Modelling centralization of pancreatic surgery in a nationwide analysis

Background: The benefits of centralization of pancreatic surgery have been documented, but policy differs between countries. This study aimed to model various centralization criteria for their effect on a nationwide cohort. Methods: Data on all pancreatic resections performed between 2014 and 2016 were obtained from the Italian Ministry of Health. Mortality was assessed for different hospital volume categories and for each individual facility. Observed mortality and risk-standardized mortality rate (RSMR) were calculated. Various models of centralization were tested by applying volume criteria alone or in combination with mortality thresholds. Results: A total of 395 hospitals performed 12 662 resections; 305 hospitals were in the very low-volume category (mean 2·6 resections per year). The nationwide mortality rate was 6·2 per cent, increasing progressively from 3·1 per cent in very high-volume to 10·6 per cent in very low-volume hospitals. For the purposes of centralization, applying a minimum volume threshold of at least ten resections per year would lead to selection of 92 facilities, with an overall mortality rate of 5·3 per cent. However, the mortality rate would exceed 5 per cent in 48 hospitals and be greater than 10 per cent in 17. If the minimum volume were 25 resections per year, the overall mortality rate would be 4·7 per cent in 38 facilities, but still over 5 per cent in 17 centres and more than 10 per cent in five. The combination of a volume requirement (at least 10 resections per year) with a mortality threshold (maximum RSMR 5 or 10 per cent) would allow exclusion of facilities with unacceptable results, yielding a lower overall mortality rate (2·7 per cent in 45 hospitals or 4·2 per cent in 76 respectively). Conclusion: The best performance model for centralization involved a threshold for volume combined with a mortality threshold.