Perioperative mortality after pancreatectomy: A risk score to aid decision-making

• Published in: Surgery 2012-09, Vol.152 (3), p.S120-S127
• Main Authors: Ragulin-Coyne, Elizaveta, MD, MPH, Carroll, James E., MD, Smith, Jillian K., MD, MPH, Witkowski, Elan R., MD, Ng, Sing Chau, MS, Shah, Shimul A., MD, Zhou, Zheng, MD, PhD, Tseng, Jennifer F., MD, MPH
• Format: Article
• Language: English
• Subjects: Aged; Aged, 80 and over; Decision Support Techniques; Female; Hospital Mortality; Humans; Male; Middle Aged; Models, Statistical; Pancreatectomy - mortality; Pancreatic Neoplasms - surgery; Pancreatitis - surgery; Risk Factors; Surgery
• ISSN: 0039-6060; 1532-7361
• DOI: 10.1016/j.surg.2012.05.018

Background Undergoing a pancreatectomy obligates the patient to risks and benefits. For complex operations such as pancreatectomy, the objective assessment of baseline risks may be useful in decision-making. We developed an integer-based risk score estimating in-hospital mortality after pancreatectomy, incorporating institution-specific mortality rates to enhance its use. Methods Pancreatic resections were identified from the Nationwide Inpatient Sample (1998–2006), and categorized as proximal, distal, or nonspecified by the International Classification of Diseases, 9th edition. Logistic regression and bootstrap methods were used to estimate in-hospital mortality using demographics, diagnosis, comorbidities (Charlson index), procedure, and hospital volume; 80% of this cohort was selected randomly to create the score and 20% was used for validation. Score assignments were subsequently individually fitted to risk distributions around specific mortality rates. Results Sixteen thousand one hundred sixteen patient discharges were identified. Nationwide in-hospital mortality was 5.3%. Integers were assigned to predictors (age group, Charlson index, sex, diagnosis, pancreatectomy type, and hospital volume) and applied to an additive score. Three score groups were defined to stratify in-hospital mortality (national mortality, 1.3%, 4.9%, and 14.3%; P < .0001), with sufficient discrimination of derivation and validation sets (C statistics, 0.72 and 0.74). Score groups were shifted algorithmically to calculate risk based on institutional data (eg, with institutional mortality of 2.0%, low-, medium-, and high-risk patient groups had 0.5%, 1.9%, and 5.4% mortality, respectively). A web-based tool was developed and is available online ( http://www.umassmed.edu/surgery/panc_mortality_custom.aspx ). Conclusion To maximize patient benefit, objective assessment of risk for major procedures is necessary. We developed a Surgical Outcomes Analysis and Research risk score predicting pancreatectomy mortality that combines national and institution-specific data to enhance decision-making. This type of risk stratification tool may identify opportunities to improve care for patients undergoing specific operative procedures.