According to the literature, the rate of deep venous thrombosis after total hip arthroplasty (THA) can be high (45%-63%) without prophylactic anticoagulation.1-6 A meta-analysis of 13 studies found a rate of 51%.7 As lower extremity deep venous thrombi are the initial source of symptomatic pulmonary emboli in about 90% of cases,8 THA patients are usually given medication postoperatively focused on prevention of these thromboembolic events.9 Chemoprophylaxis may involve warfarin, enoxaparin, or their combination in an anticoagulation bridge. Enoxaparin is one of many low-molecular-weight heparins (LMWHs). All LMWHs exert their anticoagulant effect by binding to antithrombin III.10 The binding of LMWH to antithrombin III catalyzes the inhibition of factor Xa by antithrombin III, disrupting clot formation.11
In its hydroquinone form, vitamin K is essential as a cofactor for carboxylation of the glutamic acid residues of the amino-terminals of the coagulation proteins II, VII, IX, and X, leading to their activation. Anticoagulation by warfarin is achieved by the inhibition of the reductase enzymes that produce vitamin K hydroquinone in the liver from vitamin K epoxide.12 This inhibition prevents activation of the clotting proteins.12,13 Prophylaxis with enoxaparin or warfarin can reduce the rate of venous thromboembolic disease to 3.6% and 3.7%, respectively.2 However, these medications inhibit the clotting cascade, and their use risks prolonging the healing process.9 The delay increases the risk for wound infection,14 which can lead to a longer hospital stay and therefore higher costs.
We conducted a study to compare patients who received warfarin only with patients who received warfarin bridged with enoxaparin as antithrombotic chemoprophylaxis after THA. Outcomes of interest were number of days until a dry wound was observed and length of hospital stay. We hypothesized that, compared with warfarin-only therapy, bridged therapy would increase the risk for prolonged wound healing and result in longer hospital stays.
Materials and Methods
At our 746-bed academic medical center, 121 THAs were performed between January 1, 2008 and December 31, 2009. This study was approved by the center’s Office for Human Subjects Protections institutional review board (IRB). The research involved collecting or studying existing data, documents, and records recorded anonymously by the investigator in such a manner that subjects could not be identified, directly or through identifiers linked to the subjects, and therefore patient consent was not needed. Therefore, the IRB waived the need for consent. Relevant data included in this study were extracted from patient medical records, given within 35 days of surgery. For each patient, discharge notes provided data on the hospital course, and nurses’ notes provided data on wound status after THA.
Propensity Score Matching
For accurate analysis, it was important to consider confounding factors in both patient groups. Some covariates that may influence accurate analysis are age,15 diabetes,16 sex,15,17 hypertension,18 and body mass index.15,19Propensity score, defined as the conditional probability of receiving treatment, given the observed background covariates, was initially defined by Rosenbaum20 and Rubin.21 The motivation behind propensity scores can be understood by considering an idealized situation in which the 2 groups are similar on all background characteristics. In nonexperimental studies, researchers aim to find for each treated individual a comparison individual who looks exactly the same as the treated individual with respect to observed pretreatment covariates. Thus, assuming no hidden bias, any difference in outcomes within these pairs can be attributed to the variable of interest and not to any other differences between the treated and comparison individuals. Our study is a typical nonexperimental retrospective study in which the 2 groups being compared are patients receiving warfarin only or warfarin bridged with enoxaparin. To minimize the influence of background covariates, we used matching procedures and present our results both with and without the use of matching techniques.
Data and Results
There are different matching algorithms aimed at matching groups. In our study, the optimal matching procedure alone could not produce adequately matched data, so we used both optimal matching20 and genetic matching.22,23 Genetic matching procedure with replacement22 can produce well-matched data—it matched each patient in the warfarin-only group with a patient in the bridged-therapy group and allowed different patients to be matched with 1 similar patient in the control group. However, as the same patients in the bridged-therapy group might be matched multiple times, it would complicate the after-matching analysis. We therefore used a 2-step matching procedure to obtain well-matched data, and a simplified analysis procedure after matching. In the first step, we implemented genetic matching with replacement, as introduced by Abadie and Imbens,22 to match each warfarin-only patient with 1 bridged-therapy patient. In the second step, we applied optimal matching to the 2 groups. This 2-step matching turned out to produce better matched pairs, as denoted by Rubin.21 Both matching steps were implemented using the MatchIt function in R.24