Author + information
- Received July 13, 2018
- Revision received September 26, 2018
- Accepted October 16, 2018
- Published online January 7, 2019.
- Arash Salemi, MDa,
- Art Sedrakyan, MD, PhDb,
- Jialin Mao, MD, MSb,
- Adham Elmously, MDa,
- Harindra Wijeysundera, MD, PhDc,
- Derrick Y. Tam, MDc,
- Antonino Di Franco, MDa,
- Simon Redwood, MDd,
- Leonard N. Girardi, MDa,
- Stephen E. Fremes, MD, MScc and
- Mario Gaudino, MDa,∗ ()
- aDepartment of Cardiothoracic Surgery, Weill Cornell Medical College, New York, New York
- bDepartment of Healthcare Policy and Research, Weill Cornell Medical College, New York, New York
- cSchulich Heart Centre, Sunnybrook Health Science, University of Toronto, Toronto, Ontario, Canada
- dBHF Centre of Excellence, Cardiovascular Department, King’s College St. Thomas’ Hospital, London, United Kingdom
- ↵∗Address for correspondence:
Dr. Mario Gaudino, Weill Cornell Medicine, Department of Cardiothoracic Surgery, 525 East 68th Street, New York, New York 10065.
Objectives The aim of this study was to assess the impact of individual operator experience on transfemoral transcatheter aortic valve replacement (TAVR) outcomes.
Background TAVR volume-outcome relationships have not been evaluated at the individual operator level.
Methods New York Statewide Planning and Research Cooperative System data from 8,771 transfemoral TAVR procedures performed by 207 operators between 2012 and 2016 were analyzed. Operator volume was defined as the number of TAVR procedures performed during 1 year prior to the index procedure. Hierarchical and restrictive cubic spline regression models were used to evaluate the impact of individual operator experience on risk-adjusted in-hospital outcomes. The primary outcome was a composite of in-hospital mortality, stroke, and/or acute myocardial infarction. Secondary outcomes were the individual components of the primary outcome.
Results After adjusting for hospital and physician characteristics, patients undergoing TAVR performed by high-volume physicians (≥80/year) had a significantly lower risk for death, stroke, or acute myocardial infarction (odds ratio: 0.59; 95% confidence interval: 0.37 to 0.93) compared with those treated by low-volume physicians (<24/year). Being treated by operators who performed 200 procedures during the prior year was associated with significantly lower risks for post-procedural stroke (odds ratio: 0.41; 95% confidence interval: 0.17 to 0.97) and composite events (odds ratio: 0.45; 95% confidence interval: 0.26 to 0.78). This relationship was nonlinear, and a sensitivity analysis excluding the first 10, 20, and 30 procedures for each operator mitigated the effect of the initial learning curve.
Conclusions Increased TAVR experience of operators is associated with improved risk-adjusted in-hospital outcomes. These results have potentially important implications for individual training and hospital programs in TAVR.
Transcatheter aortic valve replacement (TAVR) for severe aortic stenosis has emerged as a successful alternative to surgical aortic valve replacement, demonstrating a significant benefit compared with medical therapy in patients deemed to be inoperable, as well as noninferiority to surgical aortic valve replacement in patients deemed to be at intermediate or high risk for surgery (1–5). In the era of commercialization of TAVR technology after U.S. Food and Drug Administration approval, results with “real-world” commercial use and clinical trials have been comparable (6,7).
The relationship between hospital procedure volume and outcomes has been a topic of considerable interest within the academic community and health care systems and has been studied extensively, with a substantial body of research supporting it with a variety of surgical procedures, including cardiac surgical procedures (8–11). The Leapfrog Group, a consortium of private organizations and public agencies that purchase health care, has established minimum volume standards for several surgical procedures (12).
In the realm of TAVR, quantifiable technical learning curves at the hospital level have been demonstrated, and thus 1 of the important structure measures of quality has been center experience (13–17). There have been numerous efforts by multiple governing bodies to attempt to diminish the TAVR learning curve, as evident by site selection criteria based on volume metrics for reimbursement (18–20). However, data regarding the relationship between individual operator experience and TAVR outcomes have not been previously studied.
Recent evidence is supportive of a hospital volume–outcome relationship in TAVR practice (13–17). Hospital-level studies are based on the hypothesis that increasing experience will eventually lead to proficiency, with proficiency being a certain asymptote of improved outcomes on the learning curve (21–23). In the seminal trials leading to the Food and Drug Administration approval of TAVR, proficiency in technical performance (i.e., procedure time, fluoroscopy time, contrast load, device success) as well as a plateau in post-operative clinical outcomes and complications were achieved by centers after an average of approximately 25 to 30 procedures (24,25). However, the number of cases required to achieve proficiency progressively decreased for late-entering trial centers (24,25).
The aim of the present study was to systematically evaluate the relationship between individual operator experience, as measured by sequential case number, and risk-adjusted in-hospital outcomes in patients undergoing transfemoral TAVR.
We used data from New York State Department of Health Statewide Planning and Research Cooperative System (SPARCS). SPARCS is an all–age group, all-payer database that collects patient and treatment information for every hospital discharge, outpatient and ambulatory surgery, and emergency department visit in New York State. International Classification of Diseases-Ninth Revision-Clinical Modification and International Classification of Diseases-Tenth Revision-Clinical Modification procedure codes were used to identify transfemoral TAVR procedures performed between January 2012 and December 2016 (International Classification of Diseases-Ninth Revision: 35.05; International Classification of Diseases-Tenth Revision: 02RF37Z, 02RF38Z, 02RF3JZ, 02RF3KZ, and X2RF332). Procedures for which the operating physician could not be identified and those procedures associated with a nonphysician license were excluded. The primary cohort for analyses was further restricted to elective procedures (Figure 1).
Physician volume was calculated using sequential case number, as each physician has a unique identifier in the SPARCS database. Sequential case number was quantified as the number of procedures performed during the 1 year prior to the procedure of interest. When calculating volume, all procedures were taken into account, regardless of their elective, urgent, or emergent status. Because procedure codes for TAVR were instituted on October 1, 2011, the calculation of sequential procedure volume started from that time point.
Patient characteristics included age, sex, race, insurance payer status, comorbidities, recent conditions, and previous and concurrent procedures received. Comorbidities, recent conditions (recent myocardial infarction [MI], carotid stenosis, endocarditis, atrial fibrillation or flutter, and renal failure), and procedural variables to be examined were chosen on the basis of variables included in the calculation of the European System for Cardiac Operative Risk Evaluation and the Elixhauser comorbidity index. Previous procedures examined were previous valve interventions, previous coronary artery bypass grafting or congenital defect repair, and previous implantation of pacemaker and defibrillator. Additional hospital and physician information included hospital volume and the first year in which each physician began performing TAVR.
The primary outcome was a composite of in-hospital mortality, stroke, and/or acute MI. Secondary outcomes were the individual components of the primary outcome (Online Table 1). The incidence of acute MI following TAVR procedure was very low (0.4% in the entire cohort and 0.2% in the primary cohort) and therefore was not analyzed alone.
Patient characteristics and in-hospital outcomes were examined within the entire cohort as well as the primary cohort for analyses limited to elective cases only. Means and SDs were obtained for age, and events and percentages were obtained for categorical variables.
Physician volume was analyzed using 2 methods: 1) as a categorical variable defined by tertiles (low, 1 to 23; medium, 24 to 79; high, ≥80); and 2) as a continuous variable. To account for clustering of cases performed by the same physician, a hierarchical logistic regression model was used and included a random effect for physicians. The fully adjusted model included covariates for patient demographics, comorbidities, recent conditions, previous and concurrent cardiac procedures, procedure year, hospital volume, and the first year in which each physician began performing TAVR. Hospital volume was adjusted as a continuous variable.
In the risk-adjusted analyses using categorical volume measure, the low-volume group was defined as reference group. In the risk-adjusted analyses of continuous volume measure, physician volume was modeled using restrictive cubic spline regression. Models with 3 knots were chosen on the basis of fitness statistics of the Akaike information criterion. Odds ratios (ORs) and 95% confidence intervals (CIs), with physicians performing 1 case as annually as the reference level, were obtained and examined graphically. These analyses were performed within the primary cohort of elective procedures and repeated within the entire cohort.
Sensitivity analyses were performed using the entire cohort of all procedures using average annual volume as the physician volume measure. We further performed an analysis excluding the initial 10, 20, and 30 TAVR cases to assess whether the exclusion of initial cases would mitigate the impact of the initial learning curve and change the volume-outcome relationship. All analyses were performed using SAS version 9.3 (SAS Institute, Cary, North Carolina). A 2-sided p value was used, and significance was determined at α = 0.05.
A total of 8,771 TAVR procedures were conducted by 207 operators in New York State during the study period. Of these procedures, 5,916 were elective and were included in the primary cohort. Patients and procedural characteristics are listed in Table 1. Patients in the primary cohort of elective procedures had an average age of 83 ± 7.7 years. One-half of the patients were male, and 85.3% were white. Fourteen percent of patients had undergone previous valve interventions. The majority of procedures were conducted by physicians who had started performing TAVR in calendar year 2011 or 2012 (77.6%). Median duration of hospitalization was 4 days (interquartile range: 2 to 6 days).
In the primary cohort, the in-hospital mortality rate was 1.9%. Stroke occurred in 1.6% of patients and MI in 0.2%. The total event rate of the composite outcome of in-hospital mortality, stroke, and MI was 3.4%. When analyzed with tertiles and using the low-volume group as reference, patients undergoing TAVR performed by high-volume physicians had a trend toward lower mortality (OR: 0.59; 95% CI: 0.32 to 1.08; p = 0.08) and a significantly lower risk for death, stroke, or acute MI (OR: 0.59; 95% CI: 0.37 to 0.93) (Table 2).
When analyzing volume as a continuous variable and after adjusting for patient demographics, comorbidities, recent conditions, and previous and concurrent procedures, as well as hospital and physician characteristics, there was an inverse relationship between physician volume and adverse composite and individual death and stroke outcomes among their patients, whereby adverse outcomes decreased with higher volumes (Figure 2). The relationship was nonlinear, with the improvement in patient outcomes being most pronounced at the lower end of volume, followed by a continued inverse relationship with no clear plateau. Being treated by operators who performed 200 procedures during the prior year was associated with significantly lower risks for post-procedural stroke (OR: 0.41; 95% CI: 0.17 to 0.97) and composite adverse events (OR: 0.45; 95% CI: 0.26 to 0.78) compared with those treated by physicians who performed only 1 procedure during the prior year (Table 3). Being treated by operators who performed 300 procedures during the prior year was associated with a 73% reduction in the risk for post-procedural stroke (OR: 0.27; 95% CI: 0.09 to 0.85) and a 60% reduction in the risk for composite adverse events (OR: 0.40; 95% CI: 0.19 to 0.96) compared with those treated by physicians who performed only 1 procedure during the prior year.
Secondary analysis within the entire cohort, including both elective and nonelective procedures, showed a similar trend but less significant results (Online Table 2, Online Figure 1). A sensitivity analysis was performed to mitigate the initial learning curve by excluding the initial 10, 20, and 30 cases performed by every physician. The analysis excluding first 10 procedures yielded similar results to the main analysis. When excluding the first 20 or first 30 procedures, the association between operator volume and improvement in patient outcomes became more linear (Figure 3). Sensitivity analyses using averaged annual volume as a volume measure demonstrated similar results (Online Figure 2).
In this study, we included all 8,771 TAVR procedures (207 operators) conducted in the state of New York from January 2012 to December 2016. We found a statistically significant and clinically important inverse relationship between increasing individual operator experience and a risk-adjusted composite outcome of in-hospital mortality, stroke, and/or MI, with this relationship being driven primarily by a reduction in stroke with increasing operator volume. The association with operator volume was most pronounced for the first 20 sequential procedures, after which a more linear and gradual improvement in risk-adjusted outcomes was observed with increasing individual operator volume. The present data demonstrate that for physicians performing transfemoral TAVR, there is a substantial initial steep learning curve for operators, followed by a more protracted and persistent volume outcome relationship.
Carroll et al. (13) conducted the first study that assessed the relationship between TAVR site volume, by sequential case number, and in-hospital adjusted outcomes, demonstrating significant reductions in mortality, vascular complications, and bleeding when comparing the highest volume (138 to 602 cases) centers with low-volume centers (1 to 30 cases). However, they also demonstrated an estimated hospital threshold of 100 cases to achieve a plateau in procedural success and complications, beyond which outcomes continued to improve, albeit less dramatically. Our study is in line with this, in which the inverse relationship between volume and stroke continues beyond 200 procedures. The French TAVR experience including more than 12,000 patients demonstrated that TAVR outcomes have evolved and improved over time, with marked decreases in cardiac and cerebrovascular events, as well as a 50% decrease in in-hospital mortality over the 6-year study period (15). At the operator level, in-hospital stroke seems to be fundamentally related to operator volume, in the case of this analysis beyond the point of 200 operations. Although our study focused on the volume-outcome relationship at the level of the individual operator, our findings are congruent with those of Carroll et al. at the hospital level. Microembolization from the calcified native aortic valve and/or aorta during prosthesis positioning and implantation is a known cause of early post-TAVR stroke. This highlights the importance of careful, precise, and quick positioning of the TAVR valve and avoidance of recaptures and second valves. All these factors can reduce cerebral embolization of valvular and/or aortic debris. This may also highlight the importance of appropriate valve selection, another skill developed with experience (26,27).
There have been numerous efforts by governing organizations and professional societies to attempt to diminish the effect of the learning curve on TAVR outcomes (18–20). The Centers for Medicare and Medicaid Services state in their national coverage determination that for TAVR operations to be reimbursed by the federal government, they should be performed by heart teams (i.e., centers) completing at least 20 such procedures annually or 40 procedures biannually. Although these numbers appear to be generally consistent with our study results in general terms, perhaps these guidelines are more appropriately directed toward specific operators. A substantial body of research in multiple disciplines has shown a concordant and highly dependent relationship of outcomes not with hospital volume but with the operating surgeon volume. For example, in a national study of surgical aortic valve replacement, surgeon volume accounted for 100% of the effect of hospital volume on mortality (8).
Operator skill has been shown to be strongly related to procedure volume (28). With TAVR, a high level of skill may be essential in performing meticulous groin cannulation, crossing and deploying valves, or controlling intraoperative complications such as bleeding or arterial dissection. A high level of skill may also be associated with shorter procedure time, which in a learning curve analysis of the PARTNER-I (Placement of Aortic Transcatheter Valve) trial had a strong inverse linear relationship with 30-day major adverse events (24,25). According to our analysis, if the assumption is made that outcomes are in part a surrogate for operator skill, then skill can be achieved with higher volumes. Although high-volume hospitals demonstrate improved outcomes for select procedures, the experience of the operator at the high-volume center can be paramount. Although a clear learning curve as well as an association between hospital volume and TAVR has been demonstrated (13,14,16,17,29–31) at a hospital level, our report demonstrates the important effect of operator experience at an individual level with relation to longitudinal outcomes, independent of hospital volume.
The reported volume-outcome relationship has general implications for TAVR programs. Our findings support creating a partnership with an established program to minimize the initial learning process by enhancing patient selection and refining procedural technique. Although TAVR was initially performed in high-risk or inoperable patients, it is now approved for use in moderate-risk patients and being investigated for use in low-risk patients. We adjusted for case mix but did not specifically test whether the volume-outcome relationship was evident in lower risk TAVR candidates. Formal TAVR proctoring programs for new centers may also have implications for short-term and long-term outcomes, but that is speculative (32). Our findings regarding the steep early learning and protracted late learning curves might be typical for new complex technology, but this has not been previously shown. During our study period, TAVR technology was evolving, newer iterations of devices were introduced, and rapid site expansion occurred, factors that likely contributed to the developing volume-outcome relationship. It is important to note that mortality was not strongly linked to operator volume, and the difference in the composite endpoint was mostly linked to stroke.
This was a retrospective observational study that used a statewide database for data collection. Although SPARCS is a powerful tool that provides reliable cross-sectional outcome statistics including individual operator volume, it relies on International Classification of Diseases-Ninth Revision-Clinical Modification and International Classification of Diseases-Tenth Revision-Clinical Modification codes to describe characteristics of patients and periprocedural outcomes. However, the reported rates of in-hospital mortality and stroke were 2.5% and 1.7%, respectively, and the rate of the composite endpoint of stroke, mortality, and/or MI was 4.2%, which is comparable with TAVR registry reports. Also, data specifically relevant to TAVR practice such as the make and type of the device used, Society of Thoracic Surgeons score, and frailty index could not be elucidated from this database. Another important confounder is that although most TAVR teams have a primary operator, this is not uniformly true. Most procedures are performed by a team, and post-procedural care might involve additional teams; the individual assignment of cases and outcomes are therefore not perfectly associated. Furthermore, the inclusion of only elective cases might have blunted volume-outcome associations, because the more complex the procedure with higher risks tends to bring out this association and the related learning curve.
Additionally, although risk adjustment was extensive, the potential for unmeasured confounding factors exists. Because of the considerable advances in TAVR technique and technology over the 6-year study period, numerous iterations of devices were included in the analysis, and their contribution to outcomes could not be accounted for. Because of the rapid evolution of TAVR technology, the findings of this study may not be fully relevant to current practice, in which more advanced devices are used. With regard to illustrating an operator-level volume-outcome relationship, it is important to note that similar to the rest of the United States, most centers in New York State are considered low volume (7,14,16,17). Our data demonstrated that one-third of centers performed <83 cases annually, and one-third of individual operators performed <24 procedures annually; lower procedure volumes make estimates of operator volume-outcome relationships more imprecise.
In this large-scale report of individual operator TAVR experience, we demonstrate a statistically significant inverse relationship between operator experience and risk-adjusted post-operative complications, most prominently for in-hospital stroke. A clear and quantifiable learning curve on the operator level of at least 20 procedures exists, after which a less significant albeit continuous improvement in outcomes is demonstrated. As TAVR continues to expand and its indications broaden, our data demonstrate the value of a specialized approach to the delivery of TAVR therapy as well as the importance of proctoring and partnership with experienced TAVR practitioners during the early adoption period.
WHAT IS KNOWN? A learning curve has been demonstrated for TAVR at the hospital level; however, volume-outcome relationships have not been evaluated at the individual operator level. We aimed to assess the impact of individual operator experience on transfemoral TAVR outcomes.
WHAT IS NEW? Increased TAVR experience of operators improves risk-adjusted in-hospital outcomes.
WHAT IS NEXT? Our data demonstrate the value of a specialized approach to the delivery of TAVR therapy as well as the importance of proctoring and partnership with experienced TAVR practitioners during the early adoption period.
This study was funded in part by the U.S. Food and Drug Administration through grant U01FD005478. The funder had no influence on 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. The authors have reported that they have no relationships relevant to the contents of this paper to disclose.
- Abbreviations and Acronyms
- confidence interval
- myocardial infarction
- odds ratio
- New York State Department of Health Statewide Planning and Research Cooperative System
- transcatheter aortic valve replacement
- Received July 13, 2018.
- Revision received September 26, 2018.
- Accepted October 16, 2018.
- 2019 American College of Cardiology Foundation
- Popma J.J.,
- Adams D.H.,
- Reardon M.J.,
- et al.
- De Biasi A.R.,
- Paul S.,
- Nasar A.,
- Girardi L.N.,
- Salemi A.
- Gonzalez A.A.,
- Dimick J.B.,
- Birkmeyer J.D.,
- Ghaferi A.A.
- Carroll J.D.,
- Vemulapalli S.,
- Dai D.,
- et al.
- Kim L.K.,
- Minutello R.M.,
- Feldman D.N.,
- et al.
- Auffret V.,
- Lefevre T.,
- Van Belle E.,
- et al.
- Bestehorn K.,
- Eggebrecht H.,
- Fleck E.,
- Bestehorn M.,
- Mehta R.H.,
- Kuck K.-H.
- ↵U.S. Department of Health and Human Services, Centers for Medicare and Medicaid Services. Available at: https://www.cms.gov. Accessed January 11, 2017.
- Bavaria J.E.,
- Tommaso C.L.,
- Brindis R.G.,
- et al.
- Holmes D.R.,
- Mack M.J.,
- Kaul S.,
- et al.
- Webb J.G.,
- Chandavimol M.,
- Thompson C.R.,
- et al.
- Webb J.G.,
- Pasupati S.,
- Humphries K.,
- et al.
- Tamburino C.,
- Capodanno D.,
- Ramondo A.,
- et al.
- Alli O.,
- Rihal C.S.,
- Suri R.M.,
- et al.
- Patel P.A.,
- Patel S.,
- Feinman J.W.,
- et al.
- Young M.N.,
- Elmariah S.,
- Kennedy K.F.,
- Inglessis I.,
- Yeh R.W.
- Schelbert E.B.,
- Vaughan-Sarrazin M.S.,
- Welke K.F.,
- Rosenthal G.E.
- Eggebrecht H.,
- Bestehorn M.,
- Haude M.,
- et al.
- Gurevich S.,
- John R.,
- Kelly R.F.,
- et al.