FormalPara Key Points for Decision Makers

This paper highlighted that community paramedics reduce emergency department (ED) transportation by approximately half compared with standard paramedic models for non-emergency patients.

These models are effectively a simple substitution of labor, as the (relatively expensive) ED is replaced by the (relatively cheaper) community paramedic, which in turn leads to net savings from the healthcare system perspective.

However, these models increase emergency medical services costs (because of community paramedic training and increased time at the scene with patients; this is offset by greater system savings from reduced ED costs). Redistribution of benefits from the ED to emergency medical services may be necessary to incentivise appropriate investment in models to generate net savings for the healthcare sector.

Decision makers could consider introducing community paramedics where sufficient non-emergency demand exists; the threshold for this is yet to be determined and would primarily depend on the cost of introducing a community paramedic model compared to the savings from reduced ED presentations.

1 Background

Emergency medical services (EMSs) around the world report that non-emergency presentations (including primary or urgent care categorisations) now dominate their demand [1,2,3,4,5,6]. In Australia, each state or territory has a single government-approved EMS whose role includes responding to the national emergency number (‘000’), and which are staffed primarily by undergraduate-degree paramedics trained in resuscitation [7,8,9]. In Australia, these services cost $5.4 billion Australian dollars (AU) in 2022–23 [10]. During that period, the EMSs delivered healthcare to over 4 million patients, and transported 85% of these patients to an emergency department (ED) [10]. Of the 4 million patients, 57% were not emergency presentations, and a steady shifting of EMS demand to primarily low urgency has been widely observed [10,11,12,13,14,15,16]. Numerous multifaceted innovations have been described in the literature to address non-emergency requests for EMSs, including hospital-at-home and telehealth [17,18,19,20,21,22,23,24,25], alternative care [26,27,28,29,30] and community paramedics [31,32,33,34].

Community paramedics are in use, or are being developed, in every Australian EMS [9, 34]. These specialist paramedics have different names, such as extended care paramedics, general care paramedics, community paramedics, low acuity paramedics, paramedic practitioners and generalist paramedics [9]. For the purposes of this review, the term community paramedic will be adopted [9, 34]. Community paramedics undergo further education compared with standard emergency paramedics, most commonly a Master’s degree on top of the entry-level Bachelor’s degree, and have a correspondingly increased scope of practice [36]. We have provided an example of how community paramedics complement existing paramedic roles and their different scopes of practice in Fig. 1 [9, 34,35,36].

Fig. 1
figure 1

An illustration of how community paramedic interventions compare to other paramedic roles’ interventions. IM intramuscular, IN intranasal, IO intraosseous, IV intravenous, MDI metered-dose inhaler, PEG percutaneous endoscopic gastrostomy, PO oral, PR per rectal, SC subcutaneous, SL  sublingual

Full size image

Community paramedic roles were recently defined via a Delphi process:

A community paramedic provides person-centred care in a diverse range of settings that address the needs of the community. Their practice may include provision of primary healthcare, health promotion, disease management, clinical assessment and needs-based interventions. They should be integrated with interdisciplinary healthcare teams that aim to improve patient outcomes through education, advocacy and health system navigation [31].

These roles may be either proactive (e.g. scheduled home visits to high-risk recent ED discharges to prevent relapse) or reactive (e.g. dispatched instead of usual emergency paramedics after an EMS request) [34, 35]. The goal for the community paramedic within an EMS is to provide appropriate and safe care to anyone who calls the national emergency number (‘000’) without the need for ED care; this may often involve treatment at the scene with either no transport or transport to a non-ED service such as a community health centre or a general practitioner (GP) [34].

Previous research has found that paramedics specialising in primary-urgent care transport 14–78% fewer patients to EDs [20, 37,38,39,40,41,42,43]. There may therefore be economic benefits to the healthcare system by reducing unnecessary ED presentations. For example, a 20–40 minute GP appointment costs the Australian healthcare system AU$83, compared with a mean cost of $789 for an ED visit (notwithstanding the patient value of receiving immediate rather than potentially delayed care, limited GP availability and patient out-of-pocket GP expenses) [44, 45]. Further benefits may include increased availability of the EMS for medical emergencies [16, 20, 46], a reduction in ED crowding [16, 18, 46], a presumed reduction in iatrogenic harm associated with ED transportation (infection, missed medications, discomfort, loss of routine, unfamiliar surroundings, delirium) [14, 16, 19, 46,47,48,49,50] and increased convenience for patients [16].

Recent research has mapped where and how these roles are being used; however, a synthesis of their effect on cost to the healthcare system remains outstanding. [34, 35] Despite the known limitations of synthesis of economic evidence [51,52,53,54], this is necessary for policymakers given the inconsistent funding and delivery of community paramedic models [34, 35]. As many countries grapple with increasing ED demand, ED wait times, EMS wait times, non-emergency ‘000’ (or equivalent emergency number) calls and prolonged offload times (commonly called ‘ramping’), these community paramedic models may present one mechanism for assisting and consequently merit investigation [55,56,57,58,59,60].

The processes that EMS policy makers would need to employ to gain approval to expand their model of care to include a community paramedic model are unclear in the published literature, although these are likely to vary across services and require, as a minimum, a suitable business case. Such a business case would be greatly assisted by robust economic evaluation evidence [61,62,63,64,65]. A full economic evaluation includes both costs and consequences measured across intervention and comparison groups and covers several methodologies including a cost-effectiveness analysis (CEA) and its utility-specific subset of a cost-utility analysis (CUA), a cost-benefit analysis (CBA), or a cost-minimisation analysis (CMA) [66]. Other methodologies that incorporate some but not all of these elements are known as partial economic evaluations: these include studies considering only costs (costing studies), or those considering only consequences (effectiveness studies) [66].

Seasonal variation and heterogeneous contexts complicate the decisions made regarding delivery of service and in turn decisions to expand service based on data that do not account for these factors. This review seeks to provide guidance on economic outcomes only for those considering reactive community paramedic models within EMSs; other facets of the healthcare quadruple aim, including patient health outcomes, patient experience and practitioner experience, are addressed elsewhere [34, 35], and reported as almost entirely positive. For example, Thompson et al. found a rate of adverse events of community paramedics of 0.001% (1 in 892 patients) [41], while Mason et al. found 99.2% of 2025 patients were treated optimally [67]. Economic guidance, conversely, has been reported as under-researched [33] and previously identified as a key barrier to understanding these community paramedic models [62].

1.1 Review Question

What is the impact of reactive community paramedics within EMSs on healthcare system economic outcomes compared to standard emergency paramedics for similar non-emergency cases?

2 Materials and Methods

A multidisciplinary research team was established including expertise in health economics, service models and paramedicine. This systematic review was conducted in accordance with the Joanne Briggs Institute (JBI) methodology for systematic reviews [68, 69]. Prior to commencing the review, to avoid duplication of efforts, a preliminary search of PROSPERO, MEDLINE, the Cochrane Database of Systematic Reviews, the JBI Database of Systematic Reviews and Implementation Reports and Open Science Framework was conducted and no current protocols on the topic were identified. External input was encouraged by publishing the protocol both with PROSPERO and on Open Science Framework [70, 71]. Patients were not involved in the design or conduct of this research.

2.1 Inclusion and Exclusion Criteria

2.1.1 Types of Studies

This review includes experimental and analytical observational study designs. Descriptive observational studies were excluded given they do not provide a quantifiable effect size.

2.1.2 Population

This review included all patients requesting EMSs via an emergency telephone line (‘000’, ‘111’, ‘999’, ‘911’ or equivalent). In some instances, new emergency contact numbers were established and provided to prospectively enrolled participants for the study (such as a 24-hour palliative care emergency line answered by the EMS provider); these have been included as the number was considered ‘equivalent’ to calling the standard EMS number. All population characteristics, all pathologies and all countries were eligible for inclusion in this review.

2.1.3 Intervention

This review included all paramedic service delivery models that featured a community paramedic working reactively (i.e. in response to ‘000’ requests, rather than proactively scheduled appointments) within an EMS. Studies outlining proactive models, including those identifying high-risk patients post-ED discharge and scheduling follow-ups, were excluded. This exclusion was based on the reasonable expectation of differing effect sizes between reactive and proactive models.

2.1.4 Comparison/Control

This review included articles that feature comparisons with patients attended by standard paramedics (i.e. usual care). Studies without a comparison group, such as those only describing outcomes from a community paramedics model, were excluded because they cannot demonstrate the relative impact compared with standard care.

2.1.5 Outcomes

The primary outcome of this review were any economic evaluation outcomes (those combining both costs and consequences within a single measure), such as a CEA, CUA, CBA, and CCA. Secondary outcomes additionally incorporated were those from partial economic evaluations, including those considering costs only (costing studies) or those that evaluated consequences only [66].

2.2 Search Strategy

A four-step search strategy was utilised (Appendix I of the Electronic Supplementary Material [ESM]) [69, 72]. First, using the initial PICO (population, intervention, comparison, outcome) elements from the research question, a preliminary search of PubMed and Embase was undertaken to develop a comprehensive list of alternate search terms. The results of this preliminary search were reviewed by a professional librarian using the peer review of electronic search strategies (PRESS) checklist [73]. Second, this full string was tested on MEDLINE (Ovid, 1946-present) on 6 February, 2023. Third, the full string was applied on 8 February, 2023 to five databases: Embase (Ovid, 1883–present), MEDLINE Complete (Ovid, 1946–present), Cochrane CENTRAL Register (Cochrane Library), CINAHL (EBSCO, 1961–present) and Scopus (Elsevier, 1966–present) [72]. Fourth, all articles that progressed to a critical appraisal (Appendix II of the ESM) had their reference lists screened for any additional potentially relevant studies (Appendix III of the ESM).

2.3 Study Selection

All citations identified by the search were collated and uploaded into JBI SUMARI (JBI, Adelaide, SA, Australia) and duplicates removed. Titles and abstracts were screened by a sole reviewer for prima facie relevance with a low threshold for inclusion; this stage only aimed to remove all obviously immaterial results. All potentially relevant citations were retrieved in full and assessed in detail against the inclusion criteria by two independent reviewers, with disagreements discussed openly and resolved by consensus among the entire research team (Appendices II and III of the ESM).

2.4 Assessment of Methodological Quality

Any eligible studies selected for retrieval were assessed by two independent reviewers for methodological quality using standardised critical appraisal instruments from JBI (Appendix IV) [74]. The economic evaluation tool was used if the study was at least a partial economic evaluation defined using the criteria of Drummond et al., including comparative incremental analysis, costs, and consequences [66]. Authors of papers were contacted for clarification where methods were unclear (Appendix V of the ESM).

Any studies deemed not to be of satisfactory quality were excluded. This was assessed using pre-determined criteria, including the validity criteria of JBI: provision of adequate data (to identify the cost per patient or other relevant outcome), appropriate rigor in methodology (including comprehensiveness of costs and consequences included, credible valuation, suitable time period, and risk of biases) and potential for selection or information bias [75]. Any disagreements that arose between the reviewers were discussed and resolved by a third researcher, with that third researcher having the opportunity to question the reasoning behind the decisions of the initial two conflicting authors.

2.5 Data Extraction

Data were extracted from studies by two independent reviewers using a comprehensive extraction tool designed by the research team (Appendix VI of the ESM). The data extracted involved a minimum of 52 unique characteristics for each study, including specific details about the publication (authors, year, location), design (methodology, methods), data collection, intervention (training, working alone or in a team, response vehicle capacity to transport), population (characteristics, size), sample group (intervention and comparison characteristics and size), outcomes measured, limitations and findings. Of the 52 unique characteristics extracted from each included study, 21 were outcomes. These 21 outcomes include intervention group size, comparison group size, the absolute numbers within each group that did or did not have the outcome measured, the outcome rate, difference across groups, probability value, confidence intervals, odds ratio, risk ratio, absolute risk reduction and time ratio (ratio of time on scene between intervention and comparison groups) as appropriate. Authors of papers were contacted to request missing or additional data where required (Appendix V of the ESM). Reviewers were also able to separately record any comments on any aspect of the papers, with these openly discussed and resolved by consensus among the entire research team (Appendix VII of the ESM).

2.6 Data Synthesis

Given the high level of variation in populations and interventions, the meta-analysis was considered inappropriate for this review; this was confirmed by the standard X2 and I2 tests [76,77,78,79]. The X2 test was assessed using a probability value of 10% [80] and the I2 test was interpreted using the 0–40%, 30–60%, 50–90% and 75–100% method [81, 82]. Their results were p < 0.001 and > 99% respectively, demonstrating extreme heterogeneity from a statistical perspective [80,81,82]. Therefore, narrative and tabular synthesis was used [69, 83]. Studies were compared based on methodology, population (including recruitment, and population characteristics such as gender, mean age, and disease type and severity), intervention and comparison (including paramedic type), outcomes (including point estimates, interval estimates and a discussion of how the estimates were reached), strengths, limitations and potential for bias. Where relative risk, number needed to treat and confidence intervals could be calculated but were not provided by the authors, these statistics were calculated by the research team using the methods of Altman [84, 85]. Where zeros caused problems with computation of the relative risk or its standard error, 0.5 was added to all cells as per the methods of Pagano and Gauvreau, Deeks and Higgins, and Altman [86, 87]. Relative risk was preferred to odds ratios as per the guidance of Sackett et al. [88].

3 Results

3.1 Study Inclusion

Eleven peer-reviewed published journal articles were included [14, 16, 18,19,20, 40, 41, 46, 47, 89, 90]. The results of the search are presented in a Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA) flow diagram in Fig. 2 [91]. There are minor variations from the PRISMA diagram to capture our additional step of full-text screening for all relevant studies found in the reference lists of included studies.

Fig. 2
figure 2

Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA) diagram of the search results

Full size image

A total of 567 citations were identified and screened. One citation (a government report) was unable to be retrieved in full by the reviewers, professional librarians or the authors of the report themselves [92]. One set of three papers [47, 67, 93] and a second set of two papers [14, 94] were found to use the same underlying data sets; in these cases, the most relevant paper was included, with the data cross-checked for any inconsistency against the other papers and the authors contacted for confirmation where inconsistency was identified. One study claimed to have collected and analysed EQ-5D data; however, it did not present any data, only an unsupported conclusion, and this was considered insufficient reporting by reviewers to support its inclusion in the review [67]. Search results are reported in detail in Appendices I–III of the ESM.

3.2 Methodological Quality

The methodological quality of all but one study was found to be moderate. In seven studies, the intervention group was limited to only low-acuity calls; however, the comparison group (standard paramedic response) included all calls regardless of acuity, a different population [14, 18, 20, 41, 46, 89, 90]. Ten studies identified were found to have inadequate blinding resulting in a high risk of selection bias [14, 16, 18,19,20, 40, 41, 46, 47, 90]. Eight studies used a re-presentation rate, sometimes within a period as short as 24 or 48 hours, as a proxy for treatment success or ‘appropriateness’, with no data presented to justify this assumption [14, 19, 20, 40, 41, 46, 47, 90]. In these papers, representation is usually measured by either a review of ambulance records or the records of a single local ED; patients may have represented elsewhere, such as general practice or community services. Four studies appear to have used the national average ED cost to calculate savings; however, the assumption that these patients have average costs is not evidenced [16, 18, 40, 41]. Of the 11 studies reviewed, all of the studies were noted to have poor reporting of at least one of the precise definition, measurement, collection and analysis of reported outcomes, which hinders accurate interpretation and application of the findings. The quality assessment of individual studies is summarised in Appendix IV of the ESM.

3.3 Study Design

Of the 11 studies included, one can be described as an economic evaluation, using a CUA with EQ-5D (with the UK time-trade-off tariff) [47]. Ten are best described as partial economic evaluations (incorporating comparison of either costs or consequences) rather than economic evaluations. Of these ten partial economic evaluations, four are costing studies (costs alone measured, with no consequence outcomes) and six are effectiveness studies (consequences alone measured, with no cost outcomes).

Although three of the costing studies [16, 18, 41] refer to ‘cost effectiveness’ (a type of economic evaluation), the review team consider that the effectiveness outcome (ED or GP transportation rate) does not reflect health outcomes [16, 18, 40, 41]. The full reasoning behind this is provided in Appendix VII of the ESM. All studies are from countries with developed economies. All studies included an incremental analysis (a comparison with service-as-normal), and three included a marginal analysis (the effect on outcomes per single additional intervention unit) [16, 41, 47]. Characteristics are summarised in Table 1.

Table 1 Study designs of included studies
Full size table

3.4 Patient Population

The models varied significantly in the population eligible for community paramedic attendance. Six models included all EMS calls (‘000’, ‘111’, ‘999’, ‘911’, or equivalent) with a condition that was prima facie appropriate for the community paramedic scope of practice [14, 16, 18, 20, 40, 41]. Two models exclusively responded to specific populations (palliative care [89] and elderly callers [47]), two exclusively responded to residential aged-care facilities [46, 90], and one to pre-enrolled chronic illness patients and provided them with an alternative 24-hour emergency number to call [19].

3.5 Intervention

The interventions provided by the different models also varied significantly. At least seven models had community paramedics responding solo [14, 18,19,20, 46, 47, 90], with two having a team of a community paramedic and nurse [16, 89]; the results of the systematic review were unaffected if the team models were included or excluded from the solo models. Training provided to the community paramedics was extraordinarily heterogenous, this varied from a 10-day course [14] to a 2-year Masters’ degree with additional ED and GP placements [20]. The vehicles were not able to transport patients in at least eight studies [14, 19, 20, 41, 46, 47, 89, 90], with only one site consistently using an ambulance [16]. The populations and interventions of studies are summarised in Table 2.

Table 2 Populations and interventions of studies
Full size table

Oversight was reported in five studies and varied significantly [14, 18, 19, 46, 90]. Community paramedics with low amounts of training generally had more oversight, such as compulsory teleconsultation with a physician during every case [46, 90]. Those with higher levels of training had lower levels of oversight, including no oversight and provision of three days per year for self-directed further education [18].

3.6 Economic Evaluation Outcomes

Only one study provided a full economic evaluation, based on a data set from 2003 to 2004 [47]. This intervention specifically focused on a population aged > 60 years, and the community paramedics underwent three weeks of full-time theory training by medical specialists followed by 45 days of supervised practice (with training costs included in the analysis). Consequences were calculated as quality-adjusted life-years (QALYs) using an EQ-5D completed 28 days post-care. While the study found a > 95% chance of cost effectiveness using a standard UK QALY of £20,000, the difference in costs was not statistically significant with a reduction of AU$454 (inflated and converted to AU 2023, calculations in Appendix VIII of the ESM) in the intervention group. There was no statistically significant difference in QALYs or in patient’s subjective ratings of their own health across the intervention and control groups.

This economic evaluation, while finding a positive outcome from the intervention, is likely to have understated the true effect size; as the authors used intention-to-treat, nearly 30% of the patients included in the intervention group received normal service, leading to a statistically insignificant effect size. The authors of that study chose to use intention-to-treat to reflect real-world practice, where the limited number of community paramedics may mean all eligible patients do not receive the intervention. However, this approach has the unfortunate impact of rendering it difficult to conduct a marginal analysis on scaling up the intervention services to an optimal level (i.e. introducing more community paramedics to treat the nearly 30% of patients who were eligible for the intervention but did not receive treatment). Furthermore, considering a dataset that is nearly two decades old and recognising the advancements in paramedic education during that time, it is conceivable that the same programme could yield even more substantial benefits if implemented in the contemporary context.

3.7 Cost Outcomes

There were an additional four costing studies [16, 18, 41, 93]. Two of these had solo community paramedics attending all eligible emergency calls, both of which found savings, with a range from AU$338 to AU$1227 per patient (inflated and converted to AU$2023, calculations in Appendix VIII of the ESM) if there was suitable demand for services [18, 41]. A further two studies looked at specific models. One found a saving of AU$473 per patient when a paramedic and a nurse attended together in an ambulance vehicle [16]. A second found a saving of AU$900 per patient, although the model, training, vehicle and oversight are not reported [93]. Findings for the economic evaluation and four costing studies are summarised in Table 3.

Table 3 Summary of costing outcomes
Full size table

The main drivers of increased cost in these studies appear to be increased time at the scene, initial training (previously recognised as a disincentive to introducing community paramedics [33, 95, 96]) and paramedic salaries for community paramedics. However, these were offset by the reduced ED attendance and the lower cost per minute of a community paramedic vehicle compared with a standard emergency vehicle. As stated by Thompson et al., ‘the ambulance service bears the cost of setting up and running it while other stakeholders of the health care system realise some of the benefits.’ [41, p. 61]

However, one of these studies’ reported results is based on a scenario analysis (a hypothetical cost of the models, rather than the actual cost observed) [41]. Their scenario analysis assumed that community paramedics attending six patients per shift; in reality, this did not occur. This study was run over five sites, and at one of these sites the community paramedics attended 0.1 patients per shift (2134 fewer per annum than the six per shift predicted in a scenario analysis), resulting in a cost per attendance of AU$10,580. This single outlier site significantly distorted the data from all other sites, and highlights that if community paramedics do not attend patients, they do not provide an economic benefit; there must be sufficient demand.

The cost of the triaging process was not considered by any of the costing studies [16, 18, 41, 47, 93]. Previous research has shown that accurate dispatch of community paramedics to appropriate cases often requires a community paramedic in the communications centre, and sometimes ‘calling back’ the patient after their initial EMS request to obtain further information and determine if the patient is suitable for community paramedic attendance [64, 97, 98]. The cost of a community paramedic dispatching is likely to be different to a standard call taker, in addition to the further costs of any ‘call back’; however, this may result in a more appropriate allocation of a valuable resource and a net benefit. The effect of this on outcomes remains unknown, although from a costing perspective only, it can reasonably be assumed it would minimise the beneficial cost reduction described above.

3.8 Effectiveness Outcomes

Surrogate outcomes investigated included the admission rate, admission length of stay, paramedic time on the case and the 48-hour re-presentation rate. Findings showed mixed results across studies. These findings are summarised in Table 4; unless otherwise noted, all calculations including relative risk, 95% confidence intervals and the probability value were calculated by the research team.

Table 4 Findings of effectiveness measures included in more than one study
Full size table

Emergency department transportation rate was the only outcome universally calculated, and all studies reported a benefit [14, 16, 18,19,20, 40, 41, 46, 47, 89, 90]. A reduction in the ED transportation rate ranged from 14 to 78% [19, 20], likely owing to heterogeneity in models and studies; therefore, determining an absolute effect size is not feasible. However, assuming even a conservative improvement, these models are likely to have a marked impact, illustrated graphically as a forest plot in Fig. 3. When only the higher quality studies are considered, these found relative reductions of 50–54% [18, 41]. The absolute rates of ED transportation in those studies prior to the intervention were 93% and 81%, respectively; the objective effect size, as well as the relative effect, is therefore substantial [18, 41].

Fig. 3
figure 3

Relative risk of emergency department (ED) transportation by community paramedics compared to standard paramedics. CIs confidence intervals

Full size image

A series of other outcomes were investigated in individual studies; these, and the direction and reliability of key findings, have been summarised in Appendix IX of the ESM. An intuitive simple summary of all findings is provided in Table 5.

Table 5 Concise summary of key outcomes
Full size table

The evidence on ED re-presentation is inconclusive: a single study found a 25% reduction in 24-h representation [20], while two studies by the same author found a conflicting 11% reduction and a 6% increase in 48-h representation [46, 90]. In Hoyle et al., 3% of patients re-presented for the same issue within 7 days; these 18 cases were reviewed by an emergency consultant, and in each case, the initial community paramedic management was considered to be appropriate [14]. Longer term 28-day outcomes were similarly limited by weak evidence, although none was negative: individual studies found non-inferiority in EMS presentations [67], and 25% and 6% reductions in EMS representations and hospital admission rates, respectively, in those aged > 60 years [47]. One study considered mortality over 28 days for those aged > 60 years, finding no difference [47].

4 Discussion

This review captured, evaluated and combined existing data on the economic outcomes of reactive community paramedic models within EMSs. Economic evaluations acknowledge that resources necessary for health interventions are scarce and seek to determine how they may best be allocated to minimise the opportunity cost when comparing alternative interventions [66]. Despite the limitation of imperfect information and system heterogeneity, among others, systematic reviews of healthcare aim to help meet this need for policymakers [51,52,53,54]. In this review, economic evidence is limited to one cost-utility study [47], with four additionally providing costing evidence [16, 18, 40, 41], and 11 evaluating effectiveness measures including the ED transportation rate and re-presentation rate [14, 16, 18,19,20, 40, 41, 46, 47, 89, 90].

The single economic evaluation found a > 95% chance of cost effectiveness. The interpretation of this study is critical for users of this review, as it is the only study to assess both costs and health outcomes together [47]. One narrow interpretation is that providing community paramedics with limited additional training over several weeks to specialise in gerontology could be a cost-effective model of care. Furthermore, considering a dataset that is nearly two decades old and recognising the advancements in paramedic education during that time, it is conceivable that the same programme could yield even more substantial benefits if implemented in the contemporary context.

In addition to this, four costing studies found net healthcare system savings of AU$338–AU$1227 per attendance, provided a minimum threshold of demand is met; given the barriers to generalisability of these findings across systems, these should only be considered examples of potential outcomes [16, 18, 40, 41]. However, one subgroup of one study found a negative effect on cost, in a rural area where the community paramedics saw no patients for 4 consecutive months [41]. This created a shortfall in predicted cases of 2134 and a corresponding cost overrun. This study, although skewing the data, highlights the most important finding of this review; economic outcomes for these practitioners are highly dependent on a consistent number of patients seen per shift.

When considering all models, ED transportation rate was the only outcome universally calculated, likely a result of the known phenomenon of health services routinely measuring what is easy to measure, rather than what is most meaningful for health outcomes [33, 99]. All studies reported a reduction in the ED transportation rate [14, 16, 18,19,20, 40, 41, 46, 47, 89, 90] described as ‘dramatic’ [89, p. 367], ‘large’ [90, p. 114], ‘marked’ [91, p. 242), and having ‘clear benefits’ [18, p. 52]. This is the driver behind the costing and economic benefits, suggesting that the introduction of these models is ‘a simple substitution of labor’ [47, p. 449], in which the (more expensive) ED is replaced by the (relatively cheaper) community paramedics [18, 47]. However, there remains insufficient evidence of whether this represents an absolute cost reduction or simply cost shifting onto the consumer. For instance, decreased public ED use could lead to increased referrals to privately funded allied health, which remains uninvestigated.

This has important policy implications: many EMSs are remunerated based on transportations to the ED, with either lower or no funding to offer on-site services [10]. As noted by Breyre et al., these models ‘decrease revenues for EMS systems that are reimbursed only for transported patients [however they may be] ethically the right thing to do for the patient’ [89, p. 369]. To incentivise EMS may involve remuneration for non-transportation, or the shifting of funding from EDs (where the economic benefit of community paramedics is felt) to EMSs (who bear the cost of the services but not the benefit).

There are further potential benefits to these models that have not been measured in the existing evidence. This includes increased availability of EMSs for medical emergencies (reduced response time for ‘high acuity’ requests for service), [16, 20, 46, 47] a reduction in ED crowding (and, consequently, reduced waiting room time and total time in ED for other patients), [16, 18, 46] a presumed reduction in iatrogenic harm associated with ED transportation (infection, missed medications, discomfort, loss of routine, unfamiliar surroundings, delirium), [14, 16, 19, 46] and increased convenience for patients [16]. These factors all may have a genuine benefit but are simply not measured in existing evidence.

As the evidence included narrowly focuses on reactive community paramedics performing that role full time, the benefit of community paramedics in mixed roles, such as additionally assisting in a local ED or providing proactive primary care services whenever there are no pending EMS cases [95, 97, 100], is an unknown benefit. Such mixed roles have been discussed as being beneficial in rural and remote areas where there may be a lower demand for EMS [96, 97, 100]. Further research would be needed to quantify this benefit. Critically, this review also does not consider the ethical or moral reasons why EMS may choose to provide EMS in areas with lower demand, such as providing care to under-serviced rural and remote communities, which may be socially desirable regardless of cost.

None of the research included in this systematic review quantified the effect of any delay in accessing GP care or any out-of-pocket expenses (such as gap fees) for patients accessing GP care. In some locations, such as Australia, limited GP availability and increasing out-of-pocket costs are currently major policy considerations at the national level, and there is research suggesting that lack of alternative options (such as GP availability) is a driver of non-emergency EMS requests [101, 102]. Any significant delay in patient care has the potential to negatively affect health outcomes, and patient out-of-pocket fees are an unappreciated cost, both of which would be expected to reduce the economic benefit of community paramedics.

The health outcomes used in community paramedic research can be split into two broad sub-types: those investigating the degree of benefits (such as a mmHg change in systolic blood pressure) and those adopting a binary non-inferiority approach (such as the number of adverse events compared to usual care). The former are rare, and despite their more rigorous approach are arguably of limited utility; paramedics invariably treat an almost unlimited range of pathologies, each of which have their own pertinent clinical endpoints, and which makes choosing and measuring outcomes across a heterogenous patient set impractical. This is further complicated by the fact that their baseline state is almost universally unknown; these patients are often seen after an EMS call, and prospectively identifying patients before the call to EMS is similarly impractical. Use of universal outcomes, such as 28-day neurologically intact survival, are largely inapplicable because of the low acuity nature of the pathologies being treated, and more simplistic measures such as the ED transportation rate do not reflect health status or treatment suitability.

Therefore, almost all research in this field comes with a long list of confounders and limitations. These routinely include an unknown baseline for both comparison and intervention groups, selection bias (as the community paramedics are not randomly dispatched, but instead intentionally allocated to cases that they think they can help with), different comparison and intervention group severity (community paramedics are highly unlikely to be dispatched to Code 1 cases in practice, which are more likely to require ED transportation and consequently more likely to be high cost, skewing the results to over-stating a benefit), and a wide variety of conditions attended inhibiting use of outcome measurements.

Despite this multitude of difficulties, as Drummond et al. state, ‘decisions can and must be made, regardless of the limitations of available information’ ([66], p. 2). In the case of community paramedics operating reactively within EMSs, appropriate policy based on current evidence appears to be EMS negotiating with their health department, local ED, or insurers to introduce a rebate for successful community paramedic non-ED transportations and, following this, geographical areas with suitable non-emergency demand identified and community paramedic models piloted with prospective economic evaluations.

4.1 Limitations

Across almost all studies, patients were commonly not randomised and instead purposefully allocated, risking selection bias: community paramedics may have been preferentially allocated to patients inherently less likely to require ED transportation. Similarly, at least four studies’ comparison groups included high-acuity patients, skewing outcomes towards overstating the effect size of the intervention [18, 20, 46, 90].

Eight studies measured the re-presentation rate, often as a proxy for health outcomes [14, 18,19,20, 40, 46, 47, 90]. As previously mentioned, this may over-simplify findings as re-presentation may be affected by many factors unrelated to an objective health status or treatment quality (such as patient reassurance or relief of symptoms without correction of underlying pathology) and is not appropriate as a health measure.

The single economic evaluation used a CUA approach on patients aged > 60 years with a data set from 2003 to 2004, and with paramedics undertaking only 3 weeks’ additional training [41]. Modern community paramedics commonly hold a Master’s degree and carry, in some instances, dozens of additional medications [9, 31, 35, 36, 103]. It is reasonable to suggest that this vastly expanded education and scope of practice will in turn influence the practitioner’s ability to treat at the scene, and result in cost savings to the healthcare system. However, with an almost complete absence of full economic evaluation data on modern community paramedic systems, this could not be evaluated in this review.

As discussed above, the sole economic evaluation ever produced on these models used a CUA using the EQ-5D [47]. The use of EQ-5D in these circumstances, while commendable, has limitations. Isolating the specific QALY gains attributable to the community paramedic intervention was not possible in this study design, as patients are likely to have seen multiple health practitioners over the month, confounding the measured effect. Furthermore, there was neither randomisation nor baseline utility values collected to support this, and significant baseline differences were reported between groups with the intervention group being older and less likely to live at home [47]. Collectively, these limitations — a highly specific population, minimal paramedic education, heterogenous comparison and control groups, dated data, and an outcome that may be confounded by other interventions—mean that the generalisability of findings cannot be assumed.

Given the limitations of a CUA for this research aim, alternative economic evaluation methods, such as a CEA, CBA or CMA may be preferable, and each of these will be briefly considered [66]. A CEA is likely of minimal use in evaluating community paramedics because of the multitude of pathologies these patients present with; there is no universal natural unit to measure health outcomes with. These patients have chronic conditions, generally managed by a multidisciplinary team, and the impact of a single episode of care on a patient’s trajectory is difficult to isolate. Even a measure as basic as mortality may not be a useful outcome measure for this intervention, as the non-emergency nature of these patients means death is unlikely to be affected during the brief single episode of care provided by community paramedics. Additionally, one of the subgroups routinely treated by these specialists are palliative patients, for whom a dignified death at home is routinely a treatment goal [19, 46, 89, 90]. While some studies self-identify as using ED transportation as a natural unit [16, 18], the research team consider that this does not reflect health outcomes and may be due to many factors unrelated to treatment quality, such as patient satisfaction (i.e. a patient may feel reassured and decide they do not need transport, despite having been given the incorrect treatment). A CMA, already widely accepted as having minimal use in modern health economics [66, 104], is not currently supported as there is insufficient non-inferiority health outcome evidence (and, given the heterogeneity of models and populations, such data would need to be region specific). Given these limitations to CUA, CEA, and CMA methodologies, the most appropriate methodology for future researchers is likely to instead be a CBA using a patient-reported outcome such as willingness to pay [66]. While the monopoly granted to most EMSs means no competitive market exists, contingent valuation methods appear feasible [66].

4.2 Recommendations for Practice

There are three recommendations for policymakers.

  1. 1.

    Emergency medical service policymakers could consider negotiating with their health department, local ED or insurers to introduce a rebate for successful, community paramedic non-ED transportations.

  2. 2.

    Following this, geographical areas with suitable non-emergency demand could be identified.

  3. 3.

    In these areas, community paramedic models could be piloted with a prospective economic evaluation or, where this is not feasible, with sufficient data collection to enable a post hoc analysis.

4.3 Recommendations for Research

From a methodological and policy-guiding perspective, there are 12 recommendations and areas for future research.

  1. 1.

    The production of further economic evaluations. Currently, policymakers are forced by a lack of evidence into making decisions without adequate economic evaluation evidence, such as decisions based on surrogate outcomes such as the ED transportation rate that may not correlate directly with cost benefits. The sole economic evaluation included in this study uses a cost-utility methodology of a specific population subgroup from data collected two decades ago [47]. These data have limited function owing to their lack of contemporaneity (especially given the rapid evolution of these models) and the specific population subgroup. Additionally, the costing study methodology used by four other studies does not incorporate health outcomes, and consequently provides an incomplete picture of the consequences of an intervention. Adoption of a CBA, CUA or CEA approach will avoid this pitfall. Completion of further economic evaluations will ensure policymaker decisions are informed by evidence.

  2. 2.

    A CBA appears the most appropriate future methodology in these circumstances. As discussed earlier, multi-attribute utility instruments are less appropriate for this cohort, ED transportation should not be considered a natural unit representing health status, mortality has limited use and there are no other suitable natural units identified. Use of a CBA with contingent valuation appears the most appropriate methodology. This point will allow policymakers to have the most appropriate evidence informing their decision making.

  3. 3.

    Three costing studies here either incorrectly described themselves as cost-effectiveness analyses or are otherwise ambiguous as to their exact methodology [16, 40, 41]. It would be appropriate for future authors to consider the criteria of Drummond et al. when describing their methods [66].

  4. 4.

    While costing studies are likely to retain a place within EMS evaluations of these models because of their relative ease of use, their accuracy compared to economic evaluations remains unknown and should be tested. Policymakers report currently being forced to make decisions based on costing studies because of inadequate economic evaluations; testing the accuracy of costing studies against a ‘gold standard’ economic evaluation will assist in determining how reliable this approach is, and if it should be retained by policymakers going forward.

  5. 5.

    A marginal analysis is included in only three studies [16, 41, 47]. A marginal analysis is important for programme directors to evaluate the optimal level of investment, and future research is needed in this area. Without this, policymakers will be unable to determine the supply of community paramedics necessary to meet the demand of their area.

  6. 6.

    Most studies lack appropriate blinding, and several are at a high risk of selection bias as patients are prospectively allocated to either the community paramedics or normal care, in some cases, explicitly based on their perceived likelihood of requiring ED transportation [14, 16, 18,19,20, 40, 46, 47, 90]. While blinding is difficult in the emergency setting, potential methods include comparison group selection via propensity score matching. Studies of community paramedics without blinding are at a high risk of skewed findings and may be misleading for policymakers and result in inappropriate policy.

  7. 7.

    Future studies could limit the comparison group to patients of a similar acuity to the intervention group. It is reasonable to expect high-acuity patients are more likely to require ED transport and incur higher costs; inclusion of these patients within the comparison group will inappropriately skew results to overstating the intervention effect size. As above, studies of community paramedics using inherently different intervention and control groups have a risk of misleading results.

  8. 8.

    Future research may be able to identify a population density at which there are sufficient appropriate cases to warrant the model being introduced. Currently, policymakers have no guidance on how to determine if a local area is suitable for a community paramedic introduction or not, limiting their ability to make informed decisions.

  9. 9.

    No studies considered the costs of triaging calls to appropriate EMS models of care. As previous community paramedic research has recommended a community paramedic remain in the communications centre to oversee dispatch, and in some cases, initiating a second call back to the patient to gain further clinical information, this will have costing implications. Without this information, researchers may be skewing results to overstating the positive impact of models, and in turn providing inaccurate guidance to decision makers.

  10. 10.

    All studies used ED transportation rate as a surrogate outcome, and in some instances stated that this was used as a proxy for health outcomes. As previously discussed, ED transportation is inappropriate as a health outcome measure. Policymakers should not be provided with data that assume that reduced ED transportation rate is, of itself, reflective of cost and health outcomes, as it remains unproven that it directly correlates with either.

  11. 11.

    Studies were noted in some circumstances to take differing interpretations of their results. The rate of inpatient admission of community paramedic-attended patients was lower in one study and higher in another; however, both studies claimed that their findings were positive [19, 47]. The study with a lower rate suggested that community paramedics are minimising patient deterioration, while the study with the higher admission rate suggested that community paramedics are accurately identifying the sickest patients and transporting them appropriately. Pre-stating interpretation intentions in a protocol may reduce this inconsistency and minimise the risk of confirmation bias; policymakers may be misled by any researcher confirmation bias in the presentation of results.

  12. 12.

    The included studies calculate costs largely from the healthcare system perspective. This means it is unclear if the reduction in costs demonstrated in four studies is a genuine reduction, or a simple shifting of costs onto the healthcare consumer. Future research could investigate economics from the patient perspective. Policymakers could be informed on the holistic impact of models, including on the consumer cost, to ensure informed decision making.

5 Conclusions

There are two conclusions from this review of relevance to policymakers. For community paramedic models within EMSs to be economically beneficial, clinicians would require a minimum number of patients per shift. Or, put more simply, if introduced in an area with no demand, they will not have a benefit. The population threshold necessary to achieve this demand is unclear and would primarily depend on the region’s saving per ED attendance avoided compared with the cost of the community paramedic response. In rural or remote regions with lower demand, community paramedics may nonetheless still be economically beneficial by performing other community services such as assisting in primary care, a topic beyond the scope of this review. They additionally may be desirable in rural or remote communities for ethical reasons, such as providing appropriate care to underserved communities.

Second, community paramedic models within EMSs have been observed to increase costs to EMSs (owing to the additional training costs and increased time at the scene), although this is outweighed by downstream savings (from reduced ED presentations). As stated by Abrashkin et al., ‘lack of reimbursement for non-transported paramedic services is an obstacle’ [19, p. 4]; redistribution of benefits may be necessary to incentivise EMSs to invest in these programmes for a whole-of-system economic gain. Therefore, policymakers for EMSs could consider negotiating with their health department, local ED or insurers to introduce a rebate for successful community paramedic non-ED-transportations. Following this, geographical areas with suitable non-emergency demand could be identified, and community paramedic models introduced and tested with a prospective economic evaluation.