What evidence is there for the identification and management of frail older people in the emergency department? A systematic mapping review

Stage 1: search of evidence retrieved for earlier review and scoping search

An initial search (in May 2016) was undertaken using the evidence base retrieved for the Turner et al. 3 review. These references were filed in an EndNote library (version 8; Clarivate Analytics, Philadelphia, PA, USA) and this was searched using terms for older people and frail older people. The purpose of this search was to provide an initial idea of the size and scope of the available literature and to refine search terms for the database search. The following keywords were searched for in the title of the references: ‘ageing’, ‘aged’, ‘elderly’, ‘frail’, ‘old’ and ‘geriatric’.

Additionally, a search was conducted in May 2016 in MEDLINE (via OvidSP) for reviews and other relevant literature; it was developed using pre-existing search strategies, used for reviews in the same topic area, devised by information specialists at the University of Sheffield. The search was structured using terms for population (frail older people) and setting (ED). The search was not limited by intervention type as an a priori decision about which interventions were to be included could have limited our understanding of the scope of the topic. The search was limited to evidence published from 2005 to 2016 to ensure currency of the included research and limited to English-language-only papers as time constraints meant that it would not have been feasible to translate non-English-language papers. The search was not limited to any specific geographical region as published search filters to identify evidence from specific countries are not always successful. The MEDLINE search strategy is provided in Appendix 1.

Stage 2: search of health and medical databases

The second search, undertaken in July 2016, involved a wider range of health and medical databases. The following databases were searched, with the MEDLINE search adapted appropriately for the different databases:

• EMBASE via OvidSP

• The Cochrane Library via Wiley Online Library

• Web of Science via Web of Knowledge via ISI

• Cumulative Index to Nursing and Allied Health Literature via EBSCOhost

• Health Management Information Consortium via OpenAthens

• PROSPERO.

Stage 3: complementary searching

We also undertook a number of complementary searches (in autumn 2016) to ensure that we had retrieved all relevant evidence for the review. These included scrutiny of reference lists of included papers and relevant reviews. Any relevant papers that were within our date range were obtained and, if they met the inclusion criteria, were included in the review. The reviews used for this exercise are detailed in Appendix 2. In addition, we undertook citation searching of included primary studies that focused on a frail or at-risk population.

Screening criteria

We limited the evidence included in our review to that published from 2005 to 2016. The reason for this was related to the volume of evidence in the area and the need to retrieve a manageable evidence base. In addition, earlier evidence would have been identified and included in the many evidence reviews published in this area. Restricting the date ensured that the evidence included was relevant to the current clinical environment.

Notably, the review does not include ‘frail older people’ as an inclusion criteria. Throughout the process of the review, from the development of the protocol onwards, it became clear that identifying papers that had a population of frail older people according to predefined criteria would be challenging. Had we included only evidence from papers in which the authors had defined their population as frail, or their intervention as targeted at frail older people, then we would have limited the review, as scrutiny of titles and abstracts often did not reveal the population included. Therefore, we took the approach at the screening stage to include all studies in which the population was aged ≥ 65 years and then, at a later stage, further divided these studies into those including frail older people and those including a general population of older people.

Screening process

Screening was undertaken by three reviewers (LP, AC and DC). All titles and abstracts retrieved by the searches were entered into EndNote and EndNote was used for screening. All titles and abstracts were screened by one reviewer (either LP or AC), with DC screening 50% of the titles and abstracts screened by LP and 50% of the titles and abstracts screened by AC (i.e. 50% of all titles and abstracts). The decisions made about whether articles should be ‘included’, ‘excluded’ or ‘queried’ were noted in EndNote. Any queries were discussed with a fourth reviewer (JT) until consensus was reached. The inclusion and exclusion criteria were used to guide this discussion. Queries tended to be around the setting of an intervention and whether or not it was delivered in an ED setting. Articles that met the inclusion criteria that were (systematic) reviews were also marked as ‘include’ and background articles were also identified. To check the screening consistency of the two reviewers, a third reviewer screened approximately 50% of the references as detailed above and a kappa coefficient was calculated.

Characteristics of included studies: screening

Thirty-seven studies (40 publications18–23,27–60) dealt with strategies aimed at identifying patients with frailty or distinguishing higher- from lower-risk patients in the ED. The great majority of these studies assessed the diagnostic or prognostic accuracy of tools using a prospective or retrospective cohort design, which is an appropriate design for this type of study. Only one study (published as a conference abstract) used a randomised trial design18 and one was a secondary analysis of data from a randomised trial. 19 Both of these studies were conducted in the USA.

The largest group of studies came from the USA (n = 1218,19,35,36,38–40,45,46,48,57,59), followed by Canada (n = 528,37,47,51,52). Among European countries, the largest numbers of studies were performed in Switzerland (n = 322,33,42,53,54) and Italy (n = 421,27,49,58). The Netherlands (n = 232,55) was the only other European country with more than one included study. The only study included from the UK was a study of a screening tool that was reported in abstract form only. 20 Outside Europe, studies were included from Australia,44,60 New Zealand,29,30 Turkey43 and South Korea. 23

The numbers of patients included in screening studies ranged from 6918 to 2057. 21 Two other studies22,23 recruited > 1000 patients. Most studies recruited patients aged ≥ 65 years, but the average age of patients actually recruited was considerably older, typically in the mid-seventies or older (see data extraction tables in Appendices 4 and 5). The proportions of men and women included varied among the included studies.

Characteristics of included studies: interventions

Fifty-three studies (63 articles24–26,61–120) examined changes made to how ED services were delivered to populations of (frail) older people. These studies tended to investigate changes to the structure of the ED (n = 1186–97), changes to staffing in the ED (n = 2126,61–85) or changes to how care is delivered (n = 1824,25,98–117), such as the introduction of comprehensive geriatric assessment (CGA) or similar assessment-type interventions. There were also a number of unique interventions (n = 3119–121), which are also reported here.

The majority of the studies reported here were observational studies; predominantly before-and-after studies or cohort studies. Three studies reported results from randomised controlled trials (RCTs). 24–26

All of the studies reported either patient or health service outcomes that were derived from patient data, with the exception of one study,120 which reported changes in ED clinician prescribing behaviour. The main patient-related outcome measures were mortality, functional status, frailty and place of residence (own home or residential/nursing care). The main health service outcomes were admissions, readmissions, ED reattendance and length of stay.

The largest group of studies came from the USA (n = 1564,66–68,70,84,91–94,96,97,113,114,116,119,120), followed closely by the UK (n = 1361,62,72,74,83,86,88,89,95,98,102–106,117). Nine studies were undertaken in Australia. 63,73,76–80,82,110

Most studies reported outcomes for patients aged ≥ 65 years, as these patients were considered to be ‘older people’ and, therefore, the target age for identification of frailty or at risk of adverse outcomes. However, when a mean age was reported, this tended to be > 75 years (see Appendix 6; more detailed reporting of age is not possible because of variable reporting in the included articles). The proportion of men and women included varied among the studies.

Detailed analysis of study and intervention characteristics was hindered by the limited data in the included papers, many of which were conference abstracts.

Diagnostic tools to identify frailty

We included seven studies (nine publications20,22,27–33) of diagnostic tools to identify frailty (Table 5). These were studies that recruited a sample of older people attending the ED and assessed the accuracy of a screening tool against a reference standard.

The included studies evaluated a wide variety of screening tools. The Identification of Seniors at Risk (ISAR) tool was the only tool to be evaluated in two studies. 27,28 A diagnostic accuracy study27 reported that the ISAR tool had a sensitivity of 94% and specificity of 63% relative to a frailty measure, the Deficit Accumulation Index (DAI). The area under the receiver operating characteristic (ROC) curve was 0.92, indicating a good performance in identifying frailty based on the DAI definition. However, a study of the implementation of the ISAR tool in a Canadian ED setting found that only 51.6% of eligible patients actually received an ISAR screen. 28 This was attributed to the fast-paced nature of emergency care and lack of staff resources at night.

Other screening tools have been evaluated in single diagnostic accuracy studies. The Brief Risk Identification for Geriatric Health Tool (BRIGHT), developed in New Zealand, is an 11-item tool that showed a good ability to identify older people with ‘decreased function’ relative to a reference standard of CGA. 29,30 The limitations of this study, identified by the authors, include that it was a small, single-centre study and that 18% of patients who completed BRIGHT were lost to follow-up, raising the possibility of follow-up bias. BRIGHT is designed to be suitable for completion by the patient or a carer and used in combination with a particular type of CGA.

The only other fully published study of this type evaluated FRESH, which is a five-item tool (subsequently reduced to four items) that was specifically designed to screen for frailty. 31 FRESH was evaluated using a range of frailty indicators as reference standards and performed well, with both sensitivity and specificity being around 80%. The test takes only a few minutes to administer and requires minimal input from the older person. However, the tool has been evaluated in only one small study to date (n = 161) and the data were not collected during the ED visit but during a subsequent visit to patients at home. 31

Finally, of three diagnostic accuracy studies published only as conference abstracts, one was carried out in a UK setting. 20 This study used the Clinical Frailty Scale (CFS), a rapid and simple case-finding tool, to assess 118 older patients admitted to geriatric wards from the ED. The CFS performed well in comparison with established frailty scales at appropriate cut-off points. The authors suggested that use of the CFS as a triage tool in the ED could increase the proportion of frail older people admitted directly to geriatric wards (i.e. admitted earlier rather than later). However, although this was a study of a relevant population, data were not actually collected in the ED and patient management and outcomes were not evaluated. Thus, the value of this study by itself appears to be limited.

The other two conference abstracts evaluated an ED screening tool32 and an emergency geriatric screen (EGS). 22,33 The ED screening tool performed well, with an area under the curve (AUC) of 0.83 relative to a reference standard, described as a safety management screening bundle. However, few details of either tool were reported in the abstract. The second study used actual ED diagnoses as the reference standard and reported an increase in the detection of potentially overlooked geriatric problems compared with a control period.

Overall, the evidence for the diagnostic accuracy of tools for identifying frail older people is limited. None of the tools has been evaluated extensively using this methodology and differences in terminology make it unclear whether or not different studies are examining the same phenomenon. In addition, individual studies have different methodological features and settings, which may limit their internal or external validity. However, the evidence base using follow-up to evaluate the predictive abilities of these tools is more extensive and the evidence summarised here should be read alongside Prognostic tools for adverse events after discharge.

Diagnostic tools to identify specific frailty-related issues

We identified seven diagnostic accuracy studies of tools to screen for specific frailty-related issues (as distinct from frailty as a general overall condition) in the ED (Table 6). All of the studies evaluated screening for cognitive impairment/dysfunction and most used the Mini Mental State Examination (MMSE) as a reference standard. Two studies did not use a standard diagnostic accuracy design. 18,34 In a randomised trial published as a conference abstract,18 physicians were either informed or not informed of the results of screening for mental status and delirium. The study found that information about screening results did not appear to influence physicians’ decisions in relation to documentation, disposition or management. 18 This is a potentially important finding, but the study was small (69 patients).

Hadbavna et al. 34 also did not use a conventional diagnostic accuracy study design in their study evaluating the six-item screener (SIS) test and triage risk screening tool (TRST). Instead, repeat screening with the SIS was used to confirm whether or not patients met the criteria for cognitive impairment. The authors found that there was considerable variation between nurses in the implementation of screening. 34 This adds to the study of Asomaning and Loftus28 in identifying potential problems in administering screening tools in normal clinical practice.

Prognostic tools for adverse events within the emergency department

We included five studies evaluating the accuracy of screening tools for assessing patients’ risk of adverse events within the ED itself (Table 7). Each study used a different tool, suggesting that there is currently no consensus around which tools to use. Follow-up was limited to the time that the patient was in hospital, with the exception of one study that included a 30-day follow-up. 40 This study40 found that a delirium prediction rule based on age, prior stroke or transient ischaemic attack, dementia, suspected infection and acute intracranial haemorrhage had good predictive accuracy for delirium determined by the confusion assessment method (CAM).

One study carried out in France used a brief geriatric assessment (BGA) method to identify patients in the ED who were at high risk of a long hospital stay. 41 The BGA consisted of six items and the authors concluded that a history of falls, male sex, cognitive impairment and age of < 85 years identified patients at increased risk of a long hospital stay (≥ 13 days). The authors noted that this group of patients would require geriatric care and planning for discharge. Further evidence on the management of patients following geriatric assessment in the ED is presented in Narrative synthesis of service delivery intervention papers.

The other studies in this group evaluated tools for predicting the risk of hospital or intensive care unit (ICU) admission or the need for an immediate life-saving intervention. Emergency Severity Index level 1 had low sensitivity (46.2%) but high specificity (99.8%) for predicting the need for a life-saving intervention. 42 The index level was also correlated with resource consumption, disposition, ED length of stay and survival. The Canadian Triage and Acuity Scale (CTAS) showed both high sensitivity (97.9%) and high specificity (89.2%) for the need for life-saving intervention. 23 The results of a Turkish study evaluating the Rapid Emergency Medicine Score (REMS) and Hypotension, Oxygen saturation, low Temperature, ECG changes and Loss of independence (HOTEL) tools indicated that these tools cannot be efficiently used to identify older ED patients requiring hospital admission. 43 However, the tools had reasonable validity for predicting ICU admission and in-hospital mortality. The HOTEL score was a stronger predictor than REMS or REMS without taking age into account.

These studies focus on the short-term outcomes of older patients attending the ED. The exception is the study by Beauchet et al. ,41 which may be read alongside other studies of geriatric assessment in the ED. The limited number of studies identified makes it difficult to draw conclusions about which tools may be of most value in the setting of the UK NHS.

Prognostic tools for adverse events after discharge

Eighteen studies (19 publications19,21,44–60) assessed the ability of screening tools to predict adverse outcomes following a patient’s discharge from the ED (Table 8). The studies evaluated a wide range of tools, with follow-up ranging from 28 days to 12 months. The ISAR tool and TRST were most commonly evaluated, with one study44 evaluating a tool derived from the ISAR tool. None of the included studies was performed in the UK. Four studies were published as conference abstracts only. 45–48

The ISAR tool was developed in Canada in the 1990s. 122 It is a self-report screening tool with six questions related to functional dependence, recent hospitalisation, impaired memory and vision and polypharmacy. A score of ≥ 2 (i.e. positive answers to two or more items) is the normal cut-off point for being considered high risk. Two studies in this review evaluated the ISAR tool alone for screening older patients in the ED. 49,50 Both studies concluded that the ISAR tool was a valid and reliable screening tool in their setting. Singler et al. 50 used a cut-off point of ≥ 3 rather than ≥ 2 in their study, which would have the effect of increasing the specificity of the tool. A study of a screening tool derived from the ISAR used a before-and-after design and found a decrease in re-presentation to the ED after introduction of the tool. 44 The authors suggested that this was attributable to an increase in referrals to community-based services, which diverted patients away from attending the ED.

The TRST is a risk-screening tool designed to be applied to patients aged ≥ 75 years in the ED. Like the ISAR tool, it includes six items and a score of ≥ 2 indicates high risk. Three studies in the review evaluated the TRST alone and two of them51,52 cast doubt on the predictive ability of the tool. By contrast, a study in the USA concluded that the TRST was a valid measure for assessing functional status in the ED and may be useful in identifying patients requiring referral or monitoring after discharge. 19 Thus, the evidence base for the TRST when evaluated alone is limited and mixed.

Although evaluation of single screening tools appears most feasible for delivery in the ED and the least burdensome for patients, many studies have compared two or more tools using the same sample of patients. Three studies compared the ISAR tool and the TRST. Salvi et al. 21 and Graf et al. 53,54 both concluded that the tools are useful for risk stratification in the ED and have similar properties. However, Salvi et al. 21 emphasised use of the screening tools to select patients who could benefit from geriatric interventions, whereas Graf et al. 53,54 favoured their use to avoid unnecessary intervention. By contrast, a US study45 found that neither tool successfully distinguished patients at high and low risk for adverse outcomes at 1 and 3 months. Once again, it is difficult to draw firm conclusions from this group of studies.

Two further studies compared the performance of ISAR and TRST with that of two other tools, the Runciman123 and Rowland124 questionnaires. Moons et al. 56 highlighted the value of the Rowland questionnaire for predicting repeat ED visits, whereas Buurman et al. 55 found that none of the screening instruments distinguished between patients with and without poor outcomes over 120 days of follow-up. These similarly designed studies were carried out in Belgium and the Netherlands, respectively, and so their relevance to UK settings is uncertain.

Other screening tools have been evaluated in single studies. We included seven studies of this type, all of which reported positive results. The emergency screening instrument (ESI),57 rapid screening assessment46 and Survey of Health, Aging and Retirement in Europe Frailty Instrument (SHARE-FI)59 are short question-based tools similar to those discussed above. Eagles et al. 47 evaluated the timed up and go test (TUGT) and reported that scores were associated with frailty, functional decline and fear of falling. Limited details of this study are available as it was published as a conference abstract only. Two studies described tools to predict specific frailty-related outcomes: falls60 and ED readmissions. 48 Finally, the Silver Code differs from other risk screening tools by being derived from administrative data. When compared with the ISAR tool, the Silver Code showed a similar ability to predict ED return visits, hospital admission and mortality over 6 months of follow-up. 58 The concept of using administrative data to support initial triage in the ED seems attractive, but in this study the Silver Code was derived retrospectively several months after the patient was enrolled for the study. As noted by the authors, improved processing and flow of administrative data would be necessary for the data to be used for real-time triage in the ED.

Summary of screening papers

The evidence on tools to support the identification and management of patients with frailty in the ED is extensive but inconclusive. The ISAR tool and the TRST are the most extensively evaluated tools, but many other tools are available, including non-question-based tests and, potentially, tools using administrative data. Limitations of the included studies include the small sample sizes, that most were conducted at a single centre and that many were published as conference abstracts with limited details. Contradictory results obtained in different prognostic studies using the same tool reflect the fact that health service use-related outcomes, in particular outcomes such as repeat ED visits and hospital admission, will be influenced by the health and care system as well as by patient factors. Hence, the results of studies performed in one country cannot be readily generalised to another country. The lack of UK studies in this body of evidence limits the relevance of the evidence to NHS settings. There are other studies that examine screening tools for conditions that are common in frail older people; however, these have not been included in the review as they were not identified through the literature searches as they were not specifically limited to a frail or older population.

Emergency department staffing initiatives

We identified 21 studies (26 publications26,61–85) reporting on instances in which the staffing of the ED had been modified to better meet the needs of an older population. These staffing modifications varied; there were examples of initiatives in which a single individual was located in the ED or added to an existing multidisciplinary team (MDT) or in which a new MDT was established. Differentiating between staffing initiatives and care initiatives (e.g. when CGA was introduced to an ED and delivered by a newly established geriatric liaison nurse) was problematic. The description of the interventions was often brief, reflected in the fact that a number of the studies were reported in conference abstracts only. Details of these interventions are given in Table 9.

Changes to the physical infrastructure of the emergency department

Eleven studies (12 articles86–97) reported changes to the ED in terms of the physical infrastructure of the ED. These interventions ranged from the creation of geriatric emergency departments (GEDs) (which will also have included changes to staffing), through to making EDs ‘frail friendly’, making general changes to the ED to benefit all patients but with specific benefits for frail and older people and the establishment of specific units in the ED to meet the needs of frail and older patients. These studies are presented in Table 11.

A UK study by Silvester et al. 86 reported on the redesign of the system of care for older people. This consisted of the formation of a MDT with a clinical systems engineer who facilitated changes in discharge, 7-day working and the designation of a medical assessment unit as a frailty unit with a co-located MDT. Analysis of data over 2 years (before and after the changes) demonstrated a fall in bed occupancy rates, a fall in mortality rates and unchanged rates of readmission.

A key UK study by Ellis et al. 88 evaluated an ACE unit. The four-bedded ACE unit undertook CGA, with the aim of admissions avoidance or direct specialty admission. The study was a non-randomised trial comparing three groups of patients: (1) patients admitted before the ACE unit was set up, (2) patients admitted to the ACE unit and (3) patients admitted to the medical receiving unit outside the hours that the ACE unit was open. The study measured a number of outcomes for patients receiving ACE care and found that there was an increase in same-day discharge, mixed findings on length of stay and no significant findings in terms of 7- and 30-day readmission, 12-month mortality, admission to residential care or living at home.

Another key UK intervention was the establishment of an emergency frailty unit (EFU) within an ED in the UK. 89 The study was a before-and-after study and the outcome measures were admission rate from the ED, readmission following an ED visit, length of stay for admitted patients and total bed-day use. The EFU, which had 8–12 beds and undertook CGA, was staffed by geriatricians, emergency physicians, physiotherapists, occupational therapists and ‘primary care co-ordinators’. Analysis of the pre- and post-intervention data indicated that, although there was a pattern of increased ED attendances over the period of the study, admission rates fell significantly, from 69.6% in 2010 to 61.2% after the EFU was implemented. Readmission rates also decreased (4.7% vs. 3.3% at 7 days; 12.4% vs. 9.2% at 30 days; and 19.9% vs. 26.0% at 90 days). The EFU demonstrated a clear improvement in service delivery outcomes; however, no data on patient outcomes, such as mortality, were collected.

Salvi et al. 90 reported on the patterns of use of a GED (an ED with a six-bed elderly observation unit staffed by geriatricians) by frail older people. Comparing patients who had used the GED (n = 200) with those who had used a conventional ED, the patients using the GED had a small but significantly lower mortality rate.

Pareja-Sierra et al. 87 described the impact of an emergency department observation unit (EDOU) on admissions and length of stay. The EDOU was a small, six-bed unit staffed by geriatricians and targeted at frail older people. The author compared data from before and after the EDOU was set up. Although data were limited, the authors reported that an initial increase in admissions was followed by a decrease in admissions and length of stay.

Genes et al. 91 reported on patient satisfaction with a GED that combined structural enhancements with service delivery changes. Analysing patient satisfaction data from 286 patients (67 of whom were described as geriatric) surveyed both before and after the GED was established, the authors found that, although overall satisfaction scores did not change significantly for either group, the geriatric group saw significant improvements in satisfaction relating to specific aspects of the GED.

Admissions data were analysed by Karounos et al. 92 following the introduction of a GED. Examining data from pre and post introduction of the GED, there were significantly fewer admissions post introduction [2.9% fewer (n = 1130); p < 0.001]. This was a large data set (n = 27,838), although the authors caution that a further analysis on readmissions and costs is required.

Keyes et al. 93 also looked at admissions, length of stay and ED visits following the introduction of a senior ED and compared data with those from before the senior ED was introduced. The senior ED comprised a number of changes including staff education, changes to physical infrastructure and screening. Study outcomes demonstrated that the senior ED was associated with decreased admissions but not with ED return visits or length of stay.

A rapid access centre (RAC), a six-bed consultant-led ward, was introduced to a hospital in the UK in an intervention reported by Tang et al. 95 Data on admissions from the RAC were compared with data on admissions from the ED for two 7-month periods before and after the introduction of the RAC. Simple analysis of data from 441 patients showed that patients admitted from the RAC had a shorter length of stay and were discharged earlier.

Ng et al. 96,97 reported on a GEDI-WISE intervention, the introduction of a GED and a before-and-after evaluation of admission rates. They found that admission rates declined by a statistically significant amount following introduction of the ED, from 58.9% in January 2011 to 50.7% in May 2013.

Wilber et al. 94 reported on a senior ED, which was a 15-bed unit, with assessment by a nurse care co-ordinator, interventions and discharge follow-up. This was a pilot intervention and quality assurance data were analysed from before and after the intervention. Statistically significant results were seen for the outcome of admissions, which significantly decreased (from 55.5% to 51.2%, difference –4.3 percentage points, 95 CI –7.2 to –1.4 percentage points). There was a small but insignificant decrease in length of stay and revisits resulting in admission or observation at 7 and 30 days.

Overall, the interventions reporting changes to the physical infrastructure of the ED were also highly variable. Predictably, in most of the studies reported here, changes to the physical infrastructure were made alongside changes to staffing as part of an overall reconfiguration of how care was delivered. Again, the outcomes measured and reported across the studies were variable; however, the majority of studies reported improvements in admissions-related outcomes, although whether or not these were planned outcomes of the interventions and the wider implications for patients of reduced admissions are not reported.

Care delivery interventions

The studies reported in this section are those that describe and evaluate changes to the whole care package delivered to (frail) older people within the ED. The interventions reported in this section take the form of CGA, which combines interventions to identify frail or at-risk older people and interventions to deliver targeted care to them. ‘Comprehensive geriatric assessment has become the internationally established method to assess elderly people in clinical practice. It is a process of specialist elderly care delivered by a MDT to establish an elderly person’s medical, psychological and functional capability, so that a plan for treatment and follow-up can be developed’. 125

The majority of studies that we identified in this review were descriptive reports of CGA and CGA-type interventions introduced to ED settings. Details of the 18 CGA studies (22 publications24,25,98–117) included in this review are provided in Table 12.

Ismail et al. 98 reported on an interface geriatrician delivering CGA in the UK, which was positively received by staff and patients and led to a non-significant fall in admissions. Three additional UK studies evaluated CGA delivered by an older people assessment and liaison (OPAL) team in Manchester. 102–104 In these very small-scale service evaluations, there were no significant changes in outcomes following the intervention, although the studies reported a non-significant decrease in admissions and in length of stay compared with age-matched control patients or patients not given CGA by the OPAL team. A similar evaluation of a CGA OPAL team elsewhere in the UK reported similar, improved patient outcomes, although these were also non-significant. 105

A study undertaken in Taiwan100,101 of CGA introduced to older people visiting the ED three times in 30 days found that ED revisits were reduced but that the intervention increased admissions. Whether or not this was a positive outcome for patients and the health service overall was not reported.

Identification of at-risk older people followed by CGA was reported in five studies. Beirne et al. 107 reported on the use of the ISAR tool to screen patients, who then received a CGA intervention if they were at risk. A convenience sample of the ‘at-risk’ patients received CGA in the ED with community follow-up. There was a small, insignificant, reduction in ED reattendance in the intervention group. Foo et al. 112 reported on a quasi-RCT undertaken in Singapore that had functional status as its primary outcome measure. Despite the fact that those in the intervention group were frailer than those in the control group, there was a significantly better outcome of functional preservation at 12 months in the intervention group than in the control group. There were also improvements for the control group in terms of avoiding admissions and ED reattendance, but these were not significant. Yim et al. 25 developed a Hong Kong version of the ISAR tool to screen patients and then delivered a CGA-type intervention to those identified as being at high risk. High-risk patients were identified through a cohort study of the Hong Kong ISAR tool and then patients were randomised to the intervention or the control. Limited information on the methods used in the RCT were provided and there were no significant differences between the intervention group and the control group in any of the individual or composite outcomes. Grudzen et al. 113 reported on an intervention that combined screening to identify patients in need of an intervention to prevent inappropriate admissions with appropriate referrals to palliative care services. This was part of the wider GEDI-WISE intervention. With the premise that admission to acute services is not appropriate for patients who require palliative care services, ICU admissions significantly declined.

The development of a screening plus intervention ‘elder alert’ in the USA was described by Warburton. 114 The aim of elder alert was to develop a strategy to identify and manage high-risk ED patients aged > 75 years. This included screening patients using the ISAR tool. Screening was found to be accurate and referral to appropriate management appeared to have a positive impact. Comparing groups of patients showed that screening needed to be followed by an intervention for patient outcomes to be improved. Notable cost savings were projected by the evaluation.

The Hong Kong-based ‘We Care’ CGA programme108,109 delivered CGA with the aim of admissions avoidance. The authors reported positive results, with only 15% of patients admitted; however, they did not compare these data with any other admissions data.

Limited evidence from Ngyuen et al. 110 indicated that a self-administered intervention, the synthesised geriatric assessment, was feasible as it was completed within 20 minutes (n = 25 patients). Lo Storto et al. 111 reported on the introduction of a social health and triage team to deliver CGA and found that inappropriate admissions were avoided, although data to confirm this finding were not provided.

Four studies24,115–117 reported on discharge interventions, which was a smaller number than anticipated. In the study by Foo et al. ,115 additional geriatric assessment prior to discharge was delivered by a geriatric nurse, with interventions delivered as appropriate. Positive outcomes for this assessment were reported in terms of hospitalisation and ED reattendance.

Interventions that started in the ED but that included substantial post-ED follow-up were reported in three studies. 24,116,117 Eklund et al. 24 evaluated the ‘continuum of care’, which was designed to help frail older people to remain in their home environment. The intervention was initiated in the ED by geriatric nurses and followed up in the community by a multiprofessional team. The outcomes studied were frailty and activities of daily living (ADL), with the intervention demonstrating improvements in ADL compared with the control group; there were no differences in frailty scores.

Mahony et al. 116 examined an intervention in which patients were screened by a nurse for palliative care triggers and, if they screened positive, received an intervention that consisted of an assessment of needs, consultation and follow-up. Although the results were not significant, there were small reductions in length of stay, which were attributed to better links with home care services.

A before-and-after cohort study by Wright et al. 117 reported on the UK triage and rapid elderly assessment team (TREAT) intervention, consisting of assessment, CGA and pre- and post-discharge support. This complex intervention saw improvements in a number of outcomes. Median and mean length of stay were significantly reduced. The same-day discharge rate significantly increased for those who had been given the TREAT intervention.

To summarise, the evidence base describing CGA and CGA-type interventions is larger than that for other types of service delivery innovations, which is suggestive of the acceptability and feasibility of these types of interventions. More data on outcomes were provided by study authors, with most of these studies measuring outcomes in terms of either admissions avoidance or ED reattendance. Only one study focused on patient outcomes alone (ADL and frailty). 24 There appears to be a general trend for these interventions to improve admissions avoidance. Notably, there was little evidence on discharge interventions that are delivered in the ED to prevent readmission.

Other interventions

Three additional interventions were identified and are reported in Table 13.

Terrell et al. 120 reported a RCT from the USA of computer-aided decision support to reduce prescribing errors for older people by reducing the prescription of potentially inappropriate medicines on discharge from the ED. The intervention was delivered to 32 ED physicians (with 31 acting as a control group). The RCT found that the proportion of potentially inappropriate medicines being prescribed significantly decreased from 5.4% to 3.4%.

A screening intervention to identify patients at a high risk of falls119 and a screening intervention to reduce waiting times118 were identified. Huded et al. 119 reported on the use of the TUGT, which was performed on 443 of 1135 patients by a geriatric nurse. These patients had not presented with falls but those who screened positively were referred to fall prevention interventions.

In the study by Lovato et al. ,118 the Silver Code prognostic tool was demonstrated to result in reduced waiting times for frail older people. On arrival in the ED, patients were allocated a colour code and those who received a specific code were seen more quickly. The observational retrospective data showed that waiting times for frail older people decreased, without waiting times for other groups increasing.

Summary of service delivery interventions

Staffing initiatives tended to take the form of either a specialist geriatric member of staff (doctor, nurse or pharmacist) working in the ED or the development of a geriatric MDT. Roles tended to be in care-co-ordination, assessment or medication management. Differentiating between studies of staff initiatives and the introduction of CGA-type initiatives was not always straightforward and interventions to change the physical infrastructure of the ED were often delivered in addition to a change in staffing.

The evidence for improved outcomes for individual staff interventions was limited. Across a broad range of outcomes there was limited evidence that the interventions had improved patient outcomes. Study authors attribute this to problems with study design and the lack of community follow-up of the ED intervention.

Evidence from MDT-type interventions largely showed beneficial outcomes in terms of reducing avoidable admissions and improving early discharge rates from the ED. The CCT interventions had mixed results, with a borderline improvement in reducing avoidable admissions in one study,78 but a higher risk of ED reattendance and a much higher risk of hospital readmission in another. 82

Structural changes to the ED took the form of the development of GEDs (all studies were from the USA), adapting ED environments to better meet the needs of older or frail patients or establishing units on the ED to meet the needs of these patients. A number of UK studies were reported in this category, with largely positive outcomes in terms of decreased admissions and improved discharge times and rates.

The evidence base for CGA-type interventions was much larger than that for other types of service delivery changes. Of the 18 studies24,25,98–117 that described CGA and assessment interventions in the ED, six reported results that had statistical significance. 24,100,101,112,113,115,117 These results were again highly variable but there was a general trend towards improved outcomes in terms of admissions avoidance.

Identification of frail or at-risk older people

The review by Sutton et al. 121 focused on screening tools to identify older patients presenting to EDs who are at risk of functional decline. Five separate screening tools were identified: the hospital admission risk profile (HARP), ISAR tool, TRST, complexity prediction instrument (COMPRI) and Evaluation du Risque de Perte d’Autonomie (SHERPA). None of these tools was recommended as a gold standard screening tool. Thiem et al. 126 also examined the same five tools, plus the Index of Functional Decline. They found that, even though the ISAR tool has been examined the most frequently and tested the most widely, even for this tool the evidence is weak or conflicting. The review authors also cautioned that management approaches need to be considered alongside screening tools, as there is no value in identifying frail or high-risk patients unless interventions can be tailored to meet their needs.

In 2012, McNamara et al. 127 examined six screening tools used during triage of older patients to identify those at risk. These tools comprised three ‘general’ tools (Manchester Triage, Emergency Severity Index and the CTAS) and three specific tools [ISAR, TRST and variables indicative of placement (VIP)]. The review found that the ISAR tool and TRST performed best, with good sensitivity, a high negative predictive value, low specificity and a low positive predictive value. The VIP had low sensitivity. The review cautioned that clear distinctions need to be made between those who are in need of acute medical care and those who are in need of discharge follow-on care.

Bissett et al. 128 looked at the functional assessment tools used in ED practice. They identified 14 different assessments, four of which were developed specifically for the ED [TRST, ISAR, Runciman questionnaire123 and Functional Status Assessment of Seniors in Emergency Departments (FSAS-ED)]. The review examined the validity of the tools rather than their outcomes and found that the ISAR tool and TRST were most suitable for fast screening and the Runciman questionnaire123 and FSAS-ED were most suitable for comprehensive screening.

The review by Yao et al. 122 looked at the ISAR tool only, in terms of its predictive validity in identifying adverse outcomes for older patients following an ED visit. The review included 10 studies and found that the ISAR tool is quick to use and inexpensive and recommended its use. However, although it was found to have value in identifying high-risk patients with frailty, it has poor, or poor to fair, predictive validity for adverse health outcomes for patients discharged from the ED.

The review by Carpenter et al. 129 looked at what might predict short-term adverse outcomes in GED patients and examined the prognostic value of individual risk factors and ED screening instruments. Seven tools were examined: ISAR, TRST, VIP, Silver Code, Mortality Risk Index and Rowland124 and Runciman123 questionnaires. Various predictors of vulnerability were also considered. The review found that adverse outcomes often occurred post discharge and so identification of these outcomes is critical. However, the review found that there were no risk factors or screening instruments that had sufficient prognostic accuracy to distinguish patients at risk.

The findings of these reviews broadly reflect the findings of our mapping of the primary research – that there are a wide number of tools to identify older people at high risk of adverse outcomes following ED and to identify older patients with frailty. There is no clear recommendation on which tools to use in practice – the ISAR tool is used widely and has been extensively evaluated, but the evidence base for use of the tool is not strong.

Service delivery innovations

Looking specifically at the population of cognitively impaired older people, Parke et al. 130 examined screening and service delivery interventions to better manage this population in the ED. They found that the contextual details and characteristics of interventions were poorly reported and that there were no interventions that were effective and the screening tools identified were inconsistently used and it was therefore difficult to measure effectiveness.

Schnitker et al. 131 also examined evidence on interventions for cognitively impaired older people and identified 12 studies of their management in the ED. These 12 studies were categorised into four groups: (1) those designed to improve recognition of cognitive impairment (and subsequent provision of care); (2) those designed to prevent delirium; (3) those designed to manage behavioural or psychological symptoms; and (4) ‘other interventions’. They proposed the routine inclusion of screening and assessment into care practices and stressed the importance of both screening patients quickly to recognise cognitive dysfunction and using risk tools on discharge.

Two reviews examined the use of CGA in the ED. Graf et al. 132 looked at how best to screen to identify eligible patients for CGA as well as the use and value of CGA. They found that routinely using CGA without screening first was too time-consuming and that an approach that screened for high-risk patients who were then given CGA was most effective. The most effective tool was found to be the ISAR tool. CGA was found to be effective at decreasing functional decline, ED readmission and possibly nursing home admission. Conroy et al. 133 looked at whether or not CGA improved outcomes for frail older people who received this intervention at the point of discharge when they had been discharged rapidly. Using formal systematic reviewing methods, the review looked at a number of outcomes and did not find clear evidence for the benefits of CGA at the point of discharge for this specific population. The review highlighted the lack of trial evidence in this population and the limitations of the small amount of trial evidence that does exist.

Fan et al. 134 reviewed interventions to reduce ED utilisation. The scope of their review was wider and looked at community interventions as well as those delivered in the ED. Although the community interventions were generally more effective than those delivered in the ED for reducing ED utilisation, five ED interventions significantly reduced ED utilisation. These interventions were varied and incorporated risk screening, assessments or discharge planning and referral co-ordination.

The review by Lowthian et al. 10 looked at ED-to-community transition strategies. The review identified nine low-quality research studies that examined interventions that tended to consist of ED assessment with community follow-up. The assessments took a number of forms, including use of the ISAR tool, CGA and discharge planning. There was limited evidence of effectiveness with regard to the outcomes of interest, which included unplanned reattendance, admissions or mortality.

In 2015, Karam et al. 135 undertook a review of interventions delivered within EDs in relation to four outcomes (ED revisits, hospitalisations, nursing home admissions and deaths following discharge). Nine studies met the review inclusion criteria. The effectiveness of interventions was found to be related to intervention intensity – the more intensive an intervention, the more frequently it resulted in reduced adverse outcomes. For the less intensive interventions, effectiveness was enhanced when a screening tool for identifying high-risk patients was used. Karam et al. 135 argued that the specific choice of screening tool to use may be less important than how the tool is actually used.

Others

Tran et al. 136 examined interventions to prevent ED returns in a population of older ED patients. The intensive interventions that they examined, alongside risk factors for ED return, resulted in a reduction in short-term ED returns, but this pattern did not hold in the long term. It was not clear whether or not this long-term pattern was anticipated.

Sinha et al. 137 used a systematic review to develop a geriatric emergency practice model to improve patient outcomes. Examining 28 outcome measures, their review included eight model characteristic components that were seen to contribute to improved outcomes. These eight components were evidence-based practice, nursing clinical involvement/leadership, risk screening, focused geriatric assessments, discharge planning within the ED, interprofessional working practices, discharge follow-up and evaluation/monitoring.

Both screening and service delivery interventions were considered in the review by Fealy et al. ,138 who looked at the effectiveness of nursing interventions for older ED attendees. Although no statistically significant effects were found on patient or health service outcomes, improved effectiveness was demonstrated when interventions incorporated post-ED discharge planning and/or referral.

Summary of review-level evidence

The systematic and other types of reviews that we identified in the searches for the review of primary evidence encompassed both screening and intervention studies.

The screening reviews tended to identify the tools that were available and aimed to assess their value and determine whether or not a single tool could be recommended for use in the ED. The evidence for screening tools was found to be very mixed. The ISAR tool and TRST were found to be the best-performing tools for triage,127 frailty screening122 and rapid functional assessments. 128 However, other reviews that examined these tools did not find sufficient evidence to recommend their use. A prognostic review by Carpenter et al. 129 found that there were no risk factors or screening instruments that had sufficient prognostic accuracy to distinguish patients at risk.

In terms of service delivery interventions, there was mixed evidence on the outcomes of CGA. 132,133 There was evidence that specific interventions reduced short-term ED returns, but this did not hold in the long term. 136 Fan et al. 134 identified five ED interventions that reduced ED utilisation, which included elements of the following: risk screening, assessments, discharge planning and referral co-ordination. Karam et al. 135 examined the effect of interventions on ED revisits, hospitalisations, nursing home admissions and deaths following discharge and found that the intensity of an intervention was a greater predictor of effectiveness than the intervention itself and that the choice of which tool to use was less important than how the tool was used.

Focusing on interventions delivered by a specific health-care professional, Fealy et al. 138 examined nursing interventions, none of which was found to be significant in terms of patient or health service outcomes, although outcomes were improved when post-ED discharge planning was incorporated in the intervention.

Focusing on a specific population, Parke et al. 130 and Schnitker et al. 131 examined screening and service delivery interventions to better manage cognitively impaired older people. No specific tools or interventions were found to be effective – the inconsistent application of the screening tools limited any conclusions that could be drawn from the evidence.

Research designs and their strengths and limitations

The majority of included studies used a prospective observational research design. The screening papers generally measured the accuracy of a tool by gathering follow-up data at a particular time point (that varied across studies) from different sources, including medical records, patient and carer interviews or return visits to the ED. This type of study design was also used in the majority of service delivery intervention studies. These studies, although valuable for descriptive purposes, do not provide direct evidence on the clinical usefulness of a screening tool (in comparison with another tool or an unscreened control group) or the effectiveness of an intervention. The lack of a control group means that it is not possible to determine if the intervention or tool is more clinically effective or cost-effective than usual care.

There are a number of other weaknesses of these study designs that may influence the reliability and validity of their findings. The selection of the study population depends on whether or not patients were exposed to the screening tool or the intervention. Selection bias may therefore influence which patients were given the assessment tool or the intervention. In some studies, for example, the use of a tool was limited to specific times of the day,21 or particular groups were excluded such as those presenting with trauma. 90 This might lead to an under- or over-representation of particular groups of patients and limits the generalisability of findings.

The methods of measuring follow-up outcomes may also introduce bias in studies of this design. Studies may rely on the collection of data that may not have been undertaken consistently. Patients may, for example, return to different ED departments or hospital records may not be consistently coded. The length of follow-up also varied in the included studies, meaning that comparisons between studies may be limited. The number of patients lost to follow-up was also poorly reported but is likely to be very high. This may result in bias if there are differences in follow-up between those who had the outcome being measured and those who did not.

Prospective studies in which data on explanatory and confounding variables are collected before outcomes are known to have an advantage over other study types in determining whether or not outcomes might be associated with the outcome of a tool or the effect of an intervention, as there is less risk of selection or information bias relating to outcomes. In contrast, a retrospective design, used in 12 studies in this review, may affect outcome classification if the exposure to the tool or intervention is known by the person assessing the outcome status (observer bias).

Of the included studies, only a small number used designs that would be appropriate for testing diagnostic accuracy of screening tools or the clinical effectiveness of interventions. The studies that were used to evaluate clinical effectiveness included RCTs, a quasi-randomised study and before-and-after studies.

These research designs are also at risk of bias. Diagnostic accuracy studies in this review may be vulnerable to selection bias if the sample of patients chosen to receive the screening tool or intervention is not random. It may be that the expertise of the individuals using the tools may influence how they are used, leading to measurement bias. The reference standards may also have limitations.

Before-and-after studies offer a valuable method of evaluating clinical effectiveness when a randomised trial may not be feasible. They can provide a historical control against which outcome data may be compared. A weakness of this type of study design is attributing change in outcomes solely to an intervention. It may be that other factors might also influence outcomes, for example staff changes or initiatives in the community for frail older people by a voluntary organisation.

There was a limited attempt to measure the costs of screening and interventions, in terms of either the cost of the intervention or the costs saved as a result of improved outcomes. The nature of many of the interventions reported here is that patients are more appropriately cared for in community settings on discharge, rather than in the acute setting, which is why reduced admissions are a frequently reported outcome. However, there is little evidence on measuring how this ‘care’ is displaced from the ED to the community and the effect that this has on costs.

Compared with service delivery intervention studies more generally, there was a lack of long-term follow-up of individual study participants in the studies reported here. This may be because of the nature of frail older people. Interventions delivered to general older populations, on the other hand, can be followed up over a longer period of time because of their relative better health. The nature of research in the ED means that short-term outcomes are more straightforward to measure. These shorter-term outcomes, such as admissions and length of stay, are outcomes for both the health service and patients. Longer-term patient outcomes (such as mortality) often assumed lesser importance in the reporting of study findings. However, shorter-term outcomes are much more appropriate for an older population, especially given that frail older people are often nearing the end of life. Not only are longer-term outcomes harder to measure, they also assume less importance for this population group, whose outcomes may be better reported in terms of intervention acceptability, for example.

Much of the evidence was not experimental; the majority of the study designs were retrospective or prospective before-and-after cohort studies, with very few trials (either randomised or non-randomised). The sample sizes tended to be small, particularly for the experimental research.

The ED is a challenging place to deliver care, let alone undertake experimental research. The difficulties in undertaking research are reflected in the fact that many of the described interventions are limited in terms of the hours during which they are offered to patients, the staff who are available to deliver them or unintended events (such as outbreaks of infectious diseases), which hamper their evaluation.

There was not much evidence on staff education, which is surprising. When looking at interventions that introduced staff changes, the extent to which staff members had received additional education and training is unclear. The lack of evidence on staff education in this review may in part be related to the outcomes of interest to this review; patient and health service outcomes may be too distal for staff education and the outcomes of interest to staff education interventions, such as increased knowledge and confidence, were outside the scope of this review.

One of the key limitations of the evidence base was the inability to distinguish the frail population from the general population of older adults. Much of the evidence included in the review was not specific to frail older people, instead including a target population aged ≥ 65 years. However, on closer scrutiny, although the interventions tended to be taken up by populations with a mean or median age of ≥ 65 years, in many cases the population was much older. However, the lack of consensus around the definition of frailty makes designing interventions for this population and monitoring their outcomes, and the effectiveness of their outcomes, challenging. In addition to the lack of ‘frail’ population groups, only two studies used changes to frailty as an outcome measure. 24,112

Author-reported limitations

The authors of studies routinely highlighted limitations of their methodology, which led them to caution the extent to which their findings can be generalised and compared with those of other studies. The self-reported limitations of the studies in which frail older people were the focus and full data extraction was undertaken were extracted and a narrative summary of these is given below. A major limitation is the number of conference abstracts that were included in the review; not only is detail missing on the study methods for our assessment of the limitations of the evidence, but conference abstracts rarely contain data on study limitations.

In terms of the sources of data used, data were often collected from routine sources not specific to the studies. 86,88 Authors commented on the seasonal fluctuations in ED attendance,61,106 which may affect the generalisability of the results. There was a general lack of data from outside the ED on participants or service use (community or use in other settings),86 which limited follow-up, and many of the data collected were retrospective. 53

Many of the studies were retrospective or prospective before-and-after studies and there were a limited number of controlled trials. Jones and Wallis61 and Fox et al. 106 noted the lack of a control group, with Silvester et al. 86 and Conroy et al. 89 commenting on the lack of a contemporaneous control group. In addition, a number of authors cautioned that their study had a small sample size;67,71,98,106 that both participants and staff were not blinded (when there was some element of controlling interventions);21,24,112 that not all variables were controlled for;27,78 that the study was conducted at a single centre therefore limiting generalisability;29,33,49,53 and that there may have been some selection bias. 52

The way in which screening and interventions were delivered was cited as a limitation; examples included that the intervention was not delivered 24 hours a day so not all potential participants were included;21,27,49,78,106 only non-urgent attenders were included; the screening tool used was amended for a local setting;78 the difficulty in recruiting patients;112 and unforeseen circumstances, such as an infectious disease outbreak, which may have influenced the study results (before-and-after study). 88

The impact of study findings was limited by the lack of long-term follow-up patient outcome data73,86,89 and the fact that, when there was follow-up, there were high rates of dropout24,29 because of the nature of the population. Two authors also mentioned that study findings would have been enhanced if qualitative data on staff or carer satisfaction with the intervention had been collected. 86,112

In line with the limitations that we identified, there was no clear definition of frail older people27,61 and there was a lack of cost data. 72,89

Relevance of the evidence to the current NHS setting

The consideration of the relevance of the included studies to the NHS setting lies largely in whether or not they have reported research undertaken within the same health system, whether or not the health systems in which the studies were undertaken can be compared with the NHS and whether or not the interventions and screening tools used could be used within the NHS.

In terms of the screening papers, it is noteworthy that only one screening paper20 reported research undertaken in UK EDs. This is in contrast with the depth of research being undertaken in community settings on screening for frailty. In contrast, the intervention papers more widely reported UK research. In particular, interventions for frail populations in the UK combined screening and interventions, perhaps suggesting that it is considered more effective to combine these interventions rather than consider them separately.

In terms of whether or not the included studies could provide models that could be used in the NHS, it is unlikely, given the current NHS landscape, that the GED is a model that would be adopted. Indeed, the current pressures on the ED service make it a challenge (both financial and logistical) to introduce any new interventions and evaluate them.

Key priority areas for further research

A number of areas warranting further examination have emerged throughout this review.

There is a lack of UK evidence relating to how to identify frail older people, compared with the volume of evidence on service delivery interventions from the UK. It is unclear whether this is because there is a greater consensus around how to identify older people who are frail or at risk, because older people are treated as a homogeneous group or because of some other reason. Research is needed in the UK on which tools are currently used in practice and how does identifying patients as frail or high risk subsequently link to their treatment and management. It would be useful to have evidence on whether the purpose of identifying frail and older patients is to identify those needing further assessment or to rule out those not needing it. Consensus on a tool to identify frailty needs to consider which aspects of frailty are more important in the ED, for example patients with dementia or delirium may need very different treatment from patients who are prone to falls and identifying them simply as frail or as being at high risk does not reflect these subtleties. The acceptability of tools to patients and the usefulness of tools to clinicians also needs to be examined. This could include a comparison between tools that are question based and those that employ different methods.

Looking at the complexity of the health-care system, it is unclear whether or not interventions to reduce inappropriate admissions are displacing care to elsewhere in the health-care system; the implications of ‘displaced’ care have not been considered. Another issue that has not been addressed in any detail is whether or not the staff member delivering an intervention has any effect on the outcomes or acceptability of the intervention (i.e. whether it is delivered by a doctor, nurse or other health-care professional and whether or not this staff member requires specific geriatric expertise). These models appear to have been evaluated in the literature; however, the reasons for why a specific clinician was chosen remain unclear and whether or not it is thought that this may have influenced the costs and outcomes of an intervention is not reported.

This review did not identify many interventions that were delivered both within and outside the ED. This may be related to our search approach; however, it would be interesting to further examine interventions that combine ED intervention with home follow-up and compare different models of discharge management and follow-up and the cost implications of these interventions. Community screening to identify those older patients at greater risk of admission to hospital or nursing homes may provide an opportunity for patients who present at the ED to be ‘prescreened’ and identified as frail and high risk, so that their care can be managed accordingly. It may be that interventions that divert frail older people away from presenting at the ED may be more effective than trying to improve outcomes for the proportion who will inevitably attend the ED with an acute medical condition.

In terms of service delivery interventions, it has been argued that it is unlikely that the GED model will become widespread11 because of the cost and resource implications required to develop this model. However, there is a precedent with the use of paediatric EDs in the UK. With an ageing population, further exploration of the GED may be of use.

One area that was not covered in this research and will undoubtedly be of interest and importance to patients, carers and the health service are which outcomes are important for patients and how long should we measure these outcomes for. Bearing in mind that many frail patients may be nearing the end of life, how important is it to measure long-term outcomes? The quality of the experience of ED care may well be more important to patients than how quickly they are discharged. Knowing more about which outcomes are important will help us to determine which interventions should be considered.

Key design features of research

A standard evidence review approach has allowed us to discover what evidence is available on the identification and management of frail older people in the ED. The variability in the types of interventions used, the outcomes reported and the standard of reporting more generally has meant that it is not possible to make overarching conclusions about which interventions are more effective. Limited qualitative data were identified on the feasibility and acceptability of interventions, so it would be useful to understand the views of patients, carers and clinicians about the ED more generally and the appropriateness of interventions. One way of doing this would be through a qualitative review or a realist synthesis of evidence, which would allow for disparate types of evidence to understand more about how and why interventions work and would be less restricted by the requirement of this review to focus on interventions and their outcomes.

Data on anything other than patient and health service outcomes were rare; a basic cost analysis was undertaken by Leah and Adams,72 who estimated cost savings from reducing avoidable admissions. Many of the studies reported increased engagement with health services as a beneficial outcome of the intervention, for example an increase in appropriate admissions, consultation with a geriatrician in the ED, increased community follow-up and more people referred for care according to guidelines and protocols. However, there is very little evidence on the cost of these outcomes. Although there may be a cost benefit in reducing admissions, there is no evidence on the displacement of these costs and the increased cost of community-based interventions when inappropriate admissions are avoided. This is specifically the case for older people generally, and frail older people in particular, as their engagement with the health service differs in its cost and frequency from that of the wider population.