Associations between in-hospital bed occupancy and unplanned 72-h revisits to the emergency department: a register study

The study was conducted as a retrospective register study, including all visits to the ED of a 420-bed emergency hospital in southern Sweden between 1 January 2011 and 31 December 2012, not resulting in admission, death, or transfer to another hospital. In order to avoid selection bias, no further selection was made.

The ED of Helsingborg Hospital serves a population of around 250,000. Due to tourism, the population expands to nearly 300,000 during summer. It is one of the four emergency hospitals in the region of Skåne in southern Sweden. The annual ED census is around 60,000, with approximately 15% of patients arriving by ambulance. Patients are registered in the information system Patientliggaren® by a secretary upon arrival. Patients who arrive by ambulance or are referred by a physician gain access to the ED directly after registration. Other patients gain access to the ED in accordance with predefined guidelines, or are further evaluated by a nurse in primary triage. Patients could be referred elsewhere from primary triage (e.g., to primary care). Patients who gain access to the ED undergo secondary triage, which is performed by a nurse. The following is controlled upon secondary triage: Airway, respiratory rate, and SpO₂ (pulse oximetry), heart rate, and blood pressure (non-invasive), alertness (Reaction Level Scale (RLS)), and body temperature. The four-level triage system ‘Medical emergency triage and treatment system’ (METTS) was used for secondary triage during the study period. The triage priority is registered in Patientliggaren® directly after the secondary triage. Only physicians may down-prioritize patients (Table 1).

Patients are directed to separate units for Surgery, Orthopedics, Medicine, and Otolaryngology in a triage-to-specialty model after the secondary triage. A complementary unit staffed by emergency physicians capable of handling various complaints except for psychiatric, otolaryngologic, ophthalmologic, and pediatric (medicine) complaints was introduced in 2010 and operates from 8 am to 11 pm daily. There are separate EDs for children with medical conditions (<18 years of age) and for patients with obstetric/gynecologic, psychiatric, or ophthalmologic complaints. Visits to these EDs were not included in the study. Patients with suspected hip fractures or ST elevation myocardial infarction diagnosed in the ambulance bypass the ED in fast tracks and were not included either. Hand surgery, neurosurgery, and thoracic surgery are not available in the hospital. The availability of endovascular surgery and percutaneous coronary intervention is limited from 5 pm to 08 am. Patients with such needs are referred to Skåne University Hospital and were not included in the study.

Physical ED records for patients who are advised to revisit the ED are stored at each specialty desk. Nurses indicate whether a visit is a planned revisit in Patientliggaren® upon patient arrival. Swedish national reimbursement systems are tied to a goal of 80% of visits with ED LOS ≤4 h. At in-hospital bed occupancy close to 100%, the hospital utilizes full-capacity protocols.

Data was retrieved from the ED information system Patientliggaren® and the in-hospital information system PASiS®. Data matching was performed by the hospital Informatics Unit using QlikView® software.

Occupancy was defined as the overall proportion of occupied beds in the hospital at whole-hour intervals. All patients registered in Patientliggaren® during an interval were assigned the same occupancy.

The proportion of visits resulting in an unplanned 72-h revisit was computed for in-hospital occupancy levels of <85%, 85% to 90%, 90% to 95%, 95% to 100%, 100% to 105%, and ≥105%. Subgroup analysis was performed for each specialty unit. Computations were repeated for unplanned 72-h revisits resulting in admission.

Adjusted analysis was performed in an attempt to take confounding factors into account, using binary logistic regression models. Perceived clinical significance governed the decision of screened predictors but was inevitably tainted by data availability. Screened variables were the following: specialty unit, presenting complaint at index, referral status at index, triage priority at index, age group, sex, index presentation on an intense shift, index presentation on a night shift and during weekends, leaving without being seen (LWBS) at index, entering ED via primary triage at index, time to physician at index, and in-hospital occupancy at index. The variable indicating presentation on an intense shift was constructed as a dichotomous variable indicating presentation on one of the 25% of shifts subject to most visits (adjusted for shift type and unit). Night shift was set from 12:00 mn to 08:00 am. Presenting complaint was constructed as a nominal variable indicating the ten most common complaints, using the remainder as reference.

The medicine unit was used as reference among the specialty units. Age was grouped into intervals 0 to 18 years, 18 to 40 years, 40 to 65 years, and ≥65 years. Age ≥65 years was used for reference. Youths in Sweden become of age at 18 and pension age is 65 years.

For the multivariate models, in-hospital occupancy was categorized as <85%, 85% to 90%, 90% to 95%, 95% to 100%, 100% to 105%, and ≥105%. The reference interval was set to <85%. Sensitivity analysis was performed using occupancy <95% as reference.

Predictors were tested for crude association with the outcome before entering the preliminary primary effects model. Associations weaker than P = 0.25, but of clinical importance were still included [[18]]. Multicollinearity testing was performed using Spearman correlation [[19]]. Selection of interaction terms screened for inclusion in the final models was governed by perceived clinical significance and made a priori to analysis. Variables were manually added to the models, rather than stepwise [[18]]. Missing data was indicated by a separate category and included in the models [[20]].

Model fit was evaluated through Nagelkerke’s R². The association between each predictor and the outcome was addressed by the −2LL and the Wald statistics. The final models were the models with the highest explanatory value and the fewest number of predictors [[19]]. Additionally, the models were screened for influential cases by addressing standardized residuals and Cook’s distance.

Statistical analyses were performed in IBM® SPSS® Statistics 19. Data was anonymized before analysis. The Regional Ethical Review Board in Lund granted ethical approval for the study.

Out of the 81,878 cases, 4,753 cases resulted in an unplanned 72-h revisit, and 1,213 cases resulted in an unplanned 72-h revisit and admission. Proportions of unplanned 72-h revisits/unplanned 72-h revisits resulting in admission were 5.8%/1.5% overall, 6.2%/1.4% for occupancy <85%, 6.4%/1.5% for occupancy 85% to 90%, 5.8%/1.4% for occupancy 90% to 95%, 6.0%/1.6% for occupancy 95% to 100%, 5.4%/1.6% for occupancy 100% to 105%, and 4.9%/1.4% for occupancy ≥105% (Figure 1).

All predictors screened for inclusion in the multivariate models were included, except from time to physician, which was omitted as it violated the assumption of linearity in the logit [[19]]. No significant associations between in-hospital bed occupancy at the index visit and the probability for unplanned 72-h revisits was observed in the model using occupancy <85% as the reference. In the sensitivity analysis, a trend to lower odds for unplanned 72-h revisits was observed among patients being discharged from the ED at occupancy ≥105% relative to at <95%, OR 0.88 (CI 0.76 to 1.01, P = 0.062). No significant associations between in-hospital bed occupancy and the probability for unplanned 72-h revisits resulting in admission were seen in either model. A full account of the models (including coefficients of overall fit) is shown in Additional file 1 (Figures 2 and 3).

The Nagelkerke R² coefficients (given in Additional file 1) indicate that the variables that were not available for study, e.g., diagnosis and co-morbidity, influenced the outcome in the adjusted analyses. This is also supported by the presence of some influential cases. The relatively large sample size is thought to have balanced some of this effect. As diagnosis and IPLOS vary across specialties, it might have been better to model occupancy in different in-hospital units separately. Unfortunately, this was not possible. The external validity of the results is limited, as the study was performed in a single hospital. The fact that some groups of patients are cared for in separate EDs (children with medical conditions and patients with obstetric/gynecologic, psychiatric, or ophthalmologic complaints) and others bypass the ED in fast tracks (patients with STEMI diagnosed in the ambulance and patients with suspected hip fractures) is important to note when comparing results to other EDs. Another limitation is that patients making an unplanned revisit to another ED in the region are not included in the study, but empirical knowledge suggests that this fraction is small. The authors also recognize that the chosen outcomes are not designed to evaluate the appropriateness of ED discharges. Their selection was motivated by frequent use in other studies and that the Swedish National Board for Health and Welfare made unplanned 72-h revisits subject to national follow-up from April 2013.