Introduction: Although mortality rates from pancreatectomy have decreased worldwide, death remains an infrequent but profound event at an individual practice level. Root-cause analysis is a retrospective method commonly employed to understand adverse events. We evaluate whether emerging mortality risk assessment tools sufficiently predict and account for actual clinical events that are often identified by root-cause analysis.
Methods: We assembled a Pancreatic Surgery Mortality Study Group comprised of 36 pancreatic surgeons from 15 institutions in 4 countries. Mortalities after pancreatectomy (30 and 90 days) were accrued from 2000 to 2010. For root-cause analysis, each surgeon "deconstructed" the clinical events preceding a death to determine cause. We next tested whether mortality risk assessment tools (ASA, POSSUM, Charlson, SOAR, and NSQIP) could predict those patients who would die (n = 218) and compared their prognostic accuracy against a cohort of resections in which no patient died (n = 1,177).
Results: Two hundred eighteen deaths (184 Whipple's resection, 18 distal pancreatectomies, and 16 total pancreatectomies) were identified from 11,559 pancreatectomies performed by surgeons whose experience averaged 14.5 years. Overall 30- and 90-day mortalities were 0.96% and 1.89%, respectively. Individual surgeon rates ranged from 0% to 4.7%. Only 5 patients died intraoperatively, while the other 213 succumbed at a median of 29 days. Mean patient age was 70 years old (38% were >75 years old). Malignancy was the indication in 90% of cases, mostly pancreatic cancer (57%). Median operative time was 365 min and estimated blood loss was 700 cc (range, 100-16,000 cc). Vascular repair or multivisceral resections were required for 19.7% and 15.1%, respectively. Seventy-seven percent had a variety of major complications before death. Eighty-seven percent required intensive care unit care, 55% were transfused, and 35% were reoperated upon. Fifty percent died during the index admission, while another 11% died after a readmission. Almost half (n = 107) expired between 31 and 90 days. Only 11% had autopsies. Operation-related complications contributed to 40% of deaths, with pancreatic fistula being the most evident (14%). Technical errors (21%) and poor patient selection (15%) were cited by surgeons. Of deaths, 5.5% had associated cancer progression-all occurring between 31 and 90 days. Even after root-cause scrutiny, the ultimate cause of death could not be determined for a quarter of the patients-most often between 31 and 90 days. While assorted risk models predicted mortality with variable discrimination from nonmortalities, they consistently underestimated the actual mortality events we report.
Conclusion: Root-cause analysis suggests that risk prediction should include, if not emphasize, operative factors related to pancreatectomy. While risk models can distinguish between mortalities and nonmortalities in a collective fashion, they vastly miscalculate the actual chance of death on an individual basis. This study reveals the contributions of both comorbidities and aggressive surgical decisions to mortality.