Benefits of the volume-outcome association in pancreatic cancer surgery depend on patient level risk

Abstract

Presenter: Sasha Hornock DO | William Beaumont Army Medical Center Background: Research has consistently demonstrated an association between hospital procedural volume and improved perioperative outcomes following high-risk cancer operations. Proponents argue this evidence supports centralization of high-risk surgery. However, concerns exist that centralization may inherently be associated with inequity of care in the US healthcare system. Our objective was to examine patient and hospital-level characteristics of patients undergoing pancreatectomy for cancer at low-, medium- and high-volume hospitals and determine if outcome benefits persist after controlling for differences in patient, administrative and hospital characteristics. Methods: The 2012-2014 Nationwide Inpatient Sample was queried for all patients undergoing pancreatectomy for cancer. Hospitals were stratified by volume proportion based on the annual percentage of hospital discharges for pancreatic surgery. Baseline patient demographics and hospital characteristics were compared. Propensity score matching (PSM) was performed using a generalized boosting model for patient, administrative and hospital-related covariates. Primary outcomes included inpatient mortality, composite morbidity, and length of stay (LOS). Results: 2,748 patients underwent pancreatectomy for cancer. Hospitals with high volumes of pancreatic surgery more commonly treated patients that were white (81% vs 70%) with private insurance (40% vs 31%) and lower Elixhauser Index (EI) comorbidity risk scores (52% vs 37% for EI1) compared to low-volume centers (all p0.05). Though high volume centers do not indicate strong evidence of a reduction in risk of inpatient death (HR 0.40; CI 0.12-1.10; p=0.07), compared to low-volume centers, application of a global model comparison test provides evidence of a statistically significant interaction between the EI and volume that affects mortality (p<0.001), implying that volume effect on mortality depends on patient EI score. Conclusion: Considerable disparities exist among patients treated at high vs low volume centers, with higher-risk and socioeconomically vulnerable patients more commonly treated at lower volume centers. Improvement in inpatient mortality rates appear to be derived from the highest-risk patients receiving treatment at high-volume centers. Development of predictive models to guide selective referrals of high-risk patients to high-volume centers is needed.