Predictive modeling of lower extreme deep vein thrombosis following radical gastrectomy for gastric cancer: based on multiple machine learning methods – Scientific Reports
A recent study aimed to develop and validate a predictive diagnostic model for lower extremity deep vein thrombosis (DVT) formation in patients who underwent radical surgery for gastric cancer (GC). By utilizing clinical and pathological data, combined with machine learning algorithms, researchers hoped to predict the likelihood of DVT development in postoperative GC patients.
The retrospective case‒control study focused on patients who had undergone radical gastrectomy for GC between May 2020 and February 2023 at the Second Affiliated Hospital of Zhejiang University School of Medicine. Inclusion criteria were strict, including factors such as age, preoperative diagnosis of GC, absence of distant metastasis, and complete clinical data. Patients with a history of venous thrombosis, coagulation dysfunction, distant metastasis, or incomplete data were excluded from the study.
Symptoms of lower limb DVT following surgery for GC typically include pain, swelling, and increased tissue tension in the affected limb. Diagnostic guidelines based on venous Doppler ultrasound and plasma D-dimer level measurements were used to categorize patients into DVT and non-DVT groups.
A wide range of clinical, preoperative, postoperative, surgical, and pathological data was collected, totaling 49 clinical observation indicators. These variables were analyzed using a combination of logistic regression, the Boruta algorithm for feature selection, and multivariate logistic regression to identify factors associated with lower extremity DVT formation and create a predictive model.
Multiple methods were then employed to evaluate the predictive model, including discrimination, calibration, and the application of decision curve analysis, clinical impact curve analysis, and different machine learning techniques such as decision trees, random forests, support vector machines, extreme gradient boosting, and light gradient boosting machines.
Overall, the study found that machine learning methods enhanced the prediction of lower extremity DVT following radical gastrectomy for GC. By incorporating various clinical parameters and advanced algorithms, researchers were able to develop a reliable diagnostic model to assess the risk of DVT in postoperative GC patients. This approach fills a crucial gap in understanding and managing postoperative risks for individuals undergoing radical surgery for gastric cancer.
