A Hierarchical/Mixed Model for Calculating Standardized Infection Ratios in Pennsylvania Acute Health Care Facilities

Background:

Hospital acquired infections (HAIs) are major causes of morbidity, mortality, and healthcare costs in the US.  The optimal way to calculate rates of HAIs, especially in the context of state-mandated reporting, has been controversial and challenging.  Traditionally, crude rates have been used to compare infection rates. However, these can lead to erroneous conclusions for facility comparisons because they neglect important confounding hospital characteristics which could greatly influence infection rates.  Indirect standardization is a way to adjust for these risk factors using the Standardized Infection Rate (SIR), the number of observed cases divided by the number of expected.  A 95% confidence interval is used to determine the significance of SIR values. How best to assess potential variables and characteristics into calculation of expected numbers of infections to determine SIRs is unclear.  Recent models look at these factors through Poisson regression and retain those that are statistically significant. This approach has been shown to be good for analyzing rare events and count data.  However, this method neglects potential nested (hierarchical) structures within the data which could affect correlations between the risk factors and the outcome measure. 

Objective:

To utilize hierarchical/mixed models for health care acquired infections at various facilities in Pennsylvania (PA) and consequently obtain SIRs scores for each of the facilities. Compare this approach to determine SIR scores with the more standard Poisson non-hierarchical approach. 

Methods:

In PA, although all HAIs are required to be reported through the National Healthcare Safety Network (NHSN), a smaller number are being used to calculate infection rates (which require denominator data) for obtaining SIRs.  This analysis focuses on calculating SIRs for catheter-associated urinary tract infections (CAUTIs) and central line-associated bloodstream infections (CLABSIs).  Data from July 2008-March 2009 (the baseline period) were used here. Factors assessed in the modeling include: medical school affiliation, number of licensed beds, and the device utilization ratio (DUR) which is the number of device days divided by the number of patient days. The nine months of data were also collapsed to one point of time to increase the sample sizes of infection counts per facility.  A Poisson distribution for these variables was assessed and compared to a hierarchical/mixed model approach in finding expected values for the SIRs.

Results:

We used Poisson models to assess relevant risk factors for both CAUTI and CLABSI. Here, the DUR was generally found to be significant but most of resulting SIRs were not significantly different than a chance result.

Conclusions:

From this exploratory study, we hypothesize that the hierarchical/mixed model is more efficient than other models for evaluating risk factors for the HIAs.