Upload
others
View
4
Download
0
Embed Size (px)
Citation preview
The Pennsylvania State University
College of Medicine
Department of Public Health Sciences
IMPACT OF SURGEON AND HOSPITAL VOLUME ON MORTALITY, LENGTH OF
STAY, AND COST OF PANCREATICODUODENECTOMY
A Thesis in
Public Health Sciences
by
Laura M. Enomoto
© 2014 Laura M. Enomoto
Submitted in Partial Fulfillment
of the Requirements
for the Degree of
Master of Science
May 2014
ii
The thesis of Laura M. Enomoto was reviewed and approved* by the following:
Christopher S. Hollenbeak
Professor of Surgery and Public Health Sciences
Thesis Adviser
Niraj J. Gusani
Associate Professor of Surgery, Medicine, and Public Health Sciences
Douglas L. Leslie
Professor of Psychiatry and Public Health Sciences
Director, Master of Science in Public Health Sciences Program
* Signatures are on file in the Graduate School
iii
ABSTRACT
Improved mortality rates following pancreaticoduodenectomy by high volume surgeons and
hospitals have been well documented, but less is known about the impact of such volumes on
length of stay and cost. This study uses data from the Healthcare Cost and Utilization Project
(HCUP) National Inpatient Sample (NIS) to examine the effect of surgeon and hospital volume
on mortality, length of stay, and cost following pancreaticoduodenectomy while controlling for
patient specific factors. Data included 3,137 pancreaticoduodenectomies from the NIS
performed between 2004-2008. Using logistic and generalized linear regression models, the
relationship between surgeon volume, hospital volume, and post-operative mortality, length of
stay, and cost was estimated while accounting for patient factors. Propensity score matching was
used to address potential covariate imbalance. After controlling for patient characteristics,
patients of high volume surgeons at high volume hospitals had a significantly lower risk of
mortality compared to any other group (2.7% vs. 6.0%, p = 0.016). Patients of high volume
surgeons at high volume hospitals also had a five day shorter length of stay (p < 0.001), as well
as significantly lower costs ($12,159, p < 0.001). The results of this study, which simultaneously
accounted for surgeon volume, hospital volume, and potential confounding patient
characteristics, suggest that both surgeon and hospital volume have a significant effect on
outcomes following pancreaticoduodenectomy, affecting not only mortality rates but also lengths
of stay and costs.
iv
TABLE OF CONTENTS
List of Tables ...................................................................................................................................v
List of Figures ................................................................................................................................ vi
Acknowledgements ....................................................................................................................... vii
Introduction ......................................................................................................................................1
Methods............................................................................................................................................2
Data ............................................................................................................................................2
Statistical Analysis .....................................................................................................................5
Results ..............................................................................................................................................7
Patient Characteristics ................................................................................................................7
Mortality ....................................................................................................................................9
Length of Stay ..........................................................................................................................12
Hospital Costs ..........................................................................................................................15
Propensity Score Matching ......................................................................................................18
Discussion ......................................................................................................................................20
References ......................................................................................................................................24
v
LIST OF TABLES
Table 1. Summary statistics of patients undergoing pancreaticoduodenectomy
stratified by surgeon and hospital volume (Low volume, LV; high volume, HV) ..........................8
Table 2. Results of logistic regression model of effect of surgeon and hospital
volume on mortality, controlling for other covariates ...................................................................11
Table 3. Results of generalized linear model of effect of surgeon and hospital
volume on length of stay, controlling for other covariates ............................................................14
Table 4. Results of generalized linear model of effect of surgeon and hospital
volume on cost, controlling for other covariates ...........................................................................17
vi
LIST OF FIGURES
Figure 1. Determination of study cohort ..........................................................................................3
vii
ACKNOWLEDGEMENTS
This work has previously been published in edited form (Enomoto LM, Gusani NJ, Dillon PW,
Hollenbeak CS. Impact of Surgeon and Hospital Volume on Morality, Length of Stay, and Cost
of Pancreaticoduodenectomy. Journal of Gastrointestinal Surgery: Official Journal of the
Society for Surgery of the Alimentary Tract. 2014 April; 18(4): 690-700.). The final publication
is available at http://dx.doi.org/10.1007/s11605-013-2422-z
1
INTRODUCTION
Population studies have demonstrated an association between procedure volume and
patient outcomes for complex surgical procedures including pancreaticoduodenectomy.1-3
When
compared with high volume (HV) hospitals, data have shown that low volume (LV) hospitals
have poorer operative outcomes and lower long-term survival.4,5
Previous studies have also
suggested that surgeon volume is related to post-operative mortality following
pancreaticoduodenectomy with HV surgeons demonstrating lower in-hospital mortality.6 As a
result of these studies, there is a growing trend towards referral of patients to HV centers for their
care.7
Patient-related factors also contribute to poor post-operative outcomes. Recent studies
have identified racial minorities and the underinsured as less likely to receive their
pancreaticoduodenectomy at HV institutions, despite efforts to encourage centralization.8,9
Notably, even with the trend toward centralized delivery of complex operations, patients with
more comorbid disease may have worse surgical outcomes, independent of the expertise of the
surgeon.10
Post-operative outcomes arise from a complex interplay between surgeon expertise,
hospital volume, and patient factors.
While the relationship between volume and mortality has been extensively studied, the
relationship between volume and other outcomes, such as length of stay (LOS) and costs, has not
been as well documented. Therefore, the purpose of this study was to examine the effects of
surgeon and hospital volume on outcomes for pancreaticoduodenectomy using data from the
Healthcare Cost and Utilization Project (HCUP) National Inpatient Sample (NIS).
2
METHODS
Data
This was a retrospective cohort study that included patients in the NIS data set from 2004
to 2008 who underwent pancreaticoduodenectomy. The NIS is the largest publically available,
all-payer inpatient care database in the United States, constituting 20% of hospital discharges,
and including both academic and specialty hospitals.11
There were 3,290 patients in the NIS
with a primary International Classification of Diseases, 9th
Revision, Clinical Modification (ICD-
9-CM) procedure code for pancreaticoduodenectomy (52.7). After excluding pediatric and
trauma patients as well as 140 patients for missing covariates (Figure 1), 3,137 patients were
included in the final analysis. Due to the large number of patients with a missing race or
admission source variable (N = 559 and N = 507, respectively), the analysis was performed with
missing race and admission source as additional covariates rather than excluding those patients
altogether from the analysis.
3
Figure 1: Determination of study cohort.
4
To evaluate volume, the cohort was categorized into groups for purposes of comparison
before any volume-outcome data analysis was performed. Surgeons were classified as either HV
if they performed ≥5 pancreaticoduodenectomies in a given year or LV if they performed <5
pancreaticoduodenectomies in a given year as previously described.1,6
Using the Leapfrog cut-
off for high-risk surgical procedures, hospitals were defined as HV if a center performed ≥ 11
pancreaticoduodenectomies per year and LV if they performed <11
pancreaticoduodenectomies.12
Analyses controlled for several patient level variables, including demographics (age,
gender, and race/ethnicity), diagnosis of cancer, co-morbidities (history of acute myocardial
infarction, congestive heart failure, peripheral vascular disease, cerebrovascular disease, chronic
obstructive pulmonary disease, peptic ulcer disease, mild liver disease, moderate or severe liver
disease, diabetes with and without complications, and renal disease), primary payer (Medicare,
Medicaid, commercial, and other), and type of admission (emergent, urgent, and elective). The
analyses also controlled for region of the country (northeast, midwest, south, and west), teaching
hospital status, and year of operation (2004 through 2008). Diagnosis of cancer was determined
using ICD-9-CM diagnosis codes (151.0 – 197.8, 200.00 – 209.30, 230.7 – 258.0). Co-
morbidities were assigned using ICD-9-CM codes as previously described.13
Most hospitals
appeared in the data set in one (64.3%) or two (26.6%) years.
Three outcomes were studied: mortality, LOS, and cost. Mortality was assessed as in-
hospital mortality. LOS was measured as the total length of admission to the hospital where the
pancreaticoduodenectomy was performed. Cost analysis was performed from the perspective of
the hospital and costs were estimated using a cost to charges ratio (CCR) methodology. This is a
standard accounting technique where costs are estimated as a percentage of hospital charges. In
5
this methodology, individual hospital departments calculate a yearly CCR. The next year,
departmental charges are multiplied by the previous year’s CCR and used as an estimate of costs.
The total cost for an admission is the sum of all departmental costs. Although this only provides
an estimate of costs, previous research has shown that CCR’s are an acceptable approach for
estimating costs across hospitals for patients with a common diagnosis-related group.14
Costs
were inflation adjusted to 2008 US dollar values using the medical care component of the
consumer price index.
Statistical Analysis
Statistical analysis was performed primarily to determine whether surgeon and hospital
volume were significantly associated with outcomes after controlling for important covariates.
Univariate analysis using 2 tests for binary and categorical variables was performed to
determine whether there were differences in patient characteristics across surgeon and hospital
volume groups. Although weights are available to produce nationally representative descriptive
statistics, our interest was not in producing nationally representative estimates. Therefore, in our
descriptive statistics and in our multivariate models described subsequently, we did not apply
national or regional weights.
Logistic regression was used to model the effects of surgeon and hospital volume on in-
hospital mortality after controlling for patient, procedure, and provider characteristics. LOS and
costs were fit to generalized linear regression models assuming a gamma family of distributions
and log link function. These models were chosen because LOS and cost data were highly skewed
and did not meet the normality assumption of the classical linear model. We report the marginal
6
effects from the generalized linear models, which show the effect of a one-unit change in the
independent variable on the outcome.
If a significant imbalance in patient covariates existed between different surgeon and
hospital volumes, then a regression model may not adequately control for covariates. Therefore,
a propensity score matching analysis that dealt with potential covariate imbalance was
performed. The propensity score models for mortality, LOS, and cost were estimated using
logistic regression with HV surgeons at HV hospitals as the dependent variable and covariates as
previously described. Using the estimated propensity scores from this model, patients in the HV
surgeon, HV hospital group were matched 1:1 to patients in the other groups using a k-nearest
neighbor match with a max-min common support restriction.
The primary metric for the propensity score analysis was the average effect of treatment
on the treated (ATT). For all three outcomes, this is the difference between the outcome for the
patient treated by a HV surgeon in a HV hospital and the outcome for the patients in the three
remaining groups. To deal with the uncertainty induced by both the selection process and the
data, a standard bootstrapping algorithm was used to compute 95% confidence intervals. All
reported inferences for the ATT were based on 1000 bootstrap replicates. All statistical analyses
were performed using STATA (version 12.1, StataCorp LLP, College Station, TX) and the
psmatch2 routines.15
Statistical significance for all analyses was defined as a p-value < 0.05.
7
RESULTS
Patient Characteristics
There were 3,137 pancreaticoduodenectomies recorded in the NIS database between
2004 and 2008 after excluding 153 patients. Data from all participating institutions in the
database were included. Of these procedures, 866 were performed by LV surgeons at LV
hospitals, 451 were performed by LV surgeons at HV hospitals, 179 were performed by HV
surgeons at LV hospitals, and 1,641 were performed by HV surgeons at HV hospitals.
Demographic characteristics of patients who underwent pancreaticoduodenectomy stratified by
surgeon and hospital volume and are shown in Table 1. Compared with other groups, patients
who underwent pancreaticoduodenectomy by HV surgeons at HV hospitals were more likely to
be diagnosed with a disease other than cancer (p = 0.036), elective admissions (p < 0.001), and
treated in a teaching hospital (p < 0.001).
8
Table 1: Summary statistics of patients undergoing pancreaticoduodenectomy stratified by
surgeon and hospital volume. (Low volume, LV; high volume, HV)
LV Surgeon HV Surgeon
LV Hospital HV Hospital LV Hospital HV Hospital
Variable (N=866) (N=451) (N=179) (N=1,641) P-Value
Age
0.374
18-55 26.0% 22.4% 17.9% 24.0%
56-65 25.8% 26.6% 31.3% 27.0%
66-75 29.9% 28.6% 31.3% 30.0%
76+ 18.4% 22.4% 19.6% 19.0%
Female 48.6% 46.3% 45.8% 50.9% 0.226
Race/Ethnicity
< 0.001
White 55.8% 62.3% 68.7% 66.0%
Black 10.0% 8.2% 14.5% 5.0%
Hispanic 7.4% 10.9% 1.1% 6.9%
Other 5.7% 6.9% 2.8% 3.8%
Missing 21.1% 11.8% 12.8% 18.3%
Any Diagnosis of Cancer 86.7% 86.0% 85.5% 82.6% 0.036
Co-morbidities
Acute Myocardial Infarction 4.0% 4.9% 5.6% 4.3% 0.764
Congestive Heart Failure 3.7% 3.8% 4.5% 2.4% 0.159
Peripheral Vascular Disease 2.4% 0.9% 2.2% 1.7% 0.238
Cerebrovascular Disease 1.8% 0.9% 0.6% 1.2% 0.300
COPD 16.2% 10.0% 10.6% 11.3% 0.001
Peptic Ulcer Disease 4.5% 3.3% 1.1% 1.6% < 0.001
Mild Liver Disease 1.5% 1.8% 0.6% 1.5% 0.726
Moderate/Severe Liver Disease 0.8% 1.3% 1.1% 0.6% 0.459
Diabetes 23.9% 24.2% 27.4% 19.4% 0.007
Diabetes with Complications 1.6% 0.9% 0.0% 1.2% 0.289
Renal Disease 1.8% 2.9% 0.6% 0.7% 0.002
Payer
0.005
Medicare 47.9% 51.9% 48.0% 49.5%
Medicaid 7.9% 7.3% 3.4% 5.7%
Commercial 37.6% 34.1% 46.4% 40.0%
Other 6.6% 6.7% 2.2% 4.8%
9
Table 1 (continued): Summary statistics of patients undergoing pancreaticoduodenectomy.
LV Surgeon HV Surgeon
LV Hospital HV Hospital LV Hospital HV Hospital
Variable (N=866) (N=451) (N=179) (N=1,641) P-Value
Admission Source
< 0.001
ER 18.9% 13.3% 3.9% 2.8%
Other hospital 1.8% 5.1% 2.2% 2.4%
Long term care 0.7% 0.4% 0.0% 0.5%
Other 68.2% 65.2% 74.3% 75.3%
Missing 10.3% 16.0% 19.6% 19.0%
Admission Type
< 0.001
Elective 67.3% 66.5% 86.0% 87.6%
Urgent 10.4% 14.2% 9.5% 7.5%
Emergent 22.3% 19.3% 4.5% 4.9%
Teaching hospital 56.7% 91.4% 73.7% 93.4% < 0.001
Region
< 0.001
Northeast 29.0% 28.8% 24.6% 29.5%
Midwest 8.0% 20.6% 17.9% 9.6%
South 42.7% 27.5% 38.0% 25.9%
West 20.3% 23.1% 19.6% 35.0%
Year
< 0.001
2004 21.6% 29.9% 25.1% 18.3%
2005 17.6% 19.1% 10.6% 18.3%
2006 21.7% 12.9% 24.0% 15.1%
2007 18.4% 17.3% 20.7% 24.3%
2008 20.8% 20.8% 19.6% 24.1%
Mortality
Table 2 provides the results of the logistic regression model of in-hospital mortality
controlling for covariates. There was a significant difference in mortality between LV surgeons
at LV hospitals and LV surgeons at HV hospitals, as well as HV surgeons at HV hospitals (OR
0.54, p = 0.018; and OR 0.32, p < 0.001; respectively). Additionally, several other patient
characteristics were associated with increased mortality. Patients who were 76 years of age or
older had an almost three times higher odds of mortality compared to those of younger age (OR
10
2.76, p = 0.003). Several comorbidities were significantly associated with greater risk of
mortality, including a history of acute myocardial infarction, congestive heart failure, and
cerebrovascular disease. Diabetes and PVD were significantly associated with a lower risk of
mortality. Urgent and emergent admissions had significantly higher odds of mortality compared
to patients who were admitted electively (OR 1.77, p = 0.033; and OR 2.07, p = 0.04;
respectively). Other covariates were not significantly associated with risk of mortality.
11
Table 2: Results of logistic regression model of effect of surgeon and hospital volume on
mortality, controlling for other covariates.
95% Confidence Interval
Variable Odds Ratio Lower Upper P-value
Surgeon and Hospital Volume
Low/Low Reference
Low/High 0.54 0.33 0.90 0.018
High/Low 0.56 0.27 1.19 0.130
High/High 0.32 0.20 0.49 < 0.001
Age
18-55 Reference
56-65 1.07 0.60 1.89 0.825
66-75 1.24 0.64 2.41 0.524
76+ 2.76 1.41 5.37 0.003
Female 1.01 0.73 1.40 0.958
Race/Ethnicity
White Reference
Black 1.39 0.79 2.45 0.250
Hispanic 1.26 0.69 2.30 0.444
Other 0.52 0.20 1.38 0.187
Missing 0.76 0.45 1.28 0.300
Any Diagnosis of Cancer 1.68 0.95 2.98 0.074
Co-morbidities
Acute Myocardial Infarction 2.35 1.29 4.26 0.005
Congestive Heart Failure 3.35 1.84 6.08 < 0.001
Peripheral Vascular Disease 0.12 0.02 0.97 0.047
Cerebrovascular Disease 3.89 1.58 9.62 0.003
COPD 1.21 0.77 1.91 0.401
Peptic Ulcer Disease 1.73 0.82 3.68 0.152
Mild Liver Disease 2.40 0.95 6.11 0.066
Moderate/Severe Liver Disease 2.88 0.76 10.90 0.120
Diabetes 0.57 0.37 0.88 0.012
Diabetes with Complications 0.98 0.26 3.67 0.977
Renal Disease 1.79 0.65 4.96 0.262
Payer
Medicare Reference
Medicaid 1.28 0.60 2.75 0.525
Commercial 0.65 0.38 1.11 0.118
Other 1.11 0.49 2.48 0.808
12
Table 2 (continued): Results of logistic regression model of effect of surgeon and hospital
volume on mortality, controlling for other covariates.
95% Confidence Interval
Variable Odds Ratio Lower Upper P-value
Admission Source
ER Reference
Other hospital 1.41 0.50 3.97 0.517
Long term care 0.99 0.11 8.96 0.995
Other 1.04 0.50 2.17 0.908
Missing 1.04 0.45 2.43 0.924
Admission Type
Elective Reference
Urgent 1.77 1.05 2.98 0.033
Emergent 2.07 1.04 4.12 0.040
Teaching hospital 0.71 0.48 1.06 0.096
Region
Northeast Reference
Midwest 1.00 0.53 1.88 0.993
South 0.97 0.63 1.48 0.880
West 0.88 0.53 1.47 0.633
Year 0.95 0.83 1.09 0.459
Length of Stay
The results of a generalized linear regression model of LOS controlling for covariates are
shown in Table 3. Patients treated by LV surgeons at HV hospitals, HV surgeons at LV
hospitals, and HV surgeons at HV hospitals had a significantly shorter LOS relative to patients
treated by LV surgeons at LV hospitals (-2.44 days, p < 0.001; -2.97 days, p < 0.001; and -5.65
days, p < 0.001; respectively). A significantly increased LOS was found in patients between 66
and 75 years of age and over 76 years of age compared to those of younger age (1.54 days, p =
0.032; and 2.85 days, p = 0.001; respectively). Race/ethnicity was significantly associated with
LOS, with Hispanic and other race/ethnicity groups having significantly longer stays relative to
13
patients of white race (2.53 days, p = 0.004; and 4.25 days, p < 0.001; respectively). Patients
with congestive heart failure had significantly increased LOS (2.93 days, p = 0.024) while
patients with diabetes had significantly decreased stays compared to those without those co-
morbidities (-1.67 days, p < 0.001). Urgent and emergent admissions were associated with a
longer LOS relative to elective admissions (4.91 days, p < 0.001; and 7.47 days, p < 0.001;
respectively). Patients treated at teaching hospitals had slightly longer stays than patients treated
at non-teaching hospitals (1.19 days, p = 0.024). Gender, cancer diagnosis, payer, admission
source, and region of the country were not significantly associated with LOS.
14
Table 3: Results of generalized linear model of effect of surgeon and hospital volume on length
of stay, controlling for other covariates.
Marginal 95% Confidence Interval
Variable Effect Lower Upper P-value
Surgeon and Hospital Volume
Low/Low Reference
Low/High -2.44 -3.59 -1.30 < 0.001
High/Low -2.97 -4.42 -1.51 < 0.001
High/High -5.65 -6.70 -4.60 < 0.001
Age
18-55 Reference
56-65 0.93 -0.17 2.04 0.098
66-75 1.54 0.13 2.95 0.032
76+ 2.85 1.21 4.50 0.001
Female -0.38 -1.12 0.36 0.313
Race/Ethnicity
White Reference
Black 0.86 -0.69 2.41 0.278
Hispanic 2.53 0.81 4.24 0.004
Other 4.25 2.00 6.49 < 0.001
Missing 0.09 -0.97 1.14 0.873
Any Diagnosis of Cancer -0.47 -1.54 0.60 0.387
Co-morbidities
Acute Myocardial Infarction 0.70 -1.20 2.60 0.469
Congestive Heart Failure 2.93 0.38 5.47 0.024
Peripheral Vascular Disease -0.94 -3.59 1.70 0.486
Cerebrovascular Disease 2.66 -1.20 6.52 0.177
COPD 0.43 -0.74 1.60 0.469
Peptic Ulcer Disease 1.89 -0.71 4.49 0.154
Mild Liver Disease 2.83 -0.75 6.42 0.122
Moderate/Severe Liver Disease 0.27 -3.97 4.50 0.902
Diabetes -1.67 -2.53 -0.82 < 0.001
Diabetes with Complications -1.56 -4.67 1.56 0.327
Renal Disease 0.84 -2.63 4.31 0.635
Payer
Medicare Reference
Medicaid -0.92 -2.64 0.81 0.297
Commercial -0.73 -1.84 0.38 0.195
Other -0.49 -2.32 1.34 0.600
15
Table 3 (continued): Results of generalized linear model of effect of surgeon and hospital
volume on length of stay, controlling for other covariates.
Marginal 95% Confidence Interval
Variable Effect Lower Upper P-value
Admission Source
ER Reference
Other hospital -0.40 -3.32 2.53 0.791
Long term care 0.94 -4.92 6.80 0.753
Other -1.74 -4.03 0.56 0.139
Missing -0.89 -3.15 1.36 0.438
Admission Type
Elective Reference
Urgent 4.91 3.18 6.64 < 0.001
Emergent 7.47 4.77 10.18 < 0.001
Teaching hospital 1.19 0.16 2.22 0.024
Region
Northeast Reference
Midwest -1.16 -2.46 0.14 0.080
South -0.61 -1.60 0.37 0.222
West 0.81 -0.27 1.89 0.141
Year -0.54 -0.86 -0.23 0.001
Hospital costs
Results of a generalized linear regression model of cost are presented in Table 4. Patients
treated by LV surgeons at LV hospitals showed a significant increase in cost compared to
patients treated by LV surgeons at HV hospitals, HV surgeons at LV hospitals, and HV surgeons
at HV hospitals (-$4,800, p = 0.009; -$9,087, p < 0.001; and -$12,275, p < 0.001; respectively).
A significantly increased cost was associated with patients over 76 years of age relative to those
of younger age ($5,630, p = 0.023). Women had a significantly lower cost compared to men (-
$3,406, p = 0.003). Similar to the trends observed with LOS, Hispanic and other race/ethnicity
groups had significantly higher costs of $8,853 and $16,100, respectively compared to white race
16
(p = 0.001 and p < 0.001, respectively). Patients with a history of acute myocardial infarction
and congestive heart failure had increased costs ($7,650, p = 0.021; and $11,346, p = 0.007;
respectively) while diabetic patients had lower costs relative to patients without those co-
morbidities (-$5,870, p < 0.001). Urgent and emergent cases had higher costs than elective cases
($8,421, p = 0.001; and $18,223, p < 0.001; respectively), as did teaching hospitals and hospitals
in the Midwest and South. Diagnosis of cancer, payer, admission source, and year of operation
were not significant predictors of costs.
17
Table 4: Results of generalized linear model of effect of surgeon and hospital volume on cost,
controlling for other covariates.
Marginal 95% Confidence Interval
Variable Effect Lower Upper P-value
Surgeon and Hospital Volume
Low/Low Reference
Low/High -4,800 -8,382 -1,217 0.009
High/Low -9,087 -13,355 -4,820 < 0.001
High/High -12,275 -15,444 -9,107 < 0.001
Age
18-55 Reference
56-65 1,780 -1,550 5,110 0.295
66-75 3,112 -1,131 7,354 0.151
76+ 5,630 774 10,486 0.023
Female -3,406 -5,656 -1,156 0.003
Race/Ethnicity
White Reference
Black 3,062 -1,729 7,852 0.210
Hispanic 8,853 3,429 14,276 0.001
Other 16,100 8,667 23,534 < 0.001
Missing 4,057 624 7,490 0.021
Any Diagnosis of Cancer -1,192 -4,451 2,067 0.473
Co-morbidities
Acute Myocardial Infarction 7,650 1,162 14,138 0.021
Congestive Heart Failure 11,346 3,055 19,636 0.007
Peripheral Vascular Disease -163 -8,667 8,340 0.970
Cerebrovascular Disease 7,059 -4,677 18,795 0.238
COPD 2,486 -1,144 6,116 0.180
Peptic Ulcer Disease 3,147 -4,456 10,750 0.417
Mild Liver Disease 9,803 -1,635 21,242 0.093
Moderate/Severe Liver Disease 12,330 -4,062 28,722 0.140
Diabetes -5,870 -8,411 -3,329 < 0.001
Diabetes with Complications 738 -10,005 11,481 0.893
Renal Disease 2,704 -7,991 13,400 0.620
Payer
Medicare Reference
Medicaid 291 -5,288 5,870 0.919
Commercial -1,763 -5,145 1,620 0.307
Other -1,765 -7,296 3,765 0.532
18
Table 4 (continued): Results of generalized linear model of effect of surgeon and hospital
volume on cost, controlling for other covariates.
Marginal 95% Confidence Interval
Variable Effect Lower Upper P-value
Admission Source
ER Reference
Other hospital -973 -9,858 7,912 0.830
Long term care 6,590 -12,746 25,926 0.504
Other -3,067 -10,004 3,870 0.386
Missing -3,704 -10,425 3,016 0.280
Admission Type
Elective Reference
Urgent 8,421 3,525 13,316 0.001
Emergent 18,223 10,072 26,374 < 0.001
Teaching hospital 4,869 1,795 7,943 0.002
Region
Northeast Reference
Midwest -4,638 -8,475 -801 0.018
South -3,170 -6,117 -223 0.035
West -904 -4,122 2,314 0.582
Year -600 -1,543 344 0.213
Propensity score matching
The results from multivariate models were confirmed by the propensity score analysis.
Among propensity score matched groups, patients treated by HV surgeons at HV hospitals (N =
1,641) had a mortality rate of 2.7%, as compared to the three other groups’ (N = 1,496) rate of
6.0%, which is a significant difference (ATT) of 3.3% (p = 0.016). Patients treated by HV
surgeons at HV hospitals had a significantly lower LOS among the propensity score matched
groups. LOS of patients treated by HV surgeons at HV hospitals was 13.6 days, as compared to
18.8 days for the remaining three groups. This 5.2 day difference (ATT) was significant (p <
0.001). Finally, costs for patients treated by HV surgeons at HV hospitals were $36,919
19
compared to $49,078 for the three other groups. This difference (ATT) in cost of $12,159 was
statistically significant (p < 0.001).
20
DISCUSSION
Post-operative outcomes of pancreaticoduodenectomy have been widely studied, and like
other complex surgical procedures, correlated with both surgeon and hospital volume. Single
institution studies as well as studies incorporating large, national databases have demonstrated
significantly higher mortality rates in patients undergoing pancreaticoduodenectomy by LV
surgeons as compared to HV surgeons.1,2,6,16
Studies have also shown longer lengths of stay and
cost differences greater than $5,000 associated with patients treated by LV surgeons.17
Even after
controlling for patient related factors, the benefits of treatment by HV surgeons remain evident.
However, as Birkmeyer et al. noted, outcomes from complex surgical procedures may
depend not only on how well the operation is performed but also on the available resources at the
hospital itself.1 Not surprisingly, several studies of hospital volume and surgical outcome have
demonstrated significantly higher mortality rates in patients undergoing pancreatic resection at
LV hospitals as compared to HV hospitals.2,18
By their nature, HV hospitals may tend to provide
more resources to specialty care, and may benefit from the availability of other specialists,
consistent post-operative care processes, and established intensive care unit protocols. Likewise,
LOS and cost of pancreaticoduodenectomy has been compared between low and high volume
hospitals. Glasgow et al. showed that length of hospital stay did not vary significantly based on
hospital volume, but that total hospital charges were significantly higher at LV centers.19
This
observation is in contrast, however, to the findings by Ho and Aloia who found no significant
cost differences by hospital volume.17
Because of these findings, there has been a growing trend to centralize care by
concentrating select operations in HV hospitals. The Leapfrog group has advocated for improved
outcomes through regionalization and volume based referral.12,20
As a result, many studies have
21
described increasing centralization of care,21,22
with subsequent improvement in outcomes.
Although factors other than hospital volume were responsible for the observed decrease in
mortality for pancreatic resection between 1999-2008, Finks et al. demonstrated that 67% of the
overall decrease in mortality for pancreatectomy was attributable to increased hospital volume.21
While better outcomes are associated with high surgeon volume and high hospital
volume, it is generally observed that HV surgeons operate at HV centers while LV surgeons
operate at LV centers. Complexities arise when HV surgeons operate at LV hospitals and LV
surgeons operate at HV hospitals. This confounds whether the surgeon or hospital volume effect
is the driving force behind improved surgical outcomes following pancreaticoduodenectomy.
The results of our study, which simultaneously account for surgeon volume, hospital
volume, and potential confounding patient characteristics in the analysis, suggest that both
surgeon and hospital volume have a significant effect on outcomes following
pancreaticoduodenectomy. In this cohort of the NIS, patients treated by LV surgeons at LV
hospitals had a mortality rate more than three times as high as patients of HV surgeons at HV
hospitals. Additionally, LV surgeons at HV hospitals had significantly lower mortality rates than
LV surgeons at LV hospitals. This supports previous studies which demonstrated that patients at
HV centers had lower mortality rates than those at LV centers, regardless of surgeon volume.1
Patients of HV surgeons at LV centers, however, did not have significantly lower mortality rates
compared to patients of LV surgeons at LV hospitals suggesting that the protective effect of HV
surgeons on patient outcomes demonstrated in previous studies was not enough to overcome the
poorer patient outcomes found in LV hospitals.6
Our results also demonstrate significant differences in LOS and cost following pancreatic
resection when stratified by surgeon and hospital volume. LOS of patients treated by HV
22
surgeons at HV centers was two days shorter than the LOS of patients of any other group. These
results are similar to the findings of Kennedy et al. who demonstrated that average hospital stays
for patients of HV surgeons were significantly shorter than the LOS for patients of LV
surgeons.23
In our study, patients of HV surgeons at HV centers also had significantly lower
costs associated with their stay. Compared to LV surgeons at LV hospitals, the difference in cost
of patients of HV surgeons at HV hospitals was $12,275 less. This finding is consistent with
previous studies that demonstrated cost differences between high and lower volume
surgeons.17,23
There are limitations to any study involving administrative data. Although the NIS
database is the largest all-payer database in the United States, it does not ensure that our cohort is
representative of all pancreaticoduodenectomies being performed in this country. Additionally,
outcomes were only collected in the hospital setting during the index admission, and thus only
in-hospital mortalities were captured. Lengths of admission to acute care nursing or
rehabilitation facilities were not included in the data set so we could not study whether shorter
lengths of in-hospital stay were associated with longer lengths of stay at sub-acute facilities.
Similarly, we also could not address whether cost results were influenced by early discharge to
nursing or rehabilitation facilities. The number of patients treated by HV surgeons at LV
institutions was relatively low compared to the rest of the groups. Lastly, despite the strengths of
propensity score matching, it cannot account for selection bias related to unmeasured
confounders.
In conclusion, our study confirms previous findings of improved post-operative outcomes
following pancreaticoduodenectomy in patients treated by HV surgeons at HV centers and also
highlights the complex interplay between surgeon and hospital volume while adding additional
23
information about the relationship between volume and resource use. With growing trends
toward centralization, care must be taken to ensure that both surgeon and hospital volume are
optimized at centers providing pancreaticoduodenectomy.
24
REFERENCES
1. Birkmeyer JD, Stukel TA, Siewers AE, Goodney PP, Wennberg DE, Lucas FL. Surgeon
volume and operative mortality in the United States. The New England journal of
medicine. Nov 27 2003;349(22):2117-2127.
2. Birkmeyer JD, Siewers AE, Finlayson EV, et al. Hospital volume and surgical mortality
in the United States. The New England journal of medicine. Apr 11 2002;346(15):1128-
1137.
3. Begg CB, Cramer LD, Hoskins WJ, Brennan MF. Impact of hospital volume on operative
mortality for major cancer surgery. JAMA : the journal of the American Medical
Association. Nov 25 1998;280(20):1747-1751.
4. Al-Refaie WB, Muluneh B, Zhong W, et al. Who receives their complex cancer surgery
at low-volume hospitals? Journal of the American College of Surgeons. Jan
2012;214(1):81-87.
5. Fong Y, Gonen M, Rubin D, Radzyner M, Brennan MF. Long-term survival is superior
after resection for cancer in high-volume centers. Annals of surgery. Oct
2005;242(4):540-544; discussion 544-547.
6. Eppsteiner RW, Csikesz NG, McPhee JT, Tseng JF, Shah SA. Surgeon volume impacts
hospital mortality for pancreatic resection. Annals of surgery. Apr 2009;249(4):635-640.
7. Stitzenberg KB, Meropol NJ. Trends in centralization of cancer surgery. Annals of
surgical oncology. Nov 2010;17(11):2824-2831.
8. Epstein AJ, Gray BH, Schlesinger M. Racial and ethnic differences in the use of high-
volume hospitals and surgeons. Arch Surg. Feb 2010;145(2):179-186.
25
9. Liu JH, Zingmond DS, McGory ML, et al. Disparities in the utilization of high-volume
hospitals for complex surgery. JAMA : the journal of the American Medical Association.
Oct 25 2006;296(16):1973-1980.
10. Janssen-Heijnen ML, Maas HA, Houterman S, Lemmens VE, Rutten HJ, Coebergh JW.
Comorbidity in older surgical cancer patients: influence on patient care and outcome. Eur
J Cancer. Oct 2007;43(15):2179-2193.
11. HCUP Nationwide Inpatient Sample (NIS). 2004 - 2008; www.hcup-
us.ahrq.gov/nisoverview.jsp.
12. Birkmeyer JD, Finlayson EV, Birkmeyer CM. Volume standards for high-risk surgical
procedures: potential benefits of the Leapfrog initiative. Surgery. Sep 2001;130(3):415-
422.
13. Quan H, Sundararajan V, Halfon P, et al. Coding algorithms for defining comorbidities in
ICD-9-CM and ICD-10 administrative data. Medical care. Nov 2005;43(11):1130-1139.
14. Shwartz M, Young DW, Siegrist R. The ratio of costs to charges: how good a basis for
estimating costs? Inquiry : a journal of medical care organization, provision and
financing. Winter 1995;32(4):476-481.
15. Leuven E, Sianesi, B. PSMATCH2: Stata module to perform full Mahalanobis and
propensity score matching, common support graphing, and covariate imbalance testing.
http://ideas.repec.org/c/boc/bocode/s432001.html.
16. Schmidt CM, Turrini O, Parikh P, et al. Effect of hospital volume, surgeon experience,
and surgeon volume on patient outcomes after pancreaticoduodenectomy: a single-
institution experience. Arch Surg. Jul 2010;145(7):634-640.
26
17. Ho V, Aloia T. Hospital volume, surgeon volume, and patient costs for cancer surgery.
Medical care. Jul 2008;46(7):718-725.
18. Teh SH, Diggs BS, Deveney CW, Sheppard BC. Patient and hospital characteristics on
the variance of perioperative outcomes for pancreatic resection in the United States: a
plea for outcome-based and not volume-based referral guidelines. Arch Surg. Aug
2009;144(8):713-721.
19. Glasgow RE, Mulvihill SJ. Hospital volume influences outcome in patients undergoing
pancreatic resection for cancer. The Western journal of medicine. Nov 1996;165(5):294-
300.
20. Milstein A, Galvin RS, Delbanco SF, Salber P, Buck CR, Jr. Improving the safety of
health care: the leapfrog initiative. Effective clinical practice : ECP. Nov-Dec
2000;3(6):313-316.
21. Finks JF, Osborne NH, Birkmeyer JD. Trends in hospital volume and operative mortality
for high-risk surgery. The New England journal of medicine. Jun 2 2011;364(22):2128-
2137.
22. Gasper WJ, Glidden DV, Jin C, Way LW, Patti MG. Has recognition of the relationship
between mortality rates and hospital volume for major cancer surgery in California made
a difference?: A follow-up analysis of another decade. Annals of surgery. Sep
2009;250(3):472-483.
23. Kennedy TJ, Cassera MA, Wolf R, Swanstrom LL, Hansen PD. Surgeon volume versus
morbidity and cost in patients undergoing pancreaticoduodenectomy in an academic
community medical center. Journal of gastrointestinal surgery : official journal of the
Society for Surgery of the Alimentary Tract. Dec 2010;14(12):1990-1996.