Application value of prediction model established based on limb muscle strength status combined with clinical data for occurrence of deep vein thrombosis in acute stage of stroke

Objective  To explore the application value of the prediction model established based on limb muscle strength combined with clinical data in the occurrence of deep vein thrombosis (DVT) in acute stage of stroke.

Methods  Clinical data of 697 stroke patients were retrospectively analyzed, and they were randomly divided into modeling group (n=488) and validation group (n=209) according to a ratio of 7∶3, and the data of the two groups were analyzed for differences. The modeling group was divided into DVT group and non-DVT group according to whether patients had DVT in the acute stage. Multiple Logistic regression analysis was used to screen out the influencing factors of DVT in acute stage of stroke, and R software was used to obtain the prediction model expressed in line graph. Receiver operating characteristic (ROC) curve was used to evaluate the model differentiation. The calibration degree of the model was evaluated by Bootstrap method (self-sampling method); the decision curve was used to evaluate the clinical effectiveness of the model.

Results  There was no significant difference in clinical data between the modeling group and the verification group (P>0.05). In the modeling group, 77 of 488 stroke patients developed DVT during acute hospitalization (DVT group), with an incidence of 15.78% (77/488). Logistic regression analysis showed that age, diabetes mellitus, dyslipidemia, Padua score, D-dimer and limb muscle strength were all influencing factors of DVT in acute stage of stroke (P < 0.05). The area under the curve of DVT risk in the modeling group was 0.890 (95%CI, 0.866 to 0.923), and was 0.851 (95%CI, 0.781 to 0.911) in the verification group, which indicated that the model was well differentiated; the average absolute error of deviation calibration curve of the modeling group and the verification group was 0.012 and 0.015, respectively, which indicated that the calibration degree of the prediction model was high. The threshold probability value in the decision curve was set at 33%, and the net clinical benefit for the modeling group and the validation group was 62% and 64%, respectively, which indicated that the prediction model was effective in clinic.

Conclusion  The prediction model established based on limb muscle strength status, age, diabetes and dyslipidemia, Padua score and D-dimer index has certain value in predicting the risk of DVT in acute stage of stroke.