2009 HSR&D National Meeting Abstract
3039 — Comparing Safety Climate between Different Settings of Care: Methodological Considerations
Hartmann CW (CHQOER), Rosen AK
(CHQOER), Zhao S
(CHQOER), Hanchate A
(Boston University), Singer SJ
(Harvard University), Meterko M
(COLMR), Shokeen P
(CHQOER), Gaba D
(VA Palo Alto)
Prior efforts to measure safety climate have not considered the influence of sample characteristics on differences in safety climate between different settings of care. Straight comparisons between settings may be misleading because different settings’ characteristics may have different effects. Understanding these effects can help target improvements. We undertook a methodological analysis to explore how respondent and hospital characteristics influence safety climate in two settings.
We administered the Patient Safety Climate in Healthcare Organizations (PSCHO) survey at 30 VA hospitals (n = 9,309) and 69 non-VA hospitals (n = 19,881). Data for hospital variables were obtained from the 2004 American Hospital Association Annual Survey. All data were weighted. The unit of analysis was the individual. The dependent variable was the overall percent problematic response (PPR) across 39 PSCHO survey items. All models included respondent and hospital characteristics. We estimated separate hierarchical linear regression models (HLM) for each setting. Recognizing that differences in PPR may arise from systematic differences in distributions of demographic characteristics associated with safety climate and from differential effects of characteristics in the model. We used Oaxaca-Blinder decomposition to quantify the extent to which predicted difference in PPR between settings was due to each.
Survey response rates were 50% and 38% in the VA and non-VA, respectively. All demographic characteristics were significantly different between settings (p < 0.001). The HLM analyses of individual and hospital-level predictors of PPR produced similar results in both settings. Using the non-VA model as reference, decomposition analysis indicated that demographic characteristics were favorable to a lower predicted PPR in the VA compared to non-VA. However, this was overwhelmed by substantially larger differences related to model effects, favoring the non-VA, arising from the differential impact across settings of urban location and region on PPR.
Decomposition analysis demonstrates how differences between settings are influenced by demographic and model effects.
The combination of analyses described here are useful when comparing the influence of respondent characteristics from a single hospital with those of a group, or of respondent and hospital characteristics between two settings. By achieving a more thorough understanding of the predictors of safety climate, we can develop effective interventions to improve it.