Construction Of A Predictive Model For Post-Thrombotic Syndrome In Patients With Lower Extremity Deep Vein Thrombosis

Background:Post-thrombotic syndrome(PTS)is the most common chronic complication in patients with lower extremity deep vein thrombosis(DVT),which will seriously reduce the quality of life because of its poor prognosis and irreversible pathophysiological changes.Prevention of PTS has become the focus of clinical work.Risk factors of PTS have the characteristics of diversity,complexity and instability,which makes it difficult to identify high-risk patients.As an important tool to evaluate the risk of disease occurrence,the risk predictive model provides new ideas for PTS prevention.Scholars around the world have begun to carry out studies on PTS predictive model to assess the risk of PTS quickly and accurately.Nonetheless,the models have some problems,such as unscientific research design and single model construction method.Whereas all the limitations,it is necessary to carry out a prospective cohort study to explore the model predictors based on risk factors of PTS in lower extremity DVT patients,and build a PTS predictive model by using machine learning.The establishment of predictive model can assist doctors and nurses to identify high-risk patients quickly and give preventive measures,so as to reduce the incidence of PTS and promote the development of PTS prevention and management.Objective:To explore the model predictors based on risk factors of PTS in patients with lower extremity DVT;to build and verify the PTS predictive model for lower extremity DVT patients.Method:This study is divided into three parts.Part one:we determined predictors through a combination of literature review and expert meeting.Part two:a prospective cohort study method was used to conduct a 2-year follow-up study on patients with lower extremity DVT in outpatient,emergency and inpatient departments from December 1,2018 to December 31,2019.The Villalta scale was used to evaluate the incidence of PTS at 6,12,and 24 months after DVT.Data on 10 predictors including patient-specific factors,DVT characteristics-related factors,and DVT treatment-related factors were collected.We obtained the risk factors of PTS through univariate analysis and multivariate binary logistic regression analysis.Part three:random forest,a machine learning algorithm,was used to construct the PTS prediction model,and carry out internal verification.Finally,discrimination and calibration were used to evaluate the prediction ability of PTS prediction model.Results:A total of 675 patients were included in this study,and a total of 518 patients finally entered the model development cohort.A total of 113 patients developed PTS within 2 years,with a cumulative incidence rate of 21.81%.Multivariate analysis showed that advanced age(OR=1.02,95%CI=1.01～1.04),proximal DVT(OR=3.14,95%CI=1.80～5.49)and recurrent DVT(OR=4.75,95%CI=1.78～12.70)were risk factors,and induced DVT(OR=0.36,95%CI=0.18～0.72)was protective factor.The random forest model based on machine learning model has good prediction effect(AUC=0.722,Acc=0.770).The predictors in the random forest model were ranked from large to small:proximal DVT,induced DVT,recurrent DVT,Age,BMI,gender,number of DVT symptoms and signs(baseline),GCS treatment,history of varicose veins,and adequate anticoagulation.The 5-fold cross-validation confirmed that the random forest model was practical and feasible.Conclusion:In this study,a predictive model of PTS was constructed by random forest algorithm.The model has high reproducibility,universality and feasibility,and can effectively identify patients at high risk of PTS.The model still needs external validation before clinical application.