Session number: 1127
Abstract title: Preop Risks and Postop Adverse Events from Administrative Data Bases
Author(s):
WR Best - Midwest Center for Health Services and Policy Research, Hines VA, and Loyola’s Stritch School of Medicine
SF Khuri - VA Boston Health Care System, West Roxbury, MA, and Harvard Medical School
M Phelan - Cooperative Studies Program Coordinating Center, Hines VA, and (now) Center for Disease Control, Atlanta, GA
K Hur - Cooperative Studies Program Coordinating Center, Hines VA
WG Henderson - Cooperative Studies Program Coordinating Center, Hines VA
JG Demakis - HSR&D Service, VACO, Washington, DC
J Daley - VA HSR&D Progr(was Sr Resch Assoc),West Roxbury,MA, Harvard Med Sch, and Ctr for Hlth Syst Design and Eval,Inst for Health Policy,Mass Genl Hosp/Partners HealthCare Syst
Objectives: To determine whether routine hospital discharge abstracts contain sufficiently accurate information to replace trained nurse data collectors in obtaining preoperative patient characteristics and 30-day postoperative outcomes on most major operations in 123 DVA hospitals for the National Surgical Quality Improvement Program (NSQIP). Despite poor showing in prior published studies, some have suggested sufficiently accurate information might be obtained from existing databases at much lower cost.
Methods: With preoperative risks and 30-day outcomes recorded by trained data collectors as criteria, ICD-9-CM hospital discharge diagnostic codes in VA’s Patient Treatment File (PTF) were tested for sensitivity and positive predictive value. ICD-9-CM codes for 61 preoperative patient characteristics and 21 adverse postoperative events were identified.
Results: Moderately good ICD-9-CM matches of descriptions were found for 37 NSQIP preoperative patient characteristics (61%); good data were available from other automated sources for another 15 (25%). ICD-9-CM coding was available for only 13 (45%) of the top 29 predictor variables. In only 3 of these (23%) was sensitivity and in only 4 (31%) was positive predictive value greater than 0.500. There were ICD-9-CM matches for all 21 NSQIP postoperative adverse events; multiple matches were appropriate for most. Postoperative occurrence was implied in only 41%; same breadth of clinical description in only 23%. In only 4 (7%) was sensitivity and in only 2 (4%) was positive predictive value greater than 0.500.
Conclusions: Sensitivity and positive predictive value of administrative data in comparison to NSQIP data were poor. At this time we cannot recommend substitution of administrative data for NSQIP data methods.
Impact statement: VA’s NSQIP leads the world in assessing quality of care in surgery on the basis of outcome. Unfortunately, there is a cost in achieving such excellence, and this study indicates that such cost cannot be compromised.