E. M. Gleeson1, A. P. Johnson2, M. E. Kilbane3, H. A. Pitt4 1Drexel University College Of Medicine,Philadelphia, Pa, USA 2Thomas Jefferson University,Philadelphia, PA, USA 3Indiana University School Of Medicine,Indianapolis, IN, USA 4Temple University,Philadelpha, PA, USA
Introduction: The National Surgical Quality Improvement Program (NSQIP) provides a unique resource to researchers at participating institutions through the de-identified national Participant Use Files (PUF). In an effort to improve data collection, some definitions for the outcome variables have changed over time. Despite these changes, researchers often combine the PUFs across years to increase the power of their studies. This analysis aims to determine if these definition changes have affected the rates of the outcomes recorded.
Methods: We reviewed cases collected by NSQIP from 2005-2014. Control charts were utilized to evaluate the association between variable definition changes and special cause variation in the occurrence rates of surgical outcomes over each admission quarter. Mortality, known to have decreased since the institution of NSQIP without any variable changes, was used as a negative control variable. We particularly focused on postoperative occurrences with the most dramatic variable changes: myocardial infarction (2009), bleeding requiring transfusion (2010), , and sepsis (2013).
Results: We found that the majority of postoperative occurrences decreased over time, consistent with mortality. However, special cause variation was noted for myocardial infarction (2009), bleeding requiring transfusion (2010), and sepsis (2013) with violation of control chart rules at the time of significant variable definition changes (Figure).
Conclusion: This study demonstrates that special cause variation in postoperative occurrences may correlate with significant variable definition changes. This issue severely limits conclusions drawn by comparing these outcomes over time periods with significant definition changes. Authors and readers should remain vigilant to the limitations of NSQIP data and be aware of definition changes when comparing outcomes across years.
