Effects of large-scale quality improvement (QI) projects on hospital mortality and length of stay (LOS) are poorly understood. An initial evaluation of the Michigan Keystone Intensive Care Unit (ICU) QI initiative found reductions in healthcare-associated infections. This dissertation extends this evaluation to assess its impact on LOS and mortality for elderly patients admitted to Michigan ICUs. The study evaluates whether: (1) implementation of the Project was associated with reductions in hospital mortality and LOS compared to states located adjacent to Michigan; and (2) differences in ICU catheter-related bloodstream infections following Project implementation were associated with differences in outcomes.

Study aims were addressed using Medicare claims data for ICU patients aged ≥65 years old. Retrospective, comparative study methods were used to evaluate outcomes over time for hospital episodes in Michigan compared to the surrounding region, and outcomes from episodes in participating hospitals during quarters with infection rates greater than zero compared to zero. Analyses accounted for patient/hospital characteristics and clustering of episodes within hospitals.

Compared to baseline, adjusted odds of death (95% confidence interval [CI]) decreased during implementation year, months 1-12 following implementation, and months 13-22 following implementation for the intervention group by 7.7% (3.0-12.1%), 11.7% (6.8-16.3%), and 16.5% (10.5-22.0%), respectively, and for the comparison group by 7.1% (3.8-10.4%), 7.4% (3.9-10.8%), and 9.1% (5.2-12.9%), respectively. Mortality reductions were significantly greater for the intervention than comparison group during the 13-22 month post-period (P=0.038). Compared to baseline, adjusted LOS (95% CI) decreased for the intervention group by 0.25 (0.12-0.37), 0.28 (0.10-0.46) and 0.29 (0.10-0.48) days, respectively, while the comparison group showed no significant reductions. However, differences in LOS reductions between groups were not significant.

Adjusted odds of death were 26.4% higher for patients hospitalized during quarters when infection rates were greater than zero compared to zero, without hospital clustering (95% CI: 19.8%-33.4%), but were not significant when episodes were clustered by hospital (P=0.608). LOS was significantly longer for ICUs reporting positive infection rates when clustering was and was not taken into account (clustered adjusted hazard ratio: 0.834, 95% CI: 0.750-0.918).

Investments in large-scale, successful QI projects may maximize both patient- and financial-related benefits.