Skip to main page content
U.S. flag

An official website of the United States government

Dot gov

The .gov means it’s official.
Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

Https

The site is secure.
The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
. 2025 Oct;104(10):5441-5451.
doi: 10.1007/s00277-025-06628-z. Epub 2025 Sep 26.

Development and validation of deep vein thrombosis diagnostic model based on machine learning methods

Bin Yan  1   2   3 , Haijun Guo  4 , Tianxi Hu  1   2   3 , Yu Zhang  5 , Zhixin Zheng  5 , Weipeng Du  6   7   8 , Yuhan Gu  9
Affiliations

Development and validation of deep vein thrombosis diagnostic model based on machine learning methods

Bin Yan et al. Ann Hematol. 2025 Oct.

Abstract

The D-Dimer testing has limited diagnostic value for patients with a deep venous thrombosis (DVT) probability based on clinical prediction rules. There are still patients with normal D-Dimer levels (< 500 ng/mL) diagnosed with DVT. Some new predictive marker may improve the predictive power of D-Dimer, especially in DVT patients with normal levels of D-Dimer. All subjects were from Nanyang Central Hospital. The demographic data and laboratory test data were collected. Multiple models were used to evaluate and calculate the importance rank. Multivariate logistics was used to establish a DVT diagnostic model. Compared to D-Dimer and other markers, this combined model has better performance. The von Willebrand factor Gain-of-function mutant GPIb binding assays (VWF: GPIbM) can improve the diagnostic capability of D-Dimer, which has higher diagnostic value and clinical benefits. In addition, the model still has good diagnostic capability in DVT patients with normal D-Dimer levels. The combined model has better diagnostic performance than D-Dimer, and it is valuable for some patients whose clinical prediction rules cannot be evaluated due to difficulties in obtaining medical history information. VWF: GPIbM can be used to assist in the diagnosis of DVT in the future.

Keywords: D-Dimer; Deep venous thrombosis; Von willebrand factor.

PubMed Disclaimer

Conflict of interest statement

Declarations. Informed consent: has been obtained from the patients and subjects. This study was approved by the Ethics Committee of Nanyang Central Hospital (KYLW-2025-0126). The rights and privacy of all patients and subjects are fully protected. Competing interests: The authors declare no competing interests.

Figures

Fig. 1
Fig. 1
Importance rank of clinical factors based on multiple models. (A) The β values of different clinical factors in Lasso model. (B) Gini index of different clinical factors in Random Forest model. (C) The importance of clinical factors calculated by Boruta algorithm based on Random Forest model. (D) SHAP value of clinical factors in XGBoost model. (E) SHAP value of clinical factors in SVM model. (F) Comprehensive ranking of clinical factor importance scores calculated by Robust Rank Aggregation algorithm
Fig. 2
Fig. 2
Diagnostic capabilities of the combined model in train dataset and test dataset. (A) Odds ratio of variables in combined model based on multivariate logistic regression analysis. OR, odds ratio; CI, confidence interval. TG, triglyceride; PT, prothrombin time; APTT, activated partial thromboplastin time; RDW-CV, red blood cell distribution width coefficient of variation; ALP, alkaline phosphatase; AT, antithrombin; Hb, hemoglobin. (B) ROC analysis of the combined model compared with other markers in train dataset and test dataset. (C-D) Calibration curve of combined model and D-Dimer in train dataset (C) and test dataset (D). (E) Decision curve analysis of combined model and D-Dimer in train dataset. (F) Clinical impact curve of combined model in train dataset. (G-H) ROC analysis of missing data deletion (G) and replacement by median (H) in all data
Fig. 3
Fig. 3
Diagnostic capabilities of the combined model in patients with normal D-Dimer levels. (A) The ROC analysis of the combined model and D-Dimer in DVT patients with normal D-Dimer levels. (B) Decision curve analysis of the combined model and D-Dimer in DVT patients with normal D-Dimer levels. (C-D) ROC analysis of missing data deletion (C) and replacement by median (D)
See this image and copyright information in PMC

References

    1. Duffett L (2022) Deep venous thrombosis. Ann Intern Med 175(9):ITC129–ITC144. 10.7326/AITC202209200 - PubMed
    1. Arshad N, Isaksen T, Hansen JB, Braekkan SK (2017) Time trends in incidence rates of venous thromboembolism in a large cohort recruited from the general population. Eur J Epidemiol 32(4):299–305. 10.1007/s10654-017-0238-y - PubMed
    1. Heit JA (2008) The epidemiology of venous thromboembolism in the community. Arterioscler Thromb Vasc Biol 28(3):370–372. 10.1161/ATVBAHA.108.162545 - PMC - PubMed
    1. Lutsey PL, Zakai NA (2023) Epidemiology and prevention of venous thromboembolism. Nat Rev Cardiol 20(4):248–262. 10.1038/s41569-022-00787-6 - PMC - PubMed
    1. Bagot CN, Arya R (2008) Virchow and his triad: a question of attribution. Br J Haematol 143(2):180–190. 10.1111/j.1365-2141.2008.07323.x - PubMed

Publication types

LinkOut - more resources