The Geriatric Emergency Department Intervention model of care: a pragmatic trial

De-identified data for all patients aged 70 years and older, who presented to the study ED from January 2012 through to 31st August 2016, were retrieved from Emergency Department Information Service (EDIS; Healthcare Group, CSC)® and Hospital Based Corporate Information System (HBCIS; iSoft) databases. Data extraction was undertaken by health service data managers with deterministic linking of the EDIS, HBCIS and financial data. There were three time periods:

To overcome the ‘improvement-evaporation effect’, whereby the benefits reaped from new practices diminish over time [27] we used data from the immediate period post intervention (interim intervention), when the model was being developed and staffed by the most experienced and dedicated clinicians, but also included the full implementation period which represents a more standard implementation environment.

Independent variables that were used to describe the sample and to build multivariable models to compare outcomes were:

The intervention has been described previously [25, 30] and is detailed within an Implementation Toolkit [31]. The aims of the model of care are to avoid hospital admission, if appropriate, and where this is not possible, to fast-track admission and medical management. Briefly, the GEDI intervention is delivered in the ED by a multidisciplinary team consisting of an ED physician champion and advanced practice ED nurses who have at least 5 years of experience working with older adults and preferably post-graduate qualifications in gerontology. The GEDI nurses operate as a supplementary sub-speciality team assisting the primary ED nurses and physicians. They target all patients over the age of 70 years especially those transferred from RACFs. They will either receive referrals from the primary care team or will identify patients via the electronic medical records system or via routine rounding. The use of frailty screening was trialled but was not found to be of use for these experienced clinicians.

In consultation, with the ED physician champion they will undertake targeted geriatric assessment (using the aspects of a comprehensive geriatric assessment that are appropriate for the individual patient) and problem formulation. They will then work with the primary ED team to fast-track diagnostic processes and engage the multi-disciplinary team and, where possible, the family in client-centred decision-making. Where necessary GEDI will undertake early referral to specialist care and/or activate standardised fast track pathways e.g. for orthopaedic surgery or stroke management.

When necessary, GEDI will coordinate admission to a specialty in-patient ward avoiding, where possible, a stay on a medical assessment unit. When admission is not required, the GEDI team may assist the primary care team with hands on care (e.g. wound care, catheter change etc.) and/or liaise with appropriate community or RACF services, to mobilise resources within the patient’s home to ensure safe discharge. Finally, the team communicates all ED care and future requirements of care to either the ward or the community care team, including the GP.

The GEDI team also provides an ongoing staff development program for other ED staff. Through a program of in-service education sessions new and junior staff are provided with information about the model of care and are also educated about geriatric syndromes, cognitive assessment and the care pathways used in the department.

The study objectives were to test differences in disposition, ED and hospital length of stay, time to ED re-presentation, all-cause mortality and costs between the groups of patients who presented, to the study ED, in the three time periods.

This project and the development of the intervention were a co-design activity that included two members of the public who were over 70 years of age. They are volunteers who spent a lot of their time visiting older adults in aged care facilities and had both personal and proxy experience of the health service and the ED. They not only joined the External Advisory Group set up to guide the development of the intervention and the conduct of the study but also visited the ED and a local aged care facility with members of the research team to discuss issues specific to older adults.

An independent statistician, not otherwise involved in the study, provided statistical summaries and analysis. Descriptive statistics were used to describe the participants in the three time periods including frequencies, percentages, appropriate measures of central tendency and distribution.

For the primary outcome, survival analysis was used to jointly model length of stay and disposition, with the three destinations as competing risks [32]. We used survival analysis for ED re-presentations with out-of-hospital mortality as a competing risk. All models adjusted for the patient level factors of gender, age, ATS, season, day of the week and time of presentation. Survival analysis is ideal here because our primary outcomes are times that are subject to censoring from competing risks. Previous analysis have simply categorised time and used logistic regression, for example discharge within 24 h. However, this wastes valuable information and reduces statistical power. We used cumulative incidence curves to look for changes due to the intervention over all times.

Pre-post designs are vulnerable to confounding by other changes over time that may be attributed to the intervention [33]. To control for this, we included a linear trend (based on date) in all models to account for gradual changes that are not captured in the individual variables, e.g., experience of healthcare workforce. We also adjusted for season using a sinusoid with an annual cycle to control for the winter peak in morbidity [34]. The survival analyses used Cox proportional hazard survival models. The models’ residuals were checked for outliers and correlation over time. We calculated Cook’s influential statistic and examined relatively large outliers. We calculated the variance inflation factor and removed variables with a score above five on the basis that they were co-linear. The key outcome was the mean effect of the intervention together with 95% confidence intervals.

We compared patients who may have received the GEDI intervention to patients for whom it was not available. This “usual care” group could be those who were admitted before the intervention was introduced, or those who were admitted after the intervention but outside the GEDI team’s hours. We also expected the intervention to change during the interim and full period, hence we created five categories:

Hence we had three control groups and two intervention groups. We used the additional control groups because we suspected that there would be carry-over effects from the intervention that may create some additional benefit compared with the pre-intervention controls.

We compared the hospital lengths of stay and costs for the three time periods. We specified a two-stage recursive model which exploits the unidirectional causal pathway among the endogenous outcome variables (i.e. length of stay and cost) such that, for a given set of exogenous variables, the endogenous variables can be identified sequentially [35]. The exogenous variables that potentially influenced the length of stay and cost include: patient individual characteristics (age, gender and place of residence), the properties of the presentation (mode of arrival, major diagnostic category and mode of discharge) and whether or not the patient arrived during working hours. Since the recursive model does not allow variables that are linked in a causal chain to have correlated with the error terms, it can control for unobserved heterogeneity and endogeneity within the data [35, 36].

In this pragmatic trial we aimed to control for a range of factors that previous studies have shown influence outcomes for older adults who present to the ED. The results indicate that the GEDI model increases the likelihood of discharge, decreases ED and, to some extent, hospital length of stay and costs, with no effect on same cause re-presentation or mortality. Older patients who presented in the months when the GEDI model was in place, and were subsequently admitted to hospital, spent 1 day less in hospital per admission, compared with patients presenting when no GEDI model was implemented and this resulted in cost saving. Other factors also influenced these outcomes with general changes in service provision over time increasing the likelihood of discharge directly from ED.

This study capitalised on an opportunity to test the effect of a new model of service delivery over the course of its development and once the final version had been determined. Often new models of service delivery are not sustained [27]. Similarly, the final model tested in this intervention did not perform as well as the interim model (possibly representing the enthusiastic efforts of the innovative clinicians that began the change process) but it still outperformed the model employed prior to the commencement of the intervention.

While this pragmatic study used a pre-post design rather than a randomised controlled design to test the effect of the GEDI intervention, design features were incorporated to provide greater generalisability of the results. Our analysis included a variable to adjust for other changes that may have been occurring, in the study ED, over time (for example, maturation of the whole team or other interventions that were introduced to decrease ED length of stay). We also used the survival analysis taking into account major factors identified in the literature that affect patient outcomes for this cohort.

Functional decline and quality of life are important outcomes for this population. However, these variables are not routinely collected in our hospital and hence could not be examined here because of our retrospective study design.

There are a range of other models of care which aim to improve outcomes for older adults presenting to the ED. Some are outreach models or Hospital in the Nursing Home models in which ED or hospital clinicians travel to RACFs aiming to prevent transfer [15, 37‐40]. These outreach models, unlike the GEDI model, may de-skill RACF staff and result in general medical practitioner disengagement. They may also be very expensive, although it is difficult to understand cost as few robust cost analyses have been undertaken [1]. Additionally, these models only work for the small proportion of older adults living in RACFs.

Other models focus on enhancing the care within the ED. The senior work up assessment and treatment (SWAT) model [41], the Triage and Rapid Elderly Assessment Team (TREAT) model [12], the Aged Care Services Emergency Team (ASET) Program [42], the GEDI WISE transitional care nurses [43] and a range of other models that combine some level of increased geriatric assessment and liaison with community services [44] have been described. Of these models only the first three report rigorous evaluative research outcomes. Unlike our GEDI model the SWAT [41] model, of increased senior medical officer involvement, did not improve overall ED length of stay but did improve length of stay on high volume days and for discharged patients. The TREAT [12] model which involved a highly skilled specialist team of consultant geriatricians and physicians, nurse practitioners and allied health staff, did reduce hospital admission but as there was no economic evaluation it is not possible to compare the costs of this service with the GEDI model. The GEDI WISE [43] transitional care model is the model most similar to GEDI and also resulted in decreased admissions. However, there is no specific physician champion in the GEDI WISE model and the increase in re-presentation seen in their model, but not seen in our GEDI evaluation, may be because of the important role played by the senior medical officer. The physician champion role, in this nurse-led model, is unique to GEDI and ensures there is multi-disciplinary involvement and that the care of older adults is seen as being as important as the care of other cohorts within the ED. In addition, the provision of post ED support to prevent re-presentations in the community dwelling cohort is dependent upon a supportive community-based health network. The ED physician champion plays a role in ensuring collaboration between the ED setting and community health services to provide continuity of client care.

Focusing on improving the care of older adults in the ED and on strategies to prevent inappropriate ED presentation by older adults is of increasing importance in light of population ageing. Sinha and colleagues [17] identified the eight components of ED-based models of care for non-institutionalised older adults. The GEDI model was developed to include all of these components. A key finding of Sinha et al. [17] was that, “collaborative working practices are critical in model implementation and rely on the interpersonal skill sets of the clinicians delivering those initiatives and their ability to earn the trust and respect of their colleagues within and beyond the ED” (p. 680). Our findings suggest that it is possible to have advanced practice nurses, rather than nurse practitioners, in the GEDI model provided there is a strong physician-champion who supports, advises and works collaboratively with the senior nurse. This provides a relatively inexpensive yet effective model of care that reduces length of stay and increases discharge of older adults from the ED, when appropriate.