Outcome Quality Beyond the Mean – An Analysis of 43,231 Pancreatic Surgical Procedures Related to Hospital Volume
Richard Hunger, René Mantke
Abstract
OBJECTIVE: The aim of this study was to examine whether elevated in-hospital mortality rates in lower volume hospitals are only valid on average or also apply for individual hospitals. SUMMARY OF BACKGROUND DATA: Various studies demonstrated a volume-outcome relationship in pancreatic surgery with increased mortality in low volume hospitals. However, almost all studies assessed quality indicators only for groups of hospitals by averaged measures, neglecting variability of hospital performance. METHODS: The German nationwide hospital discharge data (diagnosis-related groups-statistics) was used to determine risk-adjusted in-hospital mortality for all distal pancreatectomies (DP), pancreatoduodenectomies (Whipple-proce-dure, PD), and pylorus-preserving pancreatoduodenectomies (PPD) performed between 2011 and 2015. Hospitals were stratified according to annual and 5-year total procedure volume and examined in relation to average in-hospital mortality of the highest volume quintile. RESULTS: Lowest adjusted mortality rates were observed in highest volume quintiles for each pancreatic resection procedure, with 6.2% for DP, 8.3% for PD, and 5.7% for PPD in the 5-year observation period. With these mortality rates as reference values the analysis revealed that a non-negligible proportion of hospitals performed equal or better (DP: 430/784, 54.5%; PD: 269/611, 44.0%; PPD: 255/565, 45.1%) than the hospitals of the highest volume quintile. CONCLUSIONS: High quality of care, with in-hospital mortality rates less or equal to high-volume hospitals, is also achieved in hospitals with lesser procedure volume. Therefore, mere volume seems not suitable as proximal measure for assessing individual hospital quality. instead, more sophisticated certification systems, that allow accurate quality assessment and better reflect clinical variability, should preferred to fixed minimum volume thresholds.