Evaluation of risk factors for thromboembolic events in multiple myeloma patients using multiple machine learning...

Venous thromboembolic events (VTE) is a frequent complication in multiple myeloma (MM) patients, raising mortality. This study aims to use machine learning to identify VTE risk factors in MM, helping to pinpoint high-risk individuals for better clinical management and prognosis. A retrospective analysis was conducted on the basic information, laboratory test results, treatment plans, and thrombosis prevention measures of 428 newly diagnosed MM patients at our hospital from December 2018 to December 2022. We used logistic regression (LR), random forest, and gradient boosting machine (GBM) models to identify and assess the risk factors for VTE in patients with MM. Among 428 patients with MM, 48 cases (11.21%) had concomitant VTE, including 10 cases of deep vein thrombosis with pulmonary embolism, while the remaining 38 cases were solely deep vein thrombosis. The results of the multifactorial LR analysis indicate that C-reactive protein (CRP), fibrinogen, von Willebrand factor (vWF), factor VIII (FVIII), and treatment regimen immunomodulator in the patient's treatment regimen are independent factors influencing the risk of VTE in patients with MM. In the analysis of the random forest model, we found that CRP and fibrinogen are the most important factors for predicting the risk of VTE in patients with MM, with the highest Gini indices of 12.76 and 12.31, respectively. In addition, vWF, FVIII, age, platelet count, D-dimer, β2 microglobulin, serum creatinine, and albumin were also considered key variables affecting the risk of VTE in MM patients. In the GBM model, the importance ranking of variables showed that FIB and CRP are the most important predictive factors, with influences of 36.84 and 28.56, respectively. In addition, other important variables include vWF, FVIII, age, albumin, neutrophils, β2 microglobulin, and D-dimer. We found that CRP and fibrinogen were the most important risk factors in all 3 models, while vWF and FVIII were also confirmed as significant risk factors. The identification of these common risk factors provides a clear focus for clinical practice to more accurately identify high-risk groups for VTE among MM patients.