Study Objectives: Emergency department (ED) crowding is prevalent, and existing research suggests that ED crowding is associated with increased patient mortality. However, prior studies are limited by small hospital samples, inclusion of specific patient subgroups, and inadequate case mix adjustment; furthermore, studies are not directly comparable because of different measures of ED crowding. As a result, the external validity and causal mechanisms of ED crowding measures are unclear. Identification of valid metrics is critical for the proper design of public reporting and pay-for-performance initiatives. Our aim was to assess the predictive validity of multiple ED crowding measures for inpatient mortality. Methods: We performed a retrospective cohort study using data from a regional, integrated health system (Kaiser Permanente Southern California- KPSC). We included ED visits from 2008 to 2010 occurring at 13 KPSC hospitals which resulted in hospital admission. All hospitals had implemented a systemwide, electronic medical records system during the study period. Exclusion criteria included age<18 years, visits by non-members, transfers to other hospitals, discharges from the ED, and patients receiving observation status care. The outcome was inpatient mortality. The unit of analysis was an ED visit. We assessed system- and visit-level measures of ED crowding. System metrics included exposures to ED occupancy and external length of stay, both at ED arrival and averaged over an index visitor's length of stay (LOS). Visit metrics included waiting, evaluation, and boarding time, as well as total LOS, experienced by an index visitor. To assess for non-linear effects, all ED crowding measures were categorized into quintiles. Hierarchical logistic regression models accounted for visit clustering by hospital and person. Covariates included demographic characteristics, 29 pre-existing co-morbidities noted from health service records in the preceding year, Emergency Severity Index triage level, triage vital signs, primary hospital diagnosis, time variables for year, month, day, and shift, and ED site. Results: The study cohort included 326,487 ED visits resulting in hospitalization by 201,036 unique patients. Prior to covariate adjustment, all system metrics, boarding time, and total LOS were predictive of inpatient mortality (p<0.05). In adjusted analyses, boarding time was strongly predictive of mortality, with a threshold effect seen at 1.6 hours (1.7-2.8 hours, OR: 1.1, 95%CI: 1.02, 1.2; >2.9 hours, OR: 1.2, 95%CI: 1.1, 1.3). None of the other system or person-level metrics predicted inpatient mortality after covariate adjustment (p>0.3). Conclusion: Surprisingly, system measures of ED crowding and most visit-level time intervals were poorly predictive of mortality, suggesting that threats to patient safety is driven by the experience of boarding rather than general exposure to ED crowding. Our findings suggest that measurement efforts and patient safety interventions should focus on boarding time. Other potential measures may be confounded by patient case mix or hospital level factors.
Read full abstract
Jun 1, 2015
Araz Rawshani + 5
Open Access
