Original Article
Total Center Percutaneous Coronary Intervention Volume
and 30-Day Mortality
A Contemporary National Cohort Study of 427 467 Elective, Urgent,
and Emergency Cases
Darragh O’Neill, PhD; Owen Nicholas, PhD; Chris P. Gale, BSc, MBBS, PhD, MEd, MSc;
Peter Ludman, MA, MD; Mark A. de Belder, MA, MD; Adam Timmis, MA, MD;
Keith A.A. Fox, BSc, MBChB, F Med Sci; Iain A Simpson, MD;
Simon Redwood, MBBS, MD; Simon G. Ray, MD
Downloaded from http://circoutcomes.ahajournals.org/ by guest on June 16, 2017
Background—The relationship between procedural volume and prognosis after percutaneous coronary intervention (PCI)
remains uncertain, with some studies finding in favor of an inverse association and some against. This UK study provides
a contemporary reassessment in one of the few countries in the world with a nationally representative PCI registry.
Methods and Results—A nationwide cohort study was performed using the national British Cardiovascular Intervention
Society registry. All adult patients undergoing PCI in 93 English and Welsh NHS hospitals between 2007 and 2013 were
analyzed using hierarchical modeling with adjustment for patient risk. Of 427 467 procedures (22.0% primary PCI) in 93
hospitals, 30-day mortality was 1.9% (4.8% primary PCI). 87.1% of centers undertook between 200 and 2000 procedures
annually. Case mix varied with center volume. In centers with 200 to 399 PCI cases per year, a smaller proportion
were PCI for ST-segment–elevation myocardial infarction (8.4%) than in centers with 1500 to 1999 PCI cases per year
(24.2%), but proportionally more were for ST-segment–elevation myocardial infarction with cardiogenic shock (8.4%
versus 4.3%). For the overall PCI cohort, after risk adjustment, there was no significant evidence of worse, or better,
outcomes in lower volume centers from our own study, or in combination with results from other studies. For primary
PCI, there was also no evidence for increased or decreased mortality in lower volume centers.
Conclusions—After adjustment for differences in case mix and clinical presentation, this study supports the conclusion of
no trend for increased mortality in lower volume centers for PCI in the UK healthcare system.
Clinical Trial Registration—https://www.clinicaltrials.gov. Unique identifier: NCT02184949. (Circ Cardiovasc Qual Outcomes. 2017;10:e003186. DOI: 10.1161/CIRCOUTCOMES.116.003186.)
Key Words: angioplasty ◼ cardiology ◼ mortality ◼ myocardial infarction ◼ percutaneous coronary intervention
T
he relationship between procedural volume and patient
outcomes is an important consideration in the provision
of percutaneous coronary intervention (PCI). Studies that
have suggested the existence of a relationship have informed
international guidelines1–3 on the minimum procedural limits
recommended for annual center volume to ensure safe and
effective patient care. This is also reflected in recent UK guidance on the provision of PCI.4 For acute myocardial infarction, the establishment of high-throughput, high-volume
heart attack centers within networks of acute cardiac care
has been promoted as a factor contributing to the decline in
cardiovascular mortality in developed healthcare systems.5–9
Such centers often also have high volumes of elective PCI and
in much of the published literature compare favorably with
lower volume centers where higher rates of adverse outcomes,
longer lengths of hospital stay, and increased costs have been
observed.10,11 These findings have been reflected in the reconfiguration of UK PCI services over the past decade, specifically the expectation that lower volume units undertake at
least 400 procedures per annum.
See Editorial by Kumbhani and Bittl
It is timely and of international importance to review the
evidence for and against a relationship between PCI center
volume and clinical outcomes in the UK healthcare system.
Such a national review contributes to the assessment of the
Received August 3, 2016; accepted February 20, 2017.
From the Research Department of Epidemiology and Public Health, University College London, United Kingdom (D.O., O.N.); Leeds Institute for
Cardiovascular and Metabolic Medicine, University of Leeds, United Kingdom (C.P.G.); Department of Cardiology, York Teaching Hospital, United
Kingdom (C.P.G.); Queen Elizabeth Hospital, Birmingham, United Kingdom (P.L.); The James Cook University Hospital, Middlesbrough, United Kingdom
(M.A.d.B.); NIHR Cardiovascular Biomedical Research Unit, Barts Heart Centre, United Kingdom (A.T.); Centre for Cardiovascular Science, University
of Edinburgh, United Kingdom (K.A.A.F.); Wessex Cardiac Unit, University Hospital Southampton, United Kingdom (I.A.S.); King’s College London/St
Thomas’ Hospital, United Kingdom (S.R.); University Hospitals of South Manchester, United Kingdom (S.G.R.).
The Data Supplement is available at http://circoutcomes.ahajournals.org/lookup/suppl/doi:10.1161/CIRCOUTCOMES.116.003186/-/DC1.
Correspondence to Darragh O’Neill, PhD, Research Department of Epidemiology and Public Health, University College London, 1 - 19 Torrington
Place, London WC1E 6BT, United Kingdom. E-mail [email protected]
© 2017 American Heart Association, Inc.
Circ Cardiovasc Qual Outcomes is available at http://circoutcomes.ahajournals.org
1
DOI: 10.1161/CIRCOUTCOMES.116.003186
2 O’Neill et al UK PCI Volume Outcome Relationship
Methods
WHAT IS KNOWN
• Some previous studies have suggested that a relationship exists between the volume of PCI performed in
hospitals and patient outcomes; this literature has
subsequently informed international guidelines on
minimal procedural limits.
• However, not all studies have been in agreement
regarding the presence of this theorized relationship,
and so further clarification is required.
WHAT THE STUDY ADDS
Downloaded from http://circoutcomes.ahajournals.org/ by guest on June 16, 2017
• This study has used whole country PCI registry data
with linked mortality data to examine this relationship using a Bayesian inference approach.
• After risk adjustment, no significant evidence of
worse, or better, mortality outcomes in lower volume
centers compared with higher volume centers was
found, either with the current study or in combination with results from other studies.
• An absence of support for increased mortality in
lower volume centers was also observed separately
in primary PCI-only and elective-only cohorts.
UK’s quality improvement initiative in PCI and has implications for provision of service in other countries. A meta-analysis of 10 studies (8 American and 2 Japanese, all before 2005)
on PCI comprising 1 322 342 patients in 1746 hospitals finds
in favor of an inverse relationship between center volume
and risk-adjusted in-hospital mortality after PCI (odds ratio,
0.865; 95% confidence interval [CI] 0.827–0.905) for highversus low-volume centers.12 However, evidence from studies
published since the period covered by this meta-analysis have
produced a more variable picture, with some studies supporting the presence of a relationship between center PCI volume
and risk-adjusted outcomes,10,11,13,14 and others not.15–18 A more
recent summary of the evidence by a joint task force from the
American College of Cardiology Foundation, American Heart
Association, and American College of Physicians acknowledged that the large-scale studies in this area have produced
heterogeneous findings.19 Specifically in relation to primary
PCI cohorts, a review by the National Clinical Guidance
Centre20 likewise noted that the evidence base is currently
insufficient to establish a definitive understanding of the volume–outcome relationship.
In the United Kingdom, continuous whole country data
are collected through the British Cardiovascular Intervention
Society (BCIS) registry of all PCI cases, are linked to the registry of deaths for all inhabitants in the United Kingdom collected independently by the Office for National Statistics, and
are held by the National Institute for Cardiovascular Outcomes
Research at University College London. This provides unique
opportunities to understand outcomes after percutaneous
interventions. We consequently used these mortality-linked
data to investigate the association between annual center volume of PCI cases and 30-day all-cause mortality over a 7-year
period up to 2013.
Setting and Design
This observational prognostic cohort study was based on data from
the BCIS national registry of adult interventional procedures, participation in which is mandated for all PCI operators and all National
Health Service (NHS) Trusts in England and Wales. Data for every
PCI procedure performed were collected at the time of procedure at
each hospital. These data were then encrypted and transferred online to a central database at the National Institute for Cardiovascular
Outcomes Research (NICOR), hosted at University College London,
UK. The data for each PCI procedure comprise 113 core fields
which describe the patient demographics and clinical presentation,
indications for PCI, procedural details and outcomes during the hospital stay.21 NICOR, which includes BCIS (Ref: NIGB: ECC 1–06
(d)/2011), has support under section 251 of the NHS Act 2006 to use
anonymized patient information for medical research without consent. The study involved deidentified data and formal ethical approval
was not required.
Patients
The sampling frame comprised all patients in England and Wales recorded in the BCIS database. Patients were eligible for analysis if
they had received PCI to at least one lesion or vessel over a seven
year period between 01 January 2007 and 31 December 2013, and
were aged between 18 and 100 years. We excluded a total of 68,912
(11.7%) cases from 9 private hospitals as they did not represent NHS
care, and from 14 hospitals in Scotland and Northern Ireland as mortality tracking is only consistently available for patients with a NHS
number in England and Wales. The exclusion of private hospitals was
also to account for potential differences in clinical practice; typically
they reflect independent operators rather than teams, and treat almost
exclusively elective cases. An additional 11,871 (2.0%) cases from
within English or Welsh NHS hospitals were also excluded because
of missing 30-day mortality. Ventilated patients (who mainly present
with out of hospital cardiac arrest) were also excluded as the risk
adjustment model was inapplicable to such patients. For patients with
multiple admissions, the earliest record was used with subsequent admissions being excluded. The final analytic cohort comprised 427 467
patients (Figure 1). Two subgroups were selected for analysis: (1) all
PCI cases and (2) only primary PCI cases.
Patient characteristics were examined as part of an initial descriptive analysis, including indication for intervention, urgency of
procedure (elective, urgent, emergency, or salvage), and cardiogenic
shock (based on clinical assessment including systolic blood pressure
<100 mm Hg; pulse >100 bpm, in a patient who was cool, clammy, or
requiring inotropes, intracardiac balloon pump, or cardiopulmonary
support).
Outcome
Data for all-cause mortality were extracted through linkage to the
Office for National Statistics using each patient’s unique NHS number. Patients were followed up for their vital status 30 days after the
date of their PCI procedure.
Center Volume
We took average center volume as the measure of the procedural
volume exposure. Average center volume was determined for each
center from its total number of PCIs registered in BCIS for the years
2007 to 2013, divided by the number of years where PCI activity was
nonzero. Records concerning preoperatively ventilated patients, follow-up procedures, and records with missing mortality outcome data
were retained in the calculation of overall annual volume because
these were deemed to contribute to the procedural volume. For the
purpose of stratified analysis, bands of center average volume were
established on the basis of existing standards and clinical experience.
These bands were set at 0 to 199, 200 to 399, 400 to 749, 750 to 1499,
1500 to 1999, and ≥2000 cases.
3 O’Neill et al UK PCI Volume Outcome Relationship
Figure 1. Data flow from sampling frame
to analytic cohort. BCIS indicates British
Cardiovascular Intervention Society; and
PCI, percutaneous coronary intervention.
Downloaded from http://circoutcomes.ahajournals.org/ by guest on June 16, 2017
Case Mix Adjustment
Model-based case mix standardization for 30-day mortality outcome
was performed to adjust for patient characteristics. The model22 was
developed by an independent research group using data from the
BCIS registry that comprised all eligible PCI records undertaken in
England and Wales from 2007 through 2011. Clinical validation of
the model subsequently undertaken by its developers showed that
the model performed well in estimating risk for more recent cases.
The model has 9 risk factors: age at time of procedure, sex, diabetes mellitus history, previous myocardial infarction, renal disease
history, cerebrovascular event history, cardiogenic shock, indication
for intervention, and procedural urgency. These covariates were selected for their clinical relevance, predictive utility, and low levels
of missingness. The model includes interactions for age and diabetes
mellitus, age and shock, and indication and shock. We independently
performed calibration and discrimination analyses to evaluate the
model’s suitability for our study (Data Supplement).
Statistical Analyses
A statistical analysis plan was prepared and registered on www.clinicaltrials.gov before data analysis. Statistical analyses were performed
using R version 3.1.2 (www.r-project.org) and Markov chain Monte
Carlo software written in C++. Differences in baseline characteristics
for average center volume bands were examined using ANOVA for
continuous variables and the χ2 test for categorical variables. All tests
were 2-tailed, and to correct for multiple comparisons, P values were
adjusted using the false-discovery rate method.23
Hierarchical logistic models for patient’s 30-day mortality outcome nested in centers were applied to overall, and primary, PCI
groups with adjustment for risk of adverse outcome. We included
fixed effects for year of procedure to account for year-by-year change
in risk-adjusted outcomes, resulting in estimates of center-level
effects with 95% CIs. We computed the ratio of the probability of the
estimated center level effects under the hypothesis of no relationship
with volume to the probability of the effects under the hypothesis of
a relationship. In the relationship hypothesis, we took a uniform noninformative prior for the gradient with respect to log–volume. Details
are provided in the Data Supplement. Each probability captures how
well the effects estimates match each hypothesis, and the ratio makes
a comparison. This ratio is akin to the likelihood ratio and is called a
Bayes factor.24 The smaller than the Bayes factor is, the greater our
effects estimates support the hypothesis of a relationship. We repeated
these calculations for a band-specific relationship analysis, using the
prespecified volume bands as defined above. We also estimated the
gradient of center-level effect to log–volume under the hypothesis of
a relationship.
Missing Data
Records with missing mortality status (2.0%) were excluded as part
of the sampling criteria (Figure 1). Hospital name and year of procedure were 100% complete. Missing data in covariates used to assign 30-day mortality risk were handled using the same method used
in the model’s derivation: missing categorical covariates were set to
their lowest risk value; missing age was set to sex-specific median
values. This imputation technique ensured that the model was implemented in line with its intended use in clinical practice. Additional
detail on missing data is provided in the Data Supplement.
Sensitivity Analyses
To validate our results, we first examined whether the use of alternative estimation methods in the hierarchical modeling work impacted on the coefficient estimates and CIs. Second, to explore the
categorization method used in the primary analyses, we performed an
4 O’Neill et al UK PCI Volume Outcome Relationship
alternative analysis of the center volume–outcome relationship using
equally sized quintiles of the mean annual number of cases of PCI
per year. Third, we excluded the 10% of patients who were in the
top 30-day mortality risk decile (to correct for some overprediction
of mortality identified in our calibration testing of the risk model).
Fourth, we excluded the 5% of patients who had missing data for cardiogenic shock (to remove additional uncertainty about missing data
as this field had the highest potential influence over mortality risk
estimates derived from the risk model). Fifth, we repeated the model
with elective-only patients. Finally, for the primary PCI cohort, we
reinvestigated the volume–outcome relationship using each center’s
primary PCI mean annual volume rather than the volume of all PCIs
undertaken at that center.
Results
Baseline Characteristics
Downloaded from http://circoutcomes.ahajournals.org/ by guest on June 16, 2017
Our analysis included 427 467 PCI procedures in 93 NHS hospitals in England and Wales, of which 94 022 patients (22.0%)
were primary PCI. The mean (SD) age was 64.9 (11.9), and
26.5% were female (Table 1). Overall, 17.6% had diabetes mellitus, 1.4% had a creatinine >200 µmol/L, and of the
primary PCI cohort (Table 2), 5.1% in total had cardiogenic
shock. For the overall PCI cohort, 65.1% had a drug-eluting
stent deployed and 3.1% of the PCI procedures involved the
use of intravascular ultrasound guidance. Among primary PCI
patients, a smaller proportion of procedures involved intravascular ultrasound use (1.5%), while 59.5% involved drugeluting stent deployment.
Tables 1 and 2 also show that the frequency of many patient
and procedural characteristics differed between bands of center
average volume. All characteristics were statistically significantly different between bands at P<0.001 for the overall cohort
after false-discovery rate adjustment for multiple comparison.
Most were significant at the same threshold for the primary
PCI cohort. Compared with the higher volume centers (≥2000),
the lowest (0–199) volume centers had a smaller frequency of
ST-segment–elevation myocardial infarction (STEMI) cases
(2.8% in lowest versus 24.7% in highest), as well as PCI for
non–ST-segment–elevation myocardial infarction (18.9 versus 34.4%), but greater frequency of elective (53.3% versus
38.5%), stable (47.9% versus 37.1%), and rescue (22.3% versus 1.4%) cases. The second lowest volume band (200–399)
captures a much larger number of cases than the lowest band.
Compared with the highest volume units, this second lowest
volume band had a greater level of non–ST-segment–elevation
myocardial infarction patients (52% in second lowest versus
34.4% in highest) and of urgent cases (49.6% versus 32.6%).
Although higher volume units handled a greater proportion of STEMI cases and had higher absolute levels of
patients with cardiogenic shock, results for the primary PCI
cohort (Table 2) show that the second lowest volume units in
that cohort had a higher proportion of patients with cardiogenic shock (8.4% versus 3.7%) and of creatinine levels >200
μmol/L (3.7% versus 0.5%). Higher volume centers, on the
contrary, had a greater proportion of primary PCI cases where
the patient was a smoker (30.8% versus 39%).
For the total cohort, the crude unadjusted rate for mortality at 30 days was 1.9%. From 2007 to 2013, the annual rate
increased from 1.4% to 2.2%. By contrast, for primary PCI,
the unadjusted 30-day mortality rate was 4.8% and remained
unchanged in 2013 compared with 2007. Across this same
interval, predicted mortality rates increased from 1.3% to
2.3% in the overall PCI cohort and from 4.8% to 5.2% for
primary PCI cases.
The median number of PCI cases treated annually per hospital was 659.5, and the mean was 878.5. The observed center
average volumes ranged from 61 (almost always new centers
starting up a PCI program) to 2794. Few centers (5.4%) undertook <200 PCI procedures per year, with most (87.1%) performing on average between 200 and 2000 procedures. Center
annual volumes increased over the study period from a mean
of 889 in 2007 to 917 in 2013. A smaller increase occurred for
the primary PCI sample, from 934 to 945 procedures per year.
Volume and Mortality
In both PCI and primary PCI groups, for 2007 to 2013 in the
United Kingdom, observed and predicted 30-day mortality
rates depend on center average annual volume in a largely
identical fashion (Figures 2 and 3).
Risk-adjusted estimates of 30-day mortality for each
center were derived through our hierarchical modeling. The
dose–response relationship between volume and outcome was
estimated to be 1.04 (95% CI, 1.01–1.07; P=0.01) per 2-fold
increase in center average annual volume. The Bayes factor
was 0.14 meaning that, taking into account the center-level
effects, the ratio of the chance that there is no volume–outcome
relationship compared with the chance that there is such a relationship is increased by a factor of 0.14. Our summary of existing literature yielded a prior chance of a relationship of 0.50
(denoting equal chances of the relationship being absent or
present). The current study’s center-level effect estimates revise
this down to 0.12. That is to say, the prior probability is adjusted
in light of the evidence from our data. Thus, there remains support, with probability >0.05, for the hypothesis that there is no
volume–outcome relationship in our overall PCI cohort.
We also examined each of the prespecified volume bands
in our overall PCI cohort. Bayes factors for the hypothesis that
a band-specific effect is absent relative to the hypothesis that
it is present are reported in Table 3. Bands 0 to 199, 200 to
399, and 1500 to 1999 have Bayes factors <1, signifying that
the hypothesis of no band-specific effect is less plausible25 in
the light of the center-level effects for these bands. However,
repeating the argument of the previous paragraph, under the
assumption of a 0.50 prior chance of no band-specific effect,
the chance of the band with the smallest Bayes factor, 1500 to
1999, having no effect is revised down to 0.12, that is, there
remains support, >0.05, for the hypothesis that there is no
band-specific effect for this, or any other, band.
In the primary PCI subset of cases, the dose–response
relationship of volume outcome was estimated to be 1.01
(95% CI, 0.99–1.04; P=0.36) per 2-fold increase in center
average annual volume. The Bayes factor was 1.4 meaning
that the ratio of the chance of a volume–outcome relationship
for primary PCI compared with the chance of no volume–
outcome relationship is increased by a factor of 1.4. Taking
the before-study chances of no relationship to be 0.50, our
center-level effects revise the chance up to 0.74, providing
support, with probability >0.05, for the hypothesis that there
is no volume–outcome relationship in the primary PCI cohort.
5 O’Neill et al UK PCI Volume Outcome Relationship
Table 1. Patient and Procedural Characteristics for All PCI Cases Categorized by Mean Annual Overall PCI Volumes
Mean Annual Overall PCI Volume Categories
0–199
200–399
400–749
750–1499
1500–1999
≥2000
Total
5
26
24
23
8
7
93
61–155
203–391
403–747
759–1457
1526–1728
2016–2794
61–2794
2588
37 133
67 510
148 751
76 370
95 115
427 467
64.9±11.4
65.4±11.7
65.6±11.8
65.0±11.9
63.9±11.7
64.8±12.0
64.9±11.9
Female sex*
675 (26.1)
10 076 (27.1)
17 799 (26.4)
38 618 (26)
20 243 (26.5)
25 861 (27.2)
113 272 (26.5)
Diabetes mellitus*
251 (9.7)
5890 (15.9)
12 589 (18.6)
27 218 (18.3)
12 606 (16.5)
16 779 (17.6)
75 333 (17.6)
Hospital (count)*
Annual overall PCI volume (range
of hospital means)*
Overall PCI procedures (count)*
Patient characteristics
Age, y (mean±SD)*
Downloaded from http://circoutcomes.ahajournals.org/ by guest on June 16, 2017
Hypertension*
1505 (58.2)
17 682 (47.6)
31 655 (46.9)
80 553 (54.2)
37 411 (49)
45 395 (47.7)
214 201 (50.1)
Hypercholesterolemia*
1559 (60.2)
16 535 (44.5)
32 790 (48.6)
88 195 (59.3)
40 920 (53.6)
44 434 (46.7)
224 433 (52.5)
Family history of coronary
artery disease*
1037 (40.1)
15 351 (41.3)
24 273 (36.0)
56 444 (37.9)
29 401 (38.5)
37 260 (39.2)
163 766 (38.3)
Current smoking*
415 (16.0)
7889 (21.2)
13 271 (19.7)
34 834 (23.4)
18 864 (24.7)
21 979 (23.1)
97 252 (22.8)
Previous MI*
505 (19.5)
8147 (21.9)
13 298 (19.7)
29 954 (20.1)
14 034 (18.4)
21 241 (22.3)
87 179 (20.4)
Previous CVA*
38 (1.5)
1232 (3.3)
2300 (3.4)
5614 (3.8)
2916 (3.8)
3782 (4.0)
15 882 (3.7)
Previous PCI*
484 (18.7)
5214 (14.0)
8526 (12.6)
17 926 (12.1)
7988 (10.5)
10 120 (10.6)
50 258 (11.8)
19 (0.7)
641 (1.7)
971 (1.4)
2418 (1.6)
964 (1.3)
1037 (1.1)
6050 (1.4)
9 (0.3)
137 (0.4)
666 (1.0)
1108 (0.7)
823 (1.1)
485 (0.5)
3228 (0.8)
Renal disease
Creatinine>200 μmol*
Acute or chronic*
Presentation
Indication
Stable*
1240 (47.9)
14 238 (38.3)
24 816 (36.8)
51 230 (34.4)
25 310 (33.1)
35 310 (37.1)
152 144 (35.6)
NSTEMI*
489 (18.9)
19 314 (52.0)
28 223 (41.8)
56 866 (38.2)
30 007 (39.3)
32 688 (34.4)
167 587 (39.2)
STEMI*
73 (2.8)
3110 (8.4)
12 057 (17.9)
37 169 (25)
18 464 (24.2)
23 514 (24.7)
94 387 (22.1)
Rescue*
578 (22.3)
336 (0.9)
852 (1.3)
2896 (1.9)
2101 (2.8)
1342 (1.4)
8105 (1.9)
21 (0.8)
519 (1.4)
1113 (1.6)
2867 (1.9)
1204 (1.6)
1182 (1.2)
6906 (1.6)
Shock*
Urgency
Elective*
1380 (53.3)
14 240 (38.3)
25 419 (37.7)
51 401 (34.6)
25 527 (33.4)
36 665 (38.5)
154 632 (36.2)
Urgent*
524 (20.2)
18 424 (49.6)
27 294 (40.4)
52 782 (35.5)
28 018 (36.7)
30 984 (32.6)
158 026 (37)
Emergency*
683 (26.4)
4291 (11.6)
14 593 (21.6)
44 174 (29.7)
22 450 (29.4)
27 391 (28.8)
113 582 (26.6)
0 (0.0)
33 (0.1)
108 (0.2)
357 (0.2)
217 (0.3)
559 (21.6)
6518 (17.6)
12 246 (18.1)
26 902 (18.1)
14 151 (18.5)
20 461 (21.5)
Stent deployed*
2491 (96.3)
34 679 (93.4)
63 098 (93.5)
137 430 (92.4)
70 510 (92.3)
85 196 (89.6)
393 404 (92)
DES deployed*
Salvage*
39 (0.04)
754 (0.2)
Procedural characteristics
Multiple vessels attempted*
80 837 (18.9)
1835 (70.9)
25 939 (69.9)
45 183 (66.9)
98 385 (66.1)
50 995 (66.8)
60 777 (63.9)
283 114 (66.2)
IVUS use*
31 (1.2)
779 (2.1)
1605 (2.4)
5123 (3.4)
2499 (3.3)
3080 (3.2)
13 117 (3.1)
UPLMS (where LMS attempted)*
19 (79.2)
457 (70.4)
1007 (69.5)
3179 (70.1)
1501 (62.7)
2527 (79.5)
8690 (71.1)
*Significant between-volume category difference at P<0.001.
CVA indicates cerebral vascular accident; DES, drug-eluting stent; IVUS, intravascular ultrasound; MI, myocardial infarction; NSTEMI, non–ST-segment–elevation
myocardial infarction; PCI, percutaneous coronary intervention; STEMI, ST-segment–elevation myocardial infarction; and (UP)LMS, (unprotected) left main stem.
For each prespecified volume band in this cohort, Table 3
also reports a Bayes factor for the hypothesis that there is a
band-specific volume–outcome effect relative to a hypothesis
that there is not. Band 1500 to 1999 has a Bayes factor <1,
signifying that the hypothesis of no band-specific effect is
less plausible in light of the center-level effects within that
band. Under the assumption of a 0.50 prior chance of no 1500
to 1999 band-specific effect, the chance is revised down to
6 O’Neill et al UK PCI Volume Outcome Relationship
Table 2. Patient and Procedural Characteristics for Primary PCI Cases Categorized by Mean Annual Overall PCI Volumes
Mean Annual Overall PCI Volume Categories
Hospitals (count)*
Annual overall PCI volume (range of
hospital means)*
Primary PCI procedures (count)*
0–199
200–399
400–749
750–1499
1500–1999
≥2000
Total
4
26
23
23
8
7
91
142–155
203–391
403–747
759–1457
1526–1728
2016–27 941
142–2794
69
3075
11 993
37 058
18 386
23 441
94 022
65.8±12.7
65.9±13.3
64.1±13.1
63.7±13.1
63.0±12.9
63.2±13.1
63.8±13.1
20 (29.0)
847 (27.5)
3048 (25.4)
9385 (25.3)
4840 (26.3)
6307 (26.9)
24 447 (26.0)
4 (5.8)
365 (11.9)
1547 (12.9)
5019 (13.5)
2218 (12.1)
2939 (12.5)
12 092 (12.9)
4093 (34.1)
15 681 (42.3)
6986 (38.0)
8600 (36.7)
36 434 (38.8)
Patient characteristics
Age, y , mean±SD*
Female sex*
Diabetes mellitus*
Downloaded from http://circoutcomes.ahajournals.org/ by guest on June 16, 2017
Hypertension*
27 (39.1)
Hypercholesterolemia*
21 (30.4)
793 (25.8)
3538 (29.5)
16 070 (43.4)
7230 (39.3)
7932 (33.8)
35 584 (37.8)
Family history of coronary artery
disease*
13 (18.8)
964 (31.3)
3441 (28.7)
11 540 (31.1)
5623 (30.6)
7842 (33.5)
29 423 (31.3)
Current smoking*
34 149 (36.3)
1047 (34)
20 (29.0)
946 (30.8)
3870 (32.3)
13 570 (36.6)
6597 (35.9)
9146 (39)
Previous MI*
8 (11.6)
356 (11.6)
1113 (9.3)
3644 (9.8)
1854 (10.1)
2428 (10.4)
9403 (10.0)
Previous CVA†
2 (2.9)
82 (2.7)
371 (3.1)
1360 (3.7)
693 (3.8)
789 (3.4)
3297 (3.5)
Previous PCI*
4 (5.8)
222 (7.2)
711 (5.9)
2380 (6.4)
1025 (5.6)
1168 (5.0)
551 (5.9)
Creatinine >200 μmol*
1 (1.4)
114 (3.7)
121 (1.0)
327 (0.9)
141 (0.8)
127 (0.5)
831 (0.9)
Acute or chronic*
0 (0.0)
11 (0.4)
74 (0.6)
147 (0.4)
73 (0.4)
68 (0.3)
373 (0.4)
69 (100)
3075 (100)
11 993 (100)
37 058 (100)
18 386 (100)
23 441 (100)
94 022 (100)
6 (8.7)
258 (8.4)
789 (6.6)
2043 (5.5)
786 (4.3)
870 (3.7)
4752 (5.1)
69 (100)
3064 (99.6)
0 (0.0)
11 (0.4)
Multiple vessels attempted*
10 (14.5)
Stent deployed*
DES deployed*
Renal disease*
Presentation
Indication STEMI
Shock*
Urgency
Emergency*
Salvage*
11 938 (99.5)
36 967 (99.8)
18 342 (99.8)
23 427 (99.9)
93 807 (99.8)
55 (0.5)
91 (0.2)
44 (0.2)
14 (0.1)
215 (0.2)
359 (11.7)
1122 (9.4)
3052 (8.2)
1825 (9.9)
2340 (10.0)
8708 (9.3)
64 (92.8)
2892 (94.0)
11 371 (94.8)
34 585 (93.3)
17 084 (92.9)
21 985 (93.8)
87 981 (93.6)
38 (55.1)
1908 (62.0)
7278 (60.7)
22 270 (60.1)
11 797 (64.2)
13 955 (59.5)
57 246 (60.9)
IVUS use
0 (0.0)
40 (1.3)
149 (1.2)
552 (1.5)
307 (1.7)
353 (1.5)
1401 (1.5)
UPLMS (where LMS attempted)*
0 (100)
50 (98.0)
162 (88.0)
482 (66.7)
228 (70.2)
361 (81.1)
1283 (74.2)
Procedural characteristics
CVA indicates cerebral vascular accident; DES, drug-eluting stent; IVUS, intravascular ultrasound; MI, myocardial infarction; NSTEMI, non–ST-segment–elevation
myocardial infarction; PCI, percutaneous coronary intervention; STEMI, ST-segment–elevation myocardial infarction; and (UP)LMS, (unprotected) left main stem.
*Significant between-volume category difference at P<0.001.
†Significant between-volume category difference at P<0.05.
0.41, that is, there again remains significant support for the
hypothesis that there is no band-specific effect for this, or any
other, band.
Sensitivity Analyses
We found good agreement in center-level effects between the
Markov chain Monte Carlo and R estimations. Modest differences in uncertainties for center-level effects between Markov chain Monte Carlo and R were identified, but these were
insufficient to alter our conclusions. The stratified analyses
were replicated using quintiles of center average volume, and
the results showed clear similarity to our existing findings
for both total PCI cases and primary PCI cases. Full details
are provided in the Data Supplement. Notably, there was less
variation in the lowest volume quintiles for both the overall
and primary PCI volumes compared with their equivalent
clinically based counterparts (Figures 2 and 3), which is likely
indicative of the more evenly distributed volume sizes. When
only centers’ volumes of primary PCI cases (rather than their
overall PCI volumes) were considered, we found no significant association between volume of primary PCI cases and
risk of 30-day mortality. The exclusion of the 10% of patients
7 O’Neill et al UK PCI Volume Outcome Relationship
Downloaded from http://circoutcomes.ahajournals.org/ by guest on June 16, 2017
Figure 2. Observed and predicted mortality and
relative risk of mortality by mean annual center
volume: All PCI cases. Scatter points: hospitallevel estimates (with 95% confidence interval
[CI] bars for relative risk). Colored panes: linear
estimates of mortality risk for each volume band,
with 95% CI.
who were in the top 30-day mortality risk decile, and separately the exclusion of the 5% of patients who had missing
data for cardiogenic shock, did not change the overall finding
on center volume–outcome relationship. A similar affirmation
of the hierarchical findings was obtained for the elective-only
cases and with the primary PCI cohort.
8 O’Neill et al UK PCI Volume Outcome Relationship
Downloaded from http://circoutcomes.ahajournals.org/ by guest on June 16, 2017
Figure 3. Observed and predicted mortality and
relative risk of mortality by mean annual center
volume: PCI cases. Scatter points: hospital-level
estimates (with 95% confidence interval [CI] bars
for relative risk). Colored panes: linear estimates of
mortality risk for each volume band, with 95% CI.
Discussion
This is the first whole-country, consecutive series patient-level
analysis of the relationship between total center volume of
PCI cases and risk-adjusted mortality at 30 days, undertaken
using contemporary clinical data within a modern healthcare
system. For the overall PCI cohort, the evidence for a trend
9 O’Neill et al UK PCI Volume Outcome Relationship
Table 3. Bayes Factors for Stratified Analysis of All PCI
Cases and Primary PCI Cases Categorized by Mean Annual
Overall PCI Volumes
Mean Annual Overall PCI Volume Categories
0–199 200–399 400–749 750–1499 1500–1999 ≥2000
All PCI
0.92
0.58
2.40
4.30
0.18
2.28
Primary
PCI
1.78
4.68
5.40
5.44
0.70
3.49
PCI indicates percutaneous coronary intervention.
Downloaded from http://circoutcomes.ahajournals.org/ by guest on June 16, 2017
between center average annual volume and risk-adjusted
30-day mortality was not sufficient to conclude the presence
of a relationship in the context of a priori evidence. For the
primary PCI subset of cases, no evidence was found for a
relationship between center average annual volume and riskadjusted 30-day mortality.
The findings are consistent with earlier studies that have
found no evidence of a relationship between PCI volume and
risk-adjusted mortality outcomes.16–18,26,27 Although many
other existing studies have argued in favor of higher-volume
centers, the importance of adjustment for case selection and
hospital effects cannot be understated. One of the largest studies to date15 found that although crude mortality estimates
increased exponentially at lower volume institutions, there
was no relationship between hospital volume and mortality
after consideration of confounding factors.
The current study has additionally demonstrated the utility of Bayes factors, which can be computed as a guide to
the weight of evidence in studies where alternative hypotheses
are being evaluated, in contrast to P values, which only measure how extreme a test statistic derived from a study’s data
is under one hypothesis alone. In this study, their use enabled
comparison of the evidence for the contrasting hypotheses
that a relationship between volume and center effects is either
absent or present. Bayes factors usefully enable the interpretation of study data in combination with a priori evidence.24
After the principles of Bayesian inference, we combined our
study’s Bayes factors with existing findings from comparable
observational studies, with final results showing that, under
the current UK PCI regimen, the evidence does not challenge
a prior assumption that places reasonable probability on no
volume–outcome relationship. This Bayesian approach is a
general one, with applicability beyond both the clinical context and observational design used in the present study.
We have identified important differences in the baseline cardiovascular risk and presenting phenotype between
lower and higher volume centers. For example, we found
that smaller centers had fewer patients with STEMI indications. However, STEMI cases handled at low-volume centers
were proportionally more likely to present with cardiogenic
shock—a particularly influential factor in patient survival, as
is empirically reflected by its high weighting in the risk adjustment model used in this work.22 Because the risk adjustment
accounted for these characteristics, it may be argued that there
is evidence for a volume-dependent case selection bias that
attenuates the relationship between center volume and mortality. A similar phenomenon has been observed in coronary
artery bypass grafting research,28 but we can only speculate as
to why this may be occurring. It may reflect that such cases are
undertaken emergently at the presenting hospital if the risk of
transfer to a specialist heart attack center is deemed too high
or that Emergency Medical Services bring the sickest patients
to the closest hospital. It may also reflect a different clinical
approach to the definition of cardiogenic shock in lower volume centers. An alternative interpretation may be that higher
volume centers are more selective in their choice of primary
PCI cases. These possibilities warrant further inquiry, particularly given that recent work has shown that center volume of
patients with cardiogenic shock is associated with improved
patient outcomes.29
It is important to consider wider contextual issues
regarding clinical practice and policy in understanding
the results we have established in this study. The United
Kingdom has accommodated current international guidance
with national recommendations that discourage low-volume
centers with consistently <400 PCI cases per annum and
encourage regular practice with a minimum annual operator
activity of >75 cases per year.4 This may, in part, account
for the absence of evidence for a center–volume outcome
relationship. In many parts of the United Kingdom, operators in lower volume centers also have PCI sessions in the
regional higher volume centers, thus increasing their individual annual experience and allowing greater opportunity
to select more complex cases to be performed at a higher
volume regional center.
The development of new dedicated heart attack centers
and changes in individual practice have been associated with
considerable expansion in the provision of PCI within the
United Kingdom over the past decade—the current study’s
findings reflect these developments. It has meant that our
study captures a section of the volume spectrum beyond that
typically captured in previous research. The term low volume is context specific and varies between research settings.
Many studies have used coarse volume categorizations that
arguably lack measurement precision and make comparisons
between research findings and clinical settings difficult.17,18 In
this study, we explored volume using diverse strategies such
as existing recommended volume thresholds, data-driven
quantiles, and cohort-specific volume definitions (ie, primary
PCI case volume alone). This provides a detailed insight into
the volume–outcome relationship but does not overcome the
fact that the range of volumes captured in this work may differ from previous studies. Consequently, although the current
study has provided evidence that the relationship between
hospital volumes and patient mortality is mitigated when
adjustment is made for case mix, inferences should not be
made regarding particularly low volumes not captured by this
study. We cannot exclude the existence of a lower volume
limit below which institutional competence cannot be assured.
Rather, the findings suggest that the efforts made within the
UK NHS to improve access to PCI and provide guidance for
institutional standards of service provision have resulted in a
uniformly high standard of care. We do not argue against the
promotion of volume thresholds, given that its adoption has
been a key aspect of the expansion of PCI provision in the
United Kingdom. The policy whereby the acute management
of STEMI is restricted to a limited number of dedicated heart
10 O’Neill et al UK PCI Volume Outcome Relationship
attack centers is driven largely by logistical necessity and is
not challenged by this study’s findings.
Acknowledgments
We would like to thank Professor John Deanfield for his support in
this work through his role as the director of NICOR.
Strengths and Limitations
Downloaded from http://circoutcomes.ahajournals.org/ by guest on June 16, 2017
The strengths of this study include the national data source
of consecutive PCI cases, the depth of detail of clinical data,
and the robust mortality tracking and linkage. The findings
of this work are also based on strong methodological foundations. More commonly used fixed-effect regression methods
would not have reflected the hierarchical nature of the data
under investigation,30 namely, the nesting of procedural results
within centers. Hierarchical models were used in this study to
estimate center-level effects independently of center average
volume, which were analyzed using Bayes factors to assess
evidence for, and against, a relationship between volume and
risk-adjusted outcome. Nonetheless, there are limitations to our
study. These include the categorization of the center volumes
using clinical consensus and existing guidelines. To address
this limitation, we also modeled volume as a continuous variable and undertook sensitivity analyses using quintiles of volumes; each method supported our main study findings and
provided verification that our results were not the consequence
of the a priori clinical categorization. Although BCIS data collection is mandated for all operators and all cases in England
and Wales, it is possible that some cases will not have been
recorded, although such omissions are likely to be minimal, as
case ascertainment has been shown to be 100% for the majority
of NHS hospitals in England and Wales.31 Further, some of the
cases had missing data that we treated in an identical manner to
the way in which the risk model was derived. Multiple imputation may have allowed more cases to be modeled and with
different precision.32 However, sensitivity analyses showed that
excluding records missing data in the highest weighted risk
factor made no difference to our conclusions. Operator-identifying data were not collected reliably by the BCIS registry
until 2012, and so consultant volumes were not addressed in
this study. Although our research has not supported the presence of a relationship between center volume and outcomes,
existing research33 suggests that valuable additional insights
will be obtained once sufficient data are available to replicate
the current study at the consultant level. Finally, because of the
observational nature of data, the models produced in this work
disclosed many important associations but cannot provide evidence for causation. However, it would not have been possible
to obtain such coverage of real-world practice as achieved in
this work if an experimental design had been used.34,35
Conclusions
Data covering all cases of emergent and elective PCI in England
and Wales has shown that most centers undertake between 200
and 2000 PCI procedures per year on average. Lower volume
centers undertake proportionally less STEMI cases overall but
proportionally more high risk primary PCI than higher volume
centers. Although observed and predicted mortality rates depend
on annual center volume, we have found that once adjustment
is made for patient case mix, there is insufficient evidence for
a credible volume–outcome relationship for PCI. Our analysis
suggests that the current configuration of PCI services in England and Wales delivers care of uniformly high quality.
Sources of Funding
The British Cardiovascular Intervention Society is commissioned by
the Health Quality Improvement Partnership as part of the National
Clinical Audit and Patient Outcomes Programme. The study was
funded by the British Cardiovascular Society. Dr Gale is funded by
the National Institute for Health Research (NIHR-CTF-2014-03-03)
as Associate Professor and Honorary Consultant Cardiologist.
Disclosures
None.
References
1. ACCF/AHA/SCAI Writing Committee and Task Force Members. 2011
ACCF/AHA/SCAI guideline for percutaneous coronary intervention: executive summary. Circulation. 2011;124:2574–2609.
2. Ko DT, Wijeysundera HC, Zhu X, Richards J, Tu JV; National Expert
Panel. Canadian quality indicators for percutaneous coronary interventions. Can J Cardiol. 2008;24:899–903.
3. The Task Force on Myocardial Revascularization of the ESC and the
EACTS. 2014 ESC/EACTS guidelines on myocardial revascularization.
Eur Heart J. 2014;35:2541–2619.
4. British Cardiovascular Intervention Society. Statement on the development and peer review of new PCI services. http://www.bcis.org.uk/resources/BCIS_New_Site_Guidance_31012011.pdf Published January 1,
2011. Accessed January 6, 2015.
5.Andersen HR, Terkelsen CJ, Thuesen L, Krusell LR, Kristensen SD,
Bøtker HE, Lassen JF, Nielsen TT. Myocardial infarction centres:
the way forward. Heart. 2005;91(suppl 3):iii12–iii15. doi: 10.1136/
hrt.2004.058537.
6. Fothergill RT, Watson LR, Virdi GK, Moore FP, Whitbread M. Survival
of resuscitated cardiac arrest patients with ST-elevation myocardial
infarction (STEMI) conveyed directly to a Heart Attack Centre by ambulance clinicians. Resuscitation. 2014;85:96–98. doi: 10.1016/j.
resuscitation.2013.09.010.
7.Iqbal MB, Al-Hussaini A, Rosser G, Salehi S, Phylactou M,
Rajakulasingham R, Patel J, Elliott K, Mohan P, Green R, Whitbread M,
Smith R, Ilsley C. Predictors of survival and favorable functional outcomes
after an out-of-hospital cardiac arrest in patients systematically brought to
a dedicated heart attack center (from the Harefield Cardiac Arrest Study).
Am J Cardiol. 2015;115:730–737. doi: 10.1016/j.amjcard.2014.12.033.
8. Gale CP, Cattle BA, Woolston A, Baxter PD, West TH, Simms AD, Blaxill
J, Greenwood DC, Fox KA, West RM. Resolving inequalities in care?
Reduced mortality in the elderly after acute coronary syndromes. The
Myocardial Ischaemia National Audit Project 2003-2010. Eur Heart J.
2012;33:630–639. doi: 10.1093/eurheartj/ehr381.
9. West RM, Cattle BA, Bouyssie M, Squire I, de Belder M, Fox KA, Boyle
R, McLenachan JM, Batin PD, Greenwood DC, Gale CP. Impact of hospital proportion and volume on primary percutaneous coronary intervention
performance in England and Wales. Eur Heart J. 2011;32:706–711. doi:
10.1093/eurheartj/ehq476.
10. Akin I, Hochadel M, Schneider S, Abdel-Wahab M, Zahn R, Senges J,
Richardt G, Kuck KH, Nienaber CA; DES.DE Study group. Volumeoutcomes relationship in the era of modern coronary intervention-results
from the prospective multicenter German DES.DE Registry. Catheter
Cardiovasc Interv. 2013;82:E788–E797. doi: 10.1002/ccd.24986.
11. Kontos MC, Wang Y, Chaudhry SI, Vetrovec GW, Curtis J, Messenger
J; NCDR. Lower hospital volume is associated with higher in-hospital
mortality in patients undergoing primary percutaneous coronary intervention for ST-segment-elevation myocardial infarction: a report from the
NCDR. Circ Cardiovasc Qual Outcomes. 2013;6:659–667. doi: 10.1161/
CIRCOUTCOMES.113.000233.
12. Post PN, Kuijpers M, Ebels T, Zijlstra F. The relation between volume and
outcome of coronary interventions: a systematic review and meta-analysis. Eur Heart J. 2010;31:1985–1992. doi: 10.1093/eurheartj/ehq151.
13. Navarese EP, De Servi S, Politi A, Martinoni A, Musumeci G, Boschetti
E, Belli G, D’Urbano M, Piccaluga E, Lettieri C, Klugmann S. Impact
of primary PCI volume on hospital mortality in STEMI patients: does
11 O’Neill et al UK PCI Volume Outcome Relationship
Downloaded from http://circoutcomes.ahajournals.org/ by guest on June 16, 2017
time-to-presentation matter? J Thromb Thrombolysis. 2011;32:223–231.
doi: 10.1007/s11239-011-0598-1.
14. Kuwabara H, Fushimi K, Matsuda S. Relationship between hospital volume and outcomes following primary percutaneous coronary intervention
in patients with acute myocardial infarction. Circ J. 2011;75:1107–1112.
15. Badheka AO, Patel NJ, Grover P, Singh V, Patel N, Arora S, Chothani
A, Mehta K, Deshmukh A, Savani GT, Patel A, Panaich SS, Shah N,
Rathod A, Brown M, Mohamad T, Tamburrino FV, Kar S, Makkar R,
O’Neill WW, De Marchena E, Schreiber T, Grines CL, Rihal CS, Cohen
MG. Impact of annual operator and institutional volume on percutaneous coronary intervention outcomes: a 5-year United States experience
(2005–2009). Circulation. 2014;130:1392–1406.
16. Kumbhani DJ, Cannon CP, Fonarow GC, Liang L, Askari AT, Peacock
WF, Peterson ED, Bhatt DL. Association of hospital primary angioplasty
volume in ST-segment elevation myocardial infarction with quality and
outcomes. JAMA. 2009;302:2207–2213.
17.Ohtsuka Machino T, Toyama M, Obara K, Takeyasu N, Watanabe S,
Aonuma K; Ibaraki Coronary Artery Disease Study (ICAS) Registry.
Effect of hospital case volume on treatment and in-hospital outcomes in
patients undergoing percutaneous coronary intervention for acute myocardial infarction. Results from the Ibaraki Coronary Artery Disease Study
(ICAS) Registry. Int Heart J. 2008;49:249–260.
18. Spaulding C, Morice MC, Lancelin B, El Haddad S, Lepage E, Bataille S,
Tresca JP, Mouranche X, Fosse S, Monchi M, de Vernejoul N; CARDIOARIF registry Investigators. Is the volume-outcome relation still an issue
in the era of PCI with systematic stenting? Results of the greater Paris area
PCI registry. Eur Heart J. 2006;27:1054–1060. doi: 10.1093/eurheartj/
ehi843.
19.Writing Committee and ACCF/AHA/ACP Task Force on Clinical
Competence and Training Members. ACCF/AHA/SCAI 2013 update of
the clinical competence statement on coronary artery interventional procedures. Circulation. 2013;128:436–472.
20. National Clinical Guideline Centre. Myocardial Infarction with ST-segment
Elevation: Acute Management. Clinical Guideline 167. Manchester, UK:
National Institute for Health and Care Excellence; 2013.
21.Ludman PF; British Cardiovascular Intervention Society. British
Cardiovascular Intervention Society Registry for audit and quality assessment of percutaneous coronary interventions in the United Kingdom.
Heart. 2011;97:1293–1297. doi: 10.1136/heartjnl-2011-300299.
22. McAllister KS, Ludman PF, Hulme W, de Belder MA, Stables R, Chowdhary
S, Mamas MA, Sperrin M, Buchan IE; British Cardiovascular Intervention
Society and the National Institute for Cardiovascular Outcomes Research.
A contemporary risk model for predicting 30-day mortality following
percutaneous coronary intervention in England and Wales. Int J Cardiol.
2016;210:125–132. doi: 10.1016/j.ijcard.2016.02.085.
23. Benjamini Y, Hochberg Y. Controlling the false discovery rate: a practical and powerful approach to multiple testing. J R Stat Soc Series B.
1995;57:289–300.
24. Goodman SN. Toward evidence-based medical statistics. 2: the Bayes factor. Ann Intern Med. 1999;130:1005–1013.
25. Bernardo JM, Smith AFM. Bayesian Theory. Chichester: Wiley; 2000.
26. Kimmel SE, Sauer WH, Brensinger C, Hirshfeld J, Haber HL, Localio
AR. Relationship between coronary angioplasty laboratory volume and
outcomes after hospital discharge. Am Heart J. 2002;143:833–840.
27. Shiraishi J, Kohno Y, Sawada T, Arihara M, Hyogo M, Yagi T, Shima T,
Okada T, Nakamura T, Matoba S, Yamada H, Shirayama T, Tatsumi T,
Kitamura M, Furukawa K, Matsubara H; AMI-Kyoto Multi-Center Risk
Study Group. Effects of hospital volume of primary percutaneous coronary interventions on angiographic results and in-hospital outcomes for
acute myocardial infarction. Circ J. 2008;72:1041–1046.
28.Nallamothu BK, Saint S, Hofer TP, Vijan S, Eagle KA, Bernstein SJ.
Impact of patient risk on the hospital volume-outcome relationship in coronary artery bypass grafting. Arch Intern Med. 2005;165:333–337. doi:
10.1001/archinte.165.3.333.
29.Shaefi S, O’Gara B, Kociol RD, Joynt K, Mueller A, Nizamuddin J,
Mahmood E, Talmor D, Shahul S. Effect of cardiogenic shock hospital
volume on mortality in patients with cardiogenic shock. J Am Heart Assoc.
2015;4:e001462. doi: 10.1161/JAHA.114.001462.
30. Diez-Roux AV. Multilevel analysis in public health research. Annu Rev Public
Health. 2000;21:171–192. doi: 10.1146/annurev.publhealth.21.1.171.
31. Ludman P. BCIS audit returns: adult interventional procedures. Paper presented at: BCIS Autumn Meeting, Blackpool, UK, October 2014.
32.Cattle BA, Baxter PD, Greenwood DC, Gale CP, West RM. Multiple
imputation for completion of a national clinical audit dataset. Stat Med.
2011;30:2736–2753. doi: 10.1002/sim.4314.
33.Strom JB, Wimmer NJ, Wasfy JH, Kennedy K, Yeh RW. Association
between operator procedure volume and patient outcomes in percutaneous coronary intervention: a systematic review and meta-analysis.
Circ Cardiovasc Qual Outcomes. 2014;7:560–566. doi: 10.1161/
CIRCOUTCOMES.114.000884.
34. Black N. Why we need observational studies to evaluate the effectiveness
of health care. BMJ. 1996;312:1215–1218.
35. Victora CG, Habicht JP, Bryce J. Evidence-based public health: moving
beyond randomized trials. Am J Public Health. 2004;94:400–405.
Total Center Percutaneous Coronary Intervention Volume and 30-Day Mortality: A
Contemporary National Cohort Study of 427 467 Elective, Urgent, and Emergency Cases
Darragh O'Neill, Owen Nicholas, Chris P. Gale, Peter Ludman, Mark A. de Belder, Adam
Timmis, Keith A.A. Fox, Iain A Simpson, Simon Redwood and Simon G. Ray
Downloaded from http://circoutcomes.ahajournals.org/ by guest on June 16, 2017
Circ Cardiovasc Qual Outcomes. 2017;10:
doi: 10.1161/CIRCOUTCOMES.116.003186
Circulation: Cardiovascular Quality and Outcomes is published by the American Heart Association, 7272
Greenville Avenue, Dallas, TX 75231
Copyright © 2017 American Heart Association, Inc. All rights reserved.
Print ISSN: 1941-7705. Online ISSN: 1941-7713
The online version of this article, along with updated information and services, is located on the
World Wide Web at:
http://circoutcomes.ahajournals.org/content/10/3/e003186
Data Supplement (unedited) at:
http://circoutcomes.ahajournals.org/content/suppl/2017/03/20/CIRCOUTCOMES.116.003186.DC1
Permissions: Requests for permissions to reproduce figures, tables, or portions of articles originally published
in Circulation: Cardiovascular Quality and Outcomes can be obtained via RightsLink, a service of the
Copyright Clearance Center, not the Editorial Office. Once the online version of the published article for
which permission is being requested is located, click Request Permissions in the middle column of the Web
page under Services. Further information about this process is available in the Permissions and Rights
Question and Answer document.
Reprints: Information about reprints can be found online at:
http://www.lww.com/reprints
Subscriptions: Information about subscribing to Circulation: Cardiovascular Quality and Outcomes is online
at:
http://circoutcomes.ahajournals.org//subscriptions/
Supplemental Material
Model Calibration and Discrimination
A receiver-operating characteristic (ROC) curve was generated to examine the 30-day risk
adjustment model’s predictive performance in the sample used in this study. The area under
the curve was C = 0·85 for the overall PCI cohort, suggesting a good fit. The curve is
presented in Figure A1.
Figure A1. Overall PCI Cohort: ROC Analysis
Calibration curves were also generated, specifically a logistic curve and a smoothed nonparametric curve were both fit to the observed and predicted probabilities for the overall
cohort’s data (see Figure A2). The dashed diagonal line in the plot represents the ideal
situation where there is an exact match between the predicted and observed probabilities of
mortality across the risk spectrum. While the curves are close to the diagonal, there is some
divergence evident in the upper end of the risk spectrum where the predicted probabilities
begin to overestimate patient risk relative to the actual observed mortality. This divergence
occurs at predicted probabilities of 0·4 and above, corresponding to approximately 10% of
the sample.
These analyses were repeated for the primary PCI-only subset. A C-statistic of 0·82 was
obtained for this sample (see Figure A3), again suggesting the risk adjustment model fit the
A1
data well. However the calibration curves (Figure A4) also showed that there was again
some divergence between observed and predicted risk in the upper mortality range.
Figure A2. Overall PCI Cohort: Calibration Curve
Figure A3. Primary PCI Cohort: ROC Analysis
A2
Figure A4. Primary PCI Cohort: Calibration Curve
Risk Model
The clinical procedural risk model was developed by an independent research group using
data from the BCIS registry that comprised all eligible PCI records undertaken in England
and Wales from 2007 through 2011. Validation of the model subsequently undertaken by its
developers showed that the model performed well in estimating risk for more recent cases.
The model’s covariates were selected for their clinical relevance, predictive utility, and low
levels of missingness. The model covers 9 risk domains: patient age, sex, diabetes history,
previous myocardial infarction, renal disease history, cerebrovascular event history,
cardiogenic shock, indication for intervention and procedural urgency. A table of the
prevalence of missing data in each of the covariates is presented in Table A1. The model
also includes interaction terms for age-diabetes, age-shock and indication-shock.
A3
Table A1. Missing Data Prevalence: Case Mix Adjustment Variables
Variable
Age
Female
Diabetes
Previous MI
Previous
CVA
Renal
disease
Indication
Shock
Urgency
Overall
0-199
N(%)
Mean Annual Overall PCI Volume Categories
200-399
400-749
750-1499
1500-1999
N(%)
N(%)
N(%)
N(%)
≥2000
N(%)
N(%)
0(0)
7 (0.3)
478 (18.5)
71 (2.7)
38(0.1)
48 (0.1)
2100 (5.7)
2525 (6.8)
31(0)
170 (0.3)
3595 (5.3)
7014 (10.4)
26(0)
118 (0.1)
2852 (1.9)
14795 (9.9)
19(0)
194 (0.3)
4396 (5.8)
13838 (18.1)
0(0)
59 (0.1)
5580 (5.9)
4059 (4.3)
114(0)
596 (0.1)
19001 (4.4)
42302 (9.9)
245 (9.5)
2486 (6.7)
4932 (7.3)
5529 (3.7)
829 (1.1)
8665 (9.1)
22686 (5.3)
128 (4.9)
161 (6.2)
332 (12.8)
1 (0)
1661 (4.5)
92 (0.2)
2716 (7.3)
145 (0.4)
4714 (7)
66 (0.1)
4801 (7.1)
96 (0.1)
5233 (3.5)
247 (0.2)
12009 (8.1)
37 (0)
4566 (6)
333 (0.4)
1821 (2.4)
158 (0.2)
6258 (6.6)
2192 (2.3)
3479 (3.7)
36 (0)
22560 (5.3)
3091 (0.7)
25158 (5.9)
473 (0.1)
Statistical Model
Based on the clinical procedural risk model for 30-day mortality we assumed that there
would be between center variability in risk-adjusted outcomes. This is a standard assumption
which is modelled using a random effects distribution and random effect for each center (a
log odds difference) (1). We therefore considered that center random effects might be
associated with load, which we took to be measured by center average annual volume, and
we allowed for secular change by including fixed effects for each year.
We constructed the likelihood function by assuming that the log odds of adverse outcome for
a procedure in the ith center was given by
𝑝𝑝
� + 𝛽𝛽1 𝜃𝜃1 + 𝛽𝛽2 𝜃𝜃2 + 𝛽𝛽3 𝜃𝜃3 + 𝛽𝛽4 𝜃𝜃4 + 𝛽𝛽5 𝜃𝜃5 + 𝛽𝛽6 𝜃𝜃6 + 𝜀𝜀𝑖𝑖
𝛽𝛽0 + ln �
1 − 𝑝𝑝
and that the prior distribution for 𝜀𝜀𝑖𝑖 ~𝑁𝑁(0, 𝜎𝜎 2 ), a normal distribution with mean zero and
variance 𝜎𝜎 2 . 𝑝𝑝 is the procedural risk, 𝜃𝜃1 ,… 𝜃𝜃6 are indicator variables for each year (with the
exception of 2007). 𝛽𝛽0 ,… 𝛽𝛽6, 𝜎𝜎 2 and 𝜀𝜀𝑖𝑖 are the center level effects in which we are
interested. We treat the other parameters as nuisance parameters.
Once we had estimated random effects, together with their uncertainties, we modelled the
linear association of center random effect and center average annual volume using the loglikelihood contribution from each center
A4
1
1 (𝜀𝜀�𝚤𝚤 − 𝜇𝜇 − 𝜆𝜆ln(𝑣𝑣𝑖𝑖 /𝑚𝑚))2
− ln(2𝜋𝜋) − ln(𝜏𝜏 + 𝜎𝜎𝑖𝑖 2 ) −
𝜏𝜏 + 𝜎𝜎𝑖𝑖 2
2
2
where 𝜀𝜀�𝚤𝚤 is the expectation of the marginal posterior distribution for the random effect for the
ith center, 𝜎𝜎𝑖𝑖 is a quarter of the 95% range of the marginal posterior distribution, 𝑣𝑣𝑖𝑖 is the
centre average volume, 𝑚𝑚 is the median centre average annual volume, 𝛾𝛾 is the gradient of
estimated centre random effect with respect to center volume, 𝜇𝜇 is an offset and 𝜏𝜏 is the
uncertainty in the relationship. 𝜆𝜆, 𝜇𝜇 and 𝜏𝜏 are parameters to be estimated. The parameter of
interest is 𝜆𝜆 and whether it is zero or not. 𝜆𝜆 represents the gradient of center level effects
with log-volume.
For stratified analysis we modelled the association of each strata of centers (defined by
membership of a range of center average annual volume) using the log-likelihood
contribution from each center
1
1 (𝜀𝜀�𝚤𝚤 − 𝜇𝜇)2
2)
− ln(2𝜋𝜋) − ln(𝜏𝜏 + 𝜎𝜎𝑖𝑖 −
2
2 𝜏𝜏 + 𝜎𝜎𝑖𝑖 2
The parameter of interest in 𝜇𝜇 and whether it is zero or not. 𝜇𝜇 represents the band specific
effect.
Prior Distribution
In the random effects modelling we took the prior for the inverse variance as a Gamma
distribution with parameters 𝛼𝛼 = 𝛽𝛽 = 0.001.
In the post estimation linear modelling we took a uniform prior for (𝜆𝜆, 𝜇𝜇, 𝜏𝜏) supported on
[−0.1,0.1]×[−0.2,0.2]×[0.001,0.3]. In post estimation stratified modelling we took a uniform
prior for (𝜇𝜇, 𝜏𝜏) supported on [−0.2,0.2]×[0.001,0.3].
Analytic Technique
For the random effects model we generated an MCMC run of 5 million for the overall PCI
cohort, and 20 million for the primary PCI cohort. The proposal distribution was Gaussian
with zero off-diagonal elements. The acceptance proportions were close to 0·25 and 0·23
respectively; every 3700th sample from the second half of the overall PCI run and every
5900th sample from the second half of the primary PCI run was used for analysis.
Convergence of the resulting samples was tested by estimating the potential scale reduction
of each scalar estimate.
For post random effects estimation we divided the range of 𝜆𝜆 into 200 equal intervals, 𝜇𝜇 into
400 equal intervals and 𝜏𝜏 into 299 equal intervals. Numerical integration was performed to
A5
compute posterior weights in the linear analysis for the 𝜆𝜆 = 0, 𝜆𝜆 < 0 and 𝜆𝜆 > 0 hypotheses,
and posterior weights in the stratified analyses for the 𝜇𝜇 = 0 and 𝜇𝜇 ≠ 0 hypotheses.
Bayesian Inference
We conducted a simple aggregated sample size comparison of the recent risk-adjusted
studies mentioned in the introduction of this paper to weigh the existing evidence in favor of
or against the existence of a PCI center-level volume-outcome relationship. We commenced
this process with Post et al.’s meta-analysis (2) and examined the papers used in that metaanalysis for which both inferential results and sample size data were clearly specified and
linked (one study was excluded on the basis that it did not meet this requirement). We
subsequently undertook a literature review and identified additional studies published after
the period covered by Post et al.’s meta-analysis. All of these studies used appropriate riskadjustment to investigate the relationship between PCI or primary PCI center volume and
patient mortality. The studies were grouped according to whether they provided support that
a relationship was present or absent. The sample sizes for these two groups were
aggregated as a proxy for the strength of their evidence and then compared. The resulting
values showed that there was near equivalent support both for and against the volumeoutcome relationship in these risk-adjusted data (see Table A2). Consequently, in evaluating
the evidence from our study regarding the volume-outcome relationship, we chose to set our
prior probability for the presence or absence of this relationship at 0·5.
Table A2. Support for the Absence or Presence of a PCI Volume-Outcome Relationship in
Recent Risk-Adjusted Studies
Presence
Absence
Study
Year Sample Study
Year Sample
Allareddy et al.*
2007 573072 Kimmel et al.*
2002
25222
Hannan et al.*
1997
62670 Tsuchihashi et al.* 2004
2491
Hannan et al.*
2005 107713 Vakili et al.*
2001
1342
Canto et al.*
2000
36535 Carey et al.*
2000 153755
Navarese et al.(3) 2011
2558 Shiraishi et al.*
2008
6054
Kuwabara et al.(4) 2011
8391 Badheka et al.(7)
2014 457498
Kontos et al.(5)
2013
87324 Kumbhani et al.(8) 2009
29513
Akin et al.(6)
2013
5489 Spaulding et al.(9) 2006
37848
Total
883752
713723
Ratio
0·55
0·45
* = cited in Post et al.’s meta-analysis
A6
Descriptive Tables: Significance Testing of Patient and Procedural Characteristics
A series of chi-squares and ANOVAs were undertaken to investigate whether there were any
statistically significant differences between the volume categories. The results for the overall
sample are presented in Table A3 and for the primary PCI only sample in Table A5. Posthoc pairwise comparisons are reported in Tables A4 for the overall sample and A6 for the
primary PCI only sample.
Table A3. Significance testing for baseline patient and procedural characteristics for all PCI
cases categorized by mean annual overall PCI volumes
Mean Volume
Record Count
Hospital Count
F/χ2
df
280
5, 86
175376·1 5
30·81
5
p
<0·001
<0·001
<0·001
Patient Characteristics
Age (Mean)
Female
Diabetes
Hypertension
Hypercholesterolemia
Family history of coronary artery disease
Current smoking
Previous MI
Previous CVA
Previous PCI
Renal Disease
Creatinine >200 μmol
Acute or chronic
173·7
53·56
239·67
1666·57
5488·40
345·70
691·54
72·46
103·08
607·33
144·72
319·34
5, 427461
5
5
5
5
5
5
5
5
5
5
5
<0·001
<0·001
<0·001
<0·001
<0·001
<0·001
<0·001
<0·001
<0·001
<0·001
<0·001
<0·001
Presentation
Indication
Stable
NSTEMI
STEMI
Rescue
Shock
Urgency
Elective
Urgent
Emergency
Salvage
717·61
4189·03
6622·29
6576·19
238·92
1120·53
4137·96
6439·48
205·32
5
5
5
5
5
5
5
5
5
<0·001
<0·001
<0·001
<0·001
<0·001
<0·001
<0·001
<0·001
<0·001
Procedural Characteristics
Multiple vessels attempted
Stent deployed
DES deployed
IVUS use
UPLMS (where LMS attempted)
548·49
241·91
705·69
346·85
196·57
5
5
5
5
5
<0·001
<0·001
<0·001
<0·001
<0·001
A7
Table A4. Post-hoc testing for all PCI cases: Comparisons of baseline characteristics
categorized by mean annual overall PCI volumes
Mean Volume
200-399
400-749
750-1499
1500-1999
≥2000
Record Count
200-399
400-749
750-1499
1500-1999
≥2000
Hospital Count
200-399
400-749
750-1499
1500-1999
≥2000
Age
200-399
400-749
750-1499
1500-1999
≥2000
Current
smoking
200-399
400-749
750-1499
1500-1999
≥2000
Female
200-399
400-749
750-1499
1500-1999
≥2000
0-199
12†
18†
22†
82†
32†
0-199
30043·5†
60128·2†
141164†
68945·3†
87625·2†
0-199
14·2*
12·4*
11·6*
0·7
0·3
0-199
2
2·8
0·2
4·7†
0·5
0-199
39·4†
20·6†
77·2†
101·4†
70·9†
0-199
1·2
0·1
0
0·2
1·4
Games-Howell Test
200-399
400-749
750-1499
10†
18†
12†
68†
41†
10†
30†
26†
15†
Chi-Square Test (χ2)
200-399
400-749
750-1499
†
8818·2
67023·4†
30519·1†
13563·9†
545·6†
23272†
†
†
25421·3
4685·8
11796·7†
Chi-Square Test (χ2)
200-399
400-749
750-1499
0·1
0·2
0
9·5*
8*
7·3*
10·9*
9·3*
8·5*
Games-Howell Test
200-399
400-749
750-1499
2·4
6†
10·7†
20·6†
27·4†
21·2†
†
†
7·8
12·5
3·1*
Chi-Square Test (χ2)
200-399
400-749
750-1499
†
37·3
79·1†
379·3†
†
†
165·4
525·1
45·7†
†
†
52·9
276·6
3·1
Chi-Square Test (χ2)
200-399
400-749
750-1499
6·6*
†
21·6
4·8
4·5
0·4
9·1*
0
12*
44·6†
1500-1999
†
11
1500-1999
2049†
1500-1999
0·1
1500-1999
16·7†
1500-1999
59·2†
1500-1999
8·6*
A8
Diabetes
200-399
400-749
750-1499
1500-1999
≥2000
Hypertension
200-399
400-749
750-1499
1500-1999
≥2000
Hypercholeste
rolemia
200-399
400-749
750-1499
1500-1999
≥2000
Family history
of coronary
artery disease
200-399
400-749
750-1499
1500-1999
≥2000
Previous MI
200-399
400-749
750-1499
1500-1999
≥2000
Previous CVA
200-399
400-749
750-1499
1500-1999
≥2000
Previous PCI
200-399
400-749
750-1499
1500-1999
≥2000
0-199
Chi-Square Test (χ2)
200-399
400-749
750-1499
123·7†
†
†
64
31·3
†
8·1*
106·6
41·9†
†
†
62·7
22
0·3
Chi-Square Test (χ2)
200-399
400-749
750-1499
5·1*
509†
981·1†
18·7†
63·1†
539·8†
0·1
11·1*
959·2†
Chi-Square Test (χ2)
200-399
400-749
750-1499
156·9†
2631·6†
2164·6†
†
†
818·7
359·9
672·2†
†
†
51·3
54·4
3697·3†
Chi-Square Test (χ2)
200-399
400-749
750-1499
1·6
18·1†
4·8*
2·5
0·8
295·1†
144·4†
84·4†
52·3†
0-199
34·5†
78·8†
62·4†
44·7†
63·4†
0-199
107·1†
126·4†
16·2†
83·8†
109·3†
0-199
240·1†
135·3†
0·9
44·4†
184·4†
0-199
15·7†
5·1*
7·4*
7·8*
14·4†
0-199
24·2†
26·8†
32†
30·6†
41·3†
0-199
42·4†
81·5†
104·7†
176·7†
168·4†
78·6†
99†
6·5*
173·8†
36·9†
Chi-Square Test (χ2)
200-399
400-749
750-1499
30·6†
†
21·8
3·4
15·2†
3·8
0·2
0·6
37·3†
28·7†
Chi-Square Test (χ2)
200-399
400-749
750-1499
0·9
9·9*
7*
6*
3·2
0·5
41·6†
45·2†
28·3†
Chi-Square Test (χ2)
200-399
400-749
750-1499
41·8†
†
†
107·8
14·4
311·5†
165·8†
125·3†
†
†
301·1
153·8
113·4†
1500-1999
40·2†
1500-1999
26·9†
1500-1999
798·3†
1500-1999
8·1*
1500-1999
16·3†
1500-1999
26·8†
1500-1999
1·4
A9
Renal Disease
Creatinine>200
μmol
200-399
400-749
750-1499
1500-1999
≥2000
Acute/chronic
200-399
400-749
750-1499
1500-1999
≥2000
Indication
Stable
200-399
400-749
750-1499
1500-1999
≥2000
NSTEMI
200-399
400-749
750-1499
1500-1999
≥2000
STEMI
200-399
400-749
750-1499
1500-1999
≥2000
Rescue
200-399
400-749
750-1499
1500-1999
≥2000
Shock
200-399
400-749
750-1499
1500-1999
≥2000
0-199
Chi-Square Test (χ2)
200-399
400-749
750-1499
1500-1999
13·8†
9*
11·7*
5·5*
2·9
9·4*
2·5
34·4†
77·5†
9·8*
0
10·5*
4·7*
12·4†
1·1
126·4†
60·7†
153·8†
12·7†
5·1*
9·7*
40·2†
36†
100·1†
176·5†
0-199
42·3†
2·2
75†
†
128·8
40·9†
Chi-Square Test (χ2)
200-399
400-749
750-1499
92·8†
132·3†
203·3†
244·1†
124·8†
25·6†
198·2†
297·6†
16·9†
109·4†
206·4†
2·2
37·9†
182·4†
293·7†
1060·1†
540·1†
403·2†
438·4†
268·2†
1006·4†
2333·7†
1645·1†
3484·7†
248·8†
93·9†
932·4†
24†
372·3†
442·7†
100·5†
392·9†
672·5†
634·3†
658·7†
1738†
4830·2†
4051·4†
4437·2†
1341·5†
855·3†
1087·6†
17·8†
2·2
6·8*
4932·3†
5527·4†
4704·9†
2922·6†
5714·2†
26·9†
188·2†
404·5†
54·2†
0-199
4·5*
8·6*
14·3†
6·1*
2
†
127·5
394·5†
150†
6·5*
97·3†
Chi-Square Test (χ2)
200-399
400-749
750-1499
9·4*
48·8†
22·6†
1·7
5·1*
59†
†
8·8*
59·5
204·6†
1500-1999
386†
1500-1999
30·5†
A10
Urgency
Elective
200-399
400-749
750-1499
1500-1999
≥2000
Urgent
200-399
400-749
750-1499
1500-1999
≥2000
Emergency
200-399
400-749
750-1499
1500-1999
≥2000
Salvage
200-399
400-749
750-1499
1500-1999
≥2000
Multiple
vessels
attempted
200-399
400-749
750-1499
1500-1999
≥2000
Stent deployed
200-399
400-749
750-1499
1500-1999
≥2000
DES deployed
200-399
400-749
750-1499
1500-1999
≥2000
IVUS use
200-399
400-749
750-1499
1500-1999
0-199
Chi-Square Test (χ2)
200-399
400-749
750-1499
1500-1999
226·7†
258·5†
393·7†
440·3†
230·7†
4·9*
187†
265·8†
0·4
194·3†
279·7†
13·4†
28·6†
400·8†
480·7†
835·3†
423·3†
258·1†
292·4†
174·7†
821·4†
2510·8†
1727·1†
3313·2†
487†
211·9†
1059†
31·7†
217·5†
317·2†
484·7†
33·1†
13·2†
10·8*
7*
1638·7†
5072·5†
4414·7†
4357·3†
1531·7†
1134·2†
1063†
2·2
22·6†
7·3*
1·4
3·2
5·2*
6·4*
0·3
8·5*
31·7†
42·5†
10·4*
166·4†
0-199
13·5†
24†
3·7
60·6†
140·5†
Chi-Square Test (χ2)
200-399
400-749
750-1499
26·4†
18·7†
19·1†
13·3†
0
6·7*
8·4*
21·8†
258·7†
207·6†
0-199
24·5†
35·6†
48·4†
29†
57·3†
0-199
0·2
25†
30·6†
6·8*
38·4†
0-199
9·4*
14·7†
38·3†
34·1†
0
5·9*
8·3*
270·4†
406·3†
Chi-Square Test (χ2)
200-399
400-749
750-1499
13·7†
65·7†
26·9†
2·2
7·7*
74·2†
†
†
105·8
64·4
15·7†
Chi-Square Test (χ2)
200-399
400-749
750-1499
206·5†
†
307·6
4·6*
47·3†
87·5†
174·4†
†
†
357·9
23·1
11·8†
Chi-Square Test (χ2)
200-399
400-749
750-1499
8·3*
174·7†
174·9†
122·4†
103·2†
4·5*
1500-1999
1500-1999
126·6†
1500-1999
224·6†
1500-1999
A11
≥2000
UPLMS (where
LMS
attempted)
200-399
400-749
750-1499
1500-1999
≥2000
33·4†
122·2†
0·1
0-199
104·3†
7·5*
Chi-Square Test (χ2)
200-399
400-749
750-1499
0·5
0·6
0·5
2·1
0
0·1
0
12·8†
25·6†
191·9†
0·2
18†
55·1†
38·9†
85·5†
1500-1999
*=Significant between-volume category difference at p<0·05
†=Significant between-volume category difference at p<0·001
Table A5. Significance testing for baseline patient and procedural characteristics for primary
PCI cases categorized by mean annual overall PCI volumes
Mean Volume
Record Count
Hospital Count
F/χ2
df
271·9
5, 84
60034·28 5
31·83
5
p
<0·001
<0·001
<0·001
Patient Characteristics
Age (Mean)
Female
Diabetes
Hypertension
Hypercholesterolemia
Family history of coronary artery disease
Current Smoking
Previous MI
Previous CVA
Previous PCI
Renal Disease
Creatinine >200 μmol
Acute or chronic
33·94
25·52
23·89
381·38
1203·64
98·36
204·43
22·71
13·66
67·06
309·40
21·65
<0·001
<0·001
<0·001
<0·001
<0·001
<0·001
<0·001
<0·001
0·019
<0·001
<0·001
<0·001
5, 94016
5
5
5
5
5
5
5
5
5
5
5
Presentation
Indication
Shock
Urgency
STEMI
Emergency
Salvage
Procedural Characteristics
Multiple vessels attempted
Stent deployed
DES deployed
IVUS use
UPLMS (where LMS attempted)
Significance testing not
undertaken as all volume
categories had values of 100%
on this field.
271·65
5
<0·001
60·08
5
<0·001
60·08
5
<0·001
91·67
87·47
318·80
10·89
69·00
5
5
5
5
5
<0·001
<0·001
<0·001
0·054
<0·001
A12
Table A6. Post-hoc testing for primary PCI cases: Comparisons of baseline characteristics
categorized by mean annual overall PCI volumes
Mean Volume
200-399
400-749
750-1499
1500-1999
≥2000
Record Count
200-399
400-749
750-1499
1500-1999
≥2000
Hospital Count
200-399
400-749
750-1499
1500-1999
≥2000
Age
200-399
400-749
750-1499
1500-1999
≥2000
Current
smoking
200-399
400-749
750-1499
1500-1999
≥2000
Female
200-399
400-749
750-1499
1500-1999
≥2000
Diabetes
200-399
400-749
750-1499
1500-1999
≥2000
0-199
13·3†
18†
21·9†
91†
32·1†
0-199
2874·1†
11787·6†
36851·5†
18180†
23234·8†
0-199
16·1†
13·4*
13·4*
1·3
0·8
0-199
0·1
1·1
1·4
1·8
1·7
0-199
0
0·2
1·4
1·1
2·5
0-199
0
0·4
0·4
0·2
0·1
0-199
1·8
2·6
2·6
1·9
2·4
Games-Howell Test
400-749
750-1499
12·2†
41·8†
10·4†
25·6†
15·3†
Chi-Square Test (χ2)
200-399
400-749
750-1499
5278·1†
28775·4†
12808·2†
†
†
10923·4
1345·4
6288·2†
†
†
15642·4
3698·6
3064·9†
Chi-Square Test (χ2)
200-399
400-749
750-1499
0·2
0·2
0
9·5*
7·3*
7·3*
10·9*
8·5*
8·5*
Games-Howell Test
200-399
400-749
750-1499
6·6†
9†
3·5*
†
†
†
11·1
7·2
5·3
10·4†
6·1†
3·7*
Chi-Square Test (χ2)
200-399
400-749
750-1499
2·5
41·9†
74·6†
30†
41·8†
2·9
78·2†
155·2†
35·1†
Chi-Square Test (χ2)
200-399
400-749
750-1499
5·4
7·2*
0·1
2
2·7
6·3*
0·6
8·3*
18·3†
Chi-Square Test (χ2)
200-399
400-749
750-1499
3·8
4·8
0
0·1
9·1
16·8†
2·4
0·7
1·8
200-399
9·8†
18·1†
68·3†
29·6†
1500-1999
10·7†
1500-1999
610·9†
1500-1999
0·1
1500-1999
1·6
1500-1999
43·1†
1500-1999
1·7
1500-1999
6·7*
A13
Hypertension
200-399
400-749
750-1499
1500-1999
≥2000
Hypercholeste
rolemia
200-399
400-749
750-1499
1500-1999
≥2000
Family history
of coronary
artery disease
200-399
400-749
750-1499
1500-1999
≥2000
Previous MI
200-399
400-749
750-1499
1500-1999
≥2000
Previous CVA
200-399
400-749
750-1499
1500-1999
≥2000
Previous PCI
200-399
400-749
750-1499
1500-1999
≥2000
Renal Disease
Creatinine>200
μmol
200-399
400-749
750-1499
1500-1999
≥2000
0-199
Chi-Square Test (χ2)
200-399
400-749
750-1499
0
79·5†
252†
17·4†
46·7†
94·6†
8·1*
22·5†
189†
Chi-Square Test (χ2)
200-399
400-749
750-1499
16·3†
†
†
359·3
725·2
205·6†
305·6†
82·2†
†
†
79·4
68
544†
Chi-Square Test (χ2)
200-399
400-749
750-1499
4·4
2·8
4·3
3·9
6*
8·2*
0
0·7
5·3*
0-199
0·6
0·6
0·2
0
0·1
0-199
0·5
0
4·2
1·9
0·2
1500-1999
7·5*
1500-1999
133·9†
1500-1999
38·8†
0-199
25·5†
12·3*
1·8
82·7†
35·2†
Chi-Square Test (χ2)
200-399
400-749
750-1499
11·5*
11·7*
0·2
3·1
7·6*
9·5*
10·8*
0·3
0
Chi-Square Test (χ2)
200-399
400-749
750-1499
1·7
7·1
5·8
6·8
5
0
4·2
1·5
1·9
Chi-Square Test (χ2)
200-399
400-749
750-1499
6·8*
2·8
3·7
12·7*
1·6
15·2†
†
†
26·9
13·9
53·6†
Chi-Square Test (χ2)
200-399
400-749
750-1499
0·3
0
0
0
0·1
109·2†
208·6†
195†
281·1†
5·4*
0-199
0
0·1
0
0
0
0-199
0
0
0
0
0
0-199
0
0
0
0
0
2·4
6·1*
22·7†
1·9
17·2†
1500-1999
7·5*
1500-1999
1·5
1500-1999
7·2*
1500-1999
A14
Acute/chronic
200-399
400-749
750-1499
1500-1999
≥2000
Shock
200-399
400-749
750-1499
1500-1999
≥2000
Urgency
Emergency
200-399
400-749
750-1499
1500-1999
≥2000
Salvage
200-399
400-749
750-1499
1500-1999
≥2000
Multiple
vessels
attempted
200-399
400-749
750-1499
1500-1999
≥2000
Stent deployed
200-399
400-749
750-1499
1500-1999
≥2000
DES deployed
200-399
400-749
750-1499
1500-1999
≥2000
0
0
0
0
0
2·1
0-199
10·9*
8·1*
0
19·3†
2·9
Chi-Square Test (χ2)
200-399
400-749
750-1499
13·2†
†
†
47·6
21·3
101·8†
82†
38·4†
153·6†
152·3†
101·9†
Chi-Square Test (χ2)
200-399
400-749
750-1499
0
0
0
0
0
0·4
1
1
22·6†
22·7†
0
0
0
0
0
0·4
1
1
22·6†
0-199
0
0·6
1·6
3·6
5·1*
2·8
0
0
0·1
13·1
10·1*
62·9†
0
27·6†
1500-1999
8·7*
1500-1999
22·7†
0-199
13·1†
10·1*
0
†
62·9
27·6†
Chi-Square Test (χ2)
200-399
400-749
750-1499
0·2
1·5
2·7
1·1
1·1
14·4†
43†
9·2*
9*
0
0-199
0·6
0·5
0
0·3
0·3
0-199
1·6
0·5
0·4
2·9
0·3
15·1†
2·1
41·6†
2·7
51·1†
Chi-Square Test (χ2)
200-399
400-749
750-1499
0·3
11·8*
27·4†
1·4
0·9
23·9†
2·2
2·4
21·4†
Chi-Square Test (χ2)
200-399
400-749
750-1499
10·5*
15·3†
0·6
5·5*
92·5†
175·8†
†
21·1
4·3
3·2
1500-1999
1500-1999
0·3
1500-1999
188·2†
A15
IVUS Use
0-199
0·2
0·1
0·3
0·4
0·3
200-399
400-749
750-1499
1500-1999
≥2000
UPLMS (where
LMS
0-199
attempted)
200-399 Insufficient
data
400-749 Insufficient
data
750-1499 Insufficient
data
1500-1999 Insufficient
data
≥2000 Insufficient
data
Chi-Square Test (χ2)
200-399
400-749
750-1499
0
0·6
3·8
2
8·7*
2·5
0·6
3·8
0
Chi-Square Test (χ2)
200-399
400-749
750-1499
-
1500-1999
1·7
1500-1999
-
-
-
-
-
-
20·4†
31·5†
-
-
16·4†
20†
4·4
8·1*
4
28†
3·5
12†
*=Significant between-volume category difference at p<0·05
†=Significant between-volume category difference at p<0·001
A16
Changes in Crude Mortality Over Time
Crude mortality rates in the overall PCI sample changed from 1.4% in 2007 to 2.2% in 2014
(see Table A7). There was a significant difference between years: χ2(6)=181.7, p<0.001. The
change in crude rate between 2007 and 2014 was statistically significant in post-hoc
pairwise testing with correction for multiple comparisons: χ2(6)=103.9, p<0.001.
Table A7. Crude mortality rates over time for overall PCI cohort.
2007
2008
2009
2010
2011
2012
Dead
1.4
1.5
1.8
1.9
2
2.2
Alive
98.6
98.5
98.2
98.1
98
97.8
2013
2.2
97.8
Crude mortality rates in the primary PCI sample was 5% in both 2007 and 2014 (see Table
A8). There was no significant difference between years: χ2(6)=7.4, p=0.284.
Table A8. Crude mortality rates over time for primary PCI cohort.
2007
2008
2009
2010
2011
2012
Dead
5
4.9
5.1
4.7
4.5
4.8
Alive
95
95.1
94.9
95.3
95.5
95.2
2013
5
95
A17
Changes in Volume Over Time
Comparing observed annual volumes over the study period, there was a steady increase in
the proportion of hospitals whose volume fell within the 200-399 and 400-749 bands. By the
end of the study period, twice as many patients were being treated in units of this size as at
the beginning (9,398 in 2007 vs. 19,379 in 2013), while a reverse of this trend was evident in
the lowest volume centers. This reflected new PCI centers emerging as well as volume
growth within existing units. A complete summary of the changes in annual volume on a year
by year basis is provided for the overall PCI cohort in Table A9, along with significance
statistics in Table A10. Similar results are presented for the primary cohort in Tables A11
and A12.
Table A9. Annual Center Volumes over Time for Overall PCI Cohort Categorized by Mean
Annual Overall PCI Volumes.
Year:
2007†
2008†
2009†
2010†
2011†
2012†
2013†
0-199
MEAN ANNUAL OVERALL PCI VOLUME CATEGORIES
200-399
400-749
750-1499
1500-1999
≥2000
457 (17·7)
345 (13·3)
280 (10·8)
429 (16·6)
466 (18)
362 (14)
249 (9·6)
2325 (6·3)
3700 (10)
4663 (12·6)
5188 (14)
6301 (17)
7391 (19·9)
7565 (20·4)
7073 (10·5)
8036 (11·9)
8937 (13·2)
9526 (14·1)
10555 (15·6)
11569 (17·1)
11814 (17·5)
21793 (14·7)
20844 (14)
20952 (14·1)
20985 (14·1)
21328 (14·3)
21940 (14·7)
20909 (14·1)
11816 (15·5)
11429 (15)
10788 (14·1)
10827 (14·2)
10797 (14·1)
10754 (14·1)
9959 (13)
12861 (13·5)
13855 (14·6)
14144 (14·9)
14204 (14·9)
13535 (14·2)
13615 (14·3)
12901 (13·6)
Total
56325 (13·2)
58209 (13·6)
59764 (14)
61159 (14·3)
62982 (14·7)
65631 (15·4)
63397 (14·8)
* = Significant between-volume category difference at p < 0·05,
† = Significant between-volume category difference at p < 0·001
Table A10. Significance Testing for Changes in Annual Volume for Overall PCI Cohort
Categorized by Mean Annual Overall PCI Volumes
Year
Year
2007
2008
2009
2010
2011
2012
2013
χ2
2671·87
800·65
180·31
52·74
272·66
976·47
1724·51
df
5
5
5
5
5
5
5
p
<0·001
<0·001
<0·001
<0·001
<0·001
<0·001
<0·001
A18
Table A11. Annual Center Volumes over Time for Primary PCI Cohort Categorized by Mean
Annual Overall PCI Volumes.
0-199
Year:
2007†
2008†
2009†
2010†
2011†
2012†
2013†
21 (30·4)
14 (20·3)
1 (1·4)
4 (5·8)
10 (14·5)
13 (18·8)
6 (8·7)
MEAN ANNUAL OVERALL PCI VOLUME CATEGORIES
200-399
400-749
750-1499
1500-1999 ≥2000
91 (3)
209 (6·8)
398 (12·9)
524 (17)
512 (16·7)
646 (21)
695 (22·6)
630 (5·3)
657 (5·5)
1100 (9·2)
1714 (14·3)
2329 (19·4)
2633 (22)
2930 (24·4)
2189 (5·9)
2900 (7·8)
4087 (11)
5363 (14·5)
7044 (19)
7999 (21·6)
7476 (20·2)
743 (4)
1484 (8·1)
2282 (12·4)
2908 (15·8)
3591 (19·5)
3836 (20·9)
3542 (19·3)
1115 (4·8)
2275 (9·7)
3268 (13·9)
4090 (17·4)
4072 (17·4)
4425 (18·9)
4196 (17·9)
Total
4789 (5·1)
7539 (8)
11136 (11·8)
14603 (15·5)
17558 (18·7)
19552 (20·8)
18845 (20)
* = Significant between-volume category difference at p < 0·05,
† = Significant between-volume category difference at p < 0·001
Table A12. Significance Testing for Changes in Annual Volume for Primary PCI Cohort
Categorized by Mean Annual Overall PCI Volumes
Year
Year
2007
2008
2009
2010
2011
2012
2013
χ2
219·62
217·61
220·73
122·86
51·29
76·48
236·69
df
5
5
5
5
5
5
5
p
<0·001
<0·001
<0·001
<0·001
<0·001
<0·001
<0·001
A19
Hierarchical Models: Tabulated Results
The following tables report the numeric results of the hierarchical modelling. Table A13
summarizes the baseline model for the overall sample and the model selection criteria for
this model are subsequently provided in Table A14. Equivalent results for the primary PCI
only sample are provided in Tables A15 and A16.
Table A13. Multilevel Model Coefficient Table for Overall Sample
Coef.
SE
Z
p
95% CI
Lower
Upper
Year of operation
2008
-0·04
0·05
-0·85 0·397
-0·15
0·06
(2007 as reference 2009
-0·05
0·05
-1·03 0·302
-0·15
0·05
category)
2010
-0·11
0·05
-2·27 0·023
-0·21
-0·02
2011
-0·15
0·05
-3·10 0·002
-0·25
-0·06
2012
-0·08
0·05
-1·57 0·117
-0·17
0·02
2013
-0·13
0·05
-2·69 0·007
-0·22
-0·04
Constant
0·02
0·05
0·34 0·731
-0·08
0·11
BCIS 30 day mortality risk
1·00
score (offset)
Hospital-level variance
0·05
Table A14. Model Selection Criteria for Overall Sample
Log likelihood AIC
BIC
-30511·8 61039·6 61127·3
Table A15. Multilevel Model Coefficient Table for Primary PCI Sample
Coef.
SE
z
p
95% CI
Lower
Year of operation 2008
(2007 as reference 2009
category)
2010
2011
2012
2013
Constant
BCIS 30 day mortality risk
score (offset)
Hospital-level variance
-0·002
0·02
-0·10
-0·17
-0·07
-0·12
0·04
1·00
0·09
0·09
0·08
0·08
0·08
0·08
0·08
-0·02
0·27
-1·15
-2·01
-0·91
-1·42
0·58
0·981
0·790
0·250
0·044
0·361
0·154
0·562
-0·18
-0·14
-0·26
-0·33
-0·23
-0·27
-0·10
Upper
0·18
0·19
0·07
0·00
0·08
0·04
0·19
0·03
Table A16. Model Selection Criteria for Primary PCI Sample
Log likelihood AIC
BIC
-14502·3 29020·7 29096·3
A20
Additional Bayes Factor Interpretation for Primary PCI Sample:
In the primary PCI subset, the Bayes factor for the dose-response relationship was 1.4. This
means that the ratio of the chance of a volume-outcome relationship for primary PCI
compared to the chance of no volume outcome relationship is increased by a factor of 1.4.
Taking the before study chances of no relationship to be 0·50, our center level effects revise
the chance up to 0.74. This provides support for the hypothesis that there is no volume
outcome relationship in the primary PCI cohort. For the band-specific analysis in this cohort,
the chance is revised down to 0.41 when center-level effects are taken in account, i.e. there
again remains support for the hypothesis that there is no band-specific effect for this, or any
other, band.
Sensitivity Analyses
Analysis by Quintile
The volume-outcome relationship was also analyzed using data driven quintiles. This
enabled an examination of whether the robustness of our results in our main analyses where
clinically defined volume bands were employed. As can be seen from the plots below, a very
similar result was obtained when using the quintile-based volume band definitions, for both
the overall PCI cohort (Figure A5) and the primary PCI subset (Figure A6).
A21
Figure A5. Observed and Predicted Mortality and Relative Risk of Mortality by Mean Annual
Center Volume Quintile: All PCI Cases
Scatterpoints: Hospital-level estimates (with 95% CI bars for relative risk)
Colored panes: Linear estimates of mortality risk for each volume stratum,
with 95% CI
A22
Figure A6. Observed and Predicted Mortality and Relative Risk of Mortality by Mean Annual
Center Volume Quintile: Primary PCI Cases
Scatterpoints: Hospital-level estimates (with 95% CI bars for relative risk)
Colored panes: Linear estimates of mortality risk for each volume quintile,
with 95% CI
A23
Exclusion of Records with Highest 10% Predicted Mortality Risk
We repeated the mixed-effects modelling having excluded the records with a predicted
mortality risk within the uppermost risk decile (Figures A7 and A8). This was in response to
the calibration statistics and plots which indicated that some over-prediction of mortality
occurred for at the higher end of the risk spectrum, corresponding to approximately 10% of
records included in the overall PCI analysis.
Figure A7. Overall PCI Sample Stratified Model (Excluding Top 10% Risk)
A24
Figure A8. Primary PCI Sample Stratified Model (Excluding Top 10% Risk)
A25
Exclusion of Records with Missing Cardiogenic Shock Data
A further sensitivity analysis was performed to test the impact of missing data in the most
highly weighted covariate in the 30-day mortality risk adjustment model, cardiogenic shock
(Figures A9 and A10).
Figure A9. Overall PCI Sample Stratified Model (Excluding Records Missing Cardiogenic
Shock)
A26
Figure A10. Primary PCI Sample Stratified Model (Excluding Records Missing Cardiogenic
Shock)
A27
Analysis of Elective Cases Only
A further sensitivity analysis was performed to test if a focus on a more standardized set of
non-primary PCI patients, specifically those undergoing elective procedures only, produced
comparable results to the overall sample analysis. The results are presented in Figure A11.
Figure A11. Overall PCI Sample Stratified Model (Elective Cases Only)
A28
Volume of Primary PCI Cases
Additional examination was undertaken to see if the volume of Primary PCI cases, as
opposed to the overall PCI volume, had a relationship with mortality outcomes. Primary PCI
volume was explored firstly using an a priori categorization of hospital’s mean primary PCI
volumes (Figure A12). This included a threshold at 36 cases per year, to reflect the minimum
institutional primary PCI volumes advocated in the most recent guidelines from
ACCF/AHA/SCAI. The volume band just below this cutoff shows a lower incidence of
mortality than the other volume bands, but its confidence limits overlap with neighboring
volume bands, indicating that this difference is not statistically meaningful. The profile on the
whole shows little difference in mortality risk across the volume bands. A similar picture is
evident when the volume is categorized according to data-derived quintiles (Figure A13).
Figure A12. Mortality Relative Risk by Mean Annual Primary PCI Center Volume (Primary
PCI Cases)
A29
Figure A13. Mortality Relative Risk by Mean Annual Primary PCI Center Volume Quintile
(Primary PCI Cases)
A30
References
1
Spiegelhalter DJ, Abrams KR, Myles JP. Bayesian approaches to clinical trials and
health-care evaluation. 1st ed. Chichester, UK: Wiley; 2004.
2
Post PN, Kuijpers M, Ebels T, Zijlstra F. The relation between volume and outcome
of coronary interventions: A systematic review and meta-analysis. Eur Heart J. 2010; 31,
1985–92. doi: 10.1093/eurheartj/ehq151
3
Navarese EP, De Servi S, Politi A, Martinoni A, Musumeci G, Boschetti E, Belli G,
D'Urbano M, Piccaluga E, Lettieri C, Klugmann S. Impact of primary PCI volume on hospital
mortality in STEMI patients: Does time-to-presentation matter? J Thromb Thrombolysis.
2011; 32: 223–31. doi: 10.1007/s11239-011-0598-1
4
Kuwabara, H, Fushimi K, Matsuda, S. Relationship between hospital volume and
outcomes following primary percutaneous coronary intervention in patients with acute
myocardial infarction. Circ J. 2011; 75: 1107–12. doi: 10.1253/circj.CJ-10-0556
5
Kontos MC, Wang Y, Chaudhry SI, Vetrovec GW, Curtis J, Messenger J. Lower
hospital volume is associated with higher in-hospital mortality in patients undergoing primary
percutaneous coronary intervention for ST-segment–elevation myocardial infarction: A report
from the NCDR. Circ Cardiovasc Qual Outcomes. 2013; 6: 659–67. doi:
10.1161/circoutcomes.113.000233
6
Akin I, Hochadel M, Schneider S, Abdel-Wahab M, Zahn R, Senges J, Richardt G,
Kuck KH, Nienaber CA . Volume-outcomes relationship in the era of modern coronary
intervention: Results from the prospective multicentre German DES.DE registry. Catheter
Cardiovasc Interv. 2013; 82: E788–97. doi: 10.1002/ccd.24986
7
Badheka AO, Patel NJ, Grover P, Singh V, Patel N, Arora S, Chothani A, Mehta K,
Deshmukh A, Savani GT, Patel A, Panaich SS, Shah N, Rathod A, Brown M, Mohamad T,
Tamburrino FV, Kar S, Makkar R, O'Neill WW, De Marchena E, Schreiber T, Grines CL,
Rihal CS, Cohen MG. Impact of annual operator and institutional volume on percutaneous
coronary intervention outcomes: A 5-year United States experience (2005-2009). Circulation.
2014; 130: 1392–406. doi: 10.1161/circulationaha.114.009281
8
16.
Kumbhani DJ, Cannon CP, Fonarow GC, Liang L, Askari AT, Peacock WF,
Peterson ED, Bhatt DL. Association of hospital primary angioplasty volume in ST-segment
elevation myocardial infarction with quality and outcomes. JAMA. 2009; 302(20): 2207–13.
doi: 10.1001/jama.2009.1715
9
Spaulding C, Morice MC, Lancelin B, El Haddad S, Lepage E, Bataille S, Tresca JP,
Mouranche X, Fosse S, Monchi M, de Vernejoul N. Is the volume-outcome relation still an
issue in the era of PCI with systematic stenting? Results of the greater Paris area registry.
Eur Heart J. 2006; 27(9): 1054–60. doi: 10.1093/eurheartj/ehi843
A31