A Comparison of Cross-Sectional and Longitudinal Analyses for Adverse Events Data

• Activity Number: 504
• Type: Contributed
• Date/Time: Wednesday, August 6, 2014 : 10:30 AM to 12:20 PM
• Sponsor: Health Policy Statistics Section
• Author(s): Douglas Morrison*+ and Manisha Desai and Catherine Curtin and Tina Hernandez-Boussard
• Companies: Stanford University and Stanford University and Stanford University and Stanford University
• Keywords: health care quality ; hospital quality measures ; administrative data ; longitudinal analysis ; difference score ; quality indicators

Abstract

Background: Previous studies of inpatient adverse events used cross-sectional data to associate hospital characteristics such as nurse-to-patient ratio (NPR) with outcomes such as failure to rescue from complications (FTR). These studies are vulnerable to confounding. We sought to compare such an analysis with a longitudinal analysis that associates changes in NPR with changes in FTR. Data: We took NPRs from the AHA surveys and FTR rates from the California HCUP SID data. Methods: For the cross-sectional model, we treated each year:hospital combination as a separate data point and regressed FTR on NPR. For the longitudinal model, we regressed the 2008-2011 change in FTR on the change in NPR. Results: The cross-sectional model found a significant association between NPR and FTR (est. corr. -11%, p < 0.01). The longitudinal model found no association (est. corr. -0.6%, p = 0.91). Conclusions: The longitudinal and cross-sectional analyses contradict each other; the longitudinal analysis seems less vulnerable to confounding. By pairing each hospital with its previous performance, we limit possible confounders to those that also change in the same time interval.