Moderating Factors

A Companion Guide to the National On the CUSP: Stop BSI Project Final Report

Hospital Characteristics

Section Summary

  • Characteristics of the hospital in which a unit resides may serve as a predictor of how well a unit performs in quality improvement initiatives.
  • A statistically significant reduction in CLABSI rate was found over time among adult ICUs with baseline rates greater than zero (p<0.01).
  • Greater improvement was found among teaching hospitals, within a health care system and typically registering after cohort 1. A significant interaction between rural areas and time was also found and results suggest non-government, not-for-profit hospitals may have greater improvement in the initiative.

There are a number of hospital-level characteristics that may contribute to the ability of an ICU to reduce its CLABSI rate. By better understanding how ICUs perform based on the type of hospital in which they are situated, future improvement projects can more appropriately target units that may lag as well as allocate the appropriate amount of resources (i.e. coaching sessions, site visits, etc.).



To be included in the model, units had to be an adult ICU and have baseline data and at least one post-baseline data point. In addition, units with an outlier baseline rate were excluded from analyses (defined as a value of greater than 25; n=3 units). Post baseline data rates greater than 25 were re-coded to missing (n=11 data points). However, the unit overall was not excluded. Data through quarter eight were included in the model. Finally, units were categorized as having a baseline rate of zero (n=293 units) or a baseline rate greater than zero (n=681 units), and the final model was restricted to only units with a baseline rate greater than zero. A flowchart documenting inclusion/exclusion criteria can be found in Appendix E.

Model Selection and Covariates

To examine the relationship between the change in CLABSI rates and hospital characteristics, a mixed effects regression model with random intercept and random trend was utilized with rate over time as the outcome of interest. To assess change in rates over time, an empty model of rate over time was first fitted. Although the slopes in both the linear and quadratic model were significant, the quadratic model fit was statistically significantly better than the linear model (χ2(1)=34.5, p<0.01).

Hospital characteristics of interest were added to the model simultaneously. Hospital characteristics were derived from the 2010 AHA Annual Survey. Characteristics of hospitals that did not participate in the AHA Annual Survey were coded as missing. The following a priori hospital characteristics were considered for inclusion in the model: cohort status (cohorts 1–6), control status (i.e. government, investor owned, etc.), primary service provided, teaching status, rurality, critical access status, part of a health care system, and bed size (<100, 101-75, 176-250, 251-325, 326-400, >400). Next, hospital characteristics that were not statistically significant were removed from the model and an additional interaction term (time) was created for each of these variables. Hospital characteristics and interaction terms that were not statistically significant were then removed resulting in the final, parsimonious model.

Confirmatory Analysis

As a confirmatory analysis, a count model was also utilized with number of CLABSIs per time period serving as the outcome of interest and number of central line days serving as the offset variable. Poisson, negative binomial, and zero inflated distributions were examined with a negative binomial distribution resulting in the best fit due to significant over-dispersion. Characteristics that were statistically significant in the mixed model were included in the confirmatory analysis.


CLABSI rates over time of the final sample can be found in Figure 17. Since adult ICUs included in the model were required to have a baseline rate greater than zero, the overall baseline measure of 2.147 is slightly higher than that found among all adult ICUs (1.915; see Adult ICU Outcomes section above for overall adult ICU outcomes details). Characteristics of included units can be found in Table 15.

Figure 17. CLABSI rates over time among adult ICUs by baseline rate

Table 15. Characteristics of adult ICUs included in dataset with baseline rate greater than zero

Hospital CharacteristicsUnit (n=681)Percent
Control Status
   Non-Government, Not-For-Profit42663%
   Government, Not-For-Profit8613%
   Investor Owner, For-Profit8112%
Primary Service Provided
   General Medical and Surgical58285%
Teaching Status
Critical Access
   Not Critical Access59387%
   Critical Access00%
Health Care System
   Not in a System34951%
   In a System8813%
Bed Size
   <100 6510%
   326-400 7711%
Baseline Rate Greater Than Zero

Results of the parsimonious model can be found in Table 16. A significant effect of time was found with significant reductions in CLABSI rates as the study advanced. This reduction, however, is dampened by the significant quadratic term (β=0.0297, p<0.01) indicating a slowing or "leveling-off" of the decreasing rate at later time periods. Cohort assignment was found to be significant with cohort 2 having a lower rate than cohort 1. Comparison between cohort 1 and cohort 6 was also significant; however, the rate for cohort 6 was significantly higher. There was no statistically significant difference in rate comparing cohort 1 to cohorts 2, 3, or 4. Rurality independently was not statistically significant, but the interaction between rurality and time was significant. Specifically, rural hospitals improved over time more than non-rural hospitals (β=0.117, p<0.01). Hospital teaching status was significant with non-teaching hospitals having a lower rate than teaching hospitals. Hospital control was also significant with investor-owned, for-profit hospitals having higher rates than non-government, not-for-profit facilities. Finally, health care system was significant whereby units residing in hospitals not associated with a system had a higher rate than units in hospitals associated with a system.

Table 16. Mixed regression of hospital characteristics on CLABSI rates over time among adult ICUs with a baseline rate greater than zero (n=681)

Intercept2.778 <.0001
Time (quarter)-0.5049 <.0001
Quadratic0.0297 <.0001
Cohort  0.0016
Rurality  0.328
Teaching Status  0.0058
Health Care System  0.0059
   In a Systemref  
   Not in a System0.30990.0059 
Control Status  0.0194
   Non-Government, Not-For-Profitref  
   Government, Not-For-Profit0.23120.1047 
   Investor Owned, For-Profit0.38740.0128 
Rurality x Time  0.0351
Confirmatory Analysis

Results of the confirmatory count model (negative binomial) were almost identical to the mixed model. Differences were found for the covariate cohorts (all cohorts were significantly lower than cohort 1 with the exception of cohort 6 which was significantly higher than cohort 1) and control status (no significant effect of control status). Go to Table 17 for details regarding the model.

Table 17. Negative binomial regression of hospital characteristics on CLABSI rates over time among adult ICUs with a baseline rate greater than zero (n=681)

Time (quarter)-0.2856<.0001
Cohort 0.0016
Teaching Status  
Health Care System  
   In a Systemref 
   Not in a System0.2028<.0001
Control Status  
   Non-Government, Not-For-Profitref 
   Government, Not-For-Profit0.0920.0668
   Investor Owner, For-Profit0.0930.0929
Rurality x Time  
Current as of January 2013
Internet Citation: Moderating Factors: A Companion Guide to the National On the CUSP: Stop BSI Project Final Report. January 2013. Agency for Healthcare Research and Quality, Rockville, MD.