Clostridium difficile infection as a novel marker for hospital quality, efficiency and other factors associated with prolonged inpatient length of stay

• Main Author: Miller, Aaron Christopher
• Other Authors: Polgreen, Linnea
• Format: Others
• Language: English
• Published: University of Iowa 2015
• Subjects: publicabstract; Clostridium difficile; Disease surveillance; Hospital efficiency; Hospital quality; Length of stay; Pharmacy and Pharmaceutical Sciences
• Online Access: https://ir.uiowa.edu/etd/1884; https://ir.uiowa.edu/cgi/viewcontent.cgi?article=5941&context=etd

Excess inpatient length of stay (LOS) varies between hospitals and is burdensome to patients and the overall healthcare system. Variation in LOS has often been associated with hospital-level factors, such as hospital efficiency and quality. Clostridium difficile infection (CDI) is an increasingly common hospital-acquired (HA) infection. This thesis explores the connection between hospital incidence of CDI and excess LOS in patients without a CDI. It is hypothesized that HA-CDI incidence may act as a "proxy variable" to capture unobserved hospital characteristics, such as hospital quality or efficiency, associated with prolonged LOS. In addition, hospitals with longer LOS may tend to observe more HA-CDI cases prior to discharge. This thesis analyzes the ability of CDI incidence to capture excess LOS variation across hospitals, while controlling for CDI cases that occur after discharge. We use data on hospital inpatient visits, spanning the years 2005-2011, from three data sources distributed by the Healthcare Cost and Utilization Project: the Nationwide Inpatient Sample (NIS), and the State Inpatient Databases (SID) for California and New York. The NIS provides discharge records from a nationwide sampling of hospitals in a given year. The SIDs are longitudinal populations of inpatient records in each state, and patient records can be linked across stays. We compute a variety of different measures of hospital CDI incidence and identify HA-CDI cases that occur after a patient is discharged. Various multivariable regression models are analyzed to predict LOS at an individual patient level. A generalized linear modeling approach is used, and different distributions and link functions are compared using the Akaike information criterion. A multilevel modeling approach is also used to estimate the amount of between-hospital variation in LOS that can be explained by HA-CDI incidence. We find CDI incidence to be a strong predictive factor for explaining a patient's LOS and is one of the strongest predictive variables we identified. Moreover, CDI incidence appears to primarily capture between-hospital variation in excess LOS. Although we find evidence that present-on-admission indicators may underreport cases of HA CDI, our findings suggest the connection between CDI incidence and excess LOS is driven primarily by CDI cases that are HA. In addition, when we account for HA-CDI cases that occur post-discharge, the relationship between CDI incidence and LOS appears even stronger. Our results suggest that CDI incidence may be a powerful tool for making comparisons of excess LOS across hospitals.