Targeting Early Deaths Following Pancreaticoduodenectomy to Improve Survival
- First Online:
- Cite this article as:
- Clark, W., Silva, M., Donn, N. et al. J Gastrointest Surg (2012) 16: 1869. doi:10.1007/s11605-012-1958-7
- 199 Views
There is increasing focus on disease-specific outcomes. This study was undertaken to analyze early mortality after pancreaticoduodenectomy as part of a strategy to improve long-term outcome.
One thousand thirty-one patients who underwent pancreaticoduodenectomy from 1992 to 2010 were studied. Median data are reported.
Fifty-eight (5.6 %) patients died within 90 days after pancreaticoduodenectomy. All patients had at least one significant comorbidity, commonly cardiorespiratory in nature (76 %). Sixty percent of patients had depressed serum albumin levels, and 43 % were jaundiced. The American Society of Anesthesiologists class was: 17 % class II, 72 % class III, and 10 % class IV. Seventy-four percent had malignant disease. Twenty-two percent of patients underwent a major vascular resection at the time of pancreaticoduodenectomy. Causes of death were vascular/bleeding related (26 %), cardiorespiratory causes (17 %), multiorgan failure (12 %), leak/perforation (10 %), cancer progression (9 %), infection (7 %), or indeterminate (19 %).
Death within 90 days after pancreaticoduodenectomy is uncommon, occurs in relatively older deconditioned patients, and is generally not causally related to underlying malignancy. Early death is generally associated with vascular or bleeding complications. Strategies to minimize early death should focus on careful patient selection and prompt recognition and management of herald bleeding or vascular thrombosis, as it can often result in perioperative death following pancreaticoduodenectomy.
KeywordsPancreaticoduodenectomyEarly mortalitySurvivalPancreatic cancer
Focus on perioperative outcomes and, specifically, mortality after pancreatectomy is ever increasing. Early mortality is being adopted as a metric for quality of care. Recent literature has shown that high-volume centers, high-volume surgeons, and training centers are all associated with lower 30-day mortality following pancreaticoduodenectomy.1–7 Outcomes after pancreaticoduodenectomy in elderly patients are continuously debated, scrutinized, and published.8–10 Changes in health care, political, and insurance systems have spawned terms such as “pay for performance” and “comparative effectiveness research.”11 Increasing attention to long- and short-term postoperative outcomes will be the theme for years to come. Large-scale analysis of national databases is allowing estimation of targets or complication and mortality rates that will be used as metrics for quality of care after pancreatic resection.
With such increasing scrutiny, it is easy to overlook the enormous improvement in perioperative mortality rates following pancreaticoduodenectomy that has been made in the last 30 to 40 years. Skilled high-volume surgeons in the 1970s reported 30-day mortality after pancreaticoduodenectomy of nearly 25 %.12 Today, leading centers generally report mortality rates of less than 5 %.6,13–15 Despite improvement, early deaths, including those beyond 30 days, detract from long-term survival, which is still quite poor. We undertook this study to investigate root causes of mortality within 90 days after pancreaticoduodenectomy as part of a strategy to improve long-term survival. We hypothesized that most deaths within 90 days would occur as a consequence of exacerbations of underlying systemic diseases, and complications would not directly be related to malignancy or obvious technical aspects of a pancreaticoduodenectomy surgery.
We analyzed a prospectively collected database including all patients undergoing pancreatic resection at our institution beginning in 1992. We selected the subset of patients undergoing pancreaticoduodenectomy with documented survival of less than or equal to 90 days. The database includes documentation of preoperative medical comorbidities and surgical pathology. For our analyses, cardiorespiratory disease included the following diagnoses: cardiac dysrhythmia, coronary artery disease, congestive heart failure, peripheral vascular disease, stroke, hypertension, dyspnea, COPD, and chronic bronchitis. Premalignant disease included intraductal papillary mucinous neoplasm, neuroendocrine neoplasm, and serous and mucinous cystadenomas. Cause of death is recorded in the database; however, at least one surgeon reviewed each patient's complete chart (e.g., history and physical, operative reports, progress notes, discharge or death summary, autopsy report) to identify the sentinel event leading to the cascade of events resulting in each patient's death. Causes of death are described as vascular/bleeding related, which included delayed visceral arterial hemorrhages; postoperative bleeding; intraluminal gastrointestinal bleeding; and portal vein or visceral arterial thrombosis. Cardiopulmonary causes of death included dysrhythmia, pulmonary embolus, myocardial infarction, respiratory distress syndrome, and stroke. Leak/perforation was defined as an anastomotic leak or perforation requiring reoperation or external drainage. Multiorgan failure was defined as failure in two or more major organ systems (i.e., respiratory, renal, hepatic) without a discernable cause appropriate for another category. Infection included sepsis (not secondary to an anastomotic leak) and pneumonia. Progression of cancer included patients dying while under hospice care or dying of cachexia. If the inciting event leading to the cascade of events causing death could not be identified, cause was categorized as indeterminate. All patients categorized as indeterminate died after discharge from our center without readmission. Data are reported as median (mean ± standard deviation), and statistical comparisons were made using chi-square or t tests where appropriate with significance accepted with 95 % probability. Overall survival was analyzed using Kaplan–Meier method and log rank tests.
Demographic Data, Preoperative Laboratory Data, and Comorbidities
Early mortality rates for patients undergoing pancreaticoduodenectomy
Mortality within 90 days
Preoperative demographic data, laboratory tests, and anesthesia risk scores of patients dying within versus surviving 90 days after pancreaticoduodenectomy
Died within 90 days
Survived beyond 90 days
Number of patients
72 (71 ± 10.1)
66 (65 ± 12.4)
Gender (M/F), %
26 (26 ± 4.7)
26 (26 ± 5.5)
Albumin levels lower than normal limits (%)
Albumin levels (g/dL)
3 (3 ± 0.9)
3 (3 ± 0.8)
Total bilirubin levels above normal limits (%)
ASA class II (%)
ASA class III (%)
ASA class IV (%)
ASA class V (%)
Causes of Death Within 30 Days
Causes of early mortality (i.e., <90 days)
30-day mortality (%)
90-day mortality (%)
Progression of cancer
100 (27 patients)
100 (58 patients)
Causes of Death Within 90 Days
Vascular/bleeding-related causes were most common (26 %) for patients dying within 90 days of pancreaticoduodenectomy, followed by cardiorespiratory causes (17 %), indeterminate (19 %), multiorgan failure (12 %), leak/perforation (10 %), cancer (9 %), and infection (7 %) (Table 3). Twenty-two percent of patients who died early underwent concomitant vascular resection (10 out of 13; 78 % of these patients underwent portal vein resection) at the time of surgery. Malignant disease was present in 74 % of patients experiencing early death, 16 % had premalignant disease, and 10 % had benign disease. Of note, the 11 (19 %) patients who died of indeterminate causes died after discharge to their own home or another center and despite attempts to contact family members; their cause of death could not be precisely determined.
Mortality Rates Versus Time
Early mortality rates versus time of resection
Total no. of resections
30-day mortality (%)
90-day mortality (%)
p = 0.20a
p = 0.04a
Despite increasing availability of evidence-based guidelines for the surgical treatment of pancreatic adenocarcinoma, long-term survival in patients who undergo resection remains poor (11–20 months), in fact so poor that pancreatic cancer is one of the few diseases where incidence and mortality rates mirror one another.16 If this is going to change, short-term survival needs to improve, because without short-term survival, long-term survival is impossible. Although early deaths were uncommon in our series, they did detract from long-term overall survival. When the patients who experienced early death are eliminated from survival analysis, median survival improved from 28.4 to 35.5 months (p = 0.057; log rank test). Interestingly, in this unique analysis of over 1,000 patients, we somewhat unexpectedly found that approximately one third of early deaths are a consequence of a vascular event, often bleeding. These findings are particularly relevant because patients who have acute vascular conditions (such as bleeding) are in fact “rescuable” and can be targeted for specialized management. Also, patients succumbing to the second most common cause of early death, exacerbations of cardiopulmonary disease, are another subpopulation of patients that can be targeted preoperatively with specific management strategies. Although improving long-term survival is our ultimate goal, reducing early mortality is particularly relevant in today's era of health care where quality of care is measured by early mortality rates.
By analyzing a robust single-center database of all pancreaticoduodenectomy patients, we more accurately represented cause of death than can be done using a pooled, administrative database. Large national or multicenter databases are increasingly being used to study outcomes but lack specifics, which can be problematic. An example would be a patient who suffered a delayed visceral arterial hemorrhage with consequent respiratory failure and death as a result of pneumonia. The bleeding (sentinel event) is overlooked, and focus is on the pneumonia. We sought to identify the first “domino” in the cascade leading to each patient's death and offer a unique analysis based on that. This type of analysis can only be completed by direct review of the patients' charts. Often overlooked, careful analysis of the “domino” or sentinel event leading to death is an optimal way to drive management changes that will improve survival.
A very recent publication by Vollmer et al. utilized a somewhat similar methodology to determine the “root cause” of death in patients undergoing various types of pancreatectomy.17 The study concluded that “operative-related complications” (i.e., aspiration, bleeding, bowel obstruction, anastomic leak, etc.) were responsible for the majority (26 %) of these deaths. Although the idea of the root cause of death is similar to what we have described as “first domino,” only 64 % of patients in their series underwent a pancreaticoduodenectomy. The authors concede that including patients who underwent total pancreatectomy influenced infection-related mortality. We view all complications as being “operative related,” because they happened as a consequence of surgery. By more clearly delineating or deconstructing “operative-related” complications that led to death, our analysis differs from the study discussed above. Also, all patients in our series underwent the same operation, a pancreaticoduodenectomy.
Although patients who died early in our series had frequent comorbidities, the patients who survived beyond 90 days had similar ASA class status, bilirubin levels, and BMI (Table 2). The overall 30-day mortality rate at our center was 2.6 % which is consistent with what leading centers are reporting. Our 31–60- (1.6 %) and 61–90-day (1.3 %) mortality rates have little to be compared to but seem favorable.
When early mortality rates are stratified by tertile, the results are interesting but do not suggest a clear trend (Table 4). Somewhat surprisingly, the lowest mortality was in tertile 1 (1992–1997), where out of 161 patients, 30-day mortality was 0.6 % and 90-day mortality was 1.24 %. These mortality rates are rather low, and by 1992, the surgeon was past the learning curve. The sample size of tertile 1 is relatively small, and the findings probably represent a selection bias. Higher-resolution CT imaging and more confidence with vascular reconstruction have allowed patients to be resected today that were not resected years ago.18 However, we have studied survival in our patients with pancreatic cancer and concluded that vascular resection does not affect survival. The finding could also reflect that it is more common to be referred a patient with a chronic systemic disease in the later tertiles.
A very recent multicenter trial described a similar demographic of patients who experience early death and reported the same 30-day mortality rate of 2.6 % that we have. Their analysis included 4,945 patients undergoing pancreaticoduodenectomy identified from the American College of Surgeons National Quality Improvement Program database. Patients who died within 30 days had more frequent underlying cardiorespiratory disease including dyspnea, chronic obstructive pulmonary disease, coronary artery disease, hypertension, and history of vascular disease relative to patients surviving 30 or more days. This study did not delineate cause of death or comment on frequency of vascular complications, but underscores the fact that patients with cardiorespiratory disease need to be targeted for aggressive perioperative optimization because they are most at risk.19 The limitations of this study in identifying the first “domino” in the cascade leading to death promote the findings in our report.
In patients undergoing pancreaticoduodenectomy, vascular complications, specifically acute portal vein thrombosis, and postoperative hemorrhage are believed to be rare and deadly, but their contribution to the total proportion of patients dying early is even greater than what might be expected. Using the same patient database, we have previously reported that portovenous vascular resection is not associated with increased mortality and does not appear to have a clear role in this subset of patients.18 However, looking at outcome from the perspective of who dies and why, portal vein thrombosis seems more of an issue.
After pancreaticoduodenectomy, there is a small but real number of patients who experience hemorrhagic complications, and the ability to rescue versus “failure to rescue” these patients is one key to improving long-term survival. Approximately 2–17 % of pancreaticoduodenectomy patients experience postoperative hemorrhage which can be divided into early and delayed hemorrhage with early hemorrhage representing surgical bleeding and delayed hemorrhage usually occurring secondary to a pseudoaneurysm of a visceral vessel or anastomotic ulceration.20–22 Hemorrhage secondary to an eroded or pseudoaneurysmal visceral vessel is especially lethal with incidence and mortality rates of approximately 3 and 23 %, respectively.23,24 Visceral vessel pseudoaneurysms after pancreaticoduodenectomy usually present with a sentinel bleed first (78 % of patients), and almost all of these patients have documented pancreatic leaks.23 It can be difficult to delineate the source of “pancreatic” leaks, but for discussion, almost all patients who experience visceral vessel hemorrhage have persistent drain output precluding removal with drain amylase three times greater than normal serum (ISGPF). We recommend aggressive drainage of all pancreatic and biliary leaks, and aggressive attention to sentinel bleeding including serious consideration for immediate angiography. Sentinel bleeding is a key time to rescue patients who are at exceptionally high risk of death.
Similarly, acute portal or mesenteric vein thrombosis is a potentially “rescuable” state that requires immediate and sometimes nonstandard therapy. There is a paucity of literature available to guide management decisions in patients with acute portal vein thrombosis after pancreaticoduodenectomy, although case reports exist.25 One series of 64 patients who underwent portal vein resection during pancreaticoduodenectomy reported a 5 % rate of portal vein thrombosis.26 In order to rescue these patients, early diagnosis must be made, and important signs include hypotension without bleeding, increasing abdominal distension, marked volume sequestration, elevated liver function enzymes, and tachycardia refractory to resuscitation. Operative thrombectomy can be a viable option.25 Interventional radiology can also be instrumental with this situation, wherein at least in three instances at our institution, a patient with a clotted portal vein after resection was successfully managed with a portal vein endograft inserted via a transjugular, transhepatic approach. One of these grafts was eventually occluded, but after the immediate postoperative period, and the patient did well with collateralization around his portal vein. Going forward, our approach to postoperative portal vein thrombosis will utilize transcutaneous transhepatic stenting.
Pre- and perioperatively maximizing cardiorespiratory status is the second important strategy to improve short- and consequently long-term survival after pancreaticoduodenectomy. Three fourths of patients who died early had at least one major cardiorespiratory comorbidity, and cardiopulmonary events were responsible for 21 % of deaths within 30 days of pancreaticoduodenectomy. Postoperative care must include preoperative and then postoperative medical optimization with special attention given to resuming anticoagulation such as daily aspirin or clopidogrel. Many patients now have drug-eluting cardiac stents which can thrombose without antiplatelet therapy. Pancreaticoduodenectomy can be undertaken safely when antiplatelet therapy is held for only 4 or 5 days. Aggressive management in critically ill patients with comorbidities including early tracheostomy (even at the time of pancreaticoduodenectomy) and use of pulmonary artery catheter (or newer technologies) to optimize intravascular volume status are also principles that we advocate can minimize death from exacerbations of underlying cardiorespiratory disease.
We have demonstrated that death within 90 days of pancreaticoduodenectomy is uncommon, not generally related to malignancy, and most commonly due to a vascular complication or cardiopulmonary event. Most of our patients have underlying systemic illness. Acute vascular events such as delayed hemorrhage and portal vein thrombosis, and acute cardiopulmonary problems contributed to over 50 % of early deaths after pancreaticoduodenectomy. These patients are the ideal candidates for “rescue” and are the key to further improving short-term and consequently long-term survival after pancreaticoduodenectomy. All team members need to understand the significance and signs of sentinel bleeding or shock from acute portal vein thrombosis. Mortality rates will continue to improve after pancreaticoduodenectomy, and given changes in health care, surgeons who are able to demonstrate superior postoperative outcomes will be referred to patients needing pancreaticoduodenectomy.