Continuous variables are reported as mean ± SD and categorical variables as counts and proportions. Comparison among groups (IPC, GCS, and no mechanical thromboprophylaxis) was done with analysis of variance, Kruskal-Wallis test, multinomial logistic regression, or χ2 test, as appropriate. Because of the observed imbalances in baseline characteristics, the multiple propensity scores adjustment approach21 was used as detailed elsewhere.22 Three investigators (Y. M. A., M. K., S. I. D.) selected variables for the final model to drive propensity scores, which was done according to the findings of Brookhart et al.23 These variables were age, sex, time spent in the hospital prior to enrollment, APACHE (Acute Physiology and Chronic Health Evaluation) II score, Glasgow coma score, creatinine level, international normalized ratio, partial thromboplastin time level, recent trauma, recent femur or pelvic fractures or knee or hip replacement, bedridden status, presence of malignancy, recent surgery, packed RBC transfusion, presence of central venous or hemodialysis catheter, presence of sepsis, use of vasopressors, use of prophylactic UFH or enoxaparin, and use of therapeutic anticoagulation after enrollment. Because mechanical device categories (IPC, GCS, and no mechanical thromboprophylaxis) numbered more than two (the dependent variable), multinomial logistic regression analysis was carried out with the aforementioned selected variables as independent variables. The likelihood ratio test for the model, compared with the empty model, was assessed. Independence of irrelevant alternatives assumption and goodness of fit were checked with the Hausman and McFadden24 and Hosmer-Lemeshow25 generalized goodness-of-fit tests, respectively. Three separate propensity scores were then derived from the model. Overlap of different propensity scores was checked visually with the box plot method.22 To check the balancing effect of propensity scores, two of three multiple propensity scores and their mutual interactions were used as covariates in an analysis of covariance for continuous variables and in multinomial logistic regression for categorical variables.