Machine learning analyses constructed a novel model to predict recurrent thrombosis in adults with essential
Jia Chen1,2, Huan Dong1,2, Rongfeng Fu1,2
1State Key Laboratory of Experimental Hematology, Haihe Laboratory of Cell Ecosystem, Institute of Hematology & Blood Diseases Hospital, Tianjin Key Laboratory of Gene Therapy for Blood Diseases, CAMS Key Laboratory of Gene Therapy for Blood Diseases, National Clinical Research Center for Blood Diseases, Chinese Academy of Medical Sciences & Peking Union Medical College, Tianjin, 300020, China.
Abstract:
The current study involving 318 essential thrombocythemia (ET) patients with prior thrombosis was designed to identify risk factors that were predictive of recurrent thrombosis. The whole cohort was randomly split into derivation and validation cohorts. The random forest method, support vector machine with built-in recursive feature elimination model, and logistic multivariable analysis were performed in the derivation cohort, and cardiovascular risk factor (CVF) and RBC distribution width with standard deviation (RDW-SD) were finally selected as independent predictors. Subsequently we devise a 3-tiered model (low risk: 0 points; intermediate risk: 1-1.5 points; and high risk: 2.5 points) and it showed good discrimination in all cohorts. Moreover, the model was significantly correlated with rethrombosis-free survival (rTFS) (p = 0.0007 in the derivation cohort; p = 0.0019 in the validation cohort). In the whole cohort, cytoreductive therapy was more effective than antiplatelet agents alone for 10-year rTFS (p = 0.0336). No significant difference in 10-year rTFS was observed among interferon (IFN), hydroxyurea (HU), and IFN + HU therapy (p = 0.444). The present study helps identify individuals who need close monitoring and provides valuable risk signals for recurrence in ET patients with prior thrombosis.
