- Email: [email protected]
Assessment of the learning curve for pancreaticoduodenectomy
The American Journal of Surgery (2012) 203, 684 – 690
Clinical Science
Assessment of the learning curve for pancreaticoduodenectomy William E. Fisher, M.D., F.A.C.S.*, Sally E. Hodges, Meng-Fen Wu, Susan G. Hilsenbeck, Ph.D., F. Charles Brunicardi, M.D. The Elkins Pancreas Center, Michael E DeBakey Department of Surgery, Baylor College of Medicine, 1709 Dryden, Suite 1500, Houston, TX 77030, USA KEYWORDS: Pancreaticoduodenectomy; High-volume; Learning curve; Training
Abstract BACKGROUND: Experience with the Whipple procedure has been associated with improved outcomes, but the learning curve for this complex procedure is not well defined. METHODS: Outcomes with 162 consecutive Whipple procedures during the 1st 11.5 years of practice was documented in a prospective database. A period of low (ⱕ11/y) and high (ⱖ23/y) case volume was compared using the Wilcoxon rank-sum test and Fisher exact test. RESULTS: With low case volume, blood loss was higher (800 vs 400 mL, P ⫽ .001), more patients were transfused (44% vs 18%, P ⫽ .027), there were more complications (58% vs 46%, P ⫽ .0337), and a longer length of stay (10 vs 7 days, P ⫽ .006). There was only 1 mortality (.7%). CONCLUSIONS: Frequent repetition of the Whipple procedure is associated with an improvement in quantifiable quality benchmarks, and improvement continues with extensive experience. However, with proper training and the right environment, this procedure can be performed during the learning curve with acceptable outcomes. © 2012 Elsevier Inc. All rights reserved.
The 1st successful pancreaticoduodenectomy was performed in 1909 in Germany by Kausch1 and reported in 1912. In 1935, Whipple et al2 reported their series of 3 patients in which 2 survived the operation. By the end of Whipple’s career, he had only performed 37 pancreaticoduodenectomies.3 Few surgeons attempted the procedure in the 1960s and 1970s because of a hospital mortality in the range of 25%. However, in the 1980s and 1990s, experience increased, and reports from high-volume centers of operative mortality below 5% began to appear in the literature.4 – 6 Increasing numbers of surgeons trained in the 1980s to 1990s have developed significant experiences with the procedure. A relation * Corresponding author. Tel.: ⫹1-713-798-8070; fax: ⫹1-713-798-4530. E-mail address: [email protected] Manuscript received September 5, 2010; revised manuscript January 3, 2011
0002-9610/$ - see front matter © 2012 Elsevier Inc. All rights reserved. doi:10.1016/j.amjsurg.2011.05.006
between not only hospital but also individual surgeon volume and outcome has been reported with the Whipple procedure. Lieberman et al7 studied 2,233 pancreatoduodenectomies or total pancreatectomies performed in New York State from 1984 to 1991. The hospital mortality rate was 19% in hospitals in which fewer than 10 resections were done during the period, 12% in which 10 to 50 resections were done, 13% in which 51 to 80 operations were performed, and 6% in hospitals doing more than 81 operations. These authors also found a higher operative mortality rate for low-volume surgeons (⬍ 9 resections, 16%) compared with high-volume surgeons (⬎ 41 resections, 5%). Birkmeyer et al8 also showed that individual surgeon volume is inversely related to operative mortality for the Whipple procedure. The finding that surgeons with higher volumes have better outcomes suggests the presence of a learning curve for the Whipple procedure.
W.E. Fisher et al.
Learning curve for pancreaticoduodenectomy
However, little is known about the outcome of pancreaticoduodenectomy as performed by an individual surgeon in the early phase of practice. This report documents the learning curve of a surgeon performing his 1st 162 consecutive pancreaticoduodenectomies during the first 11.5 years of independent practice.
Materials and Methods The surgeon performed 12 Whipple procedures during the chief residency year of a general surgery residency training program in a center with a high volume of pancreatic surgery. Data from every Whipple procedure performed beginning with the 1st day of academic practice after residency training for a single surgeon was reviewed. Patients who underwent pancreatic operations other than a Whipple procedure (distal, central, or total pancreatectomy) were excluded. The analysis was based on data from an institutional review board–approved prospective database and from the review of patient medical records. This prospective electronic web-based database tracks data on patient demographics, clinical history, past medical history, family and social history, physical findings, diagnostic tests, therapeutic interventions, complications, pathological data, outcomes including perioperative mortality (30-day or in-hospital mortality), and long-term survival for all patients seen in the center. Presenting symptoms were recorded included jaundice, pain, weight loss (quantified in pounds), nausea, vomiting, and diarrhea. A stated past medical history of or presence in the medical record of a history of hypertension, chronic obstructive pulmonary disease, diabetes, coronary artery disease, chronic pancreatitis, or renal insufficiency was recorded. Obesity was defined as body mass index greater than or equal to 30 kg/m2. Tobacco use was recorded in pack years (packs per day multiplied by the number of years smoked). The American Society of Anesthesiologists score was obtained from the anesthesia record. The operative time was also obtained from the anesthesia record and defined as the time from incision to application of the final wound dressing. Estimated intraoperative blood loss (EBL) was obtained from the anesthesia record and not from the surgeon’s operative report. The surgeon graded the pancreatic texture as soft/normal or hard in the operative note. The surgeon measured the pancreatic duct with a probe and recorded the size of the duct in the operative note. All specimens were submitted to pathological analysis. For cancer cases, tumor stage, the number of positive lymph nodes and total number examined, tumor grade, presence of lymphovascular or perineural invasion, and resection margin status were recorded. Later in the series, the authors engaged the hospital pathologists to implement a standardized protocol to orient the specimen, ink the margins, and section the specimen for determination of margin status. This may have resulted in an increase in R1 resection designation in the high-volume era.
685
All complications within 30 days of the date of surgery were meticulously recorded using specific and standardized definitions. We defined pancreatic fistula using the International Study Group on Pancreatic Fistula definition.9 Wound infection was defined as spontaneous or surgical drainage from the wound with a positive gram stain or culture. Gastroparesis was defined as failure to consistently tolerate solid oral intake or presence of a nasogastric tube on or after postoperative day 10. Fever was defined as any recorded temperature greater than or equal to 102.5°F. Pneumonia was defined as a positive sputum culture associated with an infiltrate on radiologic imaging requiring treatment with antibiotics. Deep venous thrombosis and pulmonary embolus were diagnosed with ultrasound or computed tomography (CT) scans. Arrhythmia was defined as any new cardiac rhythm other than sinus requiring any medical intervention or transfer to a monitored bed. Myocardial infarction was defined as 2 or more of the following: chest pain, echocardiographic changes, and/or cardiac enzyme elevation and fall consistent with myocardial infarction. Urinary tract infection was defined as urine culture with ⱖ103 colony-forming units per mL. Postoperative hemorrhage was defined as the need to return to the operating room or postoperative radiologic intervention for hemorrhage or postoperative gastrointestinal bleeding documented by endoscopy. Renal failure was defined as the need for dialysis of any duration. Biliary leak was defined as the drainage of any volume of fluid clinically consistent with bile from operatively placed drains or the wound with a bilirubin concentration greater than the serum value. Reoperation was defined as any unplanned operation within 30 days of the Whipple procedure. The length of hospital stay was calculated from the day of surgery through and including the day of discharge. Readmission was defined as an admission to any hospital within 30 days after initial discharge after surgery. Operative mortality was defined as any death within 30 days of surgery. Data were entered into the database in real time by a trained data analyst under the supervision of the surgeon. All data were backed-up by source documents, and accuracy of the data entered into the electronic database was periodically reviewed. The comparison of continuous variables was performed using the Wilcoxon rank-sum test, and the comparison of categorical variables was performed using Fisher exact test. Survival data were analyzed using the Kaplan-Meier method. Results are reported as median values unless indicated otherwise. Statistical significance was accepted for P ⬍ .05.
Results Preoperative data One surgeon performed 162 Whipple procedures in the first 11.5 years of independent practice. The annual distri-
686
The American Journal of Surgery, Vol 203, No 6, June 2012 Table 2
Presenting signs and symptoms
Abdominal pain Weight loss Jaundice Nausea Vomiting Anorexia Diarrhea
Low-volume (n ⫽ 19) (%)
High-volume (n ⫽ 143) (%)
P value
11 13 8 6 1 0 5
92 72 56 45 24 19 15
.617 .324 .808 1.000 .312 .132 .063
(58) (68) (42) (32) (5) (0) (26)
(64) (54) (39) (32) (17) (13) (11)
There were no significant differences in clinical presentation between the 2 cohorts.
Figure 1 The annual distribution of Whipple procedures performed by 1 surgeon in the 1st 11.5 years of practice.
bution of procedures appears in Figure 1. In the 1st 6.5 years, volume was low with 0 to 11 Whipple procedures annually. In the next 5 years, volume increased to ⬎22 Whipple procedures per year, with ⬎30 per year in the most recent years (Fig. 1). Data are compared between 2 time periods (low volume and high volume), dividing the time periods before and after achieving a volume of ⬎11 pancreaticoduodenectomies/year. Demographic data for the 2 groups is reported in Table 1. The median age was 55 in the low-volume era and 65 in the high-volume era (P ⫽ .033). Ages ranged from 31 years to 85 years. Ten patients were in their 80s. A total of 142 patients (88%) were white, 16 (10%) were black, and 4 (2%) were Asian or Pacific Islander. Nineteen (12%) were of the Hispanic ethnicity. There were no statistically significant differences in demographics between the 2 eras other than age. There were no clinically significant differences between the low- and high-volume eras in presenting symptoms, comorbid
Table 1
Demographics
Age (median, interquartile range) Sex (%) Male Race White Black Asian Pacific islander Ethnicity (%) Hispanic
Low volume (n ⫽ 19)
High volume (n ⫽ 143)
P value
55 (49–66)
65 (56–73)
.033
12 (63)
58 (41)
.084
16 3 0 0 1
(84) (16) (0) (0) (5)
126 13 1 3 18
(88) (9) (1) (2) (13)
.645
.702
Patients in the low-volume era tended to be younger. There were no other differences between the 2 cohorts.
conditions, or preoperative laboratories (Tables 2 and 3). Presenting symptoms included abdominal pain in 64%, weight loss in 52%, and jaundice in 40%. There were no differences in the median preoperative hemoglobin, bilirubin, creatinine, albumin, carcinoembryonic antigen (CEA), or cancer antigen 19-9. ASA scores and comorbid conditions were also similar in the 2 time periods. There was no difference in the smoking or alcohol use history between the 2 groups. Computed tomography scans were performed in almost every patient with magnetic resonance imaging being used to evaluate the pathology in only 4%. A total of 82 (53%) had preoperative biliary stents placed. These were almost exclusively placed endoscopically and usually before referral to our center. Table 3
Comorbid conditions and preoperative laboratories Low volume (n ⫽ 19) (%)
HTN 8 Diabetes 6 Chronic pancreatitis 5 CAD 6 COPD 2 Renal insufficiency 1 MI 1 Obesity (BMI ⬎30) 7 ASA ⱖ3 14 Smoking pack years 25 Yes/no 8 Active EtOH 7 Total bilirubin 1.6 Creatinine .8 Albumin 3.8 Hemoglobin 12.9 CEA 2.3 CA 19–9 70
High volume (n ⫽ 143) (%)
(42) (32) (26) (32) (11) (5) (5) (39) (74)
76 34 31 23 7 5 3 30 102
(53) (24) (22) (16) (5) (4) (2) (21) (71)
(42) (37) (.6–5.1) (.7–.9) (3.7–4.1) (12.7–13.7) (.6–4.6) (17–204)
25 79 50 .8 .9 4.1 12.6 1.5 25.2
(59) (35) (.4–2.4) (.8–1.0) (3.7–4.5) (11.5–13.9) (.7–2.5) (9–63)
P value* .465 .571 .769 .114 .285 .533 .396 .134 .942 .219 1.000 .073 .019 .127 .450 .372 .288
HTN ⫽ hypertension; CAD ⫽ coronary artery disease; COPD ⫽ chronic obstructive pulmonary disorder; MI ⫽ myocardial infarction; EtOH ⫽ ethanol ; CEA ⫽ Carcinoembryonic antigen. There were no significant differences in the comorbid conditions or preoperative laboratories between the 2 cohorts. *Fisher exact text for categoric variables and Wilcoxon rank-sum test for continuous variables (presented as median, interquartile range).
W.E. Fisher et al. Table 4
Learning curve for pancreaticoduodenectomy
Intraoperative data
PPPD Vein resection Soft pancreas PD ⬎3 mm PD stent BD stent EBL (mL) Transfusion (Y/N) OR time
Low volume (n ⫽ 19)
High volume (n ⫽ 143)
16 1 4 6 1 0 800 8
130 16 71 74 47 5 400 26
(84) (7) (22) (43) (6) (0) (500–1,000) (44)
535 (444–551)
P value*
(92) .391 (11) 1.000 (54) .022 (55) .415 (33) .015 (4) 1.000 (200–700) .001 (18) .027
444 (375–531)
.090
The surgeon characterized the pancreas as soft more frequently and used in internal pancreatic duct stent more frequently in the highvolume era. There was a significant decrease in intraoperative blood loss and transfusions in the high-volume era. PPPD ⫽ pylorus-preserving pancreaticoduodenectomy; PD ⫽ pancreatic duct; BD ⫽ bile duct; EBL ⫽ estimated blood loss. *Fisher exact test for categoric variables and Wilcoxon rank-sum test for continuous variables (median, interquartile range).
Intraoperative data Intraoperative data are presented in Table 4. A pyloruspreserving pancreaticoduodenectomy was performed in 90% with the classic Whipple procedure being reserved for patients with tumors encroaching near the pylorus. More portal or superior mesenteric vein resections were performed in the high-volume era, but this did not reach statistical significance. Seven (41%) were repaired primarily, and 10 (59%) required an internal jugular vein interposition graft. All cases were performed by the attending surgeon (W.E.F.) who allowed an assistant, usually a postgraduate year 5 general surgery resident, to perform portions of the procedure as appropriate. In cases of vascular involvement, an experienced vascular surgeon rather than a resident assisted with this portion of the case. All patients underwent reconstruction with a retrocolic pancreaticojejunostomy. One hundred fifty (97%) were performed with an end-to-side, duct-to-mucosa technique with an outer layer of 4-0 silk sutures and an inner layer of 6-0 polydioxanone suture (PDS) absorbable sutures. The texture of the gland was graded by the surgeon as soft more frequently in the high-volume era (54% vs 22%, P ⫽ .022), but there was no difference in pancreatic duct size between the 2 cohorts. An internal pancreatic duct stent (5F pediatric feeding tube cut to 3 cm in length) was used more frequently in the high-volume era (33% vs 6%, P ⫽ .015). Infrequently, in 5 cases (3%), an end-to-end intussuscepted technique was used when the pancreatic duct could not be identified. The biliary anastomosis was performed with a single layer of interrupted absorbable sutures. The afferent limb was brought antecolic for the duodenojejunostomy. This anastomosis was performed in a two-layer hand-sewn technique with an outer layer of interrupted 3-0 silk Lembert sutures and an inner running layer of 3-0 PDS suture.
687
The median operative time decreased from 535 minutes (8 hours 55 minutes) to 444 minutes (7 hours 24 minutes) from the low-volume to high-volume era, but this was not statistically significant. EBL was higher in the low-volume era, (800 vs 400 mL, P ⫽ .001), and more patients were transfused (8 [44%] vs 26 [18%], P ⫽ .027).
Postoperative data Pathological diagnoses are reported in Table 5. A total of 56 (35%) had pancreatic adenocarcinoma, and the remaining patients had ampullary carcinoma, cholangiocarcinoma, neuroendocrine cancer, intraductal papillary mucinous neoplasm, mucinous cystic neoplasm, cystadenoma, chronic pancreatitis, or an ampullary adenoma. There were no statistically significant differences in the indications for surgery between the 2 eras. Among the patients with pancreatic adenocarcinoma, there were no differences between the low-volume and high-volume eras in terms of tumor size, lymph node metastases, stage of disease, histological grade, or the presence of lymphovascular or perineural invasion (Table 6). There was a statistically significant increase in the number of lymph nodes harvested in the high-volume era (17 vs 8, P ⫽ .0003) but no difference in the number of positive nodes removed. The increase in R1 resections did not reach statistical significance. However, the implementation of a standardized method of orientation, inking, and sectioning of the specimen during the high-volume era may have confounded the determination of R1 margin status. The number of R2 resections decreased significantly with increasing experience (2 [33%] vs 1 [2%], P ⫽ .031). However, among the patients with pancreatic adenocarcinoma, there was no statistically significant difference in 2- or 5-year survival between the low- and high-volume eras. However, there were only 6 patients with pancreatic adenocarcinoma in the lowvolume era, making statistical comparisons difficult. Table 5
Diagnoses Low volume (n ⫽ 19) (%)
High volume (n ⫽ 143) (%)
9 (47) 6 (67) 3 (33)
72 50 18 4 8 26 19 1 5 1 21 8 8
P value .128
Adenocarcinoma Pancreatic Ampullary Other Neuroendocrine Cystic neoplasm IPMN MCN Cystadenoma Other Pancreatitis Adenoma Other
0 1 (5)
6 (32) 3 (16) 1 (5)
(50) (69) (25) (6) (6) (18) (76) (4) (16) (4) (15) (6) (6)
Cystic neoplasms comprised a higher percentage of the patients in the high-volume era, but this difference was not statistically significant.
688 Table 6
The American Journal of Surgery, Vol 203, No 6, June 2012 Pathology for pancreatic adenocarcinoma cases Low volume High volume P (n ⫽ 6) (n ⫽ 50) value*
Tumor size Lymph nodes number examined number positive Tumor stage Stage 0 Stage IA Stage IIA Stage IB Stage IIB Histological grade G1 (well differentiated) G2 (moderately differentiated) G3 (poorly) Lymphovascular invasion Perineural invasion Resection margin status R0 R1 R2 Survival (%) 2y 5y
3 (2.5–4.2)
3 (2.5–3.5)
5 (1–8) 1 (0-3)
17 (14–26) 2 (2.4–5)
0 0 1 0 5
0 2 8 3 37
(0) (0) (17) (0) (83)
.770 .001 .291 1.000
(0) (4) (16) (6) (74) 1.000
0 (0) 4 (67)
3 (6) 30 (61)
2 (33) 0 (0) 2 (33)
16 (33) 31 (62) 37 (74)
4 (67) 0 (0) 2 (33)
36 (72) 13 (26) 1 (2)
50 17
41 19
.06 .062 .014 .031 .961
Among the patients with pancreatic adenocarcinoma, there were fewer R2 resections as the surgeon gained experience. *Fisher exact test for categoric variables and Wilcoxon rank-sum test for continuous variables (median, interquartile range).
The exact number of cases required to attain proficiency is unknown and likely varies somewhat for individual surgeons, but a large experience may be needed. The most recent data available from the Accreditation Council for Graduate Medical Education indicates that general surgery residents perform a mean of 4 Whipple procedures during their training.10 Recently, there has been an expansion in pancreatic fellowships presumably because of effects of decreased resident work hours and a perception among training programs and trainees that additional operative experience is needed beyond general surgery residency to be a competent pancreatic surgeon. We sought to examine the learning curve associated with the Whipple procedure for an individual surgeon from the end of residency training through the 1st 11.5 years of independent academic practice to provide some insight into the volume of experience that may be necessary to achieve excellent outcomes. We found that measurable quality benchmarks such as blood loss, transfusion rate, complications, and length of stay all decreased significantly with increasing experience. The improvement may be more closely related to a frequency of greater than 11 resections per year rather than any specific career milestone number of total cases. It is difficult to provide a detailed summary of the surgical techniques and perioperative management stratTable 7
Complications and length of stay Low-volume High-volume P (n ⫽ 19) (n ⫽ 143) value*
Postoperative complications are summarized in Table 7. Patients in the low-volume era were more likely to develop one or more complications than patients in the high-volume era (11 [58%] vs 65 [46%], P ⫽ .0337). The most common complications were wound infection and pancreatic fistula. The rate of wound infection decreased (6 [32%] versus 13 [9%], P ⫽ .012) as did the rate of bile leak (2 [11%] vs 1 [.7%], P ⫽ .037) as experience was gained. Although the rate of most other common complications, such as pancreatic fistula, decreased in the high-volume era, there were no other statistically significant differences in specific complication rates between the 2 eras. The length of hospital stay decreased in the high-volume era (10 vs 7 days, P ⫽ .006). There was only 1 death in the series (high-volume era) for a mortality rate of .7% for the entire time period.
Comments A learning curve exists for most complex tasks involving innumerable variables and nuances. It is no surprise that a learning curve exists for the Whipple procedure, one of the most technically complex abdominal operations, which is typically performed on older patients with multiple comorbidities. Surgeon volume has already been shown to be associated with superior outcomes in pancreatic surgery.7,8
Any complication 11 (58) Pancreatic fistula (grade 4 (21) B/C) Wound infection 6 (32) Gastroparesis 2 (11) Readmission 1 (5) Fever 1 (5) New arrhythmia 0 (0) Reoperation 2 (11) Urinary tract infection 0 (0) Vein/artery thrombosis 0 (0) Hemorrhage 1 (5) Intra-abdominal abscess 1 (5) Adult respiratory distress 0 (0) syndrome Pneumonia 1 (5) Biliary Leak 2 (11) Myocardial infarction 0 (0) Renal failure 1 (5) Deep venous thrombosis 0 (0) Clostridium difficile colitis 0 (0) Thrombocytopenia 0 (0) Urinary retention 0 (0) Ascites 0 (0) Mortality (30-day) 0 (0) Length of stay (d) 10 (8–14)
65 (46) 19 (13) 13 6 4 4 4 2 3 3 2 2 2
.0337 .480
(9) (4) (3) (3) (3) (1) (2) (2) (1) (1) (1)
.012 .238 .469 .469 1.000 .068 1.000 .332 1.000 .314 1.000
1 (1) 1 (1) 1 (1) 1 (1) 1 (1) 1 (1) 1 (1) 1 (1) 1 (1) ⬍1 7 (7–11)
.222 .037 1.000 .222 1.000 1.000 1.000 1.000 1.000 1.000 .006
The morbidity of the Whipple procedure was significantly decreased in the high-volume era. *Fisher exact test for categoric variables and Wilcoxon rank-sum test for continuous variables (median, interquartile range).
W.E. Fisher et al.
Learning curve for pancreaticoduodenectomy
egies that lead to improved outcomes so that these improvements could be adopted by other surgeons at other centers. The problem lies in the sheer complexity of the preoperative assessment, patient preparation, surgical techniques, and perioperative management, but this series offers some insight. Although not shown in the patient demographics or comorbidities documented in this study, improved patient selection for surgery including accurate interpretation of preoperative CT scans and wise recognition of unacceptably severe comorbidities and performance status certainly played a role in improving outcomes in this series. All patients in this series with a history of cardiopulmonary risk factors were referred to medical specialists for preoperative consultation. A preoperative review of imaging studies with an experienced radiologist and repeated review just before each case helped a low-volume surgeon avoid pitfalls while simultaneously gaining personal skill in CT interpretation. An improved knowledge of anatomy and its variations, recognition of common errors, and improvements in operative technique certainly contributed as well. In particular, detailed knowledge of the anatomy of the celiac axis and variations in the origin of the hepatic arteries, the anatomy of the superior mesenteric artery and its branches, and detailed knowledge of the variations in mesenteric venous anatomy help the pancreatic surgeon avoid intraoperative blood loss. A strategy for the creation of an excellent pancreaticojejunostomy is also likely important because pancreatic leak is associated with additional complications. At least 2 modifications in the pancreaticojejunostomy technique were made in this series. Increased mobilization of the pancreatic neck for at least 2 cm by ligating small branches of the proximal splenic vein and artery allows circumferential placement of interrupted sutures. An internal pancreatic duct stent may be useful not so much to control postoperative leak but to ensure accurate placement of duct-to-mucosa sutures when the pancreatic duct is small (⬍3 mm). The development of clinical pathways and multidisciplinary care teams can certainly have an impact on outcomes, particularly the length of stay. Our study is limited by the fact that simultaneously with the achievement of high annual case volume, the surgeon implemented numerous changes in patient care delivery including a multidisciplinary pancreas clinic and tumor board and clinical care pathways. The relative impact of these programmatic infrastructure factors, the institutional learning curve, and surgical experience cannot be separated. Dissecting out the important factors that are responsible for improved outcomes seems nearly impossible, and it is likely that improved outcomes resulted from some combination of all these factors. The fact that this is a single-surgeon series presents both opportunities and problems with interpretation. Acknowledging the fact that institutions change over time, a multi-
689
center study would certainly introduce even more variables that could affect interpretation of the learning curve. Interpretation of a multi-surgeon study would be complicated by individual surgeon variables such as surgical ability and differences in the quality and quantity of previous training. The present study chronicles the experience of a single surgeon over a long period at 1 institution and does provide some insight into the changes that occurred over time that are responsible for improved outcomes. The learning curve may not be finished. Tseng et al11 reported an analysis of the 1st 120 cases performed by 3 surgeons who trained at M D Anderson Cancer Center. They observed a similar improvement in EBL, operating room time, and length of stay. The analysis of additional cases beyond 120 for one of the surgeons indicated the learning curve continues farther into a surgeon’s career. In this series, the long period of low volume likely lengthened the time required to progress through the learning curve. It also complicates interpretation of the data because the surgeon performed other complex operations during this period, and the translation of this experience into improved pancreatic surgery cannot be assessed. Ideally, a surgeon interested in pancreatic surgery should seek an environment in which a high volume of cases can be experienced immediately after training. However, no surgeon can start off his/her practice as an experienced high-volume pancreas surgeon. The learning curve for the Whipple procedure may be so lengthy that it would be extraordinarily difficult to progress completely through it even with an additional year or two of focused fellowship training. In addition to focused training, it may be critical to seek an environment with experienced colleagues and mentors who are willing to share not only their hardacquired expertise but also their referral base to ensure a steady volume of continued experience that will translate into rapid progress and excellent outcomes. However, even if willing, a mentor may not be able to share referrals because referring physicians and even patients may demand the more experienced partner. Performing cases together is one solution, but this has an impact on the junior surgeon’s productivity measures and limits the experience of other trainees. The mortality rate after the Whipple procedure is an outcome variable that is readily available. Mortality data can be derived from national databases as well as reports from expert centers and are probably quite reliable in terms of accuracy. As demonstrated in Table 8, we have reported 30-day mortality, not just in-hospital mortality. Some centers may under estimate the 30-day mortality if patients are lost to follow-up. The mortality rate in this series compares favorably with national data and meets the high benchmarks set by ultrahigh-volume centers of excellence.4 – 6,12–15 Of note is that no mortality occurred in the low-volume era, and the mortality rate for the overall series is in the acceptable range. Other data, aimed at assessing the morbidity of the procedure are less available, and variations in definitions,
690 Table 8
The American Journal of Surgery, Vol 203, No 6, June 2012 Comparison to high-volume centers Current series
Mortality Morbidity LOS (d) Reoperation OR time (min) EBL (mL)
Low volume
High volume
United States14
Hopkins1
MSK15
MDACC16
UCSFs
VM Clinic4
0% 58% 10 11% 535 800
⬍1% 46% 7 1% 444 400
9% 35% 15 NA NA NA
1% 41% 9 3% 330 700
2% 47% 14 9% 350 1,000
1% NR 14 1% 553 900
4% 59% 16 7% 402 1,167
0% NR 11 ⬍1% 450 382
NA ⫽ not applicable; LOS ⫽ length of stay; OR ⫽ operating room. Examination of the learning curve of a single surgeon with the Whipple procedure indicates that experience is associated with an improvement in measurable benchmarks. However, in the right environment, a low-volume surgeon can perform this procedure with acceptable morbidity and mortality during the lengthy learning curve.
differences in prospective and retrospective databases, and the tenacity with which every complication is documented can create problems with comparing different series. The present series was documented prospectively with extremely thorough accuracy so there should be no underreporting of the morbidity. However, there are still problems with definitions even with the simplest things like the length of stay. Even operating room time and EBL can be biased if not taken from the anesthesia record as done in this series. That being said, when we compare the present series with outcomes reported in the literature, the mortality, morbidity (percent of patients with any complication), the length of stay, the rate of reoperation, operating room time, and EBL all compare with other high-volume centers in both the low-volume and high-volume eras of the series. Therefore, this series suggests that although improvement occurs with experience and consistent high volume, an inexperienced, low-volume surgeon can embark on a career in pancreatic surgery with acceptable morbidity and mortality in the right environment. It should be emphasized that the contribution of experienced colleagues in anesthesia and perioperative care and all the support of a tertiary care academic teaching environment with a high volume of pancreatic cases certainly played a role in the outcomes achieved in this series.
Conclusions Examination of the learning curve of a single surgeon with the Whipple procedure indicates that experience is associated with decreased blood loss; number of transfusions; and complications, particularly wound infections; and a faster recovery. Strong motivation and interest in pancreatic surgery, a high-volume hospital environment, and partners capable and willing to serve as experienced mentors are necessary ingredients for a low-volume sur-
geon to perform this procedure with acceptable morbidity and mortality during the lengthy learning curve.
References 1. Kausch W. Das carcinoma der papilla duodeni und Seine Radikale entfeinung. Beitr Z Clin Chir 1912;78:439 – 86. 2. Whipple AO, Parsons WB, Mullins CR. Treatment of carcinoma of the ampulla of Vater. Ann Surg 1935;102:763–79. 3. Whipple AO. A reminiscence: pancreaticoduodenectomy. Rev Surg 1963;20:221–5. 4. Cameron JL, Riall TS, Coleman J, et al. One thousand consecutive pancreaticoduodenectomies. Ann Surg 2006;244:10 –5. 5. Winter JM, Cameron JL, Campbell KA, et al. Pancreaticoduodenectomies for pancreatic cancer: A single-institution experience. J Gastrointest Surg 1423;2006:1199 –211. 6. Cameron JL, Pitt JA, Yeo CJ, et al. One hundred and forty-five consecutive pancreaticoduodenectomies without mortality. Ann Surg 1993;217:430 – 8. 7. Lieberman MD, Kilburn H, Lindsey M, et al. Relation of perioperative deaths to hospital volume among patients undergoing pancreatic resections for malignancy. Ann Surg 1995;22:638 – 45. 8. Birkmeyer JD, Finlayson SRG, Tosteson ANA, et al. Effect of hospital volume on in-hospital mortality with pancreaticoduodenectomy. Surgery 1999;125:250 – 6. 9. Bassi C, Dervenis C, Butturini G, et al. Postoperative pancreatic fistula: an international study group (ISGPF) definition. Surgery 2005;138:8–13. 10. General Surgery (2005-2006): National Resident Report. Residency Review Committee. Accreditation Council for Graduate Medical Education. 11. Tseng JF, Pisters PWT, Lee JE, et al. The learning curve in pancreatic surgery. Surgery 2007;141:456 – 63. 12. Traverso LW, Shinchi H, Low DE. Useful benchmarks to evaluate outcomes after esophagectomy and pancreaticoduodenectomy. Am J Surg 2004;187: 604–8. 13. Trede M, Schwall G, Saeger HD. Survival after pancreaticoduodenecttomy: 118 consecutive resections without an operative mortality. Ann Surg 1990; 211:447–58. 14. Turaga K, Kaushik M, Forse RA, et al. In hospital outcomes after pancreatectomies: an analysis of a national database from 1996 –2004. J Surg Oncol 2008;98:156 – 60. 15. Schell MT, Barcia A, Spitzer AL, et al. Pancreaticoduodenectomy: volume is not associated with outcome within an academic health care system. HPB Surg 2008;2008:825940.
Clinical Science
Assessment of the learning curve for pancreaticoduodenectomy William E. Fisher, M.D., F.A.C.S.*, Sally E. Hodges, Meng-Fen Wu, Susan G. Hilsenbeck, Ph.D., F. Charles Brunicardi, M.D. The Elkins Pancreas Center, Michael E DeBakey Department of Surgery, Baylor College of Medicine, 1709 Dryden, Suite 1500, Houston, TX 77030, USA KEYWORDS: Pancreaticoduodenectomy; High-volume; Learning curve; Training
Abstract BACKGROUND: Experience with the Whipple procedure has been associated with improved outcomes, but the learning curve for this complex procedure is not well defined. METHODS: Outcomes with 162 consecutive Whipple procedures during the 1st 11.5 years of practice was documented in a prospective database. A period of low (ⱕ11/y) and high (ⱖ23/y) case volume was compared using the Wilcoxon rank-sum test and Fisher exact test. RESULTS: With low case volume, blood loss was higher (800 vs 400 mL, P ⫽ .001), more patients were transfused (44% vs 18%, P ⫽ .027), there were more complications (58% vs 46%, P ⫽ .0337), and a longer length of stay (10 vs 7 days, P ⫽ .006). There was only 1 mortality (.7%). CONCLUSIONS: Frequent repetition of the Whipple procedure is associated with an improvement in quantifiable quality benchmarks, and improvement continues with extensive experience. However, with proper training and the right environment, this procedure can be performed during the learning curve with acceptable outcomes. © 2012 Elsevier Inc. All rights reserved.
The 1st successful pancreaticoduodenectomy was performed in 1909 in Germany by Kausch1 and reported in 1912. In 1935, Whipple et al2 reported their series of 3 patients in which 2 survived the operation. By the end of Whipple’s career, he had only performed 37 pancreaticoduodenectomies.3 Few surgeons attempted the procedure in the 1960s and 1970s because of a hospital mortality in the range of 25%. However, in the 1980s and 1990s, experience increased, and reports from high-volume centers of operative mortality below 5% began to appear in the literature.4 – 6 Increasing numbers of surgeons trained in the 1980s to 1990s have developed significant experiences with the procedure. A relation * Corresponding author. Tel.: ⫹1-713-798-8070; fax: ⫹1-713-798-4530. E-mail address: [email protected] Manuscript received September 5, 2010; revised manuscript January 3, 2011
0002-9610/$ - see front matter © 2012 Elsevier Inc. All rights reserved. doi:10.1016/j.amjsurg.2011.05.006
between not only hospital but also individual surgeon volume and outcome has been reported with the Whipple procedure. Lieberman et al7 studied 2,233 pancreatoduodenectomies or total pancreatectomies performed in New York State from 1984 to 1991. The hospital mortality rate was 19% in hospitals in which fewer than 10 resections were done during the period, 12% in which 10 to 50 resections were done, 13% in which 51 to 80 operations were performed, and 6% in hospitals doing more than 81 operations. These authors also found a higher operative mortality rate for low-volume surgeons (⬍ 9 resections, 16%) compared with high-volume surgeons (⬎ 41 resections, 5%). Birkmeyer et al8 also showed that individual surgeon volume is inversely related to operative mortality for the Whipple procedure. The finding that surgeons with higher volumes have better outcomes suggests the presence of a learning curve for the Whipple procedure.
W.E. Fisher et al.
Learning curve for pancreaticoduodenectomy
However, little is known about the outcome of pancreaticoduodenectomy as performed by an individual surgeon in the early phase of practice. This report documents the learning curve of a surgeon performing his 1st 162 consecutive pancreaticoduodenectomies during the first 11.5 years of independent practice.
Materials and Methods The surgeon performed 12 Whipple procedures during the chief residency year of a general surgery residency training program in a center with a high volume of pancreatic surgery. Data from every Whipple procedure performed beginning with the 1st day of academic practice after residency training for a single surgeon was reviewed. Patients who underwent pancreatic operations other than a Whipple procedure (distal, central, or total pancreatectomy) were excluded. The analysis was based on data from an institutional review board–approved prospective database and from the review of patient medical records. This prospective electronic web-based database tracks data on patient demographics, clinical history, past medical history, family and social history, physical findings, diagnostic tests, therapeutic interventions, complications, pathological data, outcomes including perioperative mortality (30-day or in-hospital mortality), and long-term survival for all patients seen in the center. Presenting symptoms were recorded included jaundice, pain, weight loss (quantified in pounds), nausea, vomiting, and diarrhea. A stated past medical history of or presence in the medical record of a history of hypertension, chronic obstructive pulmonary disease, diabetes, coronary artery disease, chronic pancreatitis, or renal insufficiency was recorded. Obesity was defined as body mass index greater than or equal to 30 kg/m2. Tobacco use was recorded in pack years (packs per day multiplied by the number of years smoked). The American Society of Anesthesiologists score was obtained from the anesthesia record. The operative time was also obtained from the anesthesia record and defined as the time from incision to application of the final wound dressing. Estimated intraoperative blood loss (EBL) was obtained from the anesthesia record and not from the surgeon’s operative report. The surgeon graded the pancreatic texture as soft/normal or hard in the operative note. The surgeon measured the pancreatic duct with a probe and recorded the size of the duct in the operative note. All specimens were submitted to pathological analysis. For cancer cases, tumor stage, the number of positive lymph nodes and total number examined, tumor grade, presence of lymphovascular or perineural invasion, and resection margin status were recorded. Later in the series, the authors engaged the hospital pathologists to implement a standardized protocol to orient the specimen, ink the margins, and section the specimen for determination of margin status. This may have resulted in an increase in R1 resection designation in the high-volume era.
685
All complications within 30 days of the date of surgery were meticulously recorded using specific and standardized definitions. We defined pancreatic fistula using the International Study Group on Pancreatic Fistula definition.9 Wound infection was defined as spontaneous or surgical drainage from the wound with a positive gram stain or culture. Gastroparesis was defined as failure to consistently tolerate solid oral intake or presence of a nasogastric tube on or after postoperative day 10. Fever was defined as any recorded temperature greater than or equal to 102.5°F. Pneumonia was defined as a positive sputum culture associated with an infiltrate on radiologic imaging requiring treatment with antibiotics. Deep venous thrombosis and pulmonary embolus were diagnosed with ultrasound or computed tomography (CT) scans. Arrhythmia was defined as any new cardiac rhythm other than sinus requiring any medical intervention or transfer to a monitored bed. Myocardial infarction was defined as 2 or more of the following: chest pain, echocardiographic changes, and/or cardiac enzyme elevation and fall consistent with myocardial infarction. Urinary tract infection was defined as urine culture with ⱖ103 colony-forming units per mL. Postoperative hemorrhage was defined as the need to return to the operating room or postoperative radiologic intervention for hemorrhage or postoperative gastrointestinal bleeding documented by endoscopy. Renal failure was defined as the need for dialysis of any duration. Biliary leak was defined as the drainage of any volume of fluid clinically consistent with bile from operatively placed drains or the wound with a bilirubin concentration greater than the serum value. Reoperation was defined as any unplanned operation within 30 days of the Whipple procedure. The length of hospital stay was calculated from the day of surgery through and including the day of discharge. Readmission was defined as an admission to any hospital within 30 days after initial discharge after surgery. Operative mortality was defined as any death within 30 days of surgery. Data were entered into the database in real time by a trained data analyst under the supervision of the surgeon. All data were backed-up by source documents, and accuracy of the data entered into the electronic database was periodically reviewed. The comparison of continuous variables was performed using the Wilcoxon rank-sum test, and the comparison of categorical variables was performed using Fisher exact test. Survival data were analyzed using the Kaplan-Meier method. Results are reported as median values unless indicated otherwise. Statistical significance was accepted for P ⬍ .05.
Results Preoperative data One surgeon performed 162 Whipple procedures in the first 11.5 years of independent practice. The annual distri-
686
The American Journal of Surgery, Vol 203, No 6, June 2012 Table 2
Presenting signs and symptoms
Abdominal pain Weight loss Jaundice Nausea Vomiting Anorexia Diarrhea
Low-volume (n ⫽ 19) (%)
High-volume (n ⫽ 143) (%)
P value
11 13 8 6 1 0 5
92 72 56 45 24 19 15
.617 .324 .808 1.000 .312 .132 .063
(58) (68) (42) (32) (5) (0) (26)
(64) (54) (39) (32) (17) (13) (11)
There were no significant differences in clinical presentation between the 2 cohorts.
Figure 1 The annual distribution of Whipple procedures performed by 1 surgeon in the 1st 11.5 years of practice.
bution of procedures appears in Figure 1. In the 1st 6.5 years, volume was low with 0 to 11 Whipple procedures annually. In the next 5 years, volume increased to ⬎22 Whipple procedures per year, with ⬎30 per year in the most recent years (Fig. 1). Data are compared between 2 time periods (low volume and high volume), dividing the time periods before and after achieving a volume of ⬎11 pancreaticoduodenectomies/year. Demographic data for the 2 groups is reported in Table 1. The median age was 55 in the low-volume era and 65 in the high-volume era (P ⫽ .033). Ages ranged from 31 years to 85 years. Ten patients were in their 80s. A total of 142 patients (88%) were white, 16 (10%) were black, and 4 (2%) were Asian or Pacific Islander. Nineteen (12%) were of the Hispanic ethnicity. There were no statistically significant differences in demographics between the 2 eras other than age. There were no clinically significant differences between the low- and high-volume eras in presenting symptoms, comorbid
Table 1
Demographics
Age (median, interquartile range) Sex (%) Male Race White Black Asian Pacific islander Ethnicity (%) Hispanic
Low volume (n ⫽ 19)
High volume (n ⫽ 143)
P value
55 (49–66)
65 (56–73)
.033
12 (63)
58 (41)
.084
16 3 0 0 1
(84) (16) (0) (0) (5)
126 13 1 3 18
(88) (9) (1) (2) (13)
.645
.702
Patients in the low-volume era tended to be younger. There were no other differences between the 2 cohorts.
conditions, or preoperative laboratories (Tables 2 and 3). Presenting symptoms included abdominal pain in 64%, weight loss in 52%, and jaundice in 40%. There were no differences in the median preoperative hemoglobin, bilirubin, creatinine, albumin, carcinoembryonic antigen (CEA), or cancer antigen 19-9. ASA scores and comorbid conditions were also similar in the 2 time periods. There was no difference in the smoking or alcohol use history between the 2 groups. Computed tomography scans were performed in almost every patient with magnetic resonance imaging being used to evaluate the pathology in only 4%. A total of 82 (53%) had preoperative biliary stents placed. These were almost exclusively placed endoscopically and usually before referral to our center. Table 3
Comorbid conditions and preoperative laboratories Low volume (n ⫽ 19) (%)
HTN 8 Diabetes 6 Chronic pancreatitis 5 CAD 6 COPD 2 Renal insufficiency 1 MI 1 Obesity (BMI ⬎30) 7 ASA ⱖ3 14 Smoking pack years 25 Yes/no 8 Active EtOH 7 Total bilirubin 1.6 Creatinine .8 Albumin 3.8 Hemoglobin 12.9 CEA 2.3 CA 19–9 70
High volume (n ⫽ 143) (%)
(42) (32) (26) (32) (11) (5) (5) (39) (74)
76 34 31 23 7 5 3 30 102
(53) (24) (22) (16) (5) (4) (2) (21) (71)
(42) (37) (.6–5.1) (.7–.9) (3.7–4.1) (12.7–13.7) (.6–4.6) (17–204)
25 79 50 .8 .9 4.1 12.6 1.5 25.2
(59) (35) (.4–2.4) (.8–1.0) (3.7–4.5) (11.5–13.9) (.7–2.5) (9–63)
P value* .465 .571 .769 .114 .285 .533 .396 .134 .942 .219 1.000 .073 .019 .127 .450 .372 .288
HTN ⫽ hypertension; CAD ⫽ coronary artery disease; COPD ⫽ chronic obstructive pulmonary disorder; MI ⫽ myocardial infarction; EtOH ⫽ ethanol ; CEA ⫽ Carcinoembryonic antigen. There were no significant differences in the comorbid conditions or preoperative laboratories between the 2 cohorts. *Fisher exact text for categoric variables and Wilcoxon rank-sum test for continuous variables (presented as median, interquartile range).
W.E. Fisher et al. Table 4
Learning curve for pancreaticoduodenectomy
Intraoperative data
PPPD Vein resection Soft pancreas PD ⬎3 mm PD stent BD stent EBL (mL) Transfusion (Y/N) OR time
Low volume (n ⫽ 19)
High volume (n ⫽ 143)
16 1 4 6 1 0 800 8
130 16 71 74 47 5 400 26
(84) (7) (22) (43) (6) (0) (500–1,000) (44)
535 (444–551)
P value*
(92) .391 (11) 1.000 (54) .022 (55) .415 (33) .015 (4) 1.000 (200–700) .001 (18) .027
444 (375–531)
.090
The surgeon characterized the pancreas as soft more frequently and used in internal pancreatic duct stent more frequently in the highvolume era. There was a significant decrease in intraoperative blood loss and transfusions in the high-volume era. PPPD ⫽ pylorus-preserving pancreaticoduodenectomy; PD ⫽ pancreatic duct; BD ⫽ bile duct; EBL ⫽ estimated blood loss. *Fisher exact test for categoric variables and Wilcoxon rank-sum test for continuous variables (median, interquartile range).
Intraoperative data Intraoperative data are presented in Table 4. A pyloruspreserving pancreaticoduodenectomy was performed in 90% with the classic Whipple procedure being reserved for patients with tumors encroaching near the pylorus. More portal or superior mesenteric vein resections were performed in the high-volume era, but this did not reach statistical significance. Seven (41%) were repaired primarily, and 10 (59%) required an internal jugular vein interposition graft. All cases were performed by the attending surgeon (W.E.F.) who allowed an assistant, usually a postgraduate year 5 general surgery resident, to perform portions of the procedure as appropriate. In cases of vascular involvement, an experienced vascular surgeon rather than a resident assisted with this portion of the case. All patients underwent reconstruction with a retrocolic pancreaticojejunostomy. One hundred fifty (97%) were performed with an end-to-side, duct-to-mucosa technique with an outer layer of 4-0 silk sutures and an inner layer of 6-0 polydioxanone suture (PDS) absorbable sutures. The texture of the gland was graded by the surgeon as soft more frequently in the high-volume era (54% vs 22%, P ⫽ .022), but there was no difference in pancreatic duct size between the 2 cohorts. An internal pancreatic duct stent (5F pediatric feeding tube cut to 3 cm in length) was used more frequently in the high-volume era (33% vs 6%, P ⫽ .015). Infrequently, in 5 cases (3%), an end-to-end intussuscepted technique was used when the pancreatic duct could not be identified. The biliary anastomosis was performed with a single layer of interrupted absorbable sutures. The afferent limb was brought antecolic for the duodenojejunostomy. This anastomosis was performed in a two-layer hand-sewn technique with an outer layer of interrupted 3-0 silk Lembert sutures and an inner running layer of 3-0 PDS suture.
687
The median operative time decreased from 535 minutes (8 hours 55 minutes) to 444 minutes (7 hours 24 minutes) from the low-volume to high-volume era, but this was not statistically significant. EBL was higher in the low-volume era, (800 vs 400 mL, P ⫽ .001), and more patients were transfused (8 [44%] vs 26 [18%], P ⫽ .027).
Postoperative data Pathological diagnoses are reported in Table 5. A total of 56 (35%) had pancreatic adenocarcinoma, and the remaining patients had ampullary carcinoma, cholangiocarcinoma, neuroendocrine cancer, intraductal papillary mucinous neoplasm, mucinous cystic neoplasm, cystadenoma, chronic pancreatitis, or an ampullary adenoma. There were no statistically significant differences in the indications for surgery between the 2 eras. Among the patients with pancreatic adenocarcinoma, there were no differences between the low-volume and high-volume eras in terms of tumor size, lymph node metastases, stage of disease, histological grade, or the presence of lymphovascular or perineural invasion (Table 6). There was a statistically significant increase in the number of lymph nodes harvested in the high-volume era (17 vs 8, P ⫽ .0003) but no difference in the number of positive nodes removed. The increase in R1 resections did not reach statistical significance. However, the implementation of a standardized method of orientation, inking, and sectioning of the specimen during the high-volume era may have confounded the determination of R1 margin status. The number of R2 resections decreased significantly with increasing experience (2 [33%] vs 1 [2%], P ⫽ .031). However, among the patients with pancreatic adenocarcinoma, there was no statistically significant difference in 2- or 5-year survival between the low- and high-volume eras. However, there were only 6 patients with pancreatic adenocarcinoma in the lowvolume era, making statistical comparisons difficult. Table 5
Diagnoses Low volume (n ⫽ 19) (%)
High volume (n ⫽ 143) (%)
9 (47) 6 (67) 3 (33)
72 50 18 4 8 26 19 1 5 1 21 8 8
P value .128
Adenocarcinoma Pancreatic Ampullary Other Neuroendocrine Cystic neoplasm IPMN MCN Cystadenoma Other Pancreatitis Adenoma Other
0 1 (5)
6 (32) 3 (16) 1 (5)
(50) (69) (25) (6) (6) (18) (76) (4) (16) (4) (15) (6) (6)
Cystic neoplasms comprised a higher percentage of the patients in the high-volume era, but this difference was not statistically significant.
688 Table 6
The American Journal of Surgery, Vol 203, No 6, June 2012 Pathology for pancreatic adenocarcinoma cases Low volume High volume P (n ⫽ 6) (n ⫽ 50) value*
Tumor size Lymph nodes number examined number positive Tumor stage Stage 0 Stage IA Stage IIA Stage IB Stage IIB Histological grade G1 (well differentiated) G2 (moderately differentiated) G3 (poorly) Lymphovascular invasion Perineural invasion Resection margin status R0 R1 R2 Survival (%) 2y 5y
3 (2.5–4.2)
3 (2.5–3.5)
5 (1–8) 1 (0-3)
17 (14–26) 2 (2.4–5)
0 0 1 0 5
0 2 8 3 37
(0) (0) (17) (0) (83)
.770 .001 .291 1.000
(0) (4) (16) (6) (74) 1.000
0 (0) 4 (67)
3 (6) 30 (61)
2 (33) 0 (0) 2 (33)
16 (33) 31 (62) 37 (74)
4 (67) 0 (0) 2 (33)
36 (72) 13 (26) 1 (2)
50 17
41 19
.06 .062 .014 .031 .961
Among the patients with pancreatic adenocarcinoma, there were fewer R2 resections as the surgeon gained experience. *Fisher exact test for categoric variables and Wilcoxon rank-sum test for continuous variables (median, interquartile range).
The exact number of cases required to attain proficiency is unknown and likely varies somewhat for individual surgeons, but a large experience may be needed. The most recent data available from the Accreditation Council for Graduate Medical Education indicates that general surgery residents perform a mean of 4 Whipple procedures during their training.10 Recently, there has been an expansion in pancreatic fellowships presumably because of effects of decreased resident work hours and a perception among training programs and trainees that additional operative experience is needed beyond general surgery residency to be a competent pancreatic surgeon. We sought to examine the learning curve associated with the Whipple procedure for an individual surgeon from the end of residency training through the 1st 11.5 years of independent academic practice to provide some insight into the volume of experience that may be necessary to achieve excellent outcomes. We found that measurable quality benchmarks such as blood loss, transfusion rate, complications, and length of stay all decreased significantly with increasing experience. The improvement may be more closely related to a frequency of greater than 11 resections per year rather than any specific career milestone number of total cases. It is difficult to provide a detailed summary of the surgical techniques and perioperative management stratTable 7
Complications and length of stay Low-volume High-volume P (n ⫽ 19) (n ⫽ 143) value*
Postoperative complications are summarized in Table 7. Patients in the low-volume era were more likely to develop one or more complications than patients in the high-volume era (11 [58%] vs 65 [46%], P ⫽ .0337). The most common complications were wound infection and pancreatic fistula. The rate of wound infection decreased (6 [32%] versus 13 [9%], P ⫽ .012) as did the rate of bile leak (2 [11%] vs 1 [.7%], P ⫽ .037) as experience was gained. Although the rate of most other common complications, such as pancreatic fistula, decreased in the high-volume era, there were no other statistically significant differences in specific complication rates between the 2 eras. The length of hospital stay decreased in the high-volume era (10 vs 7 days, P ⫽ .006). There was only 1 death in the series (high-volume era) for a mortality rate of .7% for the entire time period.
Comments A learning curve exists for most complex tasks involving innumerable variables and nuances. It is no surprise that a learning curve exists for the Whipple procedure, one of the most technically complex abdominal operations, which is typically performed on older patients with multiple comorbidities. Surgeon volume has already been shown to be associated with superior outcomes in pancreatic surgery.7,8
Any complication 11 (58) Pancreatic fistula (grade 4 (21) B/C) Wound infection 6 (32) Gastroparesis 2 (11) Readmission 1 (5) Fever 1 (5) New arrhythmia 0 (0) Reoperation 2 (11) Urinary tract infection 0 (0) Vein/artery thrombosis 0 (0) Hemorrhage 1 (5) Intra-abdominal abscess 1 (5) Adult respiratory distress 0 (0) syndrome Pneumonia 1 (5) Biliary Leak 2 (11) Myocardial infarction 0 (0) Renal failure 1 (5) Deep venous thrombosis 0 (0) Clostridium difficile colitis 0 (0) Thrombocytopenia 0 (0) Urinary retention 0 (0) Ascites 0 (0) Mortality (30-day) 0 (0) Length of stay (d) 10 (8–14)
65 (46) 19 (13) 13 6 4 4 4 2 3 3 2 2 2
.0337 .480
(9) (4) (3) (3) (3) (1) (2) (2) (1) (1) (1)
.012 .238 .469 .469 1.000 .068 1.000 .332 1.000 .314 1.000
1 (1) 1 (1) 1 (1) 1 (1) 1 (1) 1 (1) 1 (1) 1 (1) 1 (1) ⬍1 7 (7–11)
.222 .037 1.000 .222 1.000 1.000 1.000 1.000 1.000 1.000 .006
The morbidity of the Whipple procedure was significantly decreased in the high-volume era. *Fisher exact test for categoric variables and Wilcoxon rank-sum test for continuous variables (median, interquartile range).
W.E. Fisher et al.
Learning curve for pancreaticoduodenectomy
egies that lead to improved outcomes so that these improvements could be adopted by other surgeons at other centers. The problem lies in the sheer complexity of the preoperative assessment, patient preparation, surgical techniques, and perioperative management, but this series offers some insight. Although not shown in the patient demographics or comorbidities documented in this study, improved patient selection for surgery including accurate interpretation of preoperative CT scans and wise recognition of unacceptably severe comorbidities and performance status certainly played a role in improving outcomes in this series. All patients in this series with a history of cardiopulmonary risk factors were referred to medical specialists for preoperative consultation. A preoperative review of imaging studies with an experienced radiologist and repeated review just before each case helped a low-volume surgeon avoid pitfalls while simultaneously gaining personal skill in CT interpretation. An improved knowledge of anatomy and its variations, recognition of common errors, and improvements in operative technique certainly contributed as well. In particular, detailed knowledge of the anatomy of the celiac axis and variations in the origin of the hepatic arteries, the anatomy of the superior mesenteric artery and its branches, and detailed knowledge of the variations in mesenteric venous anatomy help the pancreatic surgeon avoid intraoperative blood loss. A strategy for the creation of an excellent pancreaticojejunostomy is also likely important because pancreatic leak is associated with additional complications. At least 2 modifications in the pancreaticojejunostomy technique were made in this series. Increased mobilization of the pancreatic neck for at least 2 cm by ligating small branches of the proximal splenic vein and artery allows circumferential placement of interrupted sutures. An internal pancreatic duct stent may be useful not so much to control postoperative leak but to ensure accurate placement of duct-to-mucosa sutures when the pancreatic duct is small (⬍3 mm). The development of clinical pathways and multidisciplinary care teams can certainly have an impact on outcomes, particularly the length of stay. Our study is limited by the fact that simultaneously with the achievement of high annual case volume, the surgeon implemented numerous changes in patient care delivery including a multidisciplinary pancreas clinic and tumor board and clinical care pathways. The relative impact of these programmatic infrastructure factors, the institutional learning curve, and surgical experience cannot be separated. Dissecting out the important factors that are responsible for improved outcomes seems nearly impossible, and it is likely that improved outcomes resulted from some combination of all these factors. The fact that this is a single-surgeon series presents both opportunities and problems with interpretation. Acknowledging the fact that institutions change over time, a multi-
689
center study would certainly introduce even more variables that could affect interpretation of the learning curve. Interpretation of a multi-surgeon study would be complicated by individual surgeon variables such as surgical ability and differences in the quality and quantity of previous training. The present study chronicles the experience of a single surgeon over a long period at 1 institution and does provide some insight into the changes that occurred over time that are responsible for improved outcomes. The learning curve may not be finished. Tseng et al11 reported an analysis of the 1st 120 cases performed by 3 surgeons who trained at M D Anderson Cancer Center. They observed a similar improvement in EBL, operating room time, and length of stay. The analysis of additional cases beyond 120 for one of the surgeons indicated the learning curve continues farther into a surgeon’s career. In this series, the long period of low volume likely lengthened the time required to progress through the learning curve. It also complicates interpretation of the data because the surgeon performed other complex operations during this period, and the translation of this experience into improved pancreatic surgery cannot be assessed. Ideally, a surgeon interested in pancreatic surgery should seek an environment in which a high volume of cases can be experienced immediately after training. However, no surgeon can start off his/her practice as an experienced high-volume pancreas surgeon. The learning curve for the Whipple procedure may be so lengthy that it would be extraordinarily difficult to progress completely through it even with an additional year or two of focused fellowship training. In addition to focused training, it may be critical to seek an environment with experienced colleagues and mentors who are willing to share not only their hardacquired expertise but also their referral base to ensure a steady volume of continued experience that will translate into rapid progress and excellent outcomes. However, even if willing, a mentor may not be able to share referrals because referring physicians and even patients may demand the more experienced partner. Performing cases together is one solution, but this has an impact on the junior surgeon’s productivity measures and limits the experience of other trainees. The mortality rate after the Whipple procedure is an outcome variable that is readily available. Mortality data can be derived from national databases as well as reports from expert centers and are probably quite reliable in terms of accuracy. As demonstrated in Table 8, we have reported 30-day mortality, not just in-hospital mortality. Some centers may under estimate the 30-day mortality if patients are lost to follow-up. The mortality rate in this series compares favorably with national data and meets the high benchmarks set by ultrahigh-volume centers of excellence.4 – 6,12–15 Of note is that no mortality occurred in the low-volume era, and the mortality rate for the overall series is in the acceptable range. Other data, aimed at assessing the morbidity of the procedure are less available, and variations in definitions,
690 Table 8
The American Journal of Surgery, Vol 203, No 6, June 2012 Comparison to high-volume centers Current series
Mortality Morbidity LOS (d) Reoperation OR time (min) EBL (mL)
Low volume
High volume
United States14
Hopkins1
MSK15
MDACC16
UCSFs
VM Clinic4
0% 58% 10 11% 535 800
⬍1% 46% 7 1% 444 400
9% 35% 15 NA NA NA
1% 41% 9 3% 330 700
2% 47% 14 9% 350 1,000
1% NR 14 1% 553 900
4% 59% 16 7% 402 1,167
0% NR 11 ⬍1% 450 382
NA ⫽ not applicable; LOS ⫽ length of stay; OR ⫽ operating room. Examination of the learning curve of a single surgeon with the Whipple procedure indicates that experience is associated with an improvement in measurable benchmarks. However, in the right environment, a low-volume surgeon can perform this procedure with acceptable morbidity and mortality during the lengthy learning curve.
differences in prospective and retrospective databases, and the tenacity with which every complication is documented can create problems with comparing different series. The present series was documented prospectively with extremely thorough accuracy so there should be no underreporting of the morbidity. However, there are still problems with definitions even with the simplest things like the length of stay. Even operating room time and EBL can be biased if not taken from the anesthesia record as done in this series. That being said, when we compare the present series with outcomes reported in the literature, the mortality, morbidity (percent of patients with any complication), the length of stay, the rate of reoperation, operating room time, and EBL all compare with other high-volume centers in both the low-volume and high-volume eras of the series. Therefore, this series suggests that although improvement occurs with experience and consistent high volume, an inexperienced, low-volume surgeon can embark on a career in pancreatic surgery with acceptable morbidity and mortality in the right environment. It should be emphasized that the contribution of experienced colleagues in anesthesia and perioperative care and all the support of a tertiary care academic teaching environment with a high volume of pancreatic cases certainly played a role in the outcomes achieved in this series.
Conclusions Examination of the learning curve of a single surgeon with the Whipple procedure indicates that experience is associated with decreased blood loss; number of transfusions; and complications, particularly wound infections; and a faster recovery. Strong motivation and interest in pancreatic surgery, a high-volume hospital environment, and partners capable and willing to serve as experienced mentors are necessary ingredients for a low-volume sur-
geon to perform this procedure with acceptable morbidity and mortality during the lengthy learning curve.
References 1. Kausch W. Das carcinoma der papilla duodeni und Seine Radikale entfeinung. Beitr Z Clin Chir 1912;78:439 – 86. 2. Whipple AO, Parsons WB, Mullins CR. Treatment of carcinoma of the ampulla of Vater. Ann Surg 1935;102:763–79. 3. Whipple AO. A reminiscence: pancreaticoduodenectomy. Rev Surg 1963;20:221–5. 4. Cameron JL, Riall TS, Coleman J, et al. One thousand consecutive pancreaticoduodenectomies. Ann Surg 2006;244:10 –5. 5. Winter JM, Cameron JL, Campbell KA, et al. Pancreaticoduodenectomies for pancreatic cancer: A single-institution experience. J Gastrointest Surg 1423;2006:1199 –211. 6. Cameron JL, Pitt JA, Yeo CJ, et al. One hundred and forty-five consecutive pancreaticoduodenectomies without mortality. Ann Surg 1993;217:430 – 8. 7. Lieberman MD, Kilburn H, Lindsey M, et al. Relation of perioperative deaths to hospital volume among patients undergoing pancreatic resections for malignancy. Ann Surg 1995;22:638 – 45. 8. Birkmeyer JD, Finlayson SRG, Tosteson ANA, et al. Effect of hospital volume on in-hospital mortality with pancreaticoduodenectomy. Surgery 1999;125:250 – 6. 9. Bassi C, Dervenis C, Butturini G, et al. Postoperative pancreatic fistula: an international study group (ISGPF) definition. Surgery 2005;138:8–13. 10. General Surgery (2005-2006): National Resident Report. Residency Review Committee. Accreditation Council for Graduate Medical Education. 11. Tseng JF, Pisters PWT, Lee JE, et al. The learning curve in pancreatic surgery. Surgery 2007;141:456 – 63. 12. Traverso LW, Shinchi H, Low DE. Useful benchmarks to evaluate outcomes after esophagectomy and pancreaticoduodenectomy. Am J Surg 2004;187: 604–8. 13. Trede M, Schwall G, Saeger HD. Survival after pancreaticoduodenecttomy: 118 consecutive resections without an operative mortality. Ann Surg 1990; 211:447–58. 14. Turaga K, Kaushik M, Forse RA, et al. In hospital outcomes after pancreatectomies: an analysis of a national database from 1996 –2004. J Surg Oncol 2008;98:156 – 60. 15. Schell MT, Barcia A, Spitzer AL, et al. Pancreaticoduodenectomy: volume is not associated with outcome within an academic health care system. HPB Surg 2008;2008:825940.