Patients and Methods
Institutional review board approval was obtained prior to undertaking this study. All percutaneous transfemoral TAVR cases performed at the North Florida/South Georgia Veterans Affairs Medical Center (NF/SG VAMC) from January 2014 to December 2019 were retrospectively reviewed. TAVR patients were identified through a query of the local Veterans Affairs Surgery Quality Improvement Program (VASQIP) database, which contains baseline patient characteristics and clinical outcomes data for every cardiac surgical procedure performed. Valve-in-valve procedures and cases with missing cost data were excluded. The valve types changed with the evolving technology. The Edwards Sapien and Sapien XT valves used primarily between 2014-2015, and Edwards Sapien 3 from 2016-2019. The Medtronic Evolut R and Evolut Pro were used in a minority of cases. Patients were categorized by anesthesia type into MAC versus GA groups based on the utilization of endotracheal intubation. The decision to administer MAC or GA for each case was made in a multidisciplinary team format which included input from anesthesiologists, cardiologists, and cardiac surgeons. The analysis was performed on an as-treated basis.
Demographics, baseline medical history, and outcome data points were collected according to VASQIP definitions. Diabetes was defined as any outpatient preoperative hyperglycemia requiring oral medications or insulin. History of cardiac surgery was defined as any history of cardiac surgery, on or off pump, such as coronary artery bypass graft, valve replacement, or other procedures. Preoperative atrial fibrillation was defined as the presence of atrial fibrillation or atrial flutter in the two weeks preceding surgery. Preoperative myocardial infarction (MI) was defined as any prior history of MI, regardless of timeframe, prior to surgery. Thirty-day postoperative morbidity and mortality estimates were calculated by the VASQIP risk score.
Complications and outcomes were defined as follows: postoperative renal failure was defined as the development of new renal failure requiring renal replacement therapy, or an exacerbation of preoperative renal failure requiring initiation of renal replacement therapy within 30 days postoperatively. Postoperative atrial fibrillation included new-onset atrial fibrillation or atrial flutter requiring any treatment. Postoperative MI included any MI that occurred within 30 days postoperatively. Postoperative cerebrovascular accident (CVA) was defined as any new objective neurologic deficit lasting more than 72 hours with onset immediately post-operatively or occurring within the 30 days after surgery. Intraoperative cardiac arrest was defined as any cardiac arrest requiring external or open cardiopulmonary resuscitation (CPR) in the operating room, whereas postoperative cardiac arrest included any cardiac arrest requiring CPR within 30 days postoperatively. All lab values represent the results reported closest to the date of the surgery and exclude any lab values collected beyond 30 days before surgery.
Direct cost data was derived from the local VA Managerial Cost Accounting system, which collects and stores direct cost data associated with every VA hospitalization. Costs were reported under the following categories: direct operating room (OR) cost, direct intensive care unit (ICU) cost, and overall direct costs. Direct costs are defined as costs directly associated with patient care, such as anesthesia services, nursing services, drugs, medical supplies, and hospitalization. Indirect costs, commonly referred to as overhead costs, encompass the expenses not directly tied to a patient’s care, such as administrative expenses, medical records, and information services. Indirect costs were not analyzed in this study, and all costs subsequently mentioned are direct costs only. The types and costs of the valve prostheses used for each patient were extracted from a local VA prosthetics database. Total hospital cost was calculated, which included physician charges, labs, imaging, operating room, intensive care unit, step-down unit, and hospital floor costs. Overall cost included all these inpatient charges, plus the cost of the prosthetic valve.
A non-matched analysis was initially performed. Patient characteristics, ICU LOS, total hospital LOS, and costs were compared using Fisher’s exact test for binary data and Wilcoxon rank sum test for continuous data. Propensity score matching was then performed. The R package “matchit” was used to match patients who received GA with those who received MAC according to their propensity for receiving either. To assess propensity, a logistic regression model with MAC as the outcome and VASQIP estimated mortality, age, sex, and BMI as covariates was used. History of diabetes, history of COPD, creatinine, and hemoglobin were excluded from the model because these factors are integrated into the VASQIP estimated mortality. Additionally, only preoperative variables were utilized, and no procedure-related variables, including postoperative echocardiogram findings (gradient, presence/absence of paravalvular leaks, postoperative ejection fraction), EKG changes and/or pacer requirement, were integrated. Patients were matched using a 1-to-1 matching with a caliper width of 0.1 (meaning that matched pairs were within 0.1 standard deviations of each other on propensity score). This process resulted in the matching of 83 of 105 GA recipients with 83 of 244 MAC recipients. To assess the effect of anesthesia type on outcomes, mixed-effects models (logistic or linear regression as appropriate) with anesthesia type as a fixed factor and a random factor for a matched pair were used to account for the dependency introduced by matching. For outcomes that were too sparse for a logistic regression model, McNemar’s test for paired data was used. Finally, to assess whether OR times were significantly shorter with MAC due to the anesthesia modality itself or due to increased institutional experience (as MAC was used increasingly in later years), OR time was compared between MAC and GA groups for the year 2016 only. All analyses were performed using R (V.4.0.2, The R Foundation for Statistical Computing, Vienna, Austria).