Author + information
- Received December 17, 1998
- Revision received May 4, 1999
- Accepted July 19, 1999
- Published online November 1, 1999.
- David J Malenka, MD, FACC∗,†,* (, )
- Paul D McGrath, MD, MSc, FACC†,‡,
- David E Wennberg, MD, MPH†,‡,
- Thomas J Ryan Jr., MD, FACC‡,
- Mirle A Kellett Jr, MD, FACC‡,
- Samuel J Shubrooks Jr., MD, FACC∥,
- William A Bradley, MD, FACC§,
- Bruce D Hettlemen, MD, FACC∗,
- John F Robb, MD, FACC∗,
- Michael J Hearne, MD, FACC§,
- Theodore M Silver, MD, FACC¶,
- Matthew W Watkins, MD, FACC#,
- John R O’Meara, MD, FACC‡,
- Peter N VerLee, MD, FACC¶,
- Daniel J O’Rourke, MD, MSc, FACC∗,
- for the Northern New England Cardiovascular Disease Study Group
- ↵*Reprint requests and correspondence: David J. Malenka, Section of Cardiology, Dartmouth-Hitchcock Medical Center, Lebanon, New Hampshire 03756
The purpose of this study was to examine the relationship between annual operator volume and outcomes of percutaneous coronary interventions (PCIs) using contemporaneous data.
The 1997 American College of Cardiology (ACC)/American Heart Association task force based their recommendation that interventionists perform ≥75 procedures per year to maintain competency in PCI on data collected largely in the early 1990s. The practice of interventional cardiology has since changed with the availability of new devices and drugs.
Data were collected from 1994 through 1996 on 15,080 PCIs performed during 14,498 hospitalizations by 47 interventional cardiologists practicing at the five high volume (>600 procedures per hospital per year) hospitals in northern New England and one Massachusetts- based institution that support these procedures. Operators were categorized into terciles based on their annualized volume of procedures. Multivariate regression analysis was used to control for case-mix. In-hospital outcomes included death, emergency coronary artery bypass graft surgery (eCABG), non-emergency CABG (non-eCABG), myocardial infarction (MI), death and clinical success (≥1 attempted lesion dilated to <50% residual stenosis and no death, CABG or MI).
Average annual procedure rates varied across terciles from low = 68, middle = 115 and high = 209. After adjusting for case-mix, clinical success rates were comparable across terciles (low, middle and high terciles: 90.9%, 88.8% and 90.7%, ptrend= 0.237), as were all the adverse outcomes including death (low-risk patients = 0.45%, 0.41%, 0.71%, ptrend= 0.086; high-risk patients = 5.68%, 5.99%, 7.23%, ptrend= 0.324), eCABG (1.74%, 2.05%, 1.75%, ptrend= 0.733) and MI (2.57%, 1.90%, 1.86%, ptrend= 0.065).
Using current data, there is no significant relationship between operator volumes averaging ≥68 per year and outcomes at high volume hospitals. Future efforts should be directed at determining the generalizability of these results.
In the last four years, several studies have evaluated the relationship between the volume of interventions and outcomes for patients undergoing percutaneous coronary inter-ventions (PCIs) (1–14). They established that there is an inverse relationship between the annual volume of PCIs and the incidence of procedural complications at both the institutional level (2,4–6,10–14)and the operator level (1,3–9). These findings prompted the American College of
Cardiology (ACC) to include in their most recent PCI guidelines (15)a recommendation that hospitals should have a volume of at least 400 procedures per year and operators a volume of 75 procedures per year to help insure good outcomes. The studies used to support this recommendation relied upon data from the early 1990s, and since that time, the practice of interventional cardiology has changed.
The strongest relationship between volume and outcomes for PCIs has been for emergency coronary artery bypass graft surgery (eCABG). In general, at both the institutional and operator level, higher volumes have been associated with significant reductions in the need for postprocedure eCABG. However, recent observational studies have shown that stent use has reduced the need for eCABG following PCI (16–20). With the Federal Drug Administration’s approval of stents for a broader range of indications and a strategy of high pressure inflations to reduce the need for aggressive antithrombotic regimens, the use of stents has seen exponential growth. In addition, there have been other changes in practice patterns which might be expected to improve outcomes following PCIs that include the use of lower profile balloons, better guiding catheters and new antiplatelet agents.
Not only has equipment and drug therapy used for PCIs changed over the last several years, but the cumulative experience of individual operators has increased and there has been a growing emphasis on efforts in quality assessment and improvement. Given all these factors, we thought it appropriate to re-evaluate the relationship between operator volume and outcomes using more current data. Therefore, we used our prospective, regional, clinical database of consecutive PCIs to examine the question of whether operator volume continues to be related to in-hospital clinical success or adverse outcomes in six high volume hospitals in the more modern era of interventional cardiology.
The Northern New England Cardiovascular Disease Study Group is a voluntary research consortium composed of clinicians, research scientists and hospital administrators at the five institutions in Maine, New Hampshire and Vermont who are the sole providers of coronary revascularization in the region, and one Massachusetts-based institution. The intent of the group is to foster continuous improvement in the quality of care of patients with cardiovascular disease in northern New England through the pooling of process and outcome data and the timely feedback of data to clinicians (21,22).
Between January 1, 1994 and December 31, 1996, data were collected on 15,331 consecutive hospitalizations for a PCI. After eliminating cases associated with three operators performing <5 interventions per year (i.e., two interventional fellows in transition to new jobs and one physician who performed 13 procedures during two, three month clusters early in the three year experience and who had not performed any procedures during the last 15 months of the study), physicians performing any intervention outside the region and the few unattributed cases, there remained data on 14,498 hospitalizations, contributed by 47 interventionists performing 15,080 PCIs, which became the study cohort. Annualized procedure rates for each operator were calculated by determining the number of months during the study period in which a physician was recorded as a primary operator and the number of procedures they performed during that time period and standardizing that relationship to 12 months. After examining scatterplots of annual operator volume versus adverse outcomes and seeing no apparent relationship, operators were ranked based on their annualized rates and then categorized into low, middle or high volume terciles in an effort to maximize the power of detecting a relationship. The mean number of months during which study physicians contributed data was 30 (range 6 to 36 months, median 36 months).
The following information was collected for every PCI:
1. site and primary operator;
2. demographic data: patient age, gender, height, weight;
3. medical history: previous coronary artery bypass graft surgery (CABG), PCI or myocardial infarction (MI), family history of premature coronary artery disease, the presence of congestive heart failure, hypertension, treated diabetes, current smoking, hypercholesterolemia, chronic obstructive pulmonary disease, peripheral vascular disease, cerebrovascular disease, renal failure, baseline creatinine;
4. primary indication for PCI: stable angina, unstable angina, postinfarction angina, postinfarction anatomy, primary therapy for acute MI, cardiogenic shock;
5. priority at PCI: emergent, urgent, nonurgent;
6. therapy before during and after the procedure: intravenous heparin, intravenous nitroglycerin, thrombolytic therapy, insertion of an intraaortic balloon pump;
7. cardiac anatomy and function: percent stenosis of left main coronary artery, number of other diseased (>70% stenosis) native coronary vessels (left anterior descending, right or circumflex), dominance, number of bypass grafts (distal anastomoses), ejection fraction, left ventricular end diastolic pressure;
8. PCI procedure information: location of the lesions attempted [using the Coronary Artery Surgery Study map, (23)], pre- and poststenosis, lesion type (24)A, B1, B2 or C, location of collateral vessels, device use including balloon, directional atherectomy, transluminal extraction, rotational atherectomy, laser, or stent;
9. outcomes (all in-hospital): death, eCABG, non-eCABG, or new MI, defined as chest pain, diaphoresis, dyspnea or hypotension associated with the development of new Q-waves or ST-TW changes and a rise in creatine phosphokinase (CPK) to at least twice normal with a positive CPK-MB. Clinical success was defined as ≥1 attempted lesion(s) successfully dilated and no adverse clinical outcomes. Adverse clinical outcomes included any in-hospital death, eCABG, non-eCABG or postprocedure MI.
The number of patients in the data set were verified using hospital discharge data supplemented by cardiac catheterization laboratory logs. Any missing information was obtained. The outcomes of death and CABG were validated from the hospital discharge data set and review of the medical record. Myocardial infarction as an outcome was not independently validated.
All analyses were carried out using Statistical Analysis Software (SAS Institute Inc., Cary, North Carolina), Release 6.11 (25)or STATA Statistical Software, Release 5.0 (STATA Corp Inc., College Station, Texas) (26). Logistic regression (27)was used to assess whether changes in case-mix differed across terciles of operator volume (ptrend), except for the ordinal variables of indication, priority and lesion type where a Mantel-Haenszel chi-square statistic was used. For several patients, disease and treatment characteristics had missing data. Those with less than 2% missing values were coded as “not present.” Others, particularly the treatment variables of intravenous nitroglycerin, thrombolytics and intraaortic balloon pumping were coded as “not present” under the assumption that if they had been used, it would have been recorded.
Multivariate models were used to adjust for differences in case-mix and severity of illness across terciles of operator volume when comparing outcomes. All variables demonstrating a univariate association with the dependent variable of interest at p < 0.10 (without adjustment for multiple comparisons) were considered potential independent variables for inclusion in the multivariate analyses. Much of the data was naturally discrete. Continuous variables were examined to determine the categorization that best related them to the dependent variables. To determine whether tercile of operator volume was independently associated with outcomes, tercile was forced into the multivariate models. Tercile was used as an ordinal variable to test for a linear relationship between volume and outcomes and reported as a “p of trend.” It was also parameterized using dummy variables, with the low tercile group as a referent category, to test the independent relationship of each volume tercile to outcome, and reported as confidence intervals (CIs) around the adjusted point estimate of outcome. For death, a separate model was developed for patients at higher risk for an adverse outcome, defined as those undergoing an emergency procedure or being treated for an MI or cardiogenic shock, because previous work using a general model for death had shown that the relationship between observed and expected mortality was weakest for patients at highest risk (28). Direct standardization (29)was used to calculate adjusted rates.
Among the 47 interventionists, average annualized operator volume varied from a low of 22 to a high of 370 cases per year (Table 1). Across terciles of operator volume, the median number of cases per year increased from 75 (range 22 to 84) to 119 (range 88 to 129) to 194 (range 138 to 370). Lower tercile operators performed 14.4% of all procedures whereas middle tercile operators performed 27.9% and high tercile operators performed 57.7% of procedures.
Patient, disease and procedural characteristics by tercile of operator volume are shown in Tables 2 and 3. ⇓⇓As operator volume increased, there was a trend toward patients being older, having more diabetes, vascular disease and renal failure, having undergone previous revascularization and having more three-vessel and left main disease. High tercile operators were more likely to be intervening on patients requiring emergency procedures and in cardiogenic shock. Regardless of tercile, operators chose to intervene on a comparable number of lesions. High tercile operators were more likely to attempt ACC type B lesions and to work in bypass grafts. High tercile operators were also more likely to use rotational atherectomy and less likely to use a stent than lower volume operators.
Crude rates of outcomes (with 95% CIs) are shown in Table 4. After adjusting for case-mix and severity of illness (see Appendix Afor logistic models) there was no difference in the rates of clinical success across terciles of operator volume (90.9% low vs. 88.8% middle vs. 90.7% high volume terciles; ptrend= 0.237). The same was true for rates of all the adverse outcomes (Table 5)except for non-eCABG, which increased from 0.44% in the low volume tercile to 0.80% and 0.90% in the middle and high volume terciles, respectively, a trend of borderline significance (ptrend= 0.051), and MI, which decreased marginally between the low versus the middle and high volume terciles, though there was marked overlap among the CIs for the point estimates.
To determine whether the influx of new providers into the region influenced our results, we limited the analysis to only those interventionists who contributed to the previous study (9). Our findings did not change. Within each tercile operators were imperfectly distributed across sites. Site was therefore entered into the multivariate model to determine whether the results were confounded by site. They were not. The use of coronary stents grew markedly over the time period of this study (Fig. 1), and, therefore, we examined their relationship to our findings. After adjusting for case-mix, the use of a coronary stent was associated with a lower likelihood of CABG (odds ratio [OR] = 0.72, 95% CI 0.58, 0.91), eCABG (OR = 0.91, 95% CI 0.66, 1.24), non-eCABG (OR = 0.36, 95% CI 0.19, 0.70) and death in high risk patients (OR = 0.48, 95% CI 0.21, 1.10), no change in the likelihood of death in low risk patients (OR = 1.00, 95% CI 0.58, 1.73) and an increased likelihood of clinical success (OR = 1.19, 95% CI 1.03, 1.37), though not all these ORs reached statistical significance. Because they are used frequently in the setting of acute closures, stents were associated with an increased likelihood of an MI (OR = 1.84, 95% CI 1.43, 2.37). When stent was entered into the multivariate model, the relationship between tercile of operator volume and outcome did not change and remained insignificant.
Our major finding is that in northern New England from 1994 through 1996, the average annual volume of procedures performed by interventionists was not significantly related to the rates of successful or unsuccessful procedures. Even after adjusting for case-mix, the rates of clinical success and adverse outcomes were comparable across terciles of operator volume. Though there were possible trends between the outcomes of non-eCABG and MI and terciles of operator volume, they should be interpreted with caution as there may be a lower threshold for recommending surgical revascularization among higher tercile operators than among lower tercile operators, MI was not a validated outcome and its assessment may differ by tercile of operator volume, and these are only two associations among many that were tested.
The previous study
These results differ from what we previously reported using 1990–1993 data (9). During that time period there was a positive relationship between operator volume and success but a negative association between operator volume and the adverse outcome of CABG. Why the difference between our previous study and the current one? It is not a consequence of changing case-mix (intervening on patients who are at lower risk for an adverse outcome). Using a clinical prediction rule developed on 1994–1996 data, the average expected mortality has actually increased from 0.71% in 1990–1993 to 1.16% in 1994–1996, indicating that the more recent patient population is at higher risk for adverse events. It was not the influx of new operators into the region (see above). The average number of procedures performed by low tercile operators was slightly higher in the current study (mean = 68, median = 75) than in the previous study (mean = 62, median = 68), though it is hard to imagine that such a small increment in average volumes could make such a difference.
Hannan et al. (6)reported an interaction between operator and institutional volumes and showed that low volume operators (<75 procedures per year) working in high volume hospitals (≥600 procedures per hospital per year) have good results. We cannot rule out such an effect because in 1994 to 1996 only 3% of the patients underwent an intervention at a hospital performing <600 procedures per hospital per year by a low or moderate volume operator compared with 21.8% in 1990–1993. We cannot rule out increased cumulative experience on the part of operators or hospitals as contributing to the improvement in outcomes for all interventionists regardless of their annual volume, though this was not the case in one previous study (7). It is possible that our ongoing regional efforts at examining the process and outcomes of care (30)have resulted in improvement, as was the case for the regional study of CABG (22). These efforts include the timely feedback of accurate data on process and outcomes, regional meetings three times a year to discuss the data and an analysis of the proximate cause of death in a large, regional cohort of 12,232 patients (31).
The influence of stenting
Could the introduction of stents have something to do with our findings? The use of stents reduces the need for eCABG surgery (32–34)and may increase clinical success rates (35). In northern New England, their use grew dramatically during the study period and interventionists in the lower volume terciles were more likely to use them (Fig. 1)than those in the highest volume tercile (reflecting primarily a practice pattern). All interventionists in the region had stents available to them for use with suboptimal results. Though the use of stents represents an operator characteristic and, therefore, was not a variable we chose to adjust for when testing the volume-outcome relationship, it could help to explain the relationship. However, when stent was entered into the multivariate model, the relationship between operator volume and outcome did not change and was still insignificant. We do not yet know what effect stents might have as their increased usage by high volume operators becomes comparable with that of lower volume operators or their overall rate of use increases. Other devices and the newer antiplatelet agents were not used with any great frequency during this time period and were not significantly associated with any outcomes.
There have been three other recent studies based on registry or multicenter data that have explicitly examined the relationship between operator volume and outcomes. Using 1992 Medicare discharge abstracts, Jollis et al. (5)demonstrated an inverse relationship between post-PCI rates of CABG and operator volume. Hannan et al. (6)used 1991 to 1994 New York State clinical data and found an inverse relationship between operator volume and both the outcomes of in-hospital death and CABG. Ellis et al. (7)examined 1993 to 1994 data from five high-volume centers and determined that death and the composite end point of death, Q-wave MI or eCABG, were inversely related to the number of cases each operator performed annually. The difference in findings between these studies and our study may be a consequence of our procedure rates for the low volume operators being higher than in other studies, our use of more recent data reflecting the ongoing changes in interventional cardiology or a finding that is unique to the somewhat more conservative practice patterns in northern New England (36).
Our study has several limitations. Compared with other parts of the country, our low volume operators are not truly “low volume” and our hospitals are high volume institutions (>600 procedures per year). Using 1992 national data, Jollis et al. (5)estimated that the average annual volume of procedures for an interventionist was 26 to 39 and was 196 to 294 for hospitals. Therefore, our results may not be generalizable to operators or hospitals with lower annual volumes. The data on MIs was not validated. High volume operators, who use more devices, may be more likely to measure postprocedure cardiac enzymes and, as a consequence, to find and report MIs. This might obscure a finding of increased MI rates in low volume operators. Our results are limited to in-hospital outcomes. We cannot comment on the important long-term outcomes of target lesion revascularization and functional status. Finally, the practice of interventional cardiology continues to change. The use of stents is growing and new antiplatelet agents are being used with increased frequency. What effect this will have on practice and outcomes and their relationship to operator volume is not yet known.
Some may question why we chose to analyze death after stratifying patients into those at low versus high risk of an adverse event. In previous work (28)we determined that predicting the risk of dying post-PCI was least precise for the patients who had the highest risk of dying. Therefore, we chose to improve our ability to control for confounding by stratifying the data based on risk. The data also suggested this was a good idea because 32.6% of all deaths occurred among patients in cardiogenic shock and shock patients were 2.7 times as likely to be cared for by high volume operators than by lower volume operators. Repeating the analysis after eliminating patients in cardiogenic shock and not stratifying death by predicted patient risk did not change our results.
The ACC recommendations
Our findings do not negate the recent recommendations made by the ACC (15), suggesting that to maintain optimal proficiency in coronary interventions, physicians should be performing ≥75 procedures per year. Though the average volume of procedures in our lowest tercile (n = 68) was slightly below that limit, the median volume was 75. Our somewhat small sample size (47 operators) precluded us from eliminating the operators with rates below 75. However, the ACC recommendations suggest that it may also be acceptable for operators with annual volumes of 50 to 75 to perform coronary interventions if they work at high volume hospitals performing >600 procedures per year (as was the case in our study) and if they are “cautious in case selection.” Repeating our analysis after eliminating the three operators with volumes <50 did not change our results. It should be made clear that our findings cannot be extended to operators with even lower volumes, or working in lower volume hospitals, until the outcomes for this group of physicians are assessed. It continues to make sense that, to maintain the skills and judgment necessary for good outcomes, some minimal level of annual experience is important.
We conclude that in the modern era of interventional cardiology, after adjusting for case-mix and severity of illness, outcomes are comparable for providers across a spectrum of annual operator volumes. It remains to be seen whether these findings are generalizable beyond northern New England, for providers with lower annual volumes than the physicians in our study and in hospitals with lower annual volumes, whether they hold for long-term outcomes and what happens as the practice of interventional cardiology continues to evolve.
Logistic regression models for each outcome variable
1. Clinical success: Age, diabetes mellitus, peripheral vascular disease, creatinine >2 mg/dL, heparin before procedure, intraaortic balloon pump before procedure, number of diseased coronary arteries, indication, priority, American College of Cardiology (ACC) lesion type.
2. In-hospital death: Age, peripheral vascular disease, creatinine >2 mg/dL, indication, priority, ejection fraction, congestive heart failure, intraaortic balloon pump before procedure, ACC lesion type.
3. CABG: Age, diabetes mellitus, congestive heart failure, prior PCI, prior CABG, number of diseased coronary arteries, priority, ACC lesion type, intervention on the proximal left anterior descending coronary artery.
4. Emergency CABG: Age, diabetes mellitus, congestive heart failure, prior PCI, prior CABG, number of diseased coronary arteries, priority, ACC lesion type.
5. Non-emergency CABG: Age, diabetes mellitus, chronic obstructive pulmonary disease, prior MI, prior PCI, indication, number of diseased coronary arteries, priority, ACC lesion type.
6. Myocardial infarction: Prior PCI, intervention on a bypass graft, ACC lesion type.
- American College of Cardiology
- coronary artery bypass graft
- confidence interval
- creatine phosphokinase
- emergency coronary artery bypass graft
- myocardial infarction
- odds ratio
- percutaneous coronary intervention
- Received December 17, 1998.
- Revision received May 4, 1999.
- Accepted July 19, 1999.
- American College of Cardiology
- Bon Tempo C.P,
- Sherber H.S,
- Sheridan M
- O’Neill W.W,
- Griffin J.J,
- Stone G,
- et al.
- Hannan E.L,
- Racz M,
- Ryan T.J,
- et al.
- Ellis S.G,
- Weintraub W,
- Holmes D,
- Shaw R,
- Block P.C,
- King S.B III.
- Klein L.W,
- Schaer G.L,
- Calvin J.E,
- et al.
- McGrath P.D,
- Wennberg D.E,
- Malenka D.J,
- et al.
- GUSTO (IIb) Angioplasty Substudy Group
- Kato N.S,
- Carter G.M
- Tiefenbrunn A.J,
- Chandra N.C,
- French W.J,
- Gore J.M,
- Rogers W.J
- Zahn R,
- Vogt A,
- Neuhaus K.L,
- Schuster S,
- Senges J
- Hirshfeld J.W Jr.,
- Ellis S.G,
- Faxon D.F,
- et al.
- Roubin G.S,
- Cannon A.D,
- Agrawal S.K,
- et al.
- Lincoff A.M,
- Topol E.J,
- Chapekis A.T,
- et al.
- Herrmann H.C,
- Buchbinder M,
- Clemen M.W,
- et al.
- George B.S,
- Noorhees W.D,
- Roubin G.S,
- et al.
- Ryan T.J,
- Bauman W.B,
- Kennedy J.W,
- et al.
- ↵(1986) Statistical Analysis System (SAS) (SAS Institute Inc, Cary, NC).
- ↵(1997) StataCorp. Stata Statistical Software: Release 5.0 (Stata Corporation, College Station, TX).
- Kleinbaum D.G,
- Kupper L.L,
- Morgenstern H
- Northern New England Cardiovascular Disease Study Group,
- O’Connor G.T,
- Malenka D.J,
- Levy D.G,
- Disch D.L,
- Quinton H.B
- Kahn H.A,
- Sempas C.T
- Malenka D.J,
- O’Connor G.T
- Lincoff A.M,
- Topol E.J,
- Chapekis A.T,
- et al.
- Roubin G.S,
- Cannon A.D,
- Agrawal S.K,
- et al.