Emergency department interventions and their effect on subsequent healthcare resource use after discharge: an overview of systematic reviews

Abstract

Background

Due to the worldwide pressures on Emergency Departments (EDs), there is a focus on ED interventions to alleviate pressure. Ensuring interventions do not inadvertently impact upon other healthcare sectors is an important outcome. This overview of systematic reviews aimed to evaluate the impact of ED based interventions on subsequent healthcare resource use after ED discharge.

Methods

An overview of systematic reviews was conducted in accordance with the Cochrane Collaboration. Search criteria were devised using the PRESS standard and duplicate screening and extraction conducted for one third of systematic reviews. A primary study matrix was designed to reduce the impact of duplicate primary studies. Data was extracted in the form presented in the underlying review.

Results

After removal of overlapping primary studies, 38 systematic reviews and 213 primary studies were included. Overall confidence in the reviews was high in 12, moderate in seven, low in nine and critically low in 10 reviews. In the 38 reviews, 30 different intervention-population-resource use combinations were analysed. ED based interventions decreased subsequent healthcare resource use in 23.3% (n = 7/30) of the intervention-population-resource use combinations and had no effect in 40% (n = 12/30). The most common resource use reported was ED Revisit. The most common follow-up length from ED discharge was 12 months (n = 52/216), followed by the combined group of one month (n = 44/216).

Conclusions

ED based interventions decrease subsequent healthcare resource use in a fifth of population-intervention-resource use combinations. Future research should produce a standardised set of outcome measures for subsequent healthcare resource use.

Background

Worldwide pressures across the Emergency Care system are unprecedented [1,2,3]. In the United Kingdom (UK), healthcare pressures extend to the primary care system [4], emergency medical service (EMS) system [5] and elective care [6].

To date, policy and research efforts to combat ED pressures has focused on interventions designed to re-direct patients away from EDs, reduce ED use or improve ED flow, but there is little evidence to support these interventions [7,8,9]. Pre-hospital and ED interventions do not decrease the proportion of patients transferred to hospital [7], evidence of the effectiveness of interventions to reduce ED use remains insufficient [8] and the evidence of interventions designed to improve patient flow is weak [9]. It is therefore important to understand the resource implications of these interventions on other sectors of healthcare.

A key outcome measure, infrequently evaluated, is subsequent healthcare resource use after discharge from the ED. Interventions that increase or decrease subsequent healthcare resource use will have systems, resource and patient impacts [10]. Understanding the full impact of ED interventions will ensure the appropriate allocation of limited resources to produce a net health system benefit. Therefore, this overview of systematic reviews, aims to evaluate ED based interventions which report subsequent healthcare resource use as an outcome for interventions.

The four objectives are to (1) identify systematic reviews which report subsequent healthcare resource use as an outcome for interventions designed for ED patients; (2) evaluate interventions that been shown to decrease subsequent healthcare resource use versus interventions that have no effect; (3) identify the theoretical concepts that underpin interventions that are effective; (4) to analyse the variability in definitions of subsequent healthcare resource use in respect to resources and time elapsed from ED discharge.

Methods

Study design

This was an overview of systematic reviews and was conducted according to guidance outlined by the Cochrane Collaboration for overviews [11]. It has been reported as per the recommendations in Box V.5.b of the Cochrane guidance [11]. All references to systematic reviews, will use the term ‘review’. The protocol was registered at Prospero (ID = CRD 42021230846).

Criteria for selecting reviews for inclusion

Types of reviews

Reviews and meta-analyses of primary studies (randomised controlled trials (RCT) and/or non-randomised) which evaluated ED based interventions and reported subsequent healthcare resource use as an outcome were included. A review was defined by the five criteria defined by Cochrane [12].

Types of participants

Reviews were included if they contained primary studies with an intervention based in the ED that targeted adults (> 18 years). Interventions could focus on any target condition or symptom, ED population or ED process.

Types of interventions

Interventions were excluded if based on biomarker blood tests only. This was done to avoid the volume of biomarker diagnostic studies biasing the sample of reviews. Any other review reporting an intervention within the ED that reports subsequent healthcare resource use as a primary, composite or secondary outcome were included.

Types of outcome measures

Subsequent healthcare resource use was the outcome measure. The resource use had to be linked to the index ED attendance and a time interval of 12 months from discharge was used. Resource use was divided into the following six categories:

• Attendance to Primary Care/Family Clinician

• Re-attendance to the ED

• Referrals to secondary or tertiary speciality clinic hospital

• Referrals to community clinics

• Contact to telephone triage services (e.g., NHS 111 in the UK)

• Contact to Emergency Medical Services (EMS)

Other outcome measures not described a priori were included if they constituted healthcare-associated resource use post ED discharge. The description of the healthcare resource use was extracted in the format reported in the included review.

Search methods for identification of reviews

The search was derived using the PRESS strategy [13], with input from two independent medical librarians and the review team. The search criteria are specified in the online supplement-1. Five databases were searched, Medline, EMBASE, PsycINFO, Cumulative Index to Nursing and Allied Health Literature (CINAHL) and the CENTRAL trials registry of the Cochrane Collaboration. The search was limited to the English language. The reference lists of included reviews were scanned to identify any further reviews for inclusion.

Data collection and analysis

The search results were uploaded to Covidence, a review management software [14]. Two review authors independently screened titles, abstracts and full texts for inclusion (TR screened all, NT and DW provided independent review). Data extraction of key variables and quality assessments were performed in duplicate for a third of titles (performed by TR and CT). At this time, an inter-rater agreement (κ statistic), was assessed to allow for solo data extraction [15]. Any disagreements between reviewers were resolved with discussion between reviewers, if disagreements remained these were resolved by an independent arbitrator (EC).

Quality of included reviews

Each review was assessed using the ‘A MeaSurement Tool to Assess systematic Reviews (AMSTAR-2)’ checklist and reported narratively in the results. Each domain and a quality rating of ‘High’, ‘Moderate’, ‘Low’ or ‘Critically low’ are reported [16]. Only ‘High’ or ‘Moderate’ quality reviews are presented in the text. ‘Low’ and ‘Critically Low’ reviews are presented in data tables for reference. As above, AMSTAR-2 ratings were performed in duplicate for a third of titles (TR and CT), the remainder calculated by TR, after the calculation of a suitably high inter-rater agreement (κ statistic).

Risk of bias of primary studies included in reviews

As outlined in the Cochrane guidance, the risk of bias (RoB) of primary studies from each selected review was extracted directly and was reported narratively, as per Bialey et al. and Foisy et al. [17, 18]. Where a RoB was not reported, a RoB assessment for primary studies was not conducted.

Quality of evidence in included reviews

Reported ‘The Grading of Recommendations Assessment, Development and Evaluation’ (GRADE) ratings of each outcome in the review were extracted and reported narratively. Any other quality assessments will be reported narratively in the results. If GRADE rating or quality assessment was not done, a new assessment was not conducted.

Double counting

To account for double counting, where a primary study was included in more than one review, a mapping of primary studies was completed. This produced a corrected cover area (CCA) percentage [19]. Where a primary study overlapped, data from the higher quality review were retained. If both reviews were of the same quality, the data were retained from the newest review. If overlapping data was included in two high quality meta-analysis, the overlapping data was not removed. Once this process was completed, primary studies were re-mapped and a CCA re-calculated.

Reporting

The results of the four objectives are reported sequentially as objective one to four. Objective two, which compares interventions that have decreased resource use compared to those with no effect is reported as objectives 2a–2d. This is to facilitate easy comparison between interventions that decreased resource use (2a), those that had a mixed effect (2b), those that increased resource use but as the primary aim of the intervention (2c) and those that had no effect (2d).

Results

A total of 49 eligible reviews were identified from the search, conducted on 16/02/2021 (re-run on 26/01/2022) (Fig. 1). The 49 reviews included data from 369 primary studies. 72 primary studies overlapped. The CCA was 1.38%, demonstrating ‘slight overlap’ overall [19] (Fig. 2a). After removal of overlapping primary studies, not used in meta-analysis data, 11 reviews were removed as primary studies were reported in higher quality reviews. Of the 38 reviews remaining, 213 studies were included, 19 overlapped studies remained. The final CCA was 0.27% (Fig. 2b).

Fig. 1

PRISMA flow diagram

Fig. 2

a/b—Primary study matrix, pre and post primary study overlap

The interrater reliability between the two data extraction reviewers for the first third of reviews was κ = 0.78. This demonstrates ‘substantial’ agreements between reviewers. [15]

Description of included reviews

A detailed description of the 38 reviews is available in Table 1.

Table 1 Detailed description of reviews included in overview

Methodological quality of included reviews

The itemised results of the AMSTAR-2 assessment are outlined in Fig. 3. The overall confidence in the included reviews was defined as high in 12, moderate in seven, low in nine and critically low in 10 reviews (n = 38).

Fig. 3

AMSTAR-2 assessment

Risk of bias of primary studies included in reviews

The overall impact of the risk of bias of primary studies in each review is covered by items nine, 12, 13 and 15 in the AMSTAR-2 assessment (Fig. 3).

When analysed individually 72.7% (n = 24/31) of reviews used a satisfactory technique for assessing RoB in individual RCTs, and 56.7% (n = 17/30) for non-randomised studies of interventions (NRSI) (item nine, online supplement-2). In the 13 studies that performed a meta-analysis, 84.6% (n = 11/13) assessed the impact of RoB of individual studies on the meta-analysis (item 12 online supplement-2). Most reviews (71.1%, n = 27/38) accounted for RoB during the interpretation of the results, but only 53.1% (n = 7/13) of reviews investigated publication bias when indicated (items 13 and 15, online supplement-2).

Where available, the individual RoB assessment for the primary studies in each review is available in the online supplement-3.

Outcome 1: Reviews which evaluate ED interventions and report subsequent healthcare resource use as an outcome

In the 38 reviews, 37 unique interventions were analysed. Table 2 outlines the direction of effect of interventions, grouped by host population (n = 15) and specific resource use measured (n = 9). This resulted in 30 different intervention-population-resource use combinations. ED based interventions decreased subsequent healthcare resource use in 23.3% (n = 7/30) of the intervention-population-resource use combinations, had a mixed effect in 10% (n = 3/30), increased scheduled follow-up (aim of the interventions) in 20% (n = 6/30) and had no effect in 40% (n = 12/30). For 6.6% (n = 2/30) it was not possible to report an effect.

Table 2 Effect of interventions by population-intervention-resource use combination

The 15 populations, dictated by cohorts reported in reviews, were older adults (n = 12), frequent attenders (n = 7), ED adults (n = 3), asthma (n = 3), atrial Fibrillation (n = 2), patients on antibiotics (n = 2), alcohol related (n = 1), lower back pain (n = 1), risky behaviour (n = 1), shared decision making (n = 1), mental health (n = 1), primary care patients in ED (n = 1), ED short stay unit patients (n = 1), chest pain (n = 1) and palliative care (n = 1).

Outcome 2a: Interventions that decreased subsequent healthcare resource

Only data from high or moderate confidence reviews are reported below for all outcomes below. Table 2 includes additional data from low or critically low confidence reviews for reference.

ED revisits

Frequent attenders

Three high confidence reviews [20,21,22] demonstrated a decrease in ED revisits when care plans, case management, social work home visits, diversion strategies to non-urgent care, printout case notes were used in the patients defined as frequent attenders. A moderate GRADE was reported by Berkman et al. [21], indicating certainty that the true effect of the interventions were a reduction in ED revisits. No GRADE was reported by the other two reviews.

This data is supported by moderate confidence data from Wong et al. [23]. Data from five studies (two moderate and three low quality) demonstrated a reduction in ED revisits between 48.4% and 89.5%. Interventions were care co-ordination, pain protocols, pain contract (present twice) and behavioural interventions.

Patients presenting with chest pain

Data from a high confidence review by Flynn et al. [24] demonstrated that the provision of pre-test probability to patients and clinicians decreased 7-day ED revisit rate. Based on evidence from one study with low RoB. [25]

Hospital admissions

Frequent attenders

Based on a high confidence review by Berkman et al. [21], which reported one low RoB RCT [26] (n = 100) and one observational study [27] (n = 14 140) with “some” RoB concerns, ED-initiated patient navigation programme and decision-support were found to decrease hospital admissions in frequent attenders.

Hospital re-admissions

Older adults

Based on one high confidence review by Harper et al. [28], reporting data from two strong and four moderate quality RCTs (n = 2493), Interdisciplinary team interventions reduced hospital re-admission in older adults who fell, with a relative risk (RR) of hospital re-admission of 0.76 (95% CI 0.64–0.90). The GRADE was reported as moderate.

Testing and cost

Testing and cost were identified as additional healthcare resource use outcomes. These were not defined a-priori and are therefore presented in the online supplement-4.

Outcome 2b: Interventions that had a mixed effect on subsequent healthcare resource

ED revisit

Patients with lower back pain

A moderate confidence review from by Liu et al. [29], based on two studies with ‘low’ ‘Before and After Quality Assessment’ (BAQA) score, reported that multi-disciplinary team protocols aimed at decreasing imaging for lower back pain decreased ED revisits, whilst clinical decision support had no effect on ED revisits.

Hospital re-admissions

Older adults

A high confidence review by Hughes et al. [30] demonstrated that case management, transitions of care, medication management and discharge planning interventions did not have an effect on hospital re-admissions in a general older population. This is based on meta-analysis data from seven RCTs (n = 4806), reporting a RR of hospital re-admission of 0.96 (95% CI 0.51–1.83). The GRADE was low. Another high confidence by Elliot et al. [31], reported that MDT assessment demonstrated no effect in older adults on hospital re-admission. Based on data from two studies (1 × low RoB, 1 × Not Reported). This is in contrast to the review by Harper et al. [28], reported above, which showed interdisciplinary team interventions reduced hospital re-admission in older adults who fell.

Based on high confidence data from Galipeau et al. [32], short stay ED units resulted in decreased to no effect on hospital readmissions and ED revisits in adult ED patients (GRADE = Low, RoB = Moderate).

Outcome 2c: Interventions that increased scheduled healthcare resource as their aim

Some ED interventions were designed to increase scheduled resource use as per intervention design or national guidance. For example, UK guidelines advise GP follow-up within two days of ED attendance with asthma [33]. Detailed results are available in the online supplement-5. In summary, interventions designed for ED frequent attenders to seek more ‘appropriate’ healthcare options other than the ED, resulted in increased primary care visits as intended [21]. In patients presenting to the ED with asthma, educational interventions increased follow-up rates with a primary care practitioner as intended [34]. Care transition interventions improved the rate of follow-up with primary care or specialist providers in adult ED patients [35]. Case management interventions in ED frequent attenders increased outpatient visits as intended [22]. Finally, multi-disciplinary team protocols aimed at decreasing imaging for lower back pain, increased the use of physiotherapy and rehabilitation services as planned. [29]

Outcome 2d: Interventions that have no effect on subsequent healthcare resource

These are reported in detail in Table 2. In summary, of the 12 intervention-population-resource use combinations, 6 reported ED revisits, the other six reported ED contacts, GP contacts, GP visits, psychiatric contacts, hospital admission and hospital re-admission.

Outcome 3: Theoretical concepts that underpin successful interventions

Reviews more frequently reported increased resource use for scheduled follow-up when that was the aim of the intervention, compared to no effect for unscheduled care (i.e., no decrease in unscheduled care) when that was the aim of the intervention. This is based on eight populations (supplement Table 5) that reported scheduled follow-up, of which 87.5% (n = 7/8) reported interventions that increased scheduled follow-up. When compared to 23 unscheduled resource outcomes (from 13 populations), only 30.4% (n = 7/23) reported interventions that decreased unscheduled care.

Further analysis of the seven populations that increased scheduled follow-up, demonstrated six populations that reported both scheduled (e.g., planned GP follow-up) and unscheduled resource use (e.g., ED revisits) as outcomes from the same intervention. Interventions increased scheduled and decreased unscheduled care in two cohorts (frequent attenders and alcohol cohorts); increased scheduled resource use but no effect on unscheduled care in three cohorts (adults in the ED, asthma, alcohol) and increased scheduled resource use but had a mixed effect on unscheduled care in the lower back pain population.

In the 23 unscheduled resource use outcomes reported above, 17.4% (n = 4/23) decreased or had no effect on unscheduled resource use, 47.8% (n = 11/23) had no effect and one could not be analysed.

Outcome 4: Variability in definitions of downstream healthcare resource

The most common resource use reported was ED Revisit, reported in 36 of 38 reviews (online supplement-4). Overall, there were only nine distinct types of resources identified—ED revisit, hospital admission (including psychiatric), hospital re-admission, GP follow-up, community referral (physiotherapy, rehabilitation community psychiatry), cost, outpatient visits (including psychiatric), general resource use and testing. EMS use or telephone triage (e.g., 111 services in the UK) were not measured in any review.

There were more than 23 different time intervals for follow-up reported across primary studies. The most common was 12 months (n = 52/216), followed by the combined group of 28 days, four weeks, 30 days and one month (n = 44/216) and then six months (n = 40/216) (Table 3). Thirteen primary studies measured follow-up over a period greater than 18 months.

Table 3 The duration of follow-up reported by primary studies in reviews

Discussion

This overview provides a contemporary map of ED based interventions that impact upon subsequent healthcare resource after ED discharge.

It reports that 40% of interventions have no effect on resource use, however there is evidence within specific population-intervention cohorts (e.g. frequent attenders cohorts or shared decision making interventions) that interventions decrease subsequent healthcare resource use. The data can be practically utilised by intervention developers to review the available evidence of ED based interventions in specific patient cohorts and for specific resource outcomes. It will allow a streamlining of future efforts in those interventions where reliable evidence exists and prevent the repeated trials of interventions which have little evidence of impact.

Limitations

It is important to consider the results through the lens of overview methodology, which is to provide an overall summary of the available data.

This study was limited by two protocol deviations. Firstly, due to resource limitations data extraction was not completed in duplicate. Duplicate data extraction only occurred for the first third of reviews. At this point an inter-rater reliability was calculated and deemed sufficiently high (κ = 0.78) to continue with single data extraction. Secondly, if risk of bias assessments or GRADE ratings were not reported in the review, they were not calculated as originally specified in the protocol. Again, this was due to resource and time limitations. Both these deviations increase the possibility of bias into the overview. Finally, the search was limited to the English language which increases the chance of language bias.

Strengths

Despite the limitations, the alignment with overview methods was a key strength of this study. The use of Groove methodology, to account for primary study overlap, was a significant step forward in overview methods that has not, to the authors knowledge, been used previously in emergency care overviews [8, 9, 36, 37]. Our evidence suggests that whilst the overall confidence one can have in review evidence is improving, especially in more recent reviews, there remains consistent heterogeneity in reporting as outlined by Conneely et al. [37]

When compared to the results of previous work in this area, three of the four previous overviews of ED based interventions concluded that the evidence base was either “weak” [9, 36] or conclusions were difficult to identify due to the “significant heterogeneity in methods, intervention content and reporting of outcomes” [37].

An understanding of the subsequent healthcare resource use associated with ED based interventions remains important due to the significant pressures across the entire healthcare sector worldwide. Data from this overview highlights the need for a standardised set of outcome measures and follow-up period for ED based interventions. Importantly, future overviews, reviews and primary studies should maintain or direct their focus on patient-orientated outcomes and co-design to allow interventions to make the positive change required by patients and healthcare systems.