We conducted a retrospective analysis using data from 244 hospitals that participated in Premier Incorporated’s Perspective database. The Perspective database is a voluntary, fee-supported database developed to measure the quality and utilization of health care and is composed of mostly small to midsize non-teaching hospitals in urban locations. The database contains a date-stamped log of all items and services billed during a hospitalization, including medications, laboratory tests, and therapeutic services, as well as information about patient and hospital characteristics, primary and secondary discharge International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9) diagnostic codes, and disposition status.
We included in our analysis patients aged 18 years or older admitted to participating hospitals between October 1, 2003 and September 30, 2005 with a principal procedure code for spinal fusion surgery, subdivided into cervical fusion (codes 81.01–03, 81.31–33), lumbar fusion (81.06–08, 81.36–38), and other/unspecified spinal fusion (81.05, 81.34–35, 81.00, 81.30, 81.39). Surgical approach was categorized into anterior, posterior, circumferential, and other/unknown, based on ICD-9 codes. We also searched for secondary codes that indicated the number of vertebrae fused: fusion of 2–3 vertebrae (81.62) and fusion of 4 or more vertebrae (81.62, 81.64).
Demographic and hospital data including patient age, sex, race/ethnicity, insurance status, admission type (emergency, outpatient, or transfer from another acute hospital), and hospital region were obtained from the database, in addition to a general measure of illness severity calculated using APR-DRG software (version 15.0, 3M™
) that is used to predict mortality. We searched for diagnosis codes for major medical comorbidities using software developed by the Agency of Healthcare Research and Quality[9
], as major medical illnesses increase overall VTE risk[10
]. Finally, we searched for individual comorbid conditions that were part of a validated VTE risk score described by Caprini et al.[11
]. The VTE risk score included the following risk factors: myocardial infarction/congestive heart failure, age, varicose veins, immobility, obesity, hyperviscosity/hypercoagulable syndromes, estrogen therapy, history of venous thromboembolism, and malignancy. This VTE risk score has been validated in other surgical settings[12
Prophylaxis for Venous Thromboembolism
We searched the database for all charges corresponding to provision of either mechanical or pharmacologic VTE prophylaxis administered at any point within the first 7 days after the index spinal surgery date. This time period was chosen because the majority of patients undergoing surgery were discharged within 7 days. Mechanical means of prophylaxis was considered present when billing charges for intermittent compression devices or antiembolism stockings of the lower extremities were identified. Pharmacologic prophylaxis was considered present if any of the following medications were dispensed: injectable heparin sodium at doses between 5,000 and 7,500 units, low molecular weight heparins (specifically enoxaparin and dalteparin, as there were no other types of low molecular weight heparins administered), and fondaparinux. We excluded 6,067 patients who received either intravenous heparin or warfarin sodium within the first 7 days because these agents are not used for routine venous thromboembolism prophylaxis and because it was not possible to determine whether their use was for preexisting conditions requiring full-dose anticoagulation.
Thromboembolic and Hemorrhagic Outcomes
We searched for secondary discharge diagnosis codes consistent with deep venous thromboses (DVT) or pulmonary embolism (PE) during the index hospitalization (ICD-9 codes: 415.1x, 451.1x, 451.2, 451.81, 451.9, 453.1, 453.2, 453.40, 453.41, 453.42, 453.8, 453.9, and 997.2 with a secondary diagnosis of DVT or PE). In addition, we determined whether patients were readmitted within 30 days with a primary discharge diagnosis code for VTE. Adverse surgical outcomes were identified by searching for specific ICD-9 diagnosis and procedure codes. Post-operative hemorrhages were identified using codes for hemorrhage/hematoma complicating a procedure (998.1, 998.11, 998.12) and extradural hemorrhage (852.4, 432.0). We also searched for codes associated with other surgical complications, specifically wound disruption (998.3, 998.31, 998.32), post-operative infection (998.5, 998.51, 998.59), other complications (998.13, 998.7, 998.8), and for procedure codes associated with control of hemorrhage and re-operation (39.98, 39.99, 83.14, 83.19, 83.39, 83.44, 83.49, 86.04, 86.22, 86.28, 96.58, 96.59, 97.15, 97.16). Finally, we searched for charges indicating transfusion of ≥ 2 units of packed red blood cells within the first 4 days after the index surgery.
Rates of pharmacologic prophylaxis, mechanical prophylaxis, both forms, and no prophylaxis after surgery were determined for different categories of spinal fusion. Bivariable analysis using chi-squared tests for categorical variables, and t-tests and non-parametric tests for continuous variables, were performed to test the association between subject/hospital characteristics and receipt of prophylaxis. We then developed multivariable logistic regression models to test the independent association between clinical characteristics and receipt of VTE prophylaxis. Candidate variables significant at a p < 0.2 were considered for model inclusion. Generalized estimating equations using PROC GENMOD (SAS Institute Inc., Cary, NC) were used to account for potential clustering effects at the hospital and individual provider level. We next developed multivariable logistic regression models using generalized estimating equations modeling the association of clinical characteristics and receipt of prophylaxis with the likelihood of developing diagnosed VTE within 30 days, with the significance level set at 0.05.
We recognized that patients did not receive VTE prophylaxis at random, which introduced the threat of treatment and allocation biases unaccounted for in our base multivariable models. To address these biases, we developed a propensity score representing the likelihood that each patient received VTE prophylaxis. The propensity score was derived using a separate multivariable logistic regression model including all available patient and hospital predictor covariates, where covariates for the propensity score were retained at a significance level set at p
]. The final propensity score was then included as a separate covariate in the multivariable models modeling VTE outcomes. All analyses were performed using SAS version 9.2 (SAS Institute Inc, Cary, NC).