Hierarchical Generalized Linear Models for Time Trends and Seasonality in Nursing-Sensitive Outcomes

• Activity Number: 677
• Type: Topic Contributed
• Date/Time: Thursday, August 8, 2013 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistical Consulting
• Author(s): Jianghua He*+ and Vincent Staggs and Sandra Bergquist-Beringer and Nancy Dunton
• Companies: University of Kansas Medical Center and University of Kansas Medical Center and University of Kansas Medical Center and University of Kansas Medical Center
• Keywords: Hierarchical generalized linear models ; Time trend ; Seasonality ; Nursing-sensitive outcomes

Abstract

The National Database of Nursing Quality Indicators (NDNQI) collects quarterly unit-level nursing data from 1,900 US hospitals. This data are unique for studying recent trends and seasonality in nursing-sensitive outcomes, which are important yet have not been well-studied. We studied nursing-sensitive outcomes such as inpatient falls and hospital-acquired pressure ulcers in 2004-2011. Due to the complex data structure (units within hospitals and repeated measures within units), we used hierarchical generalized linear models with random intercepts and/or time trends at both unit and hospital levels. For inpatient fall rates, an overall decreasing trend was observed for most unit types except that an increasing trend was observed for surgical units. For the rate of hospital-acquired pressure ulcers, a unique seasonality pattern was observed besides an overall decreasing time trend; the seasonality was strong and consistent in 2004-2008, but its magnitude was greatly reduced in 2009-2011. These changes may be explained in part by multiple recent nationwide quality improvement campaigns and the 2008 reimbursement rule change by Centers for Medicare & Medicaid Services .