thromboembolism[J]. Oncologist, 2021, 26(1): e2-e7.
[13]	Cella C A, Djulbegovic B, Hozo I, et al. Comparison of Khorana vs. ONKOTEV predictive score to individualize anticoagulant prophylaxis in outpatients with cancer[J]. Eur J Cancer, 2024, 209: 114234.
[14]	Abdul Aziz N I, Palaniappan S, Kamal Rodin N S, et al. Incidence of venous thromboembolism and its associated risk factors in newly diagnosed multiple myeloma patients in the Klang Valley, Malaysia[J]. J Clin Med, 2025, 14(3): 759.
[15]	马军, 秦叔逵, 吴一龙, 等. 肿瘤相关静脉血栓栓塞症预防与治疗指南(2019版)[J]. 中国肿瘤临床, 2019, 46(13): 653-660.
	Ma J, Qin S K, Wu Y L, et al. Guidelines on prevention and treatment of tumor-associated venous thromboembolism (Version 2019) [J]. Chinese Journal of Clinical Oncology, 2019, 46(13): 653-660.
[16]	Asdahl P H, Sundbøll J, Adelborg K, et al. Cardiovascular events in cancer patients with bone metastases: a Danish population-based cohort study of 23, 113 patients[J]. Cancer Med, 2021, 10(14): 4885-4895.
[17]	Lei H K, Tao D, Zhang N N, et al. Nomogram prediction for the risk of venous thromboembolism in patients with lung cancer[J]. Cancer Cell Int, 2023, 23(1): 40.
[18]	Chen F J, Sun Y X, Zhang C, et al. Risk factors of venous thromboembolism in inpatients with colorectal cancer in China[J]. Clin Appl Thromb Hemost, 2023, 29: 10760296231221133.
[19]	ten Wolde M, Kraaijenhagen R A, Prins M H, et al. The clinical usefulness of D-dimer testing in cancer patients with suspected deep venous thrombosis[J]. Arch Intern Med, 2002, 162(16): 1880.
[20]	Jin S, Qin D, Liang B S, et al. Machine learning predicts cancer-associated deep vein thrombosis using clinically available variables[J]. Int J Med Inform, 2022, 161: 104733.
[21]	Pabinger I, van Es N, Heinze G, et al. A clinical prediction model for cancer-associated venous thromboembolism: a development and validation study in two independent prospective cohorts[J]. Lancet Haematol, 2018, 5(7): e289-e298.
[22]	Takada T, van Doorn S, Parpia S, et al. Diagnosing deep vein thrombosis in cancer patients with suspected symptoms: an individual participant data meta-analysis[J]. J Thromb Haemost, 2020, 18(9): 2245-2252.
[23]	Izuegbuna O O, Agodirin O S, Olawumi H O, et al. Plasma D-dimer and fibrinogen levels correlates with tumor size and disease progression in Nigerian breast cancer patients[J]. Cancer Invest, 2021, 39(8): 597-606.
[24]	Huang X, Chen H X, Meng S, et al. External validation of the Khorana score for the prediction of venous thromboembolism in cancer patients: a systematic review and meta-analysis[J]. Int J Nurs Stud, 2024, 159: 104867.
[25]	Chen K Y, Zhang Y, Zhang L F, et al. Machine learning models for risk prediction of cancer-associated thrombosis: a systematic review and meta-analysis[J]. J Thromb Haemost, 2025, 23(2): 610-626.
[26]	Bulut H, Maestre M J, Tomey D. A comprehensive evaluation of intramedullary devices in the management of femoral metastasis: 30-day outcomes[J]. Proc (Bayl Univ Med Cent), 2025, 38(3): 247-252.

Variable	Total (n=439)	Without VTE (n=286)	With VTE (n=153)	P value
Gender/n(%)				0.117
Female	176 (40.1)	107 (37.4)	69 (45.1)
Male	263 (59.9)	179 (62.6)	84 (54.9)
Age/year	61.9±11.4	61.3±12.1	63.0±10.1	0.155
Karnofsky performance status/score	73.5±8.2	73.7±8.0	73.1±8.5	0.465

Variable	Total (n=439)	Without VTE (n=286)	With VTE (n=153)	P value
Gender/n(%)				0.117
Female	176 (40.1)	107 (37.4)	69 (45.1)
Male	263 (59.9)	179 (62.6)	84 (54.9)
Age/year	61.9±11.4	61.3±12.1	63.0±10.1	0.155
Karnofsky performance status/score	73.5±8.2	73.7±8.0	73.1±8.5	0.465

Variable	Total (n=439)	Without VTE (n=286)	With VTE (n=153)	P value
Primary tumor site/n(%)				0.062
Breast	54 (12.3)	35 (12.2)	19 (12.4)	0.956
Lung	177 (40.3)	99 (34.6)	78 (51.0)	<0.001
Kidney	21 (4.8)	17 (5.9)	4 (2.6)	0.119
Esophagus and stomach	21 (4.8)	14 (4.9)	7 (4.6)	0.881
Colon and rectum	25 (5.7)	17 (5.9)	8 (5.2)	0.758
Other abdominal	26 (5.9)	22 (7.7)	4 (2.6)	0.032
Head and neck	21 (4.8)	18 (6.3)	3 (2.0)	0.043
Prostate	32 (7.3)	21 (7.3)	11 (7.2)	0.953
Ovary and cervix	17 (3.9)	12 (4.2)	5 (3.3)	0.631
Nasopharynx	10 (2.3)	9 (3.1)	1 (0.7)	0.176
Pancreas	6 (1.4)	4 (1.4)	2 (1.3)	0.992
Sarcoma	6 (1.4)	5 (1.7)	1 (0.7)	0.669
Bladder	5 (1.1)	3 (1.0)	2 (1.3)	0.997
Skin	8 (1.8)	4 (1.4)	4 (2.6)	0.458
Thymus	3 (0.7)	1 (0.3)	2 (1.3)	0.279
Other	7 (1.6)	5 (1.7)	2 (1.3)	0.996
Bone metastasis site/n(%)
Skull	15 (3.4)	9 (3.1)	6 (3.9)	0.670
Spine	284 (64.7)	190 (66.4)	94 (61.4)	0.297
Ribs and clavicle	149 (33.9)	103 (36.0)	46 (30.1)	0.210
Upper limbs	39 (8.9)	30 (10.5)	9 (5.9)	0.106
Pelvis	235 (53.5)	137 (47.9)	98 (64.1)	0.195
Lower limbs	126 (28.7)	65 (22.7)	61 (39.9)	<0.001
SREs/n(%)
Radiotherapy	102 (23.2)	62 (21.7)	40 (26.1)	0.291
Operation	167 (38.0)	98 (34.3)	69 (45.1)	0.026
Bone fractures	121 (27.6)	73 (25.5)	48 (31.4)	0.191
Spinal cord compression	41 (9.3)	30 (10.5)	11 (7.2)	0.258
Laboratory finding
White blood cell/(×10⁹·L^-1)	7.6 ± 3.5	7.3 ± 3.3	8.2 ± 3.6	0.008
Red blood cell/(×10¹²·L^-1)	3.9 ± 0.7	4.0 ± 0.7	3.8 ± 0.6	0.014
Hemoglobin/(g·L^-1)	116.9 ± 29.5	119.3 ± 33.2	112.2 ± 20.1	0.015
Platelet/(×10⁹·L^-1)	228.7 ± 107.1	230.9 ± 110.7	224.6 ± 100.0	0.555
Fibrinogen/(g·L^-1)	3.8 ± 1.4	3.8 ± 1.4	3.8 ± 1.3	0.844
Prothrombin time/s	12.4 ± 3.4	12.0 ± 2.0	13.1 ± 4.9	<0.001
D-dimer/(μg·mL^-1)	1.9 (0.5, 7.1)	0.9 (0.2, 3.1)	6.4 (2.4, 11.2)	<0.001

Variable	Total (n=439)	Without VTE (n=286)	With VTE (n=153)	P value
Primary tumor site/n(%)				0.062
Breast	54 (12.3)	35 (12.2)	19 (12.4)	0.956
Lung	177 (40.3)	99 (34.6)	78 (51.0)	<0.001
Kidney	21 (4.8)	17 (5.9)	4 (2.6)	0.119
Esophagus and stomach	21 (4.8)	14 (4.9)	7 (4.6)	0.881
Colon and rectum	25 (5.7)	17 (5.9)	8 (5.2)	0.758
Other abdominal	26 (5.9)	22 (7.7)	4 (2.6)	0.032
Head and neck	21 (4.8)	18 (6.3)	3 (2.0)	0.043
Prostate	32 (7.3)	21 (7.3)	11 (7.2)	0.953
Ovary and cervix	17 (3.9)	12 (4.2)	5 (3.3)	0.631
Nasopharynx	10 (2.3)	9 (3.1)	1 (0.7)	0.176
Pancreas	6 (1.4)	4 (1.4)	2 (1.3)	0.992
Sarcoma	6 (1.4)	5 (1.7)	1 (0.7)	0.669
Bladder	5 (1.1)	3 (1.0)	2 (1.3)	0.997
Skin	8 (1.8)	4 (1.4)	4 (2.6)	0.458
Thymus	3 (0.7)	1 (0.3)	2 (1.3)	0.279
Other	7 (1.6)	5 (1.7)	2 (1.3)	0.996
Bone metastasis site/n(%)
Skull	15 (3.4)	9 (3.1)	6 (3.9)	0.670
Spine	284 (64.7)	190 (66.4)	94 (61.4)	0.297
Ribs and clavicle	149 (33.9)	103 (36.0)	46 (30.1)	0.210
Upper limbs	39 (8.9)	30 (10.5)	9 (5.9)	0.106
Pelvis	235 (53.5)	137 (47.9)	98 (64.1)	0.195
Lower limbs	126 (28.7)	65 (22.7)	61 (39.9)	<0.001
SREs/n(%)
Radiotherapy	102 (23.2)	62 (21.7)	40 (26.1)	0.291
Operation	167 (38.0)	98 (34.3)	69 (45.1)	0.026
Bone fractures	121 (27.6)	73 (25.5)	48 (31.4)	0.191
Spinal cord compression	41 (9.3)	30 (10.5)	11 (7.2)	0.258
Laboratory finding
White blood cell/(×10⁹·L^-1)	7.6 ± 3.5	7.3 ± 3.3	8.2 ± 3.6	0.008
Red blood cell/(×10¹²·L^-1)	3.9 ± 0.7	4.0 ± 0.7	3.8 ± 0.6	0.014
Hemoglobin/(g·L^-1)	116.9 ± 29.5	119.3 ± 33.2	112.2 ± 20.1	0.015
Platelet/(×10⁹·L^-1)	228.7 ± 107.1	230.9 ± 110.7	224.6 ± 100.0	0.555
Fibrinogen/(g·L^-1)	3.8 ± 1.4	3.8 ± 1.4	3.8 ± 1.3	0.844
Prothrombin time/s	12.4 ± 3.4	12.0 ± 2.0	13.1 ± 4.9	<0.001
D-dimer/(μg·mL^-1)	1.9 (0.5, 7.1)	0.9 (0.2, 3.1)	6.4 (2.4, 11.2)	<0.001

Variable	OR (95%CI)	P value
Lower limb metastasis	1.89 (1.17‒3.07)	0.010
White blood cell	1.05 (0.99‒1.11)	0.137
Red blood cell	0.80 (0.59‒1.09)	0.163
Prothrombin time	1.09 (1.01‒1.17)	0.034
D-dimer	1.06 (1.02‒1.09)	<0.001
Operation	1.49 (0.87‒2.56)	0.149