Anticoagulation in Elective Spine Cases

Rates of Hematomas Versus Thromboembolic Disease

Dharani Rohit Thota, BA; Carlos A. Bagley, MD; Mazin Al Tamimi, MD; Paul A. Nakonezny, PhD; Michael Van Hal, MD


Spine. 2021;46(13):901-906. 

In This Article

Materials and Methods

After institutional review board approval, we undertook evaluation of our institutional cohort as well examination of the NSQIP cohort according to the following methodologies.

Institutional Cohort

The elective spine surgery cases that were performed at a single academic center between 2015 and 2017 were identified. A cohort from these patients was then identified based on the following exclusion criteria: emergency cases, trauma cases, and tumor cases. A retrospective chart review was conducted as it allowed for the characterization of hematoma formation that required reoperation such as those related to a confirmed intraoperative hematoma. It also allowed us to verify the protocol for the pharmacologic anticoagulation that was used.

Institutional data only included symptomatic VTE events where the patient presented with a complaint of shortness of breath, lower extremity pain, or swelling. These events were confirmed using venous duplex ultrasonography or computed tomography angiogram (CTA).

Demographic, operative, postoperative, and comorbidity factors were also characterized. Patient risk/comorbidity factors included smoking status, current smoker, American Society of Anesthesiologists' classification (ASA) classification (>3), diabetes, chronic obstructive pulmonary disease chronic obstructive pulmonary disease (COPD), heart disease, liver disease, and a history of solid tumors. Table 1 contains the demographics of this population.

A total of 3790 patients were initially identified. Two hundred fifty four patients were excluded due to either non-elective cases, trauma, tumor, or age greater than 90 or less than 18. The chart review was performed on the remaining 3536 patients. The cohort was then matched using propensity scoring (described below). This matched a single patient who did not receive pharmacological anticoagulation to a single patient who did within the institution. This technique was employed to control for differences in patient population and complexity of the case. By virtue of the propensity match technique an additional 1760 patients were excluded. This left a matched sample of 1776 patients with 888 patients in each arm which were matched according to propensity scores. This process of patient selection and matching is shown by the flow chart seen in Figure 1.

Figure 1.

Institutional cohort patient parsing tree. The tree starts with the initial patient identification and exclusion. It then moves to the propensity match exclusions and ends with the final count of patients in each arm.

NSQIP Cohort

A retrospective analysis was done using the NSQIP for years 2015 to 2017. The NSQIP is made up of more than 700 hospitals across the US Multiple factors associated with the surgery are characterized: demographic, comorbidity, intraoperation, and postoperative factors.[7] These are characterized through day 30 postoperatively for a multitude of surgical procedures.

Patients undergoing elective spine surgery were selected based on the following Current Procedural Terminology (CPT) codes: 22558, 63087, 63088, 63047, 63005, 63012, 63267, 63030, 63042, 22612, 22630, and 22633. Patients were excluded for emergency, trauma, and tumor cases due to the higher incidence of complications not directly related to the spine procedure.

Demographic, operative, postoperative, and comorbidity factors were characterized. Demographic factors included age, sex, and body mass index (BMI). Operative factors included operative time, estimated blood loss (EBL), number of levels, and time of operation. Postoperative factors included the presence of complications: VTE, PE, or unplanned return to OR for hematoma. Comorbidity factors included smoking status, heart disease, chronic obstructive pulmonary disease, diabetes, a history of cancer (not spread to the spine), and the American Society of Anesthesiologists' classification (ASA Class). The demographic and clinical characteristics for this population are shown in Table 2.

Outcome Variable

The primary outcomes were unplanned reoperation for hematoma, VTE, and PE within the 30-day perioperative period following elective spine surgery, which were operationalized as binary outcomes (yes/no). We modeled the incidence and probability of an unplanned reoperation, VTE, and PE.

The ACS-NSQIP database reports patient postoperative outcomes up to 30-days after the initial surgery. VTE, PE, and unplanned reoperations are reported if patients had an unplanned operation for any reason within the 30-days from the principal operative procedure performed either at the same hospital or at an outside hospital. Unplanned reoperations performed for hematoma formation were identified using CPT codes 22010, 22015, 63265, and 63268.


The 13 covariates listed in Table 1 were used for propensity score matching. The pool of covariates, which was selected a priori, included patient age (yrs), patient sex (male vs. female), patient BMI (kg/m2), smoking history, diabetes status, ASA classification (≥3 vs. <3), history of cardiac heart disease, history of COPD, history of solid tumor, history of liver disease. The intraoperative factors and case complexity are inherently difficult to quantify but as a surrogate we included as covariates the length of the operation (in minutes); length of anesthesia (in minutes), and number of levels involved in the operation. We did not correct for preoperative anticoagulation or antiplatelet usage as patients for elective surgery are asked to hold those medications prior to surgery.

Propensity Score Matching

Propensity score matching is a technique used to create a cohort, where exposure groups are (ideally) balanced on multiple covariates.[8] Briefly, it consists of the following steps: (1) selecting relevant covariates, (2) creating a logistic regression model using the covariates where the outcome is exposure, (3) estimating a scalar propensity score for each participant from the logistic regression model, (4) and then matching each exposed participant to unexposed participants through their nearest propensity scores. Thus, propensity score matching reduces confounding and potential selection bias by matching on observed covariates that predict exposure status. When successful, this technique creates a match where exposure groups are balanced on the predictive covariates. Thus, for the current study, a propensity score-matched sample of patients who received pharmacologic anticoagulation and those who did not was created using the aforementioned covariates (Table 1). The propensity score, defined as a participant's probability of being exposed to anticoagulation conditional on the observed covariates, was estimated in a logistic regression model (institutional cohort area under the ROC curve [AUC] = 0.810, standard error [SE] = 0.007, 95% confidence interval [CI] = 0.795–0.824). The C-Statistic of the propensity match model is the same as the area under the curve measure of 0.810. Next, each exposed patient (with pharmacological anticoagulation) was matched to one non-exposed patient (without pharmacological anticoagulation) with respect to the logit of their propensity scores using the procedures of PROC PSMATCH in SAS software (SAS Institute, Inc., Cary, NC) via the greedy nearest-neighbor matching algorithm, with a caliper of ±0.25 and no replacement. The covariates among exposed patients (with pharmacological anticoagulation) and matched non-exposed patients (without pharmacological anticoagulation) were indeed well-balanced within the institutional cohort (Figure 2); thus, indicating a tight match and mitigating confounding and selection bias on these observed covariates.

Figure 2.

Institutional propensity match figure detailing the tightness of the match between both cohorts. It shows the prematched cohort characteristics and then the matched cohort characteristics in different box plots. The box plots in conjunction with the logit of the propensity score indicates a tight and accurate match for both cohorts.

Statistical Analysis

Demographic and clinical characteristics for the matched sample of 1776 patients within the Institutional cohort and the 89,112 patients within the NSQIP cohort were described using the sample mean and standard deviation for continuous variables and the frequency and percentage for categorical variables.

Two-independent sample t test with the Satterthwaite method for unequal variances (continuous variables) and Fisher exact test (categorical variables) were used to identify any differences between the two groups (with pharmacological anticoagulation vs. without pharmacological anticoagulation) on each characteristic within the Institutional cohort.

The incidence of unplanned reoperation because of hematoma, VTE, and PE for patients who received elective spine surgery pharmacological anticoagulation and those without pharmacological anticoagulation was calculated for the NSQIP cohort and separately for the Institutional cohort. The incidence rate was calculated as the number of new cases of the event (outcome) divided by the total number of persons at risk for the event. The relative risk was also estimated.

A separate logistic regression analysis, with penalized maximum likelihood estimation along with Firth bias correction, was implemented to estimate the odds of VTE, PE, and unplanned reoperation for hematoma, respectively, from the operative indication (anticoagulation vs. no anticoagulation) for the Institutional cohort. Odds ratios along with the 95% confidence interval were reported.

Statistical analyses were carried out using SAS software, version 9.4 (SAS Institute, Inc.). The level of significance was set at α = 0.05 (two-tailed) and we implemented the false discovery rate (FDR) procedure, where applicable, to control false positives over the multiple tests.[9]