RRBI

DAgtO

R1F(fCKe

©P

elationship Between Hospital Volume, System Clinicalesources, and Mortality in Pancreatic Resection

ellal Joseph, MD, John M Morton, MD, FACS, Tina Hernandez-Boussard, PhD,lan Rubinfeld, MD, MBA, FACS, Chadi Faraj, DO, Vic Velanovich, MD, FACS

BACKGROUND: The relationship between hospital volume and perioperative mortality in pancreaticoduodenectomyhas been well established. We studied whether associations exist between hospital volume andhospital clinical resources and between both of these factors to mortality to help explain thisrelationship.

STUDY DESIGN: This two-part study reviewed publicly available hospital information from the Leapfrog Group,HealthGrades, and hospital Web sites. Hospitals were evaluated for Leapfrog ICU staffingcriteria and Safe Practice Score; HealthGrades five-star rating for complex gastrointestinalprocedures and operations; and presence of a general surgery residency, gastroenterology fel-lowship, and interventional radiology. Evaluation used trend analysis and multiple logisticregression analysis. The second part determined the mortality rate for pancreaticoduodenec-tomy using inpatient mortality data from the National Inpatient Sample and Leapfrog. Hos-pitals were categorized by low volume (� 10/year), high volume (� 11/year), strong clinicalsupport (presence of all support factors), and weak clinical support (absence of any factor). Datawere correlated by number of pancreatic resections per hospital, hospital system clinical re-sources, and operative mortality.

RESULTS: As hospital volume increased, statistically significant increases occurred in the frequency ofhospitals meeting Leapfrog ICU staffing criteria (p � 0.0001), Leapfrog Safe Practice Score(p � 0.0004), HealthGrades 5-star rating (p � 0.00001), general surgery residency (p �0.00001), gastroenterology fellowship (p � 0.00001), and interventional radiology ser-vices (p � 0.00001). No significant relationships were found between resection volume andany one of the clinical support factors and perioperative death. Presence of strong clinicalsupport was associated with lower mortality (odds ratio � 0.32; p � 0.001).

CONCLUSIONS: System clinical resources were more influential in operative mortality for pancreatic resection.This might help explain why high-volume hospitals, low-volume surgeons in high-volumeinstitutions, and some lower-volume hospitals with excellent clinical resources have lowerperioperative mortality rates for pancreatic resection. (J Am Coll Surg 2009;208:520–527.

© 2009 by the American College of Surgeons)

SdchoGcddawtoa

isclosure Information: Nothing to disclose.bstract presented at the American College of Surgeons 92nd Clinical Con-ress, Surgical Forum, Chicago, IL, October 2006, and the Paper Session ofhe American College of Surgeons 94th Clinical Congress, San Francisco, CA,ctober 2008.

eceived October 29, 2008; Revised December 13, 2008; Accepted January3, 2009.rom the Division of General Surgery, Henry Ford Hospital, Detroit, MIJoseph, Rubinfeld, Faraj, Velanovich) and the Department of Surgery, Stan-ord University, Palo Alto, CA (Morton, Hernandez-Boussard).orrespondence address: Vic Velanovich, MD, Division of General Surgery,-8, Henry Ford Hospital, 2799 West Grand Blvd, Detroit, MI 48202.

tmail: vvelano1@hfhs.org

5202009 by the American College of Surgeons

ublished by Elsevier Inc.

ince the concept of regionalization of complex proce-ures and operations was first proposed by Luft andolleagues1 25 years ago, numerous studies have shownigher hospital volume to be associated with lower post-perative morbidity and mortality.2-9 The Leapfrogroup, a consortium of large private and public health-

are purchasers, subsequently developed volume stan-ards (and other standards) for high-risk surgical proce-ures.10 Adherence to Leapfrog criteria would haveverted an estimated 8,000 deaths in 2003, 177 of whichere from pancreatectomy.4 Opinion about the volume-

o-outcomes relationship is not uniform. Ihse’s11 reviewf the volume-to-outcomes relationship in cancer oper-tions emphasized that the factors leading to this rela-

ionship might not be completely understood. Others

ISSN 1072-7515/09/$36.00doi:10.1016/j.jamcollsurg.2009.01.019

ht

wehmfbpithBuhtcpsup

MTsistsmh

PW5aLssohr

patfp

d

aiLumttfbr

eIcsSproesL

tirrlhigLlttiSrtLpwsdffia5

521Vol. 208, No. 4, April 2009 Joseph et al Pancreatic Resection Volume and Hospital Resources

ave argued that volume criteria can fall short in iden-ifying high-quality surgical centers.12

The association of hospital volume to outcomes is some-hat of a “black box,” because the mechanisms for this

ffect remain unclear. Does practice make perfect? Do theseospitals have processes that lead to lower perioperativeortality? Minimal work has been done to assess what

actors make a high-volume hospital or surgeon performetter than those at the lower end. Some low-volume hos-itals have low mortality rates, some low-volume surgeonsn high-volume hospitals have better than expected mor-ality rates, and some high-volume surgeons in low-volumeospitals have worse than expected mortality rates.13,14

irkmeyer and colleagues14 reported that a surgeon’s vol-me accounted for only 55% of the apparent effect of theospital volume for pancreatic resection. This adds fuel tohe debate on whether hospital volume versus hospital pro-ess is the prime reason for lower perioperative mortality inancreatic resections. To better elucidate these factors, wetudied whether relationships exist between hospital vol-me and hospital system clinical resources and, more im-ortantly, to operative mortality.

ETHODShis study was based on the hypothesis that, in addition to

urgical skill and volume, clinical support of the surgeon ismportant in achieving optimal outcomes in pancreatic re-ection. A two-part study was undertaken: first, to ascertainhe relationship between hospital volume of pancreatic re-ection and system clinical support; and, second, to deter-ine whether hospital volume or clinical system support

ad greater influence on operative mortality.

art 1e reviewed and collected data on all hospitals within the

0 states for pancreatic resection by using the publiclyvailable Web-based information acquired from the 2005eapfrog Group database10 and the HealthGrades Webite.15 Selection criteria for a hospital to be included in thetudy were based on two factors: hospitals had to be foundn both the Leapfrog and the HealthGrades databases, andospital data had to report the actual number of pancreaticesections.

HealthGrades is an independent private company thatrovides ratings and profiles of hospitals, nursing homes,nd physicians. Information on hospitals is derived fromhe Quality Care Medicare Provider Analysis and Reviewile, hospital data that the government keeps on Medicareatients and that includes diagnoses and procedure codes.15

The Leapfrog Group started in 1999 as an initiative

riven by organizations that purchase health care with the r

im to initiate breakthrough improvements in safety, qual-ty, and affordability of health care.The basic premise of theeapfrog Group is so-called value-based purchasing—sing the leverage of large payors to encourage improve-ents in health-care quality. Use of higher-quality hospi-

als (those meeting Leapfrog criteria) is encouragedhrough public recognition and reimbursement premiumsor hospitals. Use of lower-quality hospitals is discouragedy higher patient copayments and, in some instances, byemoving such hospitals from the health plans.

To fulfill Leapfrog criteria, hospitals must have comput-rized order-entry systems in place for inpatient care; haveCUs staffed by full-time, board-certified intensivists; meetriteria for “evidence-based hospital referral” for 5 high-riskurgical procedures; and meet the Leapfrog Safe Practicescore, which assesses a hospital’s progress on 27 of 30 saferactices endorsed by the National Quality Forum (theemaining 3 of the 30 safe practices are Leapfrog’s previ-usly mentioned first 3 criteria). For hospitals to meetvidence-based hospital referral criteria for pancreatic re-ection, volume remains the sole criterion, according toeapfrog’s 2003 standards (11� procedures per year).10

A review of all hospitals performing pancreatic resec-ions identified 434 hospitals meeting selection criteria fornclusion in the study. These hospitals were then catego-ized into four subgroups by the number of pancreaticesections: 1 to 5 (group 1, very low), 6 to 10 (group 2,ow), 11 to 30 (group 3, high), and � 30 (group 4, veryigh). Hospitals identified by group included 256 hospitals

n the 0 to 5 group, 70 each in the 6 to 10 and 11 to 30roups, and 38 hospitals in the � 30 group. Data fromeapfrog, HealthGrades, and individual hospitals were col-

ected concurrently and entered into a database. Informa-ion collected for each hospital reporting pancreatic resec-ions from the Leapfrog and HealthGrades Web sitesncluded meeting Leapfrog ICU staffing requirements andafe Practice Score and obtaining a 5-star HealthGradesating (ie, lower than expected mortality) for complex gas-rointestinal procedures and operations. Credit of theeapfrog criteria was given only to hospitals with full im-lementation of the categories. Hospitals in each categoryere also assessed for being awarded a HealthGrades five-

tar rating for mortality of complex gastrointestinal proce-ures and operations. HealthGrades awards one, three, orive stars based on risk-adjusted mortality outcomes.15 Theive-star rating indicates that actual periprocedure mortal-ty was statistically significantly better than predicted. Ofll the hospitals rated, only 10% to 15% were assigned a-star rating by HealthGrades.For the purposes of this study, additional system clinical

esources included the presence of a general surgery resi-

ddtocos(pocepwpstodceac

PAfSHParaddsoitwo

fvt

SMdButweiaprtpftr65vyhvfnv

RPcDfvff

T

V

ISHGGI

522 Joseph et al Pancreatic Resection Volume and Hospital Resources J Am Coll Surg

ency, gastroenterology fellowship, and interventional ra-iology service in the hospitals. The rationale for use ofhese factors as “system clinical resources” is based on thebservation that these services provide valuable and timelyare that could positively affect patient outcomes. Presencef a general surgery residency implies that physicians withome surgical training and interest are present at all timesbecause of the on-call schedule residents keep) in the hos-ital for immediate evaluation and treatment of any post-perative issues. In addition, this presence also implies aertain expertise in the surgical attending staff. The pres-nce of a gastroenterology fellowship implies the ability toerform diagnostic and therapeutic endoscopies, whichould lead to more accurate assessment of the patient’sancreatic disease and the ability to do interventions forurgical-related complications. Presence of an interven-ional radiologic service implies the ability to percutane-usly treat postoperative complications, such as intraab-ominal abscess, pancreatic leak, and bleeding. Face andontent validity for these measures was established throughxtensive discussion and vetting with numerous individu-ls involved with expertise in surgical quality, surgical out-omes, and complex pancreatic operations.

art 2s mortality data were not available for pancreatic resection

or the hospitals surveyed in part 1, the National Inpatientample was used to gather mortality data, the Agency forealthcare Research and Quality Cost and Utilization

roject, maintained by the Agency for Healthcare Researchnd Quality. The National Inpatient Sample is a 20% rep-esentative sample of all US hospital admissions, the largestll-payor US administrative database, which incorporatesata from approximately 1,000 hospitals and 5 to 7 millionischarges annually. The National Inpatient Sample wasurveyed for the hospitals that reported the number of casesf pancreatic resection in part 1. After these hospitals weredentified, information on in-hospital mortality was ob-ained, weighted, and stratified by hospital. These dataere then correlated to the volume and clinical resources

able 1. Relationship of Pancreatic Resection Volumes to S

ariablePa

0–5 (n � 256) 6–10 (n

CU staffing 21 46afe practice score 37 46ealthGrades five-star rating 11 17eneral surgery residency 11 17astroenterology fellowship 4 10

nterventional radiology 58 81

btained in part 1. For this part of the study, the Leap- o

rog cut-off of 11 was used to distinguish between low-olume hospitals (� 10/year) and high-volume hospi-als (� 11/year).

tatistical analysisicrosoft Excel (version 11; Microsoft Corp) was used for

ata storage. Statistical analysis was performed using SPSSase 13 for Mac (SPSS, Inc). Mantel-Haenszol test wassed for homogeneity and trend analysis of the data. Sta-istical significance was assumed at p � 0.05. Odds ratiosith 95% confidence intervals were determined for pres-

nce of each clinical support factor by incremental increasen hospital volume from 1 to 4. Multiple logistic regressionnalysis was done to determine the interdependence andredictive capabilities of hospital volume, system clinicalesources, and HealthGrades score for complex gastrointes-inal procedures and operations. In this analysis, the inde-endent variables included hospital volume categorizedrom 1 (very low; 1 to 5 resections), 2 (low; 6 to 10 resec-ions), 3 (high; 11 to 30 resections), and 4 (very high; � 30esections), system clinical resources categorized from 0 tobased on the number present, and HealthGrades score of-stars or less than 5-stars with mortality as the dependentariable. For the second part of the study, multivariate anal-sis was done to determine the risk of mortality based onospital volume and clinical support as a dichotomousariable (strong clinical support with all 6 clinical supportactors present) and as a cumulative variable (ie, by theumber of clinical support factors present from 0 to 6). A palue � 0.05 was considered significant.

ESULTSart 1: relationship of hospital volume to systemlinical resourcesata for a total of 434 hospitals that met inclusion criteria

or the study were reviewed. Assessing the data of hospitalolume against each individual system support was per-ormed for pancreatic resections. Table 1 shows the overallrequency of hospitals in each volume category having each

m Clinical Resourcestic resections (%)

p Value) 11–30 (n � 70) > 30 (n � 38)

47 74 � 0.0000160 55 0.000423 39 � 0.0000150 79 � 0.0000134 79 � 0.0000187 97 0.00001

ystencrea� 70

f the clinical support factors. The frequency of hospitals

mt0fvma(elrvs0gatch

p0ptv

otnui0chc

vtGdtdctiaGci

TpV

ISHGGI

523Vol. 208, No. 4, April 2009 Joseph et al Pancreatic Resection Volume and Hospital Resources

eeting ICU staffing requirements increased as the opera-ive volume per hospital increased (trend analysis; p �.00001). The frequency of institutions meeting the Leap-rog Group’s Safe Practice Score correlated with hospitalolume (trend analysis; p � 0.0004). HealthGrades 5-starortality rating of complex gastrointestinal procedures

nd operations also exhibited the same statistical trendp � 0.00001). System clinical resources, such as the pres-nce of a general surgery residency, gastroenterology fel-owship, and an interventional radiology service, were alsoeviewed in the same manner. Almost 80% of the high-olume (� 30) hospitals were found to have a generalurgery residency, with a statistically significant trend (p �.00001) of increase in the frequency of hospitals witheneral surgery residents with hospital volume. Presence ofgastroenterology fellowship also followed the same statis-

ical pattern, with higher-volume hospitals having an in-reased frequency of a fellowship (p � 0.00001). As theospital volumes increased, so did the frequency of hos-

able 2. Odds Ratio for the Presence of Each Clinical Sup-ort Factor for Each Increase in Resection Volumeariable Odds ratio 95% CI p Value

CU staffing 1.76 1.45–2.13 � 0.0001afe practice score 1.27 1.05–1.52 0.01ealthGrades 5-star rating 1.48 1.16–1.88 0.001eneral surgery residency 2.74 2.21–3.40 � 0.0001astroenterology fellowship 3.85 3.00–4.95 � 0.0001

nterventional radiology 2.02 1.64–2.47 � 0.0001

Figure 1. Frequency of the number of clinical

itals with an interventional radiology service (p �.00001). Table 2 presents these data as odds ratios for theresence of the clinical system support factor by incremen-al increase of hospital volume from very low (group 1) toery high (group 4).

To better understand the magnitude of the presencef clinical resources to hospital volume, Figure 1 showshe results of the relationship of hospital volume toumber of clinical resources available. As hospital vol-me increases, frequency of hospitals having more clin-

cal resources also increases. There are no hospitals in theto 5 and 6 to 10 resections per year groups with all 6

linical resource items, and more than one-half of theospitals in the � 30 resections per year had 4 or morelinical resource items.

To better understand the interdependence of hospitalolume, lower than expected mortality for complex gas-rointestinal procedures and operations (ie, the Health-rades 5-star rating), and clinical resources, Figure 2epicts the results of the logistic regression analysis ofhese factors for these tightly correlated datasets. Thisemonstrates that hospital volume predicts presence oflinical resources available and quality (as measured byhe HealthGrades five-star rating), that presence of clin-cal resources predicts both hospital volume and quality,nd that hospital quality (as measured by a Health-rades five-star rating) predicts hospital volume and

linical resources. This implies that characteristics are allnterdependent.

resources available by hospital volume.

PwOqfaotwp

essofmatpo

TSC

PISGGIH

thGra

524 Joseph et al Pancreatic Resection Volume and Hospital Resources J Am Coll Surg

art 2: relationship between perioperative mortalityith hospital volume and system clinical supportf the 28 hospitals that met inclusion criteria, the fre-

uency of performing a proximal pancreatectomy variedrom 1 to 32, with 19 hospitals performing � 10 per year,nd 9 performing � 11 per year. Table 3 shows the resultsf the univariate analysis. Although high pancreatic resec-ion volumes and presence of each clinical support resourceere associated with lower perioperative mortality, theresence of interventional radiology services and the pres-

able 3. Univariate Analysis of Frequency of Perioperative Mupport Factors and to HealthGrades Star Ratinglinical factor Weak: low volume or not pres

ancreatic resection volume 4.9CU staffing 4.0afe practice score 4.2eneral surgery residency 5.1astroenterology fellowship 6.0

nterventional radiology 13.2ealthGrades rating1 stars, 9.7%3 stars, 4.4%

Qualit

Volume

Volume Categories

(1–4)

High-volume centers predict higher-quality

scoreLogistic regression

p < .003

High-quality centers predict higher-quality

scoreLogistic regression

p < .003

Higher-volume category predicts higher resource score. Logistic

regression p < .001

Figure 2. Interdependence of hospital volume, Heal

5 stars, 1.6%

nce of a gastroenterology fellowship were associated withtatistically significant lower mortality. Table 3 demon-trates the statistically significant trend toward lower peri-perative mortality, as the HealthGrades rating improvedrom 1 through 5 stars. Table 4 shows the results of theultivariate analysis. Of the seven factors analyzed, none

chieved statistical significance as an independent predic-or of perioperative mortality. Statistical analysis of theresence of a general surgery residency and gastroenterol-gy fellowship could not yield meaningful results because

lity to Pancreatic Resection Volume and Individual Clinical

%) Strong: high volume or present (%) p Value

3.7 0.584.0 0.973.5 0.743.2 0.322.0 0.0093.4 � 0.0001

0.03

System Resources

System Resource

Score (0–6)

HealthGrades 5-Star Quality Score

(yes/no or 0–1)

Higher resource ount predicts higher volume category

Logistic regression p < .002

Higher resource count predicts higher-

volume category Logistic regression

p < .001

Higher resource count predicts higher

HG quality score Logistic regression

p < .001

des mortality rating, and clinical support resources.

orta

ent (

y

c

od61spTcstth

DMvtudsoMrwh

isrpooiiiuThvpgmvWc

tttat

TS

F

S

C

*

T

F

IVHIISGG

*

525Vol. 208, No. 4, April 2009 Joseph et al Pancreatic Resection Volume and Hospital Resources

f low numbers. Table 5 shows the multivariate analysisichotomizing system clinical resources into strong (ie, allfactors present in the hospital) and weak (ie, missing anyfactor). Again, pancreatic volume was not statistically

ignificant, but presence of all strong system clinical sup-ort was statistically significant, with an odds ratio of 0.32.able 5 shows the effect of clinical system support as aumulative variable— that is, from 0 to 6. This modelhows a statistically significant intercept and also showshat a cumulative variable, clinical system support, is sta-istically significantly associated with lower mortality, andospital volume is not.

ISCUSSIONinimal work has been done to assess what makes a high-

olume hospital or surgeon perform better than those athe lower end. In fact, a systematic review of hospital vol-me and mortality after pancreatic resection shows a greateal of heterogeneity in the data.16 Our report demon-trates that as the hospital volume increases the frequencyf hospitals with various clinical resources also increases.ore importantly, mortality was related to system clinical

esources, with a 68% reduction in perioperative deathhen all 6 clinical resources studied were present in theospital. Relationships between hospital volume and clin-

able 5. Results of Multivariate Analysis of Mortality to Hosupport Factors as a Continuous Variable

actor Coefficient

trong system clinical support*Intercept 1.77Volume 0.52Strong system clinical support �1.15

linical support factors as a continuous variableInterceptVolumeCumulative system clinical support

able 4. Results of Multivariate Analysis of Mortality to Hos

actor Coefficient SE

ntercept 0.53 1.07olume 0.93 0.51ealthGrades score 0.29 0.47

CU staffing �1.31 1.05nterventional radiology 1.53 0.93afe practice score 0.39 0.49eneral surgery residency NA* NA*astroenterology fellowship NA* NA*

Numbers of deaths too small to provide meaningful calculation.

All six support factors present.

cal, support, as measured by Leapfrog standards for ICUtaffing and Safe Practice Scores, HealthGrades five-starating for complex gastrointestinal procedures, and theresence of each a general surgery residency, gastroenterol-gy fellowship, and interventional radiology service, werebserved. Although HealthGrades five-star mortality rat-ng of complex gastrointestinal procedures and operationss not a direct measure of mortality in pancreatic resections,t does imply that hospitals are able to care for patientsndergoing these complex procedures and operations.hese data might explain why, in aggregate, high-volumeospitals have lower perioperative mortality rates than low-olume hospitals. It is known that some low-volume hos-itals have low mortality rates,17 some low-volume sur-eons in high-volume hospitals have better than expectedortality rates, and some high-volume surgeons in low-

olume hospitals have worse than expected mortality rates.e believe that this is explained by the hospital system

linical resource support.The purpose of this study was not to dispute the rela-

ionship between hospital volume and perioperative mor-ality for pancreatic resections but to determine the poten-ial causes of this association. It seems intuitive that, inddition to competent, experienced surgeons, for an insti-ution to care for these complex postoperative patients, a

Volume and Strong System Clinical Support and to Clinical

SE p ValueOddsratio

95% CIlower limit

95% CIupper limit

.46 0.0001 NA NA NA

.32 0.10 1.68 0.90 3.142

.36 0.0012 0.32 0.16 0.63

� 0.001 NA NA0.4 0.86 0.60 1.24

� 0.001 0.78 0.73 0.87

l Volume and System Clinical Support

Value Odds ratio95% CI

lower limit95% CI

upper limit

0.62 NA NA NA0.07 2.53 0.94 6.870.54 1.33 0.53 3.330.21 0.27 0.03 2.130.10 4.62 0.75 28.600.42 1.48 0.57 3.86NA* NA* NA* NANA* NA* NA* NA*

pital

000

pita

p

wtspamcchhcwm

tcotiErhivgGewawpcpsdmdmccppilistts

c

caaHaia

FhartsmhAscpfgTmonstamwsampubdm

wttsv

A

SA

526 Joseph et al Pancreatic Resection Volume and Hospital Resources J Am Coll Surg

ell-trained, experienced hospital staff with supportingechnologies is imperative. This might also explain the ob-ervation that historical measures of operative mortalityredict future performance of hospitals.18 Such hospitalsre less likely to eliminate clinical resources and more apt toake efforts to strengthen them. In addition, Gordon and

olleagues19 demonstrated a decrease in mortality of pan-reatic resection as more patients underwent resection in aigh-volume center in Maryland. This high-volume centerad a great deal of highly sophisticated clinical resources toare for these patients as well. This observation is consistentith the hypothesis that improvement in perioperativeortality is associated with clinical resources and volume.The need for strong clinical support in caring for pa-

ients after pancreatic resection cannot be surprising. Re-ent reports emphasize the value of processes and structuresf care in improving postoperative outcomes.20,21 Evenhough mortality rates for pancreatectomy have decreasedn recent years,22,23 the morbidity rate remains high.24,25

mick and colleagues26 reported a hospital readmissionate of 26% after pancreaticoduodenectomy, with nearlyalf of these a result of complications. The expeditious

dentification and management of these complications pre-ent them from being fatal. This is where the presence ofeneral surgery residents as “first responders” can be vital.eneral surgery residents, because of their constant pres-

nce in the hospital, particularly in high-volume hospitalsith complex patients, might be more adept at identifying

nd responding to complications than other institutionsithout residents. Similarly, because many postoperativeancreatectomy patients and most with postoperativeomplications will require at least a short stay in the ICU,resence of trained intensivists at the bedside, rather than aurgeon from the operating room on the telephone, ad-ressing critical care issues would seem more ideal. Thisight be where ICU staffing and safe practices make a

ifference in patient care.27 Pancreatectomy patients com-only require interventional radiologic services. Sohn and

olleagues28 reported that 44% of patients undergoing pan-reatectomy had some type of interventional radiologicrocedure, 12% in the postoperative period for some com-lication. In the univariate analysis (Table 3), presence ofnterventional radiology was statistically associated withower perioperative mortality. In addition, hospitals meet-ng the Leapfrog criteria for Safe Practice Score and as-igned a HealthGrades five-star rating would also seem ableo manage to improve the hospital course of pancreatec-omy patients. In aggregate, presence of all of these re-ources had a major impact on risk of perioperative death.

Not surprisingly, high-volume hospitals also have more

linical resources. All of these institutions are major medi-

al and referral centers. A high level of personnel expertisend technology is to be expected. Figure 2 demonstratesstatistically significant relationship among volume, aealthGrades five-star rating, and clinical resources avail-

ble. This demonstrates that volume, although important,s interdependent on the availability of clinical resources tochieve superior outcomes.

We must acknowledge several limitations of this study.irst, there might be “omitted variable bias.” As we onlyad access to hospital information that was publicly avail-ble, we cannot be certain whether our study’s findingsepresent the factors of importance or surrogates for theruly important factors. For example, the HealthGradescore is simply a report on whether the hospitals had aortality rate for complex gastrointestinal procedures

igher, the same, or lower than expected for their case mix.ll the clinical factors that lead to this score are really in the

ame “black box” as hospital volume. The same disclaimeran be made of the Leapfrog Safe Practice Score. Similarly,resence of a general surgery residency or gastroenterologyellowship can be surrogate factors to highly skilled sur-eons and gastroenterologists with excellent judgment.his could be an area of additional research. Second, theere presence of a general surgery residency, gastroenter-

logy fellowship, or interventional radiology service doesot assure the quality of the attending and resident/fellowurgeons, gastroenterologists, and radiologists. It is one fac-or to have an attending surgeon teach residents but quitenother to have a superior one. The same disclaimer can beade of the ICU staffing requirements. Third, there is noay to determine the effectiveness with which these re-

ources are used. That is, although an institution can haveparticular or even several clinical resource(s), it might oright not use these resources to their best advantage in

atient care. It might be that higher-volume hospitals alsose their resources better and that is what accounts for theetter outcomes. In fact, additional research is needed toetermine which factors within these clinical resourcesake the difference.In conclusion, this study demonstrates that hospitals

ith a high volume of pancreatic resections are more likelyo have the system clinical support to successfully treathese complex patients. What this study did not demon-trate is whether clinical support is more important thanolume in low postresection mortality.

uthor Contributions

tudy conception and design: Morton, Rubinfeld, Velanovichcquisition of data: Joseph, Morton, Hernandez-Boussard,

Faraj

A

D

C

R

1

1

1

1

1

1

1

1

1

1

2

2

2

2

2

2

2

2

2

527Vol. 208, No. 4, April 2009 Joseph et al Pancreatic Resection Volume and Hospital Resources

nalysis and interpretation of data: Morton, Hernandez-Boussard, Rubinfeld, Velanovich

rafting of manuscript: Joseph, Hernandez-Boussard, Rubin-feld, Velanovich

ritical revision: Morton, Hernandez-Boussard, Rubinfeld,Velanovich

EFERENCES

1. Luft HS, Bunker JP, Enthoven AC. Should operations be region-alized? The empirical relation between surgical volume and mor-tality. N Engl J Med 1979;301:1364–1369.

2. Lieberman MD, Kilburn H, Lindsey M, Brennan MF. Relationof perioperate deaths to hospital volume among patients under-going pancreatic resection for malignancy. Ann Surg 1995;222:638–645.

3. Birkmeyer JD, Warshaw AL, Finlayson SRG, et al. Relationshipbetween hospital volume and late survival after pancreaticoduo-denectomy. Surgery 1999;126:178–183.

4. Birkmeyer JD, Dimick JB. Potential benefits of the new Leap-frog standards: effect of process and outcomes measures. Surgery2004;135:569–575.

5. Begg CB, Cramer LD, Hoskin WJ, et al. Impact of hospitalvolume on operative mortality for major cancer surgery. JAMA1998;280:1747–1751.

6. Birkmeyer JD, Finlayson SRG, Tosteson ANA, et al. Effect ofhospital volume on in-hospital mortality with pancreaticoduo-denectomy. Surgery 1999;125:250–256.

7. Sosa JA, Bowman HM, Tielsch JA, et al. The importance ofsurgeon experience for clinical and economic outcomes fromthyroidectomy. Ann Surg 1998;228:320–330.

8. Hannan EL, Radzyner M, Rubin D, et al. The influence ofhospital and surgeon volume on in-hospital mortality for colec-tomy, gastrectomy, and lung lobectomy in patients with cancer.Surgery 2002;131:6–15.

9. Birkmeyer JD, Siewers AE, Finlayson EVA, et al. Hospital vol-ume and surgical mortality in the United States. N Engl J Med2002;346:1128–1137.

0. Leapfrog Group. Available at: http://www.leapfroggroup.org/.Accessed December 4, 2007.

1. Ihse I. The volume-outcome relationship in cancer surgery: ahard sell. Ann Surg 2003;238:777–781.

2. Christian CK, Gustafson ML, Betensky RA, et al. The Leapfrogvolume criteria may fall short in identifying high-quality surgi-cal centers. Ann Surg 2003;238:447–455.

3. Ho V, Heslin MJ. Effect of hospital volume and experience onin-hospital mortality for pancreaticoduodenectomy. Ann Surg

2003;237:509–514.

4. Birkmeyer JD, StukelTA, Siewers AE, et al. Surgeon volume andoperative mortality in the United States. N Engl J Med 2003;349:2117–2127.

5. HealthGrades. Available at: http://www.healthgrades.com/. Ac-cessed February 8, 2006.

6. van Heek NT, Kuhlmann KFD, Scholten RJ, et al. Hospitalvolume and mortality after pancreatic resection: a systematicreview and an evaluation of intervention in the Netherlands.Ann Surg 2005;242:781–790.

7. Lieberman MD, Kilburn H, Lindsey M, Brennan MF. Relationof perioperative deaths to hospital volume among patients un-dergoing pancreatic resection for malignancy. Ann Surg 1995;222:638–645.

8. Birkmeyer JD, Dimick JB, Staiger DO. Operative mortality andprocedure volume as predictors of subsequent hospital perform-ance. Ann Surg 2006;243:411–417.

9. Gordon TA, Bowman HM, Teilsch JM, et al. Statewide region-alization of pancreaticoduodenectomy and its effect on in-hospital mortality. Ann Surg 1998;228:71–78.

0. Main DS, CavenderTA, Nowels CT, et al. Relationship of processesand structures of care in general surgery to postoperative outcomes:a qualitative analysis. J Am Coll Surg 2007;204:1147–1156.

1. Main DS, Henderson WG, Pratte K, et al. Relationship ofprocesses and structures of care in general surgery to postopera-tive outcomes: a descriptive analysis. J Am Coll Surg 2007;204:1157–1165.

2. Stojadinovic A, Brooks A, Hoos A, et al. An evidence-basedapproach to the surgical management of resectable pancreaticadenocarcinoma. J Am Coll Surg 2003;196:954–964.

3. McPhee JT, Hill JS, Whalen GF, et al. Perioperative mortalityfor pancreatectomy: a national perspective. Ann Surg 2007;246:246–253.

4. Gouma DJ, van Geenen RC, van Gulik TM, et al. Rates ofcomplications and death after pancreaticoduodenectomy: riskfactors and the impact of hospital volume. Ann Surg 2000;232:786–795.

5. Glasgow RE, Jackson HH, Neumayer L, et al. Pancreatic resec-tion in Veterans Affairs and selected university medical centers:results of the patient safety in surgery study. J Am Coll Surg2007;204:1252–1260.

6. Emick DM, Riall TS, Cameron JL, et al. Hospital readmissionafter pancreaticoduodenectomy. J Gastrointest Surg 2006;10:1243–1253.

7. Milstein A, Galvin RS, Delbanco SF, et al. Improving the safetyof health care: the Leapfrog initiative. Eff Clin Pract 2000;3:313–316.

8. Sohn TA, Yeo CJ, Cameron JL, et al. Pancreaticoduodenectomy:role of interventional radiologists in managing patients and

complications. J Gastrointest Surg 2003;7:209–219.