| Objective:To explore the related risk factors of pulmonary embolism(PE)in patients with deep venous thrombosis(DVT)of lower extremities and establish logistic regression model and scoring system,which will be verified in the external population.The calibration and discrimination power of the predictive model will be tested to evaluate the clinical application value.It will provide a basis for clinicians to identify patients with high risk of pulmonary embolism.Methods:1.A cross-sectional study was conducted,a total of 776 patients with deep venous thrombosis of lower extremities were enrolled into this study between January 2016 and June 2017 in Vascular Surgery of Shanxi Dayi Hospital.The patients set was splited randomly into development database and validation database,the total number of patients in development database was 543(70%),and 233 cases(30%)in validation database.According to the results of Computed Tomography Pulmonary Arteriography(CTPA),patients with pulmonary embolism were included in the case group,and patients who did not combine pulmonary embolism were included in the control group.2.To collect all DVT patients' general and clinical data,including gender,age,heart rate,systolic blood pressure,diastolic blood pressure,risk factors for venous thromboembolism(VTE)(unprovoked,hypertensive,coronary heart disease,surgery and immobilization,superficial thrombophlebitis,stroke or paralysis,diabetes,injuryor fracture,chronic venous insufficiency,pregnancy or postpartum,nephrotic syndrome,malignant tumor,blood platelet disorders,history of VTE,hyperhomocysteinemia,family history of VTE),clinical symptoms of DVT(pain,swelling,reddening of skin),position of thrombosis(common iliac vein,external iliac vein,femoral vein,popliteal vein,fibular veins,tibial vein,muscular venous),DVT types,DVT staging,DVT limb and initial laboratory examination results of admission(blood cell analysis,coagulation seven indices and plasma D-dimer).3.All the continuous variables in development database were analyzed by the receiver operating characteristic curve(ROC),the variables with statistically significance are converted to binary classification variables according to the cutoff value determined by Youden's index,the others according to the clinical commonly used cutoff value.Variables were evaluated in univariate analysis.A multivariate logistic regression model was developed using a stepwise selection process.Variables associated with an increased risk of PE(P<0.20)in univariate analysis were included in the pool of variables for the forward stepwise regression model.The final multivariable logistic regression model include all variables that are independent risk factors of PE in patients with deep venous thrombosis of lower extremities(P<0.05).A risk score was developed based on regression coefficients from the final multivariate model.The total score for each participant was calculated by adding each component together.We chose the score that discriminated between a low-risk groups and a high-risk groups as the cut-off value and then calculated the sensitivity and specificity of the predictive model.The predictive model was verified in the verification database.Calculating the predicted probability of pulmonary embolism and total score of the patients by iterationing the validation database'data into the Logistic regression model and the scoring system.4.Calibration and discrimination of the logistic regression model and scoring system were assessed by Hosmer-Lemeshow good-of-fit test and the area under the receiver operating characteristic curve(AUC)respectively both in the development database and validation database.The AUC of the logistic regression model and scoring system was compared by U test both in the development database and validationdatabase.And the AUC of the logistic regression model and scoring system was compared by U test respectively between the development database and validation database.Results:1.Patient characteristics in the development database and validation database were well balanced,with the exception of the position of thrombosis(external iliac vein,tibial vein).2.In the development database,the final multivariate logistic regression analysis showed that the following variables were independently associated with risk of PE : the right lower extremity(OR=1.863;95%CI : 1.134 ~ 3.062),bilateral lower extremities(OR=1.781;95%CI:1.076~2.949),unprovoked(OR=2.943;95%CI:1.665~5.200),surgery and immobilization(OR=1.990;95%CI:1.253~3.160),malignant tumor(OR=1.997;95%CI:1.133~3.521),history of VTE(OR=2.591;95%CI:1.424~4.715),D-Dimer>1000ng/mL(OR=3.555;95%CI:2.311~5.470).The logistic regression model showed good calibration and discriminative power.The Hosmer-Lemeshow good-of-fit test's P value was 0.537 and the AUC was 0.735(95%CI : 0.689 ~0.781,P<0.001).3.We assigned points for the risk model based on the regression coefficients obtained from the final model,the minimum regression coefficient is 0.577 that corresponded to 1 point.The total score was between 0 and 9.The scoring system was listed: right lower extremity or bilateral lower extremities: 1 score;surgery and immobilization: 1 score;malignant tumor: 1 score;history of VTE:2 score;D-Dimer>1000ng/mL:2 score;unprovoked:2 score.We divided the population into 2 risk categories based on the score from the risk model: low risk(score<3),high risk(score?3).Scoring system showed good discriminative power,and the AUC was0.728(95%CI : 0.682 ~0.775,P<0.001).In the development database,the difference of the AUC were nonsignificant between logistic regression model and scoring system(P=0.164).Thesensitivity,specificity and the crude agreement of the scoring system respect was76.39%,55.89% and 61.33%.4.In the validation database,logistic regression model showed good calibration and discriminative power.The Hosmer-Lemeshow good-of-fit test's P value was more than0.05 and the AUC was 0.705(95%CI:0.634~0.776,P<0.001).The scoring system also showed good discriminative power,and the AUC was 0.702(95%CI : 0.631 ~0.774,P<0.001).The U test of the AUC was not statistically significant between logistic regression model and scoring system(P=0.704).The sensitivity,specificity and the crude agreement of the scoring system was 67.61%,61.73% and 63.52%respectively.5.The AUC has no statistically significant difference between the development database and validation database(P=0.486 for the logistic regression model and P=0.542 for the scoring system).Conclusion:We identified 7 clinical and laboratory parameters-the right lower extremity,bilateral lower extremities,unprovoked,surgery and immobilization,malignant tumor,history of VTE,D-Dimer>1000ng/mL –that are the risk factors of pulmonary embolism in patients with deep venous thrombosis of lower extremity.Logistic regression model and scoring system besed on the above 7 parameters showed good calibration and discriminative power.It can be applied to clinical screening of pulmonary embolism in patients with deep venous thrombosis of lower extremity due to the good sensitivity,specificity and clinical application value. |
PDF Full Text Request