Establishment And Validation Of Prediction Model Of Deep Vein Catheter-related Thrombosis In Patients With Cancer

Objective:Peripherally inserted Central Catheter(PICC),central venous catheter(CVC)and other interventional catheterization are widely used in the treatment of cancer patients with chemotherapy.However,in clinical treatment,the most serious and the most common complication of cancer patients is catheter-associated deep vein thrombosis.The purpose of this study is to explore and analyze the risk factors of thrombosis in patients with cancer undergoing catheterization,and try to establish a Logistic regression catheter-related clinical prediction model to calculate the possibility of Catheter-related deep vein thrombosis establish a quantitative tool for risk and benefit assessment.To provide an important reference for doctors,patients and medical policy makers to make decisions.Methods:A nested case-control study was used to analyze the data of 4,691 patients in the third affiliated hospital of Kunming Medical University with malignant tumors from January 01,2018 to January 31,2019 and established a Logistic regression prediction model.All patients' case data were divided into two groups:case arm(deep venous catheterization followed by ultrasonography to diagnose deep vein catheter-related thrombosis),control arm(no catheter-related thrombosis occured after deep venous catheterization).We conducted a univariate analysis of 52 predictive factors such as age,gender,patient activity,whether to undergo chemotherapy,and tumor stage of deep venous catheter-related thrombosis,etc.We included the risk factors of P?0.25 into the multivariate analysis and selected the independent influencing factors of P?0.05 to establish a Logistic regression clinical prediction model of thrombosis associated with deep vein catheterization and used Bootstrap method to verify the model.Results:We analyzed the data of deep venous catheterization cases of tumor patients and found that a total of 355 cases of thrombosis occurred,the incidence rate was 7.60%.Cervical venous thrombosis was the most,accounting for 32.50%of all cases.Univariate analysis showed that:a total of 29 factors such as age,patient activity,whether chemotherapy,tumor staging,smoking history,drinking history,blood transfusion history,surgery history,infection history,the presence or absence of glucocorticoid(dexamethasone,prednisone,etc.),hypertension,diabetes,hyperlipidemia,history of thrombosis;red blood cells count,white blood cells count,platelets count,hemoglobin,hematocrit;PT,ratio of PT,FIB,plasma antithrombin ?,fibrinogen degradation products,D2 polyme,albumin,A/G,total protein,and direct bilirubin are influencing factors(P ? 0.05).After multivariate analysis,the activities of the tube placement site,tumor stage,infection,glucocorticoid,hyperlipidemia,other comorbidities(except hypertension,diabetes,hyperlipidemia),history of thrombosis(hypercoagulable state),platelet count,PT,D2 polymer and blood glucose level 11 factors entered the prediction model(P?0.05).The prediction model is:P=ex/(1+ex).P:Probability of thrombosis associated with deep vein catheterization in cancer patients x=-8.942-2.337*x1+0.233*x2+0.483*x3+0.819*x4+0.970*x5+2.811*x6+2.588*x7+0.482*x8+0.832*x9+1.165*x10-0.479*x11.AUC=0.860(95%CI:0.843-0.887).AUC=0.860(95%CI:0.843-0.887).The positive prediction probability threshold of this model is set to 56.0%,then the sensitivity is 72.0%and the specificity is 84.0%,suggesting that this deep vein catheter thrombosis prediction research model has higher accuracy after correction.Note 1:P:Probability of thrombosis associated with deep vein catheterization in cancer patients x1=activity of the catheter site;x2=tumor stage;x3=infection;x4=glucocorticoid use;x5=hyperlipidemia;x6=other comorbidities(except hypertension,diabetes,high Outside lipidemia);x7=history of thrombosis(hypercoagulable state);x8=platelet count;x9=prothrombin time;x10=D2 polyme;x11=blood glucose level.Note 2:Activity at the catheter site:none=0,yes=1;tumor stage:stage ?=1,stage?=2,stage ?=3,stage ?=4;infection:none=0,yes=1;glucocorticoid use:none=0,yes=1;hyperlipidemia:none=0,yes=1;other comorbidities(except hypertension,diabetes,hyperlipidemia):none=0,yes=1;history of thrombosis(hypercoagulable state):none=0,yes=1;platelet count:0-99*109/L=1,100-300*109/L=2,?301*109/L=3;PT:0-10.9sec=1,11-14.5sec=2,?14.6sec=3;D2 polymer:0-0.55ug/mL=1,?0.56ug/mL=2;blood glucose level:0-3.88mmol/L=1,3.89-6.11mmol/L=2,?6.12mmol/L=3.Conclusion:Our Logistic regression thrombosis prediction model has a high accuracy and is recomended to use it in clinical practise,but needing prospective study to verify in future.