Assessing the Influence of Intensive Care Unit Organizational Factors on Outcomes in Canada: Is There Residual Confounding?

    loading  Checking for direct PDF access through Ovid


Objective:To illustrate the complexities of assessing the influence of individual organizational factors on outcomes for intensive care unit (ICU) patients in Canada.Methods:In this cross-sectional study, we used retrospectively collected data from all patients admitted to all ICUs in British Columbia between April 01, 1994, and March 31, 1999, or to 20 ICUs in Ontario during 2002. We used responses to a survey about organizational factors for all ICUs in British Columbia and Ontario that was done in 2001. Clustered data regression models were used to assess the influence of organizational factors on outcomes after adjustment for patient-level confounders.Results:Despite adjustment for age, sex, comorbidity score of patients, and limitation of the data sets to minimize confounding, we found variable relationships between ICU outcomes (hospital mortality, ICU length of stay, and hospital length of stay) and any of “closed” administrative model, daily rounds, presence of an ICU medical director, medical director trained in critical care medicine, presence of a nurse leader in the ICU, nurse-patient ratio, and presence of an ICU pharmacist. Closed administrative model was associated with lower hospital mortality (odds ratio [95% confidence interval]: 0.75 [0.66-0.85]), shorter length of ICU stay (rate ratio [95% confidence interval]: 0.76 [0.62-0.92]), and shorter length of hospital stay (rate ratio [5% confidence interval]: 0.78 [0.72-0.84]) in nontertiary ICUs but many other “advantageous” factors were associated with higher mortality.Conclusions:The lack of a consistent relationship between organizational factors and ICU outcomes, and relationships that show increased mortality associated with some advantageous factors, suggests that there is residual confounding in these relationships. Models that adjust for only conventional patient-level confounders are unlikely to provide valid estimates of the influence of organizational factors.

    loading  Loading Related Articles