1.1 Background

Historically, a patchwork of specialist healthcare support services for care homes has emerged. For the purpose of this review, these are defined as services whereby healthcare professionals, with time dedicated to this role, provide specialist clinical support to identify and address residents’ healthcare needs (Tucker et al., 2019). Some enhance standard primary care, allocating named practitioners to specific care homes or supplementing medical care with specialist nurse input; others involve partnerships between primary and secondary care and some are led by pharmacists. Core activities include comprehensive assessment of new residents; regular, structured multidimensional reviews; falls prevention; medication reviews; advance care planning and end-of-life care.

Evidence of a relationship between the provision of specialist healthcare support services and reduced avoidable hospital admissions would provide a strong argument for their expansion. However, the extent to which current models achieve this is unclear. A previous review (Clarkson et al., 2018) explored the organization, activities and responsibilities of specialist healthcare support services, but did not systematically explore outcomes. Other reviews have explored the extent to which specific activities including optimizing prescribing (Alldred et al., 2016; Forsetlund et al., 2011; Thiruchelvam et al., 2017; Wallerstedt et al., 2014), antibiotic stewardship programmes (Feldstein et al., 2018) and palliative care initiatives (Hall et al., 2011) can reduce admissions. However, the quality of evidence in these and a review of interventions to reduce acute admissions of nursing home residents (Graverholt et al., 2014) were generally poor, while the heterogeneity of interventions, methods and outcomes in reviews of the effectiveness of input from staff with geriatric expertise, integrated and partnership working (Davies et al., 2011; Goldman, 2013; Santosaputri et al., 2018) precluded robust conclusions. As such, an overall picture of the evidence to inform service development is lacking.

In addressing this gap, this review employed two taxonomies (Table 1) developed from a UK survey of specialist healthcare support services (Hargreaves et al., 2019). The first has five categories, using constructs of specialist support identified in the care home literature (Clarkson et al., 2018). The second allocated services with similar characteristics into three subgroups using latent class analysis (Hargreaves et al., 2019).

2.1 Aims

The primary aim was to identify, appraise and synthesize the evidence on the ability of specialist healthcare support services to: (1) reduce hospital admissions of older care home residents and (2) enable them to remain in the care home at end of life. Secondary aims were to: (3) determine whether particular service subgroups produced different outcomes; (4) explore how usual/standard care was defined and (5) identify service costs.

2.2 Design

A systematic review, without meta-analysis, with vote counting based on direction of effect updated and extended the review by Clarkson et al. (2018) in seeking studies published post 2010 and appraising outcomes. The review followed established guidance (CRD, 2009; Rutter et al., 2010) and the protocol was registered (PROSPERO) and published (Tucker et al., 2019). Reporting used the SWiM reporting guideline (Campbell et al., 2020) in conjunction with the 2020 PRISMA checklist (Page et al., 2021) to promote transparency.

2.3 Search methods

Searches were conducted in 14 electronic databases: AgeInfo, CINAHL Plus, the Cochrane Database of Systematic Reviews, the Database of Abstracts of Reviews and Effectiveness, EMBASE (incorporating Medline), the Health Technology Assessment Database, Health Management Information Consortium, the Joanna Briggs Foundation, the National Health Service Economic Evaluation Database, PsycINFO, PubMed, PROSPERO, Social Care Online and Web of Science. Initial searches were undertaken in November 2017 and auto-alerts monitored until January 2019. Reference lists of identified reviews, included documents and study protocols, were scrutinized for further studies.

Four search blocks (endorsed by a research librarian) were combined in EMBASE and adjusted for other databases. These pertained to: (1) care homes; (2) healthcare; (3) older people and (4) hospital admission and place of death. An example search strategy for EMBASE is given in Supplementary Appendix S1. No geographical limitations were imposed, but non-English language publications were excluded.

2.4 Study selection and quality appraisal

Two reviewers (DB and ST) independently screened the results against the study inclusion/exclusion criteria (Box 1) and quality appraised inclusions using the Effective Public Health Practice Project (EPHPP) Quality Assessment Tool (Armijo-Olivo et al., 2012; Thomas et al., 2004). This is suitable for quantitative studies and covers six domains (selection bias, study design, confounders, blinding, data collection and withdrawals/dropouts), each of which is rated strong, moderate or weak. A global rating is allocated depending on the number of weak ratings – strong (none), moderate (one) or weak (two plus) (Armijo-Olivo et al., 2012; Jackson & Waters, 2005; Thomas et al., 2004). Given the largely objective nature of the primary outcomes (raising fewer problems with reliability/validity than, say, clinical assessments), sections concerning the blinding of outcome assessors and data collection were adapted. (The modified tool is available from the authors.)

2.5 Data extraction

One reviewer extracted data on study ID, aims, design, methods, participants, interventions/services, comparison interventions/services, outcomes and costs using a specially designed Excel form. A second reviewer checked all extracted data for accuracy, consistency and agreement. Disagreements or uncertainties were resolved through discussion with the wider research group. The form was tested and refined by the same two reviewers and agreed by the review team before full data extraction. Where multiple reports were identified from the same study, information was extracted from each publication separately but combined into one study record for the purpose of synthesis.

2.6 Data synthesis

The key characteristics and results of each study were charted, described and synthesized according to the primary outcomes. No studies were discarded because of their quality rating, but weak-rated studies are highlighted in the text with the superscript*. Only studies that explicitly aimed to reduce hospitalization were included in effectiveness analyses.

A random-effects model meta-analysis (Borenstein et al., 2010) was attempted to estimate the effect size of specialist healthcare support interventions on hospitalizations. The analysis was conducted in Review Manager 5.4 (2020) and included 11 studies that reported the intervention and control group sample sizes and mean and standard deviation of hospital admissions/ED transfers post-intervention. Forest plots were presented to visualize the results. Sensitivity analyses explored the effects of study quality (excluding weak-rated studies) and outcome domains (excluding data on ED transfers).

Given the diversity of interventions, variation in effect measures and high level of missing/incomplete outcome estimates, vote counting based on direction of effect (regardless of statistical significance or testing) was used to detect evidence of an effect for study subgroups according to the foci of intervention and two taxonomies (McKenzie & Brennan, 2019). All studies reporting a positive effect on the number and/or rate of hospital admissions/ED transfers were considered beneficial regardless of any co-existing negative effects so as not to penalize studies reporting multiple outcomes. Studies reporting no change were classed as ‘no positive outcome/not beneficial’; studies reporting practitioners’ perceptions of avoided admissions or only hospital bed days were excluded. Binomial tests assessed statistical significance of the vote counting findings (McKenzie & Brennan, 2019) and sensitivity analysis explored the effect of treating studies with mixed findings as negative (reversing their handling in the main analysis). Correction for multiple testing was undertaken using the conservative Bonferroni method of multiplying the p-values by the number of comparisons (Jafari & Ansari-Pour, 2019) in cases where there appeared to be a significant outcome. This was applied to each subset of analyses, that is, considering foci of intervention as one set (six comparisons), and the models of specialist support, namely Taxonomy 1 and Taxonomy 2, as two further sets (five and three comparisons respectively).

3.1 Search outcomes

Electronic database searches identified 5466 references (Figure 1). After duplicates were removed, 3399 records were screened by title and abstract for relevance and the full texts of the 215 remaining documents, plus 66 citations identified by reference tracking, were independently assessed against the inclusion/exclusion criteria. This resulted in 92 included studies (98 references): 79 from the searches plus 19 from the review by Clarkson et al. (2018).

3.2 Study characteristics

Of the 92 included studies, 22 were undertaken in the UK, 22 in the USA and 19 in Australia (Tables 2 and 3). The earliest was published in 1991, but most (n = 82) were published between 2010 and 2019. Supplementary Appendix S2 provides more detail.

Twenty studies were randomized controlled trials, six controlled clinical trials, 16 cohort analytic studies (two- or three-group pre and post) and 30 cohort studies (one group pre and post). Other designs included eleven (two or three groups) observational cohorts and six descriptive studies in which reported outcomes commonly constituted professional judgements of avoided admissions. Participating facilities included nursing homes, residential homes, assisted living facilities and senior living communities (hereafter ‘care homes’). Sample sizes (alternatively reported as numbers of care homes, beds, residents and consultations) varied greatly; the number of care homes ranged from 1 to 405 and the number of residents from 14 to 61,636, although not all studies provided this information. Most included all older residents. However, some focused on particular subgroups, for example, residents with dementia.

3.3 Study quality

A total of 12 studies were rated strong in quality, 32 moderate and 48 weak. Of the latter, half had weak ratings on two EPHPP domains, 13 on three, 7 on four, and 4 on five. Following EPHPP guidance, 18 weak-rated studies had weak study designs. The three most common reasons for weak classifications were as follows: sample/selection bias (38/48, e.g. low probability participants represented the wider population); data collection methods (33/48, e.g. doubts about data reliability) and the potential for (uncontrolled for) confounders (24/48). Supplementary Appendix S3 provides further details.

3.4 Focus of the initiatives

Thirty-two initiatives focused on residents’ general physical health, 15 on acute illness or at risk of hospital admission, 20 end-of-life care and 6 post-hospital discharge. Ten sought to optimize residents’ medication; others focused on specific health concerns including falls and pressure ulcers.

3.5 Characteristics of the initiatives

Not all studies provided detail of participating practitioners. However, at least 46 initiatives were delivered by multidisciplinary teams (MDTs). Nurses (often advanced/specialist practitioners) participated in at least 61 initiatives and physicians at least 49. Other disciplines included pharmacists, occupational therapists, physical therapists, dieticians and emergency care practitioners. Most professional support was provided face to face. However, seven studies evaluated telemedicine services.

Table 4 shows the distribution of the initiatives according to the taxonomies. Employing Taxonomy 1, 55 initiatives provided assessment and management, 13 assessment only and 24 education or training. Using Taxonomy 2, 40 provided predominantly direct care, 26 predominantly indirect care and 26 both.

3.6 Characteristics of comparator services

Outcomes of specialist initiatives were typically compared with ‘standard’ or ‘usual’ care. However, in a minority, the comparator constituted an earlier iteration of the intervention. In most studies ‘usual care’ was simply conceptualized as the absence of the additional support the specialist service provided.

3.7 Service effectiveness

3.7.1 Effect on hospital admissions/emergency department transfers

Seventy-five studies that specifically sought to reduce hospital admissions/ED transfers provided data on hospital resource use, of which 11 were rated strong and 24 moderate in quality (Table 2; Supplementary Appendix S2). Eleven studies were entered into a meta-analysis but given the substantial heterogeneity (I² = 100%, Figure 2, Forest Plot A), the results were not meaningful. Excluding weak studies and studies only providing ED visit data did not markedly reduce the high level of heterogeneity (Forest Plots B and C).

The following subsections synthesize and describe the findings according to the foci of intervention and model of specialist support using vote counting based on direction of effect. Sensitivity analyses are reported only where they cast doubt on the robustness of the main analysis.

General health

Twenty-seven studies aiming to improve residents’ general health provided data on hospital admissions/ED transfers, of which 24 favoured the intervention: 89%, 95% CI 72% to 96%, p < .0001 (p < .001 corrected for multiple testing), indicating evidence of an effect. One was rated strong in quality, nine moderate and 14 weak. Most provided assessment and management, but two were assessment-only services (Bellantonio et al., 2008; Hex et al., 2015*) and two training and support initiatives (broad quality improvement programmes, Cool et al., 2018; Ouslander et al., 2011; Rolland et al., 2016).

Some of the strongest evidence came from New Zealand, where ED transfers fell 27% on the introduction of a multidisciplinary support service (Connolly et al., 2018). A similar initiative for residents with ‘big five’ diagnoses (cardiac failure, ischaemic heart disease, chronic obstructive pulmonary disease, stroke and pneumonia) also reported positive effects (Connolly et al., 2015, 2016), as did a nursing intervention with more emphasis on training (Boyd et al., 2014).

Three other moderately rated studies reporting beneficial outcomes involved both nurses and doctors. In Australia, specialist practice nurses provided assessment, support, education and training with input from a geriatrician (Street et al., 2015); in Hong Kong, community geriatric team nurses assessed all residents, and geriatricians subsequently assessed and reviewed those with complex needs, polypharmacy or frequent readmissions (Luk et al., 2002); in the USA, nurse practitioners and primary care physicians detected problems early and managed them aggressively (Kane et al., 2003, 2004). However, a Dutch initiative by trained nurse assistants with optional geriatrician/psychologist consultation reported increased admissions (Boorsma et al., 2011).

Four weak-rated multidisciplinary assessment and management initiatives involving doctors and nurses reported reduced admissions (Ackermann & Kemle, 1998; Hui & Woo, 2002; Hui et al., 2001; Lisk et al., 2012; Sankaran et al., 2010), as did several other weak-rated enhanced primary care initiatives (Aigner et al., 2004; Burl & Bonner, 1991; Bynum et al., 2011; Codde et al., 2010; Klaasen et al., 2009; Lacny et al., 2016; Pain et al., 2014). A similar UK intervention, however, reported increased admissions (Gordon et al., 2018*).

Two of three other nurse-staffed services (Arendts et al., 2018; El-Masri et al., 2015; Ono et al., 2015*) and a study of a UK nurse/pharmacist collaboration (Winstanley & Brennan, 2007) also reported reduced admissions.

Urgent care

Thirteen initiatives for acutely unwell residents provided data on ED transfers/hospital admissions, of which 11 reported beneficial outcomes, indicating evidence of effectiveness: 85%, 95% CI 58% to 0.96%, p = .01. However, this was no longer statistically significant after correction for multiple testing (p = .06). Four were rated strong in quality, two moderate and five weak. Sensitivity analysis treating studies reporting mixed outcomes as negative rather than positive also changed the results so the null hypothesis (no difference between groups) could not be rejected, with 10/13 initiatives favouring the intervention: 77%, 95% CI 50% to 92%, p = .09.

Two US telemedicine services reported reduced ED transfers/admissions—a nurse–doctor assessment service (Grabowski & O’Malley, 2014*) and a geriatric medicine provider assessment and management initiative (Shah et al., 2015). A UK assessment-based emergency practitioner service (Mason et al., 2012*) and Canadian paramedic assessment and management service (Jensen et al., 2016*) also reported fewer hospital referrals, albeit both were rated weak.

Of two Australian ‘Hospital in the Home’ assessment and management of admission-avoidance services, one nursing service failed to demonstrate a clear effect (Crilly et al., 2010). However, the second (strong-rated) medical and nursing initiative reported lower rates of hospital transfer/admissions (Fan et al., 2016), as did several similar services. These included two further strong-rated Australian geriatric outreach services (Chan et al., 2018; Hutchinson et al., 2015), and six moderate-rated US initiatives whereby embedded nurse practitioners provided direct patient care (Ingber et al., 2017; Rantz et al., 2017; Rantz, Popejoy, et al., 2017). A Canadian study in which registered nurses visited care homes on predefined schedules also reported reduced ED transfers (Bandurchin et al., 2011*), but was deemed weak in quality.

Two customized educational initiatives (Hullick et al., 2016; Tena-Nelson et al., 2012*) reported positive effects.

End-of-life care

Of 11 end-of-life care studies that provided data on hospital transfers/admissions, nine explicitly sought to reduce these (Chapman et al., 2018; Garden et al., 2016; Hockley et al., 2010; Kinley et al., 2018; Kuhn & Forrest, 2012; Kunte et al., 2017; Luk & Chan, 2017; O’Sullivan et al., 2016; Teo et al., 2014). Six studies favoured the intervention (Garden et al., 2016; Kinley et al., 2018; Kunte et al., 2017; Luk & Chan, 2017; O’Sullivan et al., 2016; Teo et al., 2014). However, all were rated weak and the null hypothesis could not be rejected: 66%, 95% CI 0.35 to 0.88, p = .25. Section 3.7.2 provides more information about these services’ effect on place of death.

Medication

Of ten studies aiming to improve prescribing, six sought to reduce hospitalization and provided data on ED transfer/admission numbers (García-Gollarte et al., 2014; Lapane, Hughes, Christian, et al., 2011; Lapane et al., 2011; McKee et al., 2016; Roberts et al., 2001; Zermansky et al., 2006). Of these, four favoured the intervention (one strong, two moderate and one weak). Three involved medication reviews by clinical pharmacists (Lapane, Hughes, Christian, et al., 2011; McKee et al., 2016; Zermansky et al., 2006) and the fourth a 10-hour training programme (García-Gollarte et al., 2014). However, synthesis suggested the null hypothesis could not be rejected: 66%, 95% CI 30% to 90%, p = .34.

Post-hospital discharge

All six studies evaluating post-hospital discharge services reported decreased admissions/re-admissions, suggesting a positive effect: 100%, 95% CI 61% to 100%, p = .015. However, this was no longer statistically significant after correction for multiple testing (p = .09). One was rated strong, three moderate and two weak. When studies reporting mixed findings were classed as negative rather than positive, the null hypothesis could also no longer be rejected, although 5/6 studies favoured the intervention: 83%, 95% CI 44% to 97%, p = .11.

Of three Danish studies, one evaluated a rehabilitation intervention post hip fracture provided by a geriatric orthopaedic team (Gregersen et al., 2011) and two investigated general health support services provided by nurse–physician geriatric teams (Pedersen et al., 2016, 2018). Each assessed care home residents on the first working day post-discharge, while the services available in the general health initiatives variously included telephone advice, blood tests, electrocardiograms, intravenous antibiotics, oxygen, subcutaneous fluids and blood transfusions (Pedersen et al., 2016, 2018). Two further initiatives also provided assessment and management (Harvey et al., 2014; Schubert et al., 2016), while in the sixth, a geriatrician/nurse practitioner assessment service, the initial visit was somewhat later than in the Danish studies, but support continued longer (Cordato et al., 2018).

Other specialist initiatives

Five training initiatives focused on preventing falls/hip fracture (Walker et al., 2016; Ward et al., 2010); falls, pressure ulcers and incontinence (Brownhill, 2013); pneumonia (Hutt et al., 2011) and the provision of intravenous fluids/antibiotics (Romøren et al., 2017) with a view to reducing hospitalization. Three favoured the intervention (Brownhill, 2013; Hutt et al., 2011; Romøren et al., 2017). However, synthesis suggested that the null hypothesis could not be rejected: 60%, 95% CI 23% to 88%, p = .5.

Taxonomies of specialist care home support

Taxonomy 1. Seven of nine assessment-only interventions (with/without medical practitioner involvement) (78%, 95% CI 45% to 94%, p = .09) favoured the intervention. Thus, the null hypotheses could not be rejected. There was, however, evidence that assessment and management interventions (with/without medical practitioner involvement) had an effect, with 36 of 43 studies favouring the intervention: 84%, 95% CI 70% to 92%, p < .00001 (p < .0001 when corrected for multiple testing). Syntheses of all non-training initiatives with medical practitioner involvement also showed evidence of effectiveness, with 32/36 studies favouring the intervention: 89%, 95% CI 75% to 96%, p < .00001 (p < .0001 when corrected for multiple testing). Conversely, synthesis of non-training interventions without medical involvement failed to reject the null hypothesis, 11 of 16 studies favouring the intervention: 69%, 95% CI 44% to 86%, p = .10. Synthesis of training only initiatives (with/without medical practitioner involvement) also showed evidence of effectiveness, with 11 of 14 favouring the intervention: 78%, 95% CI 52% to 92%, p = .029. However, this was no longer statistically significant after correction for multiple testing (p = .15).

Taxonomy 2. Twenty-eight of 30 interventions providing predominantly direct care favoured the intervention, showing evidence of an effect: 93%, 95% CI 79% to 98%, p < .00001 (p = .00003 when corrected for multiple testing). Predominantly indirect care also showed an effect in the main analyses, with 12/15 studies favouring the intervention: 80%, 95% CI 55% to 93%, p = .017 (p = .05 when corrected for multiple testing), but this was not sustained in sensitivity analyses: 67%, 95% CI 42% to 85%, p = .15. Synthesis of interventions providing direct and indirect care failed to reject the null hypothesis, 14 of 21 studies favouring the intervention: 67%, 95% CI 45% to 83%, p = .09.

3.8 Healthcare expenditure

Thirty-four studies provided service cost information (Supplementary Appendix S4). However, only two undertook full cost-effectiveness analyses (Lacny et al., 2016; Sackley et al., 2016). Neither had a significant effect on hospital admissions and only one explicitly aimed to reduce admissions (Lacny et al., 2016).

Of 19 moderate-/strong-rated studies with significant effects on transfers/admissions, five provided cost data. Each reported savings (Boyd et al., 2014; Chan et al., 2018; Cordato et al., 2018; Ingber et al., 2017; Kane et al., 2003, 2004), but only three undertook formal cost evaluations. The total costs (inpatient, ED presentations and physician visit costs) of an Australian post-hospital discharge service (Cordato et al., 2018) were $91,854 (mean $4175) compared with $175,517 (mean $8358) for the control group, while an Australian Acute Geriatric Outreach Service reported average savings of $2353 per person from reduced admission and ambulance transfer frequency (Chan et al., 2018). In seven US urgent care initiatives, average per resident Medicare expenditures reduced by $60–$2248 for all-cause hospitalizations and $98–$577 for potentially avoidable hospitalizations (Ingber et al., 2017).

4.1 Methodological considerations

Several methodological considerations must be taken into account when assessing these findings. First, although searches were undertaken in a broad range of databases and the review encompassed a wide spectrum of initiatives, the quality of the evidence was mixed. Many studies involved small numbers of care homes or failed to report sample sizes; most provided little information about comparator services and some lacked detail of the initiative itself. Second, although the taxonomies offered an opportunity to explore the effectiveness of different service configurations, there was inevitably still some subjectivity in the categorization of services and heterogeneity in categorizations. Third, by restricting inclusions to peer-reviewed studies in the English language, the review potentially overlooked other relevant research and the findings may be less generalizable to care home residents in parts of the world with different service structures. Fourth, the possibility of publication bias was not formally assessed and the potential for this cannot be ruled out. Fifth, given the high level of heterogeneity among studies, meta-analysis results were not meaningful. Vote counting by direction of effect facilitated inclusion of the maximum number of studies. However, while this can answer the question ‘Is there evidence of an effect?’, it provides no information on the magnitude of effects and does not account for differences in study size (McKenzie & Brennan, 2019), such that the results should be treated with caution.

Going forward, it is suggested that researchers use the Template for Intervention Description and Replication (TIDiER) checklist when reporting individual studies to improve the completeness and quality of reporting (Hoffman et al., 2014), while future reviews would also benefit from employment of the Grading of Recommendations Assessment, Development and Evaluation (GRADE) approach (Schünemann et al., 2021). This increasingly widely used tool summarizes confidence in the effects of interventions across studies based on consideration of five domains (risk of bias, inconsistency, indirectness, imprecision and publication bias) and specifies four levels of certainty for any given outcome – high, moderate, low and very low.