Association Between Antithrombotic Medication Use After Bioprosthetic Aortic Valve Replacement and Outcomes in the Veterans Health Administration System | Valvular Heart Disease | JAMA Surgery | JAMA Network
[Skip to Navigation]
Sign In
Figure.  Cohort Construction
Cohort Construction

AVR indicates aortic valve replacement; bAVR, bioprosthetic AVR; TAVR, transcatheter AVR; and VHA, Veterans Health Administration.

Table 1.  Baseline Characteristics of Patients With bAVR by Antithrombotic Strategy
Baseline Characteristics of Patients With bAVR by Antithrombotic Strategy
Table 2.  Medication Group Classification During the 1-Year Post-bAVR Period
Medication Group Classification During the 1-Year Post-bAVR Period
Table 3.  Outcome Events at 90 Days Among Patients With bAVR by Antithrombotic Strategya
Outcome Events at 90 Days Among Patients With bAVR by Antithrombotic Strategya
Table 4.  Association Between Antithrombotic Strategies and Outcomes for Unadjusted, Propensity Score–Adjusted, and Fully Adjusted Results
Association Between Antithrombotic Strategies and Outcomes for Unadjusted, Propensity Score–Adjusted, and Fully Adjusted Results
1.
Heras  M, Chesebro  JH, Fuster  V,  et al.  High risk of thromboemboli early after bioprosthetic cardiac valve replacement.  J Am Coll Cardiol. 1995;25(5):1111-1119. doi:10.1016/0735-1097(94)00563-6PubMedGoogle ScholarCrossref
2.
Whitlock  RP, Sun  JC, Fremes  SE, Rubens  FD, Teoh  KH.  Antithrombotic and thrombolytic therapy for valvular disease: Antithrombotic Therapy and Prevention of Thrombosis, 9th ed: American College of Chest Physicians Evidence-Based Clinical Practice Guidelines.  Chest. 2012;141(2)(suppl):e576S-e600S. doi:10.1378/chest.11-2305PubMedGoogle ScholarCrossref
3.
Riaz  H, Alansari  SA, Khan  MS,  et al.  Safety and use of anticoagulation after aortic valve replacement with bioprostheses: a meta-analysis.  Circ Cardiovasc Qual Outcomes. 2016;9(3):294-302. doi:10.1161/CIRCOUTCOMES.115.002696PubMedGoogle ScholarCrossref
4.
ElBardissi  AW, DiBardino  DJ, Chen  FY, Yamashita  MH, Cohn  LH.  Is early antithrombotic therapy necessary in patients with bioprosthetic aortic valves in normal sinus rhythm?  J Thorac Cardiovasc Surg. 2010;139(5):1137-1145. doi:10.1016/j.jtcvs.2009.10.064PubMedGoogle ScholarCrossref
5.
Massel  DR, Little  SH.  Antiplatelet and anticoagulation for patients with prosthetic heart valves.  Cochrane Database Syst Rev. 2013;(7):CD003464. doi:10.1002/14651858.CD003464.pub2PubMedGoogle Scholar
6.
Brennan  JM, Edwards  FH, Zhao  Y,  et al; DEcIDE AVR Research Team.  Early anticoagulation of bioprosthetic aortic valves in older patients: results from the Society of Thoracic Surgeons Adult Cardiac Surgery National Database.  J Am Coll Cardiol. 2012;60(11):971-977. doi:10.1016/j.jacc.2012.05.029PubMedGoogle ScholarCrossref
7.
Health Services Research and Development, U.S. Department of Veterans Affairs. Corporate Data Warehouse (CDW). https://www.hsrd.research.va.gov/for_researchers/vinci/cdw.cfm. Accessed November 13, 2017.
8.
Bravata  DM, Myers  LJ, Cheng  E,  et al.  Development and validation of electronic quality measures to assess care for patients with transient ischemic attack and minor ischemic stroke.  Circ Cardiovasc Qual Outcomes. 2017;10(9):e003157. doi:10.1161/CIRCOUTCOMES.116.003157PubMedGoogle ScholarCrossref
9.
Lorenz  KA, Asch  SM, Yano  EM, Wang  M, Rubenstein  LV.  Comparing strategies for United States veterans’ mortality ascertainment.  Popul Health Metr. 2005;3(1):2. doi:10.1186/1478-7954-3-2PubMedGoogle ScholarCrossref
10.
Sohn  MW, Arnold  N, Maynard  C, Hynes  DM.  Accuracy and completeness of mortality data in the Department of Veterans Affairs.  Popul Health Metr. 2006;4:2. doi:10.1186/1478-7954-4-2PubMedGoogle ScholarCrossref
11.
Cai  T, Giannopoulos  AA, Yu  S,  et al.  Natural language processing technologies in radiology research and clinical applications.  Radiographics. 2016;36(1):176-191. doi:10.1148/rg.2016150080PubMedGoogle ScholarCrossref
12.
Redman  JS, Natarajan  Y, Hou  JK,  et al.  Accurate identification of fatty liver disease in data warehouse utilizing natural language processing.  Dig Dis Sci. 2017;62(10):2713-2718. doi:10.1007/s10620-017-4721-9PubMedGoogle ScholarCrossref
13.
Mowery  DL, Chapman  BE, Conway  M,  et al.  Extracting a stroke phenotype risk factor from Veteran Health Administration clinical reports: an information content analysis.  J Biomed Semantics. 2016;7:26. doi:10.1186/s13326-016-0065-1PubMedGoogle ScholarCrossref
Limit 200 characters
Limit 25 characters
Conflicts of Interest Disclosure

Identify all potential conflicts of interest that might be relevant to your comment.

Conflicts of interest comprise financial interests, activities, and relationships within the past 3 years including but not limited to employment, affiliation, grants or funding, consultancies, honoraria or payment, speaker's bureaus, stock ownership or options, expert testimony, royalties, donation of medical equipment, or patents planned, pending, or issued.

Err on the side of full disclosure.

If you have no conflicts of interest, check "No potential conflicts of interest" in the box below. The information will be posted with your response.

Not all submitted comments are published. Please see our commenting policy for details.

Limit 140 characters
Limit 3600 characters or approximately 600 words
    Original Investigation
    Association of VA Surgeons
    December 26, 2018

    Association Between Antithrombotic Medication Use After Bioprosthetic Aortic Valve Replacement and Outcomes in the Veterans Health Administration System

    Author Affiliations
    • 1Health Services Research and Development, Center for Health Information and Communication, Richard L. Roudebush VA Medical Center, Department of Veterans Affairs, Indianapolis, Indiana
    • 2Health Services Research and Development, Precision Monitoring to Transform Care, Quality Enhancement Research Initiative, Department of Veterans Affairs, Indianapolis, Indiana
    • 3Department of Internal Medicine, Indiana University School of Medicine, Indianapolis
    • 4Department of Neurology, Indiana University School of Medicine, Indianapolis
    • 5Regenstrief Institute, Indianapolis, Indiana
    • 6VA Evidence-Based Synthesis Program, Portland, Oregon
    • 7Department of Internal Medicine, Portland VA Medical Center, Portland, Oregon
    • 8Department of Internal Medicine, Oregon Health & Science University, Portland
    • 9Department of Medical Informatics and Clinical Epidemiology, Oregon Health & Science University, Portland
    • 10Department of Biostatistics, Indiana University School of Medicine, Indianapolis
    • 11Department of Surgery, VA Boston Healthcare System, West Roxbury, Massachusetts
    JAMA Surg. 2019;154(2):e184679. doi:10.1001/jamasurg.2018.4679
    Key Points

    Question  Because the recommendations about antithrombotic medication use after bioprosthetic aortic valve replacement (bAVR) vary, how does post-bAVR antithrombotic practice across the Veterans Health Administration differ, and what are the associations between antithrombotic strategies and outcomes?

    Findings  Among 9060 veterans with bAVR at 47 facilities in this cohort study, the most commonly prescribed antithrombotic strategy was aspirin only. Adverse events were uncommon; patients receiving the combination of aspirin plus warfarin sodium had higher odds of bleeding than patients receiving aspirin only.

    Meaning  The combination of aspirin plus warfarin does not improve either mortality or thromboembolism risk but may increase the risk of bleeding compared with aspirin only for patients with bAVR.

    Abstract

    Importance  The recommendations about antithrombotic medication use after bioprosthetic aortic valve replacement (bAVR) vary.

    Objectives  To describe the post-bAVR antithrombotic medication practice across the Veterans Health Administration (VHA) and to assess the association between antithrombotic strategies and post-bAVR outcomes.

    Design, Setting, and Participants  Retrospective cohort study. Multivariable modeling with propensity scores was conducted to adjust for differences in patient characteristics across the 3 most common antithrombotic medication strategies (aspirin plus warfarin sodium, aspirin only, and dual antiplatelets). Text mining of notes was used to identify the patients with bAVR (fiscal years 2005-2015).

    Main Outcomes and Measures  This study used VHA and non-VHA outpatient pharmacy data and text notes to classify the following antithrombotic medications prescribed within 1 week after discharge from the bAVR hospitalization: aspirin plus warfarin, aspirin only, dual antiplatelets, no antithrombotics, other only, and warfarin only. The 90-day outcomes included all-cause mortality, thromboembolism risk, and bleeding events. Outcomes were identified using primary diagnosis codes from emergency department visits or hospital admissions.

    Results  The cohort included 9060 veterans with bAVR at 47 facilities (mean [SD] age, 69.3 [8.8] years; 98.6% male). The number of bAVR procedures per year increased from 610 in fiscal year 2005 to 1072 in fiscal year 2015. The most commonly prescribed antithrombotic strategy was aspirin only (4240 [46.8%]), followed by aspirin plus warfarin (1638 [18.1%]), no antithrombotics (1451 [16.0%]), dual antiplatelets (1010 [11.1%]), warfarin only (439 [4.8%]), and other only (282 [3.1%]). Facility variation in antithrombotic prescription patterns was observed. During the 90-day post-bAVR period, adverse events were uncommon, including all-cause mortality in 127 (1.4%), thromboembolism risk in 142 (1.6%), and bleeding events in 149 (1.6%). No differences in 90-day mortality or thromboembolism were identified across the 3 antithrombotic medication groups in either the unadjusted or adjusted models. Patients receiving the combination of aspirin plus warfarin had higher odds of bleeding than patients receiving aspirin only in the unadjusted analysis (odds ratio, 2.58; 95% CI, 1.71-3.89) and after full risk adjustment (adjusted odds ratio, 1.92; 95% CI, 1.17-3.14).

    Conclusions and Relevance  These data demonstrate that bAVR procedures are increasingly being performed in VHA facilities and that aspirin only was the most commonly used antithrombotic medication strategy after bAVR. The risk-adjusted results suggest that the combination of aspirin plus warfarin does not improve either all-cause mortality or thromboembolism risk but increases the risk of bleeding events compared with aspirin only.

    Introduction

    Although bioprosthetic aortic valve replacement (bAVR) is generally well tolerated, patients after bAVR are at increased risk of thromboembolism.1 However, the recommendations about post-bAVR antithrombotic medication vary.2-5 After bAVR, patients may receive a variety of antithrombotic medications, including warfarin sodium–based and antiplatelet-based strategies.

    Investigators have examined the association between alternative antithrombotic strategies and post-bAVR outcomes. For example, a large cohort study6 included 25 656 elderly patients with bAVR (2004-2006); compared with aspirin only, the aspirin plus warfarin strategy was associated with a reduced risk of death or embolic events but a higher risk of bleeding.

    The Veterans Health Administration (VHA) is the single largest health care system in the United States. Our objectives were to describe the post-bAVR antithrombotic medication practice across the VHA system from fiscal year 2005 to fiscal year 2015 and to assess the association between antithrombotic strategies and post-bAVR outcomes.

    Methods
    Data Sources

    The VHA electronic health record—Veterans Information System Technology Architecture (VistA)—includes diagnoses, procedures, medications, laboratory values, physiologic measurements, and text notes and reports. Data are aggregated from VistA to the Corporate Data Warehouse (CDW),7 a national repository of clinical and administrative data. Data from multiple domains were used, including inpatient and outpatient diagnosis codes, surgery procedure codes, laboratory data, orders, consults, allergies, health factors, and pharmacy data. The Textual Information Utilities (TIU) documents store textual information, including surgery notes, progress notes, admission and discharge summaries, and notes sent to VHA health care professionals from non-VHA health care professionals. The approach to identifying atrial fibrillation was validated previously.8 The VA Vital Status File9 contains death dates.9,10 Linked VHA–Centers for Medicare & Medicaid Services data were used to identify outcome events at non-VHA facilities. The study received institutional research (Indiana University School of Medicine Institutional Review Board) and VHA Research and Development approvals. Informed consent of participants was not needed for this retrospective cohort study.

    Cohort Construction

    We identified veterans who received a bAVR (with or without coronary artery bypass graft) in any VHA facility during the period of fiscal year 2005 to fiscal year 2015. We first identified veterans with a Current Procedural Terminology (CPT) or International Classification of Diseases, Ninth Revision (ICD-9) procedure code for aortic valve replacement (AVR) (CPT codes 33361-33369, 33404-33406, and 33410-33413; ICD-9 procedure codes 35.05, 35.06, 35.21, and 35.22). Although prior studies have identified bAVR using CPT or ICD-9 procedure codes, our medical record reviews indicated that these procedure codes cannot reliably distinguish mechanical AVR from bAVR. Because our focus was on bAVR, we needed to exclude patients with mechanical valve replacement. Therefore, we used text mining to differentiate between mechanical AVR and bAVR.

    We used TIU notes to identify patients who received bAVR. The Nurse Intraoperative Report was the primary source of valve information, including time in and out of the operating room, type of operation, names of operating room personnel, all prostheses installed, medications provided, and other text notes. The prosthesis list includes information on item name, vendor, model, lot and serial number, and size. We used this list to identify bAVR models using vendor names and model numbers. If the Nurse Intraoperative Report was not available, we used all other available text notes (eg, surgeon’s operative report, anesthesiology report, and progress notes) to identify AVR type. Our text-mining approach was iterative; we used text mining to search for bAVR and then conducted medical record reviews (n = 405) to refine and improve the text mining. Patients classified as having mechanical AVR or those who could not be classified were excluded. We further excluded patients who had the following characteristics: received bAVR at a non-VHA hospital, without any filled VHA pharmacy prescriptions in the 1 year before or 1 year after bAVR, had a prior AVR, experienced in-hospital death, were still hospitalized 30 days after bAVR, were discharged to hospice, were transferred to another VHA or non-VHA hospital, left against medical advice, were admitted more than 30 days before bAVR, had a transcatheter AVR (TAVR) procedure, or had an in-hospital outcome event (Figure).

    Classification of Antithrombotic Medications

    We used a variety of CDW data sources to classify antithrombotic medication use in the 1 year before bAVR and the 1 year after bAVR. We used the following data sources: VHA and non-VHA outpatient pharmacy data for medications filled at VHA pharmacies (ie, medications prescribed by VHA health care professionals and filled at a VHA pharmacy), VHA order data that indicated whether a VHA-prescribed medication was obtained at a non-VHA pharmacy (ie, medications prescribed by VHA health care professionals that a patient fills outside of the VHA), non-VHA pharmacy files (ie, veteran taking a medication that was prescribed by a non-VHA health care professional and was filled at a non-VHA pharmacy), and text notes. We conducted medical record reviews to confirm the medication identification and classification strategies. The medical record review–based iterative approach was helpful in identifying aspirin use because many veterans obtain aspirin outside of the VHA. The medical record reviews were instrumental in identifying data domains that were available in the CDW (eg, health factors) to identify the non-VHA aspirin prescriptions. The eAppendix in the Supplement provides a summary of medical record review findings associated with medication classification.

    Antithrombotic medications initiated within the first week after bAVR were classified into 1 of the following 6 strategies (eTable 1 in the Supplement): aspirin plus warfarin, aspirin only, dual antiplatelets, no antithrombotics, other only, and warfarin only. The 1-week postdischarge classification was used as the primary medication classification for the analyses. Because we observed that patients changed medications over the course of the 1-year period, we conducted sensitivity analyses using proportional hazards regression and censoring of observations with a change in medication before an outcome event or time point of interest (either 90 days or 180 days after discharge).

    Outcomes

    The outcomes describing potential benefits of post-bAVR antithrombotic medication use included all-cause mortality, thromboembolism risk (ie, myocardial infarction or acute coronary syndrome, ischemic stroke, pulmonary embolism, and peripheral arterial embolism), and bleeding events. The outcomes describing potential harms of antithrombotics included gastrointestinal, intracranial, genitourinary, retroperitoneal, and pulmonary bleeding events. Outcomes were identified using primary diagnosis codes from emergency department visits or hospital admissions during the 90 days after bAVR procedure hospital discharge. Both VHA and Medicare data were used to identify outcome events.

    Analyses

    The baseline characteristics included the following: age, sex, race/ethnicity, history of tobacco smoking and other medical conditions (eg, history of atrial fibrillation), concomitant coronary artery bypass graft, history of a major bleeding event, a documented allergy to aspirin or warfarin, use of aspirin plus warfarin in the 1 year before bAVR, Charlson Comorbidity Index, and past health care use. We used VHA and Centers for Medicare & Medicaid Services data from 1999 to 2015 to identify demographic and past medical, allergy, and habit data; this approach provided a look-back period of at least 5 years. Differences were examined in the baseline characteristics across the 6 antithrombotic groups. These analyses were descriptive; proportions were reported for binary variables, and means (SDs) were reported for continuous variables. χ2 Tests and Kruskal-Wallis tests were used to compare outcomes and the baseline characteristics across all antithrombotic groups.

    We performed adjusted analyses comparing the 3 most commonly observed antithrombotic medication strategies (aspirin plus warfarin, aspirin only, and dual antiplatelets) using a propensity score analysis; the aspirin-only group served as the reference category. We developed the propensity scores using a multinomial logit model to identify the baseline variables that were significantly associated with medication group. We included a random effect for the surgical facility to account for potential similarities in medication use within a facility. The predicted probabilities of each antithrombotic medication strategy were used as covariates in the final models. We used mixed-effects logistic regressions with antithrombotic medication group and the probabilities of the medication groups as the fixed effects and used a random effect for the surgical facility to assess the association between antithrombotic medication strategies and outcomes (referred to in the text as propensity scoreadjusted models). We also constructed models that included both propensity score adjustment and baseline variables that were significantly associated with the outcomes (referred to in the text as fully adjusted models). Sensitivity analyses included an offset variable equal to the log (treatment time) to account for different durations of antithrombotic medication use over time. Results were similar across methods, so we report only the propensity score–adjusted model findings and the fully adjusted model findings. Missing data were rare, and no imputations were made. A statistical software program (SAS Enterprise Guide 7.1; SAS Institute Inc) was used for data analysis. Statistical significance was identified on the basis of a 2-sided P < .05.

    Results

    The study cohort included 9060 veterans with bAVR at 47 facilities, with the number of patients per facility ranging from 1 to 580 (median, 187.5); 41 facilities performed at least 10 procedures during the study period (Figure). The number of bAVR procedures per year increased from 610 in fiscal year 2005 to 1072 in fiscal year 2015 (eFigure 1 in the Supplement).

    The patients with bAVR had a mean (SD) age of 69.3 (8.8) years, were predominantly male, and were mostly of white race/ethnicity (Table 1). In total, 6311 (69.7%) had used aspirin before bAVR; only 977 (10.8%) had used warfarin before bAVR. Key differences in the baseline characteristics were observed across the medication strategies (Table 1). As expected, patients with prior major bleeding were less likely to be prescribed aspirin or warfarin after bAVR. Patients with concomitant coronary artery bypass graft were more likely to use dual antiplatelet therapy after bAVR. Patients with a history of atrial fibrillation were more likely to receive warfarin with or without aspirin.

    The most commonly prescribed antithrombotic strategy within 1 week after discharge from the bAVR hospitalization was aspirin only (4240 [46.8%]), followed by aspirin plus warfarin (1638 [18.1%]), no antithrombotics (1451 [16.0%]), dual antiplatelets (1010 [11.1%]), warfarin only (439 [4.8%]), and other only (282 [3.1%]). During the 1-year period after bAVR, considerable medication switching occurred (Table 2) such that, by 1 year after bAVR, the patients were classified as comprising the following medication groups: 2762 (30.5%) were receiving aspirin plus warfarin, 4541 (50.1%) were receiving aspirin only, 1342 (14.8%) were receiving dual antiplatelets, 131 (1.4%) were receiving no antithrombotics, 116 (1.3%) were receiving other only (1.3%), and 168 (1.9%) were receiving warfarin only. Therefore, most patients who were discharged on no antithrombotic medication eventually received an antithrombotic medication during the 1 year after bAVR; however, some of these medication changes were made after an outcome event. It was for this reason that the 1-week postdischarge period was used to classify medication strategies for the primary analysis.

    Facility variation in antithrombotic prescription patterns was observed (eFigure 2 in the Supplement). This facility variation in medication strategies did not seem to be associated with the prevalence of atrial fibrillation. The proportion of patients receiving various antithrombotic medication strategies was constant over time.

    Table 3 lists unadjusted post-bAVR outcome data (all-cause mortality, thromboembolism risk, and bleeding events). All-cause mortality within 90 days was observed in 127 (1.4%), thromboembolism risk in 142 (1.6%), and bleeding events in 149 (1.6%). The highest rate of bleeding events was among patients receiving the combination of aspirin plus warfarin (46 of 1638 [2.8%]).

    Adjusted results are listed in Table 4. No differences in 90-day all-cause mortality or thromboembolism risk were identified across the 3 antithrombotic medication groups in either the unadjusted or adjusted models. The propensity score covariates were not significant in either the all-cause mortality or thromboembolism risk models. In unadjusted analysis, patients receiving the combination of aspirin plus warfarin had significantly higher odds of 90-day bleeding than patients taking aspirin only. After adjusting for the propensity score, the patients receiving aspirin plus warfarin still had significantly elevated odds of bleeding events; however, the odds ratio decreased from 2.58 (unadjusted) to 1.89 (propensity score model). The baseline characteristics that were associated with 90-day bleeding included age, history of coagulation defect, history of bleeding, and history of liver disease. The propensity score–adjusted model and the fully adjusted model findings were similar. With regard to the composite outcome of all-cause mortality, thromboembolism risk, and bleeding events, no significant differences between the groups were observed after adjusting for the propensity scores.

    The sensitivity analyses using proportional hazards modeling with censoring of observations if the medication was changed were in alignment with the main results (eTable 2 in the Supplement). As in the main analysis, there were no differences in 90-day mortality or thromboembolism events among the medication groups. As in the fully adjusted model, the sensitivity analyses found that the combination of aspirin plus warfarin was associated with a higher risk of bleeding at 90 days (hazard ratio, 2.58; 95% CI, 1.54-4.32) and at 180 days (hazard ratio, 2.06; 95% CI, 1.34-3.18). In the main analysis, the risk of any adverse event was higher in the group receiving aspirin plus warfarin, but this finding was not statistically significant; however, in the sensitivity analyses, the proportional hazard for the outcome of any adverse event was statistically significant at 90 days (hazard ratio, 1.47; 95% CI, 1.06-2.04) and at 180 days (hazard ratio, 1.49; 95% CI, 1.12-1.98).

    Discussion

    To our knowledge, these data provide the first national examination of antithrombotic medication use for patients receiving bAVR in the VHA. The results demonstrate that bAVR procedures are increasing (from 610 in fiscal year 2005 to 1072 in fiscal year 2015) and show that 46.8% of the patients with bAVR were prescribed aspirin only, 18.1% were taking the combination of aspirin plus warfarin, and 11.1% received dual antiplatelets.

    The antithrombotic medication pattern was similar to that reported by Brennan and colleagues6 for a US (nonveteran) bAVR population in terms of aspirin only but differed for other antithrombotic medications; they reported that 49% received aspirin only, 23% received aspirin plus warfarin, 12% received warfarin only, 8% received dual antiplatelet therapy, and 7% received no anticoagulation. Compared with the present study, their rate of using warfarin only was higher (12% vs 4.8%), whereas our no anticoagulation rate was higher (16.0% vs 7%). It may be that these differences could be explained by variations in the baseline characteristics of the cohorts, although differences in reporting of comorbidities make a direct comparison difficult.

    The variation in antithrombotic medication use across facilities warrants further attention. Although differences in patient characteristics might explain this variation, it did not appear that differences in atrial fibrillation accounted for the observed variation. Future research should include an assessment of the risk-adjusted variation across facilities.

    All-cause mortality was lower in our VHA cohort vs that reported by Brennan et al6 for aspirin plus warfarin (1.6% vs 3.1%), aspirin only (1.3% vs 3.0%), and warfarin only (1.6% vs 4.0%). The rates of thromboembolism risk were also lower in our VHA cohort vs those in the cohort by Brennan et al for aspirin plus warfarin (1.7% vs 0.6%), aspirin only (1.5% vs 1.0%), and warfarin only (1.4% vs 1.0%). However, rates of bleeding events in our VHA cohort were similar to those reported by Brennan et al for aspirin plus warfarin (2.8% vs 2.8%), aspirin only (1.1% vs 1.0%), and warfarin only (1.8% vs 1.4%). Our multivariable approach included propensity score adjustment to account for differences in patient characteristics across the medication groups. We found similar odds of 90-day all-cause mortality and thromboembolism risk across the 3 main medication groups. However, patients taking the combination of aspirin plus warfarin had higher odds of bleeding than patients taking aspirin only. These findings differ from those by Brennan et al, who reported a lower risk of death and thromboembolism but a higher risk of bleeding for patients taking aspirin plus warfarin compared with patients taking aspirin only. Our results have been shared with VHA leadership to ensure that VHA practitioners are aware of these findings.

    A contribution that the present findings offer is the inclusion of patients receiving dual antiplatelets in the multivariable modeling. The results of the propensity score–adjusted model and the fully adjusted model indicate that, although the odds of 90-day bleeding events were elevated, this elevation did not achieve statistical significance.

    An innovation of this project was the use of text mining for cohort construction. This approach is gaining acceptance and popularity given the availability of notes (which contain text data) within the VHA CDW.11 For example, Redman et al12 used natural language processing on CDW radiology reports to identify patients with fatty liver disease, and Mowery et al13 used natural language processing on radiology reports and TIU notes to identify patients with carotid stenosis.

    Limitations

    Although these data provide a robust examination of bAVR procedures across the VHA system over time, some limitations should be noted. First, these results are derived from an observational cohort constructed using administrative data. It is unlikely that a randomized clinical trial will be performed to answer questions associated with risks and harms of antithrombotic medication strategies after bAVR; therefore, a large, nationally representative cohort study like the present one will necessarily be the source of data needed to inform clinical practice. However, these results should be interpreted with the caution that is appropriately applied to observational data. Although we conducted medical record review to confirm cohort construction and medication classification, the use of administrative data may have introduced errors in medication group assignment and may have underestimated outcome events. The combination of VHA and non-VHA pharmacy data could identify the start of medication use; however, it could not reliably identify medication use end dates. Therefore, we were unable to assess medication sequences (eg, aspirin plus warfarin for 90 days, followed by aspirin only). The text-mining approach for cohort construction allowed us to exclude patients with mechanical AVR (thereby ensuring that the cohort only included the patients with bAVR), but we may have excluded some patients with bAVR; hence, our estimates of procedure volume may have been underestimated. Second, although the analyses included an assessment of bAVR across the VHA system, one facility was excluded because no TIU notes were available for that facility; therefore, patients receiving a bAVR procedure there could not be identified. Third, because we did not conduct a full medical record review or interview clinicians, we could not comment on the clinical reasoning for selecting certain antithrombotic medication strategies. Fourth, because the administrative data do not include a measure of patient preference, we could not examine the degree to which a patient’s preferences for or against a particular strategy contributed to practice. Fifth, although we included VHA and non-VHA medication data, we appreciate that this information represented an underestimate of the rate of medication use (eg, aspirin) from non-VHA sources; the degree of underestimation is unknown. Sixth, this study focuses on the postdischarge period; therefore, the potential effects of antithrombotic medication use in the immediate periprocedural (ie, inpatient) period on outcomes could not be examined. Seventh, we excluded patients with TAVR; therefore, these results should not be generalized to patients with TAVR. Eighth, we did not assess echocardiography data and thus could not comment on valve functioning in the post-bAVR period. Ninth, although we excluded patients with a history of AVR, distant prior AVRs (before our look-back period) might have been missed; therefore, some patients in this cohort may have had their bAVR as a second procedure.

    Conclusions

    These data demonstrate that bAVR procedures are increasingly being performed in VHA facilities. Aspirin only was the most commonly used antithrombotic medication strategy after bAVR. The risk adjustment models suggest that the combination of aspirin plus warfarin does not improve either all-cause mortality or thromboembolism risk but increases the risk of bleeding events compared with aspirin only.

    Back to top
    Article Information

    Accepted for Publication: September 23, 2018.

    Corresponding Author: Dawn M. Bravata, MD, Health Services Research and Development, Center for Health Information and Communication, Richard L. Roudebush VA Medical Center, 1481 W 10th St, Mail Code 11H, Indianapolis, IN 46202 (dawn.bravata2@va.gov).

    Published Online: December 26, 2018. doi:10.1001/jamasurg.2018.4679

    Author Contributions: Drs Bravata and L. J. Myers had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

    Concept and design: Bravata, Kansagara, L. J. Myers.

    Acquisition, analysis, or interpretation of data: Bravata, Coffing, J. Myers, Murphy, Homoya, Perkins, Snow, Quin, Ying, L. J. Myers.

    Drafting of the manuscript: Bravata, Coffing, Murphy, Homoya, Snow.

    Critical revision of the manuscript for important intellectual content: Bravata, Kansagara, J. Myers, Homoya, Perkins, Quin, Ying, L. J. Myers.

    Statistical analysis: Bravata, Coffing, Perkins, Ying.

    Obtained funding: Bravata, Kansagara.

    Administrative, technical, or material support: Bravata, J. Myers, Murphy, Homoya, Snow, L. J. Myers.

    Conflict of Interest Disclosures: None reported.

    Funding/Support: This work was supported by the Department of Veterans Affairs, Health Services Research and Development Service, Evidence Synthesis Program (ESP 05-225) and the Precision Monitoring to Transform Care Quality Enhancement Research Initiative (QUE 15-280). Support for Veterans Health Administration–Centers for Medicare & Medicaid Service data is provided by the VA Information Resource Center (SDR 02-237 and 98-004).

    Role of the Funder/Sponsor: The funding sources had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

    Disclaimer: The views expressed in this article are those of the authors and do not necessarily reflect the position or policy of the Department of Veterans Affairs or the US government.

    Meeting Presentation: This study was presented at the Association of VA Surgeons 2018 Annual Meeting; May 21, 2018; Miami Beach, Florida.

    References
    1.
    Heras  M, Chesebro  JH, Fuster  V,  et al.  High risk of thromboemboli early after bioprosthetic cardiac valve replacement.  J Am Coll Cardiol. 1995;25(5):1111-1119. doi:10.1016/0735-1097(94)00563-6PubMedGoogle ScholarCrossref
    2.
    Whitlock  RP, Sun  JC, Fremes  SE, Rubens  FD, Teoh  KH.  Antithrombotic and thrombolytic therapy for valvular disease: Antithrombotic Therapy and Prevention of Thrombosis, 9th ed: American College of Chest Physicians Evidence-Based Clinical Practice Guidelines.  Chest. 2012;141(2)(suppl):e576S-e600S. doi:10.1378/chest.11-2305PubMedGoogle ScholarCrossref
    3.
    Riaz  H, Alansari  SA, Khan  MS,  et al.  Safety and use of anticoagulation after aortic valve replacement with bioprostheses: a meta-analysis.  Circ Cardiovasc Qual Outcomes. 2016;9(3):294-302. doi:10.1161/CIRCOUTCOMES.115.002696PubMedGoogle ScholarCrossref
    4.
    ElBardissi  AW, DiBardino  DJ, Chen  FY, Yamashita  MH, Cohn  LH.  Is early antithrombotic therapy necessary in patients with bioprosthetic aortic valves in normal sinus rhythm?  J Thorac Cardiovasc Surg. 2010;139(5):1137-1145. doi:10.1016/j.jtcvs.2009.10.064PubMedGoogle ScholarCrossref
    5.
    Massel  DR, Little  SH.  Antiplatelet and anticoagulation for patients with prosthetic heart valves.  Cochrane Database Syst Rev. 2013;(7):CD003464. doi:10.1002/14651858.CD003464.pub2PubMedGoogle Scholar
    6.
    Brennan  JM, Edwards  FH, Zhao  Y,  et al; DEcIDE AVR Research Team.  Early anticoagulation of bioprosthetic aortic valves in older patients: results from the Society of Thoracic Surgeons Adult Cardiac Surgery National Database.  J Am Coll Cardiol. 2012;60(11):971-977. doi:10.1016/j.jacc.2012.05.029PubMedGoogle ScholarCrossref
    7.
    Health Services Research and Development, U.S. Department of Veterans Affairs. Corporate Data Warehouse (CDW). https://www.hsrd.research.va.gov/for_researchers/vinci/cdw.cfm. Accessed November 13, 2017.
    8.
    Bravata  DM, Myers  LJ, Cheng  E,  et al.  Development and validation of electronic quality measures to assess care for patients with transient ischemic attack and minor ischemic stroke.  Circ Cardiovasc Qual Outcomes. 2017;10(9):e003157. doi:10.1161/CIRCOUTCOMES.116.003157PubMedGoogle ScholarCrossref
    9.
    Lorenz  KA, Asch  SM, Yano  EM, Wang  M, Rubenstein  LV.  Comparing strategies for United States veterans’ mortality ascertainment.  Popul Health Metr. 2005;3(1):2. doi:10.1186/1478-7954-3-2PubMedGoogle ScholarCrossref
    10.
    Sohn  MW, Arnold  N, Maynard  C, Hynes  DM.  Accuracy and completeness of mortality data in the Department of Veterans Affairs.  Popul Health Metr. 2006;4:2. doi:10.1186/1478-7954-4-2PubMedGoogle ScholarCrossref
    11.
    Cai  T, Giannopoulos  AA, Yu  S,  et al.  Natural language processing technologies in radiology research and clinical applications.  Radiographics. 2016;36(1):176-191. doi:10.1148/rg.2016150080PubMedGoogle ScholarCrossref
    12.
    Redman  JS, Natarajan  Y, Hou  JK,  et al.  Accurate identification of fatty liver disease in data warehouse utilizing natural language processing.  Dig Dis Sci. 2017;62(10):2713-2718. doi:10.1007/s10620-017-4721-9PubMedGoogle ScholarCrossref
    13.
    Mowery  DL, Chapman  BE, Conway  M,  et al.  Extracting a stroke phenotype risk factor from Veteran Health Administration clinical reports: an information content analysis.  J Biomed Semantics. 2016;7:26. doi:10.1186/s13326-016-0065-1PubMedGoogle ScholarCrossref
    ×