Journal of Surgical Oncology 2014;110:400–406

Using a Modification of the Clavien-Dindo System Accounting for Readmissions

and Multiple Interventions: Defining Quality for Pancreaticoduodenectomy

MARSHALL S. BAKER, MD, MBA,1* KAREN L. SHERMAN, MD,2 SUSAN J. STOCKER, BLS,1
AMANDA V. HAYMAN, MD,2 DAVID J. BENTREM, MD,2 RICHARD A. PRINZ, MD,1
1
AND MARK S. TALAMONTI, MD
1
Department of Surgery, NorthShore University Health System, Evanston, Illinois
2
Feinberg School of Medicine, Northwestern University, Chicago, Illinois

Background: The Clavien‐Dindo system (CD) does not change the grade assigned a complication when multiple readmissions or interventions are
required to manage a complication. We apply a modification of CD accounting for readmissions and interventions to pancreaticoduodenectomy (PD).
Methods: PDs done between 1999 and 2009 were reviewed. CD grade IIIa complications requiring more than one intervention and II and IIIa
complications requiring significantly prolonged lengths of stay including all 90‐day readmissions were classified severe‐adverse‐postoperative‐
outcomes (SAPO). CD IIIb, IV, and V complications were also classified SAPOs. All other complications were considered minor‐adverse‐
postoperative‐outcomes (MAPO).
Results: Four‐hundred forty three of 490 PD patients (90.4%) had either no complication or a complication of low to moderate CD grade (I, II, IIIa).
When reclassified by the new metric, 92 patient‐outcomes (19%) were upgraded from CD II or IIIa to SAPO. One‐hundred thirty nine patients
(28.4%) had a SAPO. Multivariable regression identified age >75 years, pylorus preservation and operative blood loss >1,500 ml as predictors of
SAPO. Age was not associated with poor outcome using the unmodified CD system.
Conclusions: Established systems may under‐grade the severity of some complications following PD. We define a procedure‐specific modification
of CD accounting for readmissions and multiple interventions. Using this modification, advanced age, pylorus preservation, and significant blood
loss are associated with poor outcome.
J. Surg. Oncol. 2014;110:400–406. ß 2014 Wiley Periodicals, Inc.

KEY WORDS: pancreaticoduodenectomy; surgical complications; quality

INTRODUCTION Clavien‐Dindo system as the basis for evaluating outcomes following

The Clavien‐Dindo grading system has been the system most widely
used to evaluate early postoperative outcomes following
pancreaticoduodenectomy (PD) [1–4]. This system discriminates
complication severity on the basis of whether or not the management
of a postoperative complication requires a medical treatment (Clavien
Grade I to II), an invasive intervention without general anesthesia(grade
IIIa), reoperation under general anesthesia (grade IIIb), or the patient
experiences a life threatening event (grade IV) or death as a result of the
complication (grade V). The Clavien‐Dindo system does not otherwise
adjust the grade assigned a grade II or IIIa complication if multiple
readmissions or prolonged inpatient admission are required to manage
the complication and does not adjust the grade given a IIIa complication
if multiple invasive interventions without general anesthesia are required
to manage the complication.
For patients undergoing PD, there is substantial clinical variability in
the postoperative courses of those experiencing CD grade II and IIIa
complications. This variability creates potential to misrepresent the
clinical severity of a complication. A patient who experiences delayed
gastric emptying following PD may spend several weeks in and out of
the hospital receiving parenteral nutrition or tube feedings but will be
given the same complication grade by the CD system (grade II) as that
given a patient who receives an oral antibiotic for a superficial wound
infection and spends 7 days in the hospital in total. Likewise, a patient
who requires three separate drain placements over several weeks of in
and outpatient care to control a pancreatic fistula will be given the same
grade (IIIa) as a patient who requires a single aspiration of a sterile fluid
collection to evaluate a postoperative leukocytosis and then is sent home
on post operative Day 8. Because of this, studies using an unmodified
PD, for identifying predictors of poor outcome and pursuing quality
improvement initiatives may not capture the true impact of some
postoperative complications and may be inaccurate.
To better understand the determinants of poor outcome following
pancreaticodudodenectomy, we reviewed our series of pancreatico-
duodenectomies done between 1999 and 2009. We graded postoperative
outcomes by means of a modification of the Clavien‐Dindo system
designed to use two easily obtainable measures to better classify the
clinical severity of complications falling in the midrange of the
unmodified Clavien‐Dindo scale. We used prolonged (relative to
uncomplicated recoveries) lengths of hospitalization including all 90‐
day readmissions and the need for multiple non‐surgical invasive
interventions as easily obtainable, objective measures of the clinical
burden of the midrange complications. Stepwise multivariable logistic
regression was used to identify preoperative and operative factors
associated with poor quality outcome as defined by the modified metric.
This paper was presented as an oral presentation at the Clinical Congress of
the American College of Surgeons, Oct 3, 2012, Chicago, IL.
*Correspondence to: Marshall S. Baker, MD, MBA, Department of Surgery,
NorthShore University Health Center, Walgreen’s Building, 2nd Floor, 2650
Ridge Avenue, Evanston, IL 60201. Fax: þ847‐570‐1330.
E‐mail: mbaker3@northshore.org
Received 9 January 2014; Accepted 5 May 2014
DOI 10.1002/jso.23663
Published online 26 May 2014 in Wiley Online Library
(wileyonlinelibrary.com).

ß 2014 Wiley Periodicals, Inc.


METHODS
Patient Population, Preoperative Demographics,
and Intraoperative Variables
We queried our prospectively maintained pancreatic database to
identify all patients who underwent PD between January 1999 and
December 2009. Patient demographic characteristics and intraoperative
variables retrieved from the database were those thought, based on
author experience and literature review, to be potential determinants of
postoperative outcome. Demographic data included age and sex.
Preoperative clinical information included smoking and alcohol history,
presence or absence of presenting symptoms, presence or absence of
comorbid coronary artery disease, chronic obstructive pulmonary
disease, diabetes, and chronic pancreatitis, presence or absence of
jaundice, preoperative biliary stenting, preoperative weight loss, serum
values for CA 19‐9, albumin, liver function, coagulation parameters, and
preoperative treatment with chemoradiotherapy. Intraoperative data
included operative time, estimated blood loss, transfusion requirement,
pyloric‐preserving versus standard PD, and use of vascular resection
with reconstruction. Pathology variables included TNM stage, margin
status, and tumor differentiation.
Postoperative Complications and Readmission Events
The inpatient and outpatient medical records for each patient were
then reviewed to identify all deviations (urinary tract infection,
pneumonia, wound infection, intra‐abdominal abscess, IV site
infection, pancreatic fistula, anemia requiring transfusion, delayed
gastric emptying requiring promotility agents, supplemental enteral
nutrition or parenteral nutrition, deep venous thrombosis, cardiac
arrhythmia, respiratory failure, renal failure) from the standard recovery.
For purposes of this paper, intra‐abdominal abscess, and pancreatic
fistulas were grouped together as one type of complication. The details of
medical management and the invasive interventions (including all
endoscopic procedures, interventional radiologic procedures, and
reoperations) required to manage these events were recorded. Anti‐
nausea medications, proton pump inhibitors, acetaminophen, and
narcotic pain medication were considered standard therapies and not
identified as adverse events. Procedures that were “drain checks” or
“tube studies” done without a new drain being placed or an old drain
being replaced were not included as interventions. All readmission
events occurring within 90 days of the date of the operation and directly
due to complications from the procedure were also identified. The
number of days spent in the hospital for all procedure‐related admissions
was summed to determine the total overall length of stay for the
procedure. Days spent in the hospital prior to surgery and inpatient
readmissions related to adjuvant chemotherapy were not included.
Complication Grading
Outcomes were first graded according to the unmodified Clavien‐
Dindo system. Each patient was assigned a single Clavien‐Dindo grade
(0 to V). Patients having had more than one complication were assigned
one grade corresponding to that for the most severe complication
experienced. Each patient with a complication was then re‐graded as
either a minor (MAPO) or severe adverse post operative outcome
(SAPO) by taking into account the Clavien‐Dindo grade, the number of
non‐surgical invasive procedures required to manage the complication
and the overall length of stay including all relevant readmissions out to
90 days. Again one grade was assigned per patient. Patients who had
grade IIIb, IV, and V complications by the unmodified Clavien‐Dindo
system were classified as having had a SAPO. Those having had IIIa
complications by the Clavien‐Dindo system but requiring more than one
non‐surgical interventional procedure (endoscopy, IR angiography, or
Journal of Surgical Oncology
Quality for Pancreaticoduodenectomy 401
IR guided drain placement) to manage the complication were classified
as having had a SAPO. Those having had I, II, and IIIa complications
and requiring a total length of stay greater than three standard deviations
beyond the mean stay for those patients undergoing PD without any
identifiable complication (uncomplicated pancreaticoduodenectomies)
were also classified as having had SAPO. Patients having I, II, and IIIa
complications requiring one or fewer interventions and shorter overall
lengths of stay were classified as having had MAPOs.
The limit of three standard deviations beyond the mean for an
uncomplicated PD was chosen because that limit statistically excluded
all (99.5% of) uncomplicated recoveries from the severe adverse
outcome category. This threshold also fell (at 17 days) beyond a period
that was thought appropriate as a limit for a reasonable length for
postoperative hospitalization following PD. The limit of one drain
placement or interventional procedure was made to capture the
pancreatic fistulas that required multiple procedures in interventional
radiology but did not require a prolonged admission, a return to the
operating room or result in a life‐threatening condition.
Statistical Analysis
Data are presented as the mean  standard deviation for continuous
variables and as frequencies for all categorical variables. Multivariable
stepwise logistic regression analysis was performed using SAS v9.2
(Cary, NC). Several continuous variables were made categorical to
facilitate the regression modeling. Age was separated into two
categories: advanced age defined as 75 years, middle aged <75
years. BMI was defined as recommended by the NIH with underweight
being <18.5, normal being 18.5 to 25, overweight being 25 to 30, and
obese being >30. Albumin was divided into two categories on the basis
of literature review: normal to moderate hypoalbuminemia (>2.5 g/dl)
and severe hypoalbuminemia (2.5 g/dl). Total serum bilirubin was also
divided into three categories on the basis of literature review: normal
(<1.0 mg/dl), mild to moderate hyperbilirubinemia (1–8 mg/dl), and
severe hyperbilirubinemia (>8 mg/dl). OR time and operative blood loss
were broken down on the basis of sensitivity testing. OR time was
divided into two categories: prolonged at 8 hr and not prolonged at
<8 hr. Operative blood loss was considered excessive at 1,500 ml and
not excessive at <1,500 ml. The break points for OR time and estimated
blood loss were identified by using multiple iterations of univariate
regression. These iterations identified 8 hr and 1,500 ml as statistically
significant predictors of poor outcome.
Perioperative Management and Operative Approach
Pancreaticoduodenectomies were performed by one of four
pancreaticobiliary surgeons. Patients with adenocarcinoma who were
felt to have resectable tumors by preoperative imaging were, in general,
taken to resection without prior chemoradiotherapy. Patients who were
felt to have long segment or circumferential involvement of the superior
mesenteric/portal vein or short segment extension to the superior
mesenteric artery were treated with neoadjuvant chemoradiotherapy and
restaged prior to planned resection. Patients with evidence of metastatic
disease, encasement of the superior mesenteric artery, or involvement of
the hepatic artery were considered unresectable and are not included in
this study.
For the first 5 years in the period under study, the practice of the group
was to perform standard PD with antrectomy, to place feeding
jejunostomy tubes and to use closed suction drains. Late in the period
under review (last 5 years), there was a concerted effort to perform
pylorus preservation unless doing so would compromise the oncologic
outcome. Late in the period under review closed suction drains and
feeding jejunostomy tubes were used selectively when the operating
surgeon was concerned for a high risk of pancreatic leak. Throughout the
study period the pancreatic remnant was uniformly reconstructed by
402 Baker et al.
means of a duct‐to‐mucosa anastomosis over an internal stent. Resections TABLE I. Preoperative Demographic and Clinical Characteristics
of involved superior mesenteric vein or portal vein were performed as
N %
needed.
Early in the experience reported, the postoperative care of the patient Age
was left entirely to the discretion of the operating surgeon. Late in the <75 years 395 80.6
experience (last 5 years) an effort was made to standardize the 75 years 95 19.4
postoperative recovery using a post‐operative clinical pathway Female gender 233 47.5
embedded in the electronic medical record. Currently, when a patient Race
has a complication, the management is taken off pathway and left to the White 438 89.4
discretion of the operating surgeon. Black 44 8.9
Other 8 1.6
BMI (n ¼ 325)
Patient Anonymity Normal (18.5–25) 158 48.6
Underweight (<18.5) 11 3.3
Approval for this study was obtained from the Institutional Review Overweight (25–30) 99 30.5
Board for NorthShore University Health Systems and Northwestern Obese (>30) 57 17.5
University. The study was conducted in accordance with all institutional Comorbidities
policies protecting patient anonymity. Acute pancreatitis 61 12.5
Chronic pancreatitis 26 5.3
Diabetes 107 21.8
RESULTS CAD 61 12.5
COPD 22 4.5
Demographic Characteristics CHF 9 1.8
Four hundred ninety patients underwent PD during the years under Smoker 85 17.4
Preoperative albumin (n ¼ 422)
review. Patients undergoing PD had an average age of 62  12.4 years.
2.5 g/dl 37 8.8
The majority had a tumor that was biopsy‐proven adenocarcinoma. They >2.5 g/dl 385 91.2
were equally likely to be male or female. Body mass index tracking was Preoperative bilirubin (n ¼ 443)
begun relatively late in the course of the series and data was available and <1 mg/dl 204 46.0
is presented for 325 out of 490 patients. The majority of patients were 1–8 mg/dl 199 44.9
well nourished (normal albumin or mild hypoalbuminemia), had a >8 mg/dl 40 9.0
moderate hyperbilirubinemia and had normal to overweight BMIs. Few Neoadjuvant chemotherapy 31 6.3
patients in this series were obese. 6.3% of the total were treated with 
BMI, body mass index; CAD, coronary artery disease; COPD, chronic
neoadjuvant chemoradiotherapy prior to resection. Table I depicts
obstructive pulmonary disease; CHF, congestive heart failure.
demographic and preoperative clinical parameters used in the univariate
and multivariate models by the categorization used for the final modeling.

Complication severity was first graded by the Clavien‐Dindo system

Intraoperative and Pathologic Outcomes (Table IV). The most common grade of complication by Clavien‐Dindo
We identified a complete data set for most intraoperative and grading scheme was Grade II which occurred in 175 of 490 patients
pathologic variables including: pathologic diagnosis, whether or not the (35.7% of the total PD population). Clavien‐Dindo Grade IIIa
patients underwent neoadjuvant treatment, whether or not a given patient complications were second most prevalent with 96 of 490 or nearly
had a vein resection and underwent pylorus preservation or standard PD 20% of patients having a Grade IIIa complication. There were relatively
(Table II). A more limited data set was available for tumor size few Grade IIIb, IV, and V complications. Only 47 patients (9.6%) had a
(n ¼ 437), intraoperative blood loss (n ¼ 458), transfusion requirement Clavien‐Dindo grade IIIb or greater complication.
(n ¼ 458), and procedure time (n ¼ 477). Tumor size was not available
for many patients with benign pathology early in the series. There were
no noted absences of size data for patients with malignant disease. On TABLE II. Intraoperative Outcomes and Pathology
final pathology the average tumor size was 2.5  1.8 cm. The majority
N %
(nearly 70%) of patients undergoing PD underwent a standard PD with
antrectomy and gastrojejunostomy. The mean OR time was 7.2  1.7 hr Procedure type
and mean EBL was 781.2  745 ml. The vast majority (334 out of 458 or Standard PD with antrectomy 342 69.8
73%) required no perioperative transfusion. Eight percent of the total Pylorus preserving PD 148 30.2
underwent vein resection with reconstruction. Vein resection with reconstruction 41 8.4
Operative blood loss (n ¼ 458)
1,500 ml 408 89.1
Postoperative Complications and Outcome Severity Scores >1,500 ml 50 10.9
Perioperative transfusion (n ¼ 458)
Three hundred sixty six of 490 total PD patients (nearly 75% of 2 units 429 93.7
patients) had an identifiable complication. There were 13 postoperative >2 units 29 6.3
deaths for an overall postoperative mortality rate of 2.7%. One hundred Procedure length (n ¼ 477)
of the 490 patients required at least one postoperative readmission for <8 hr 333 69.8
an overall readmission rate of 20.4%. The most common complication 8 hr 144 30.2
was wound infection with that occurring in 136 or 27.8% of patients Tumor size (n ¼ 437)
(Table III). Pancreatic fistula and intra‐abdominal abscess were together <2 cm 129 29.5
2–4 cm 238 54.5
the second most common complication occurring in 111 of 490 patients
>4 cm 70 16.0
(22.6%). Delayed gastric emptying and anemia followed as the next Malignant pathology 361 73.7
most prevalent complications.

Journal of Surgical Oncology

 Quality for Pancreaticoduodenectomy 403
TABLE III. Complication Frequency Following Pancreaticoduodenectomy
by Type

n %a

Any 366 74.5

Wound infection 136 27.8
Pancreatic fistula/abscess 111 22.6
Delayed gastric emptying 102 20.8
Anemia 82 16.7
Pneumonia 41 8.4
Post operative intra‐abdominal hemorrhage 31 6.3
Urinary tract infection 25 5.1
Deep vein thrombosis 13 2.7
Pulmonary embolism 10 2.0
Atrial fibrillation 8 1.6
Fig. 1. Complication grade by Clavien‐Dindo and by MAPO/SAPO
a
%, percent of total number of patients undergoing pancreaticoduodenectomy
classification system. Ninety two (25%) out of a total of 366 patients
(n ¼ 490).
with complications, 35 Clavien grade II and 57 Clavien grade IIIa, fell
into what is considered to be a higher severity complication category
when reclassified by the MAPO/SAPO system.

Each patient outcome was then assessed for complication severity by

means of the modified metric using overall lengths of stay including all
90‐day readmissions and the number of invasive interventions required the tendency for readmission (Table III). The mean overall length of
to manage a complication to determine complication severity. The mean stay for an uncomplicated PD was again 9.4  2.6 days, that for a PD
overall length of stay for an uncomplicated PD was 9.4  2.6 days. The experiencing a MAPO was 11.4  2.9 days and that for a PD
cut‐off for prolonged length of stay was set to 3 standard deviations experiencing a SAPO was 28.5  16.3. Those patients experiencing a
beyond this mean. This limit fell at 17 days. When regraded, 92 patients SAPO were statistically more likely to have required readmission to
(19% of the total number of pancreaticoduodenectomies or 25% of the hospital with 41.7% of SAPO patients having at least one
pancreaticoduodenectomies with complications) moved from what we readmission. 18.5% of patients having a MAPO required a readmission
would have considered a relatively acceptable Clavien‐Dindo (P < 0.05).
complication grade (I, II, or IIIa) into the severe adverse outcome
category. When regraded, 124 patients (25.3%) had no complication,
227 (46.3%) had a MAPO, and 139 (28.4%) had a SAPO (Fig. 1). Predictors of Severe Adverse Postoperative Outcome
Thirty‐five out of 175 (20%) patients with unmodified Clavien‐Dindo
grade II complications were “upgraded” to severe adverse outcome The presence or absence of a SAPO for patients undergoing PD was
based on prolonged lengths of admission. Fifty‐seven out of a total of 96 then used to develop univariate and multivariate models identifying
(60%) patients with Clavien‐Dindo grade IIIa complications were preoperative and intraoperative factors associated with poor outcome.
upgraded to SAPO. Thirty‐seven of these were “upgraded” to SAPO Variables that demonstrated a statistical association with SAPO in the
based on prolonged length of admission alone. Five were upgraded univariate models (P < 0.2) were retained in the final multivariate
based on multiple invasive interventions alone. Fifteen of those with IIIa model. Variables retained from the univariate models included: age  75
complications who were upgraded to SAPO had both prolonged lengths years, female gender, presence of coronary artery disease, presence of
of stay and required multiple invasive interventions. chronic obstructive pulmonary disease, presence of congestive heart
As expected given our definition of SAPO, using the new metric, the failure, presence of chronic pancreatitis, prolonged operative time
mean overall length of stay varied directly with outcome severity as did (8 hr), excessive blood loss (>1,500 ml), and pylorus preservation.
Also included in the final multivariate model were several variables that
did not achieve significance in the univariate model but that were
TABLE IV. Readmission Rates, Interventions and LOS by Clavien‐Dindo thought to be potentially relevant predictors of outcome by the operating
Grade and MAPO/SAPO Classification surgeons. These included: BMI, neoadjuvant treatment and major
vascular resection. In the final multivariate model, only age >75 years
Complication Readmissions Mean number Mean LOS (odds ratio 2.7, 95% confidence interval of [1.59–4.54]), pylorus
grade n (%) (%) interventions (days)c preservation (odds ratio 1.74, 95% CI [1.07–2.84]), and blood loss
>1.5 L (odds ratio of 2.52, 95% CI [1.25–5.07]) were identified as
None 124 (25.3) 0 0 9.4  2.6
statistically relevant independent predictors of SAPO.
Clavien‐Dindo I 48 (9.8) 12.5 0 11.5  6.0
Clavien‐Dindo II 175 (35.7) 17.7 0 10.8  3.1
To determine the impact of the new outcome variable on the
Clavien‐Dindo IIIa 96 (19.6) 50 1.2  0.8 20.7  9.5 predictive modeling we also performed multivariate regression analysis
Clavien‐Dindo IIIb 14 (2.9) 35.7 0.6  1.3 31.9  15.9 predicting poor outcome defined by the Clavien‐Dindo system in an
Clavien‐Dindo IVa 16 (3.3) 25 0.7  1.1 31.8  23.3 unmodified form: with poor outcome defined as Clavien‐Dindo grade
Clavien‐Dindo IVb 4 (0.8) 75 1.3  0.5 27.5  0.8 IIIb and above (Table V). Compared to the unmodified Clavien‐Dindo
Clavien‐Dindo V 13 (2.7) 23.1 1.2  1.1 38.9  30.9 System, the MAPO/SAPO classification identified advanced age as a
MAPOa 227 (46.3) 18.5 0.1  0.4 11.4  2.9 predictor of poor outcome more clearly than the unmodified Clavien‐
SAPOb 139 (28.4) 41.7 0.9  1.1 28.5  16.3 Dindo system. When poor outcome was defined as a Clavien Grade IIIb
Overall 490 (100) 20.4 0.3  0.7 15.7  12.7
or above, age was not identified as a statistical predictor of poor
a
MAPO, mild adverse postoperative outcome. outcome. Instead, prolonged procedure time, excessive blood loss,
b
SAPO, severe adverse postoperative outcome. pylorus preservation, neoadjuvant therapy, COPD, and abnormally low
c BMI were predictors of poor outcome when Clavien grade IIIb was used
Mean LOS, mean overall length of stay including all readmissions to 90
days  standard deviation. as the cut‐off for poor outcome.

Journal of Surgical Oncology

404 Baker et al.
TABLE V. Multivariate Predictors of SAPO and Clavien Grade IIIb‐V PD without a complication. When outcomes were assessed by the modified
Complications metric, 92 patients or 19% of the total PD population and 25% of those
having any complication were found to “migrate” into the high severity
SAPO OR Clavien Grade
(95% CI) IIIB‐V OR (95% CI)
category from a Clavien‐Dindo grade that might otherwise have been
considered moderate or better (Fig. 1). Among the patients that migrated
Age were 57 who were initially given a Clavien‐Dindo grade of IIIa. This number
<75 years Referent Referent represents 60% of the patients with grade IIIa complications. Most published
75 years 2.69 (1.59–4.54) 1.67 (0.75–3.75) studies that have used the Clavien‐Dindo system to grade outcomes
Female gender 0.78 (0.50–1.23) 0.87 (0.42–1.81) following pancreatectomy have considered a CD grade IIIa as the cut‐off for
BMI labeling a complication high severity. Our findings suggest that 40% of the
Normal Referent Referent
patients in this category in our series would be inappropriately graded as
Underweight 2.35 (0.61–9.07) 5.67 (1.19–27.16)
Overweight 1.38 (0.75–2.54) 1.24 (0.45–3.41)
having a high severity complication if evaluated in that traditional manner. A
Obese 0.89 (0.40–1.96) 0.60 (0.15–2.41) quality improvement initiative using the unmodified Clavien‐Dindo system
Comorbidities in this way would identify as targets for process improvement factors
CAD 1.19 (0.64–2.20) 1.46 (0.60–3.56) associated with relatively good outcomes not factors associated with
COPD 2.46 (0.96–6.35) 5.81 (1.90–17.79) relatively poor outcomes.
CHF 3.07 (0.71–13.26) 3.53 (0.74–16.92) Using our modified metric, we identified age, pylorus preservation,
Chronic pancreatitis 2.27 (0.96–5.37) 2.05 (0.59–7.14) and excessive blood loss as predictors of poor outcome. We then
Neoadjuvant treatment 1.80 (0.78–4.16) 4.55 (1.53–13.57) attempted to determine the impact of the new grouping on predictive
Procedure
modeling by going back and using an unmodified Clavien‐Dindo based
Pylorus preservation 1.74 (1.07–2.84) 3.10 (1.48–6.50)
Vein resection 1.07 (0.49–2.35) 0.94 (0.31–2.84)
cut‐off (Grade IIIb and above) for predicted outcomes. In this analysis,
Procedure length age was statistically strongly associated with SAPO but not with the
<8 hr Referent Referent unmodified CD IIIb and above threshold. This ability to discriminate and
8 hr 1.63 (0.96–2.75) 2.33 (1.01–5.35) represent the importance of age in determining outcome, is the most
Estimated blood loss substantial difference we detected between the use of the new metric and
<1.5 L Referent Referent the unmodified Clavien‐Dindo system. This difference may be related in
1.5 L 2.52 (1.25–5.07) 3.26 (1.26–8.46) part to an effect advanced age has on post‐complication length of stay in
 general and become apparent because length of stay is such a central
SAPO, severe adverse postoperative outcome; OR, odds ratio; CI, confidence
interval; BMI, body mass index; CAD, coronary artery disease; COPD, chronic
component of the definition of poor outcome in the modified metric but
obstructive pulmonary disease; CHF, congestive heart failure. the modified metric almost certainly better represents the way elderly
people recover following PD.
We have not, in this study, attempted to use the modified metric to
predict medium or long‐term oncologic outcomes such as disease
DISCUSSION specific mortality, disease progression, or delays to chemotherapy. The
real clinical value of the metric may ultimately be determined by the
The Clavien‐Dindo grading system has been the system most widely ability of the modified metric to predict such medium and long‐term
used to evaluate early postoperative outcomes following PD. When oncologic outcomes. The current study was meant only to define the
graded by the Clavien‐Dindo system, the most prevalent complications metric for perioperative outomes and determine predictors of poor
following PD are typically those that fall in the grade II and IIIa perioperative outcome. There is substantial clinical utility in this effort
categories. For patients undergoing PD, complications that fall in this alone. We can, as clinicians, now more completely advise patients
range have substantial clinical variability and thus have potential for regarding the likely nature of their recovery, their perioperative risk of
error in representing the significance of the outcome—the real clinical significant morbidity, and better identify appropriate candidates for
burden of the complication. Most surgeons would agree that surgery. Future work will be directed at determining the ability of the
complications requiring a return to the operating room (CD grade modified metric to predict disease specific oncologic outcomes.
IIIb), a life threatening event (CD grade IV) or death (CD grade V) are There have been relatively few prior attempts to use the Clavien‐
relatively poor outcomes and to be avoided if possible. However, some Dindo system to identify predictors or correlates of poor quality
CD grade II and IIIa complications (delayed emptying requiring weeks outcomes after PD in multivariate models. The original use of the
of TPN or tube feedings, pancreatic fistula requiring three or four IR Clavien‐Dindo system in PD did attempt to develop a multivariate model
drainage procedures) should probably be considered to be relatively predicting severe complications [1]. There was, however, no description
poor outcomes while other grade II and IIIa complications (antibiotics of how a severe complication was defined. There has been one recent
for cellulitis at the wound, a single IR aspiration of a sterile effort to develop a prognostic score for major complication following PD
peripancreatic collection) should probably be considered more using the Clavien‐Dindo system as the complication grading system [4].
acceptable or better outcomes. Given this, predictive models using an The factors associated with major complication in that study were
unmodified Clavien grade IIIa or IIIb as a threshold for poor outcome for pancreatic texture and blood loss >700 ml. This paper defined major
patients undergoing PD would be expected to predict an inappropriate complication as Clavien‐Dindo grade IIIa and above. Several other
range of outcomes and fail to be clinically useful. studies have represented the incidence of complications following
We developed a modification of the Clavien‐Dindo system using two pancreaticoduodenecotmy by severity grade using the Clavien‐Dindo
easily obtainable measures (prolonged lengths of stay counting readmissions method but not done regression modeling. Most of these studies have
and the need for multiple non‐surgical invasive interventions) to better taken grade IIIa and above together as high severity complications [5],
represent the clinical impact of the complication on the patient and better others have considered grade IIIb and above as severe. Again, our results
identify which Clavien‐Dindo grade II and grade IIIa outcomes were indicate that both may be inappropriate cutoffs for defining poor
relatively poor outcomes. These measures were selected because they were outcome. Our new metric discriminates severity among CD grade II and
objective indicators of the impact of the complication, they were IIIa complications and was used here to identify age, significant
quantitative, easily obtainable, difficult to misrepresent and could be intraoperative blood loss and pylorus preservation as factor associated
standardized to the subpopulation of PD patients that were recovering from with this new definition of poor outcome.

Journal of Surgical Oncology


Many prior studies have examined the impact of advanced age on
outcomes following PD [6–9]. The more recent studies among these
have generally concluded that operating on selected elderly patients is
reasonable and safe with overall rates of morbidity that are similar to
those who are younger [6,7]. Several population based studies have
demonstrated increased rates of postoperative mortality in the elderly
and increased rates of prolonged postoperative institutionalization
among elderly patients after PD [8,9]. The bulk of the data suggest that
recovery after PD is harder in patients of advanced ages but that it can be
done with reasonable outcomes in well selected patients.
Our study links age to the incidence of severe complication in a
multivariate analysis that controls for relevant comorbid conditions,
obesity and major vascular resection. The older patients in our series
represent a selected group of fit elderly patients. To our knowledge ours
is the first study to demonstrate that advanced age is an independent
predictor of severe complication in a multivariate model attempting to
determine predictors of graded complications following PD. We
demonstrate that patients older than 75 are almost 3 times as likely as
those under age 75 to experience a severe adverse outcome. Even if there
is a relation between age and length of stay in general, advanced age in
this model does predict the incidence of severe complication. This
finding indicates that patients of advanced age are more likely to suffer
major morbidity following PD and should be counseled regarding this
risk prior to surgery. Currently, we do not absolutely preclude
performing PD at or above any discrete age cut‐off but do now use
the data obtained in this study to more concretely describe the possible
adverse perioperative outcomes for patients who are elderly so that they
can make informed choices regarding their care. We do also use the data
to make better informed judgments regarding who is appropriate for
resection versus preoperative or definitive chemoradiotherapy.
Many previous studies have examined the effects of pylorus
preservation on short‐term outcomes following PD [10–12]. Several
have examined the effect of pylorus preservation on delayed gastric
emptying as defined according to international consensus guidelines for
grading delayed emptying. None have graded outcomes according to the
Clavien‐Dindo system. The most recent randomized trial demonstrated
an increase in the incidence of mild and moderate delayed gastric
emptying with pylorus preservation but no difference in severe delayed
emptying as defined by the consensus guidelines of the international
study group on delayed emptying [11]. Our results demonstrate that
pylorus preservation is associated with severe adverse outcome when
Clavien‐Dindo is adjusted in our metric for overall length of stay. We
have no indication as to the underlying reason. It may be that pylorus
preservation leads to a delayed return of gastric function and thus to
prolonged hospitalization. It may also be that this effect represents an
outcome related to a learning curve for pylorus preservation in our
experience. In effort to clarify this, we have further broken down
outcome into two eras, first 5 years versus last 5 (data not shown). There
was no statistical difference in the incidence of SAPO by era in our
series.
Several prior studies have demonstrated the relevance of transfusion
to long‐term outcomes in patients with cancer [13–17]. To our knowledge
there has been only one study that has correlated operative blood loss to
graded short‐term outcomes following PD. This demonstrated an
association between blood loss greater than 700 ml and Clavien grade
IIIa and above complications [4]. We have examined transfusion
requirement in this study finding no correlation between transfusion itself
and SAPO in our univeriate analysis. Our findings correlating more
significant operative blood loss with clinical outcome do further support
the contention that a major intra‐operative bleeding event can impact the
short‐term morbidity after PD. Further work is needed to determine
the identifiable patient characteristics associated with this risk of
intraoperative bleeding.
Several recent studies have examined readmissions following PD and
identified the link between costs and readmissions [18–21]. Some of
Journal of Surgical Oncology
Quality for Pancreaticoduodenectomy 405
these studies have identified complications as the strongest predictors of
readmission [18,19]. Few of these studies have looked for modifiable
risk factors that are associated with post‐discharge readmission or
quality metrics accounting for readmission and thus factors that would
be potential targets for prospective quality improvement initiatives
aimed at improving rates of readmission. We attempted to incorporate
the readmission data into our outcome metric and to use preoperative
demographic and clinical information, intraoperative clinical outcomes,
and pathologic variables to predict outcome. In this way, we have
attempted to identify variables that have the potential to either be
modified prior to surgery or used to make better preoperative decisions
regarding who is treated with PD. The best example of this is the finding
of the association between age and SAPO. Again, we have defined no
absolute limits on ages for which resection is not recommended but do
now counsel elderly patients differently and use the outcomes data
generated here to make relativistic decisions on who is more
appropriately treated with resection versus chemoradiotherapy.
There are several limitations to our study. The most important is that
it is by nature a retrospective review and subject to omitted variable bias
and the possibility of imperfect clinical follow up. Many of the patients
in our series are referred from health systems outside of our own network
and may, in part, recover from surgery in the region but outside of our
network. To compensate for this, we employ a dedicated clinical
research nurse who is responsible only for maintaining the clinical
database. She makes great effort to be in direct contact with the
pancreatic patients on a continuous basis throughout their recoveries.
She documents details associated with all readmissions to facilities that
are outside our network. Because of this, we feel confident that there are
very few cases where patients are readmitted without our knowledge.
The second limitation is that this work used a relatively arbitrary metric
of poor outcome. We believe it is a clinically relevant metric, that it
better reflects the clinical recovery of patients after PD than currently
established grading systems. It was defined to be one that was consistent
with what we were observing in clinical practice. It has been our
impression that patients who are hospitalized for longer than 2 weeks are
in the midst of a substantial complication that has the potential to delay
adjuvant therapy. The metric was also defined to be one that is easy to
capture reliably and has the potential to drive quality improvement. The
metric is based on obtainable and hard to misrepresent data: overall
length of stay and number of interventional procedures. It is also defined
relative to the subpopulation of PD patients that had no complication. In
this way it is specific to our PD population and may be used to assess
quality improvement efforts on an ongoing basis.
CONCLUSIONS
Established grading systems may under grade the severity of
complications following PD. Using a procedure specific metric for
outcome quality, we demonstrate that advanced age, pylorus preservation,
and excessive operative blood loss are significant independent predictors
of poor quality outcome following PD. Such procedure‐specific systems
may have a better capacity to determine meaningful differences in
outcomes between modifications of complex gastrointestinal procedures
and to drive quality improvement than currently established complications
grading systems.
REFERENCES
1. DeOliveira ML, Winter JM, Schafer M, et al.: Assessment of
complications after pancreatic surgery: A novel grading system
applied to 633 patients undergoing pancreaticoduodenectomy. Ann
Surg 2006;244:931–939.
2. Dindo D, Demartines N, Clavien PA: Classification of surgical
complications: A new proposal with evaluation in a cohort of 6336
patients and results of survey. Ann Surg 2004;240:205–213.
406 Baker et al.
3. Clavien PA, Barkun J, Oliveira ML, et al.: The Clavien‐Dindo
classification of surgical complications: Five‐year experience. Ann
Surg 2009;250:187–196.
4. Braga M, Capretti G, Pecorelli N, et al.: A prognostic score to
predict major complications after pancreaticoduodenectomy. Ann
Surg 2011;254:702–708.
5. Assifi MM, Lindenmeyer J, Leiby BE, et al.: Surgical Apgar Score
Predicts Perioperative Morbidity in Patients Undergoing Pancrea-
ticoduodenectomy at a High‐Volume Center. J Gastrointest Surg
2012;16:275–281.
6. Lee MK, DiNorcia J, Reavey PL, et al.: Pancreaticoduodenectomy
can be performed safely in patients aged 80 years and older. J
Gastrointest Surg 2010;14:1838–1846.
7. de Franco V, Frampas E, Wong M, et al.: Safety and feasibility of
pancreaticoduodenectomy in the elderly: A matched study.
Pancreas 2011;40:920–924.
8. Riall TS, Reddy DM, Nealon WH, et al.: The effect of age on short‐
term outcomes after pancreatic resection: A population‐based study.
Ann Surg 2008;248:459–467.
9. Lightner AM, Glasgo RE, Jordan TH, et al.: Pancreatic resection in
the elderly. J Am Coll Surg 2004;198:697–706.
10. Tran KTC, Smeenk HG, Eijck CHJ, et al.: Pylorus preserving
pancreaticoduodenectomy versus standard whipple procedure: A
prospective, randomized, multicenter analysis of 170 patients
with pancreatic and periampullary tumors. Ann Surg 2004;240:
738–745.
11. Kawai M, Tani M, Hirono S, et al.: Pylorus ring resection reduces
delayed gastric emptying in patients undergoing pancreatoduode-
nectomy: A prospective, randomized, controlled trial of pylorus‐
resecting versus pylorus‐preserving pancreatoduodenectomy. Ann
Sug 2011;253:495–501.
Journal of Surgical Oncology
12. Seiler CA, Wagner M, Sadowski C, et al.: Randomized prospective
trial of pylorus‐preserving vs. classic duodenopancreatectomy.
J Gastrointest Surg 2000;4:443–452.
13. Yeh JJ, Gonen M, Tomlinson JS, et al.: Effect of blood transfusion
on outcome after pancreaticoduodenectomy for exocrine tumor of
the pancreas. Br J Surg 2007;94:466–472.
14. Acheson AG, Brookes MJ, Spahn DR: Effects of allogeneic red
blood cell transfusions on clinical outcomes in patients undergoing
colorectal cancer surgery: A systematic review and meta‐analysis.
Ann Surg 2012;256:235–244.
15. Al‐Refaie WB, Parsons HM, Markin A, et al.: Blood transfusion and
cancer surgery outcomes: A continued reason for concern. Surgery
2012;152:344–354.
16. Kooby DA, Stockman J, Ben‐Porat L, et al.: Influence of transfusions
on perioperative and long‐term outcome in patients following hepatic
resection for colorectal metastases. Ann Surg 2003;237:860–870.
17. Kneuertz PJ, Patel SH, Chu CK, et al.: Effects of perioperative red
blood cell transfuion on disease recurrence and survival after
pancreaticoduodenectomy for ductal adenocarcinoma. Ann Surg
Oncol 2011;18:1327–1334.
18. Kent TS, Sachs TE, Callery MP, et al.: Readmission after major
pancreatic resection: A necessary evil? J Am Coll Surg 2011;213:
515–523.
19. Emick DM, Riall TS, Cameron JL, et al.: Hosptial readmission after
pancreaticoduodenectomy. J Gastrointest Surg 2006;10:1243–1253.
20. Yermilov I, Bentrem D, Sekeris E, et al.: Readmissions following
pancreaticoduodenectomy for pancreas cancer: A population‐based
appraisal. Ann Surg Oncol 2009;16:554–561.
21. Reddy DM, Townsend CM, Kuo YF, et al.: Readmission after
pancreatectomy for pancreatic cancer in medicare patients. J
Gastrointest Surg 2009;13:1963–1975.