Systematic review of integrated models of health care delivered at the primary–secondary interface: how effective is it and what determines effectiveness?

Governments internationally have committed to improving the integration of healthcare services for the growing numbers of people experiencing chronic disease. Models of care that feature vertical integration of health care between primary and secondary–tertiary care (Gröne and Garcia-Barbero 2001) have become a source of interest, especially those that involve both General Practitioners (GPs) and medical specialists. Traditional, siloed, organ-based care approaches have failed to provide the holistic, accessible, ‘linked-up’ care now required – particularly for the growing numbers of community-dwelling frail elderly (Boyd et al. 2005). Canada, the United Kingdom and New Zealand have introduced legislative or policy initiatives to advance integration of primary and secondary care (Eliasoph et al. 2007; Ministry of Health 2011; Goodwin et al. 2012). In Australia, both the National Health and Hospital Reform Commission report (Australian Government National Health and Hospitals Reform Commission 2009) and the National Primary Care Strategy (Australian Government Department of Health and Ageing 2009) have strongly endorsed the need to redesign the health system towards accessible integrated services, particularly for people with chronic disease.

There are many challenges with such significant service redesign (Australian Government National Health and Hospitals Reform Commission 2009). Not least is the need to ascertain what factors improve the quality, safety or cost-effectiveness of care around the interface between community-based primary care and hospital-based specialist care, specifically for patients with chronic/complex disease like diabetes mellitus.

Different terms have been described for multidisciplinary care. Tieman et al. (2007) have described different elements of care that involve multiple providers. Coordination was seen as processes and activities that enhance the relationships, linkages, transitions and responsibility for care within the existing structural arrangements such as shared health records, case conferences or shared assessment tools, leading to improved care arrangements for the patient. Integration is the development of more comprehensive approaches to care provision that depend on formal relationships or structural arrangements to organise and deliver that care. Multidisciplinary care is a care approach that addressed complex care needs by utilising a broader set of skills in assessment and ongoing care held by providers from different disciplines, specialties and /or professions who could contribute independently. Another relevant concept is stepped care, which proposes care of increasing intensity depending on the complexity or advanced nature of the condition (Von Korff and Tiemens 2000; Smink et al. 2014). For the purposes of this project, we have elected to use the term ‘integrated care’, as we are interested in the concept of systematic, organised interaction between primary and secondary care.

Considerable work has been done examining integrated, multidisciplinary care. Most of this has focussed on describing interventions at a systematic level, seeking to define the features of this form of care. Ouwens et al. (2005) conducted a meta-review of 13 studies that focussed on identifying the characteristics of integrated care that were important in improving the care of chronically ill people, as well as their effectiveness. Martínez-González et al. (2014), building on the work by Ouwens et al. (2005), conducted a meta-review seeking to identify the principles that facilitate high-quality integrated care.

Others have focussed more on primary care. Gruen et al. (2004) conducted a Cochrane review of outreach specialist clinics in primary care and rural settings. They showed that specialist outreach clinics in general practice improved access, but not health outcomes, but more in-depth collaboration with primary care did improve health outcomes, and more efficiently. Singh (2005) sought to identify initiatives that were effective for improving the quality and cost-effectiveness of care for people with chronic illness. She found evidence to support integrated community and hospital care and a greater reliance on primary care, but high-quality evidence regarding the effects of such initiatives on clinical outcomes and healthcare costs was lacking. The Cochrane review conducted by Smith et al. (2007) assessed the effectiveness of shared care interventions for improving chronic disease management across the primary–secondary care interface, and found no improvements in patient outcomes except for better prescribing practices, and recommended that shared care models not be adopted widely until better studies, which are longer and whose design take into account the complexity of the interventions, have been conducted.

It is clear that some methods of improving care across the primary–secondary interface are more effective than others. Integrated care, as defined above, involves active negotiation of case assessment and care planning and care delivery between primary care providers and specialists. A gap in the literature exists in examining this form of integrated care between specialists and general practice care in its own right, rather than as a subset of broader cross-disciplinary care strategies.

We undertook a systematic literature review to answer the following question: For adults with chronic and/or complex chronic physical conditions, do interventions that integrate primary and secondary care improve clinical, process and cost outcomes, compared with usual care? A secondary aim was to identify, from descriptions of the workings of the models of care provided in each study, the common organisational and operational elements (design elements) present in models that produced more successful outcomes.

For the purpose of this review, integrated primary–secondary care was defined as vertically integrated models of care for managing chronic/complex chronic disease in individual patients, which involves direct interaction between primary and secondary care providers. We focussed on identifying published models of integrated primary–secondary care that reported pre-specified outcome measures, which allowed comparisons of benefit of different models. We aimed to identify how such care impacts on outcomes for patients, practitioners and the health system, and affects process of care and resource utilisation. We also sought to identify the design elements that underpin effective models of care.

Studies were included if they recruited adults with chronic/complex chronic disease and compared care models using specialist and primary care medical practitioners working collaboratively across primary and secondary care settings, with usual care. Chronic disease is defined as illness that is prolonged in duration, does not often resolve spontaneously and is rarely cured completely (Australian Institute of Health and Welfare 2012). We searched for English language peer-reviewed studies published between January 2000 and July 2012. Study designs included randomised controlled trials (RCTs), non-randomised or quasi-randomised controlled trials and pre-post studies. We excluded studies focussed on paediatric or oncology models of care because the distinctive needs of these population groups were not germane to our review. Mental health models were also excluded due to their orientation to broader societal issues that are beyond the capacity of the health system and individual patients to control (Germov 2005).

A list of terms and MeSH synonyms was developed by the authors, with reference to the above definition of integrated care models, and were categorised under the following key areas of interest: (1) integrated models of care; (2) primary and secondary care; (3) chronic/complex chronic disease; and (4) outcomes (Appendix 1). The initial list of search terms was applied to two databases (CINAHL and Medline) to test for relevance. Abstracts of potentially useful studies were read to identify any other relevant search terms. The complete search strategy with all identified search terms was then applied to Cochrane, CINAHL, Medline, PubMed, PsychINFO and Embase databases to identify all potentially relevant studies. We also hand-searched key articles and the reference lists of identified studies.

Retrieved studies were assessed independently for inclusion by two co-authors (LB, JZ). In cases of disagreement regarding whether to include a study, a decision was reached by a third reviewer with a clinical background (CJ).

Once the final set of studies was selected, information about design, participants and setting, models of care, outcome measures (clinical, process and economic outcomes) were extracted and summarised. As the studies were highly heterogeneous with regard to patients, interventions and reported outcomes, a meta-analysis of quantitative data was not possible and hence results are presented in narrative form using the method of Popay et al. (2006). From a qualitative perspective, both the operational barriers and enablers to integrated care, as identified by authors of the papers, were tabulated in an effort to identify the design elements of models of care that demonstrated improved clinical, process or economic outcomes, compared with usual care.

The Cochrane risk of bias tool (Higgins and Green 2011) was used to appraise included studies. This tool assesses the studies for the level of risk of the following forms of bias: selection bias, performance bias, detection bias, attrition bias and reporting bias. Risk is assessed as low, moderate or high, and the reviewer is required to explain how that conclusion was reached for each potential bias.

The search produced 1516 hits, and a further 50 hits were identified and screened during the assessment process, as described in Fig. 1.

Fig. 1.  Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) flow diagram of the study selection process (Moher et al. 2009).

The details of included studies are summarised in Table 1. Ten studies (14 papers) involving a total of 7697 patients were included in the review. The majority of studies related to patients with diabetes (n = 6; Simmons 2003; Nocon et al. 2004; Smith et al. 2004; Kirsh et al. 2007; Borgermans et al. 2009; Askew et al. 2010; Goderis et al. 2010; Jackson et al. 2010). Three were from New Zealand (Doughty et al. 2002; Pearl et al. 2003; Rea et al. 2004; Sheridan et al. 2009), two each from Australia (Simmons 2003; Askew Jackson et al. 2010; Jackson Tsai et al. 2010) and the UK (Nocon et al. 2004; Coast et al. 2005; Salisbury et al. 2005), with the remaining three from the US (Kirsh et al. 2007), Belgium (Borgermans et al. 2009; Goderis et al. 2010) and Ireland (Smith et al. 2004). One study was an RCT (Coast et al. 2005; Salisbury et al. 2005), four were cluster RCTs (Doughty et al. 2002; Pearl et al. 2003; Rea et al. 2004; Smith et al. 2004; Borgermans et al. 2009; Goderis et al. 2010), three reported a quasi-experimental design (Nocon et al. 2004; Kirsh et al. 2007; Askew et al. 2010; Jackson et al. 2010) and two used a pre-post design (Simmons 2003; Sheridan et al. 2009). Eight studies used quantitative methods (Doughty et al. 2002; Pearl et al. 2003; Simmons 2003; Rea et al. 2004; Smith et al. 2004; Coast et al. 2005; Salisbury et al. 2005; Kirsh et al. 2007; Borgermans et al. 2009; Askew et al. 2010; Goderis et al. 2010; Jackson et al. 2010), whereas two used a mixed-methods approach (Nocon et al. 2004; Sheridan et al. 2009).

Table 1.  Details of studies of integrated primary–secondary care for managing chronic/complex chronic disease
HbA1c, glycated haemoglobin; IDCT, interdisciplinary diabetes care team; UQIP, usual quality improvement program; AQIP, advanced quality improvement program; SBP, systolic blood pressure; LDL-C, serum low-density lipoprotein cholesterol; GP, general practitioner; GPwSI, General practitioner with a special interest; COPD, chronic obstructive pulmonary disease

Study quality was mixed, with the RCTs (Doughty et al. 2002; Pearl et al. 2003; Rea et al. 2004; Smith et al. 2004; Coast et al. 2005; Salisbury et al. 2005; Borgermans et al. 2009; Goderis et al. 2010) demonstrating low to medium risk of bias and the non-randomised trials (Simmons 2003; Nocon et al. 2004; Kirsh et al. 2007; Sheridan et al. 2009; Askew et al. 2010; Jackson et al. 2010) demonstrating medium to high risk (Table 2). A full appraisal of the risk of bias in the included studies is available in Appendix 2.

Table 2.  Assessment of quality and risk of bias (Cochrane risk of bias tool; Higgins and Green 2011)
GPwSI, general practitioner (GP) with a special interest; COPD, chronic obstructive pulmonary disease; N/A, not applicable

We categorised study outcomes into clinical, process of care and economic, and here we report the quantitative outcomes.

Eight studies reported clinical outcomes (Table 3). In both RCTs and other study designs, there were many outcomes that showed no difference between groups. For the five diabetes studies, there were a few improved outcomes in RCTs and non-randomised studies, but the magnitude of the improvements were larger in the non-randomised studies (Simmons 2003; Kirsh et al. 2007; Jackson et al. 2010). One RCT showed improvements in wellbeing (Smith et al. 2004). In programs for respiratory disease and heart failure, some of the quality of life subscale scores improved (Doughty et al. 2002; Rea et al. 2004).

Table 3.  Clinical outcomes of studies of integrated primary-secondary care for managing chronic/complex chronic disease
HbA1c, glycated haemoglobin; I, intervention group; C, control group; NS, no significant difference between intervention and control groups; SBP, systolic blood pressure; DBP, diastolic blood pressure; LDL-C, low-density lipoprotein – cholesterol; HDL-C, high-density lipoprotein – cholesterol; B, baseline; LV, last visit; CI, confidence interval; COPD, chronic obstructive pulmonary disease; FEV₁, forced expiratory volume in 1 s; ACEI, angiotensin-converting enzyme inhibitor; ARBs, angiotensin receptor blockers

Seven studies reported process of care outcomes (Table 4). Patient attendance rates improved in one study of patients with diabetes (Smith et al. 2004), and hospital attendances fell in another (Nocon et al. 2004). Reported non-attendance rates reduced in one study (Jackson et al. 2010), but were worse for intervention clinics in another study (Nocon et al. 2004). Nocon et al. (2004) also noted increased combined hospital and outpatient clinic usage (Nocon et al. 2004). While hospital admission rates fell in the intervention for complex medical patients (Sheridan et al. 2009), there was no change in admission rates for patients with chronic obstructive pulmonary disease (COPD; Rea et al. 2004) or heart failure (Doughty et al. 2002). However, falls in hospital length of stay (Rea et al. 2004) and readmission rates (Doughty et al. 2002; Sheridan et al. 2009) were reported.

Table 4.  Process outcomes of studies of integrated primary–secondary care for managing chronic/complex chronic disease
NS, no significant difference between intervention and control groups; I, intervention group; C, control group; CI, confidence interval; GP, general practitioner; HbA1c, glycated haemoglobin; ACE, angiotensin-converting enzyme; COPD, chronic obstructive pulmonary disease; 2P, two-tailed probability

There was evidence of improved clinical performance by GPs, with better recording of important clinical information, and better capture of diabetes patients on practice diabetes registers (Smith et al. 2004). There was also evidence of better patient information sharing between sectors (Smith et al. 2004). Both clinicians (Pearl et al. 2003; Nocon et al. 2004; Salisbury et al. 2005; Sheridan et al. 2009) and patients (Nocon et al. 2004; Salisbury et al. 2005; Sheridan et al. 2009) reported satisfaction with these initiatives, with clinicians holding the view that the interventions improved patient outcomes. There was no clear difference in frequency of reporting of outcomes between RCTs and studies employing other designs.

Table 5 reports health economic outcomes, with four studies reporting cost data. Jackson et al. (2010) showed a substantial cost reduction of clinic-based care for patients with diabetes compared with hospital outpatient-based care. Other programs showed modest extra costs (Coast et al. 2005; Borgermans et al. 2009) or no difference (Nocon et al. 2004). Only one study calculated an incremental cost benefit for the intervention (Coast et al. 2005). Costs were higher for the intervention in both studies that used a RCT design. For studies using other designs, costs were lower for one study and no different to controlled data for the other studies.

Table 5.  Economic outcomes of studies of integrated primary–secondary care for managing chronic/complex chronic disease
I, intervention group; C, control group; GPwSI, general practitioner (GP) with a special interest

While improvements in clinical outcomes were modest, most models showed improved process outcomes, particularly for GPs within the interventions. We reviewed the methods described in the included papers to ascertain the organisational and operational elements of each. We identified six elements that appear to facilitate models of integrated primary–secondary care. The studies in which they were described are shown in Table 6. These elements are: (1) interdisciplinary teamwork; (2) communication and information exchange; (3) the use of shared care guidelines or pathways; (4) training and education; (5) access and accessibility; and (6) a viable funding model. For each element, we identified facilitators and barriers described in each published work.

Table 6.  Design elements that underpin effective models of integrated primary–secondary care

Effective integration depends on the right mix of interdisciplinary health professionals and roles which predisposes to a well-functioning team (Doughty et al. 2002; Pearl et al. 2003; Simmons 2003; Nocon et al. 2004; Rea et al. 2004; Smith et al. 2004; Kirsh et al. 2007; Borgermans et al. 2009; Sheridan et al. 2009; Goderis et al. 2010; Jackson et al. 2010). These teams featured good coordination by personnel with an understanding of communityand specialist-based care (Doughty et al. 2002; Nocon et al. 2004) and clearly defined roles (Doughty et al. 2002; Nocon et al. 2004; Sheridan et al. 2009). Barriers to team functionality included GPs being too busy for direct involvement in comprehensive patient care, inadequate access to other key personnel and lack of role clarity (Simmons 2003; Smith et al. 2004; Sheridan et al. 2009). Nocon et al. (2004) reported clinicians’ concerns that, without role clarity for referring GPs and specialist clinics, duplications and omissions may occur in managing co-morbidities.

Effective integration involves willingness to share information, and supportive managerial and administrative staff (Rea et al. 2004; Smith et al. 2004). A high level of GP trust in specialists was regarded as important, as was improved communication between GPs and hospital specialists (Coast et al. 2005; Jackson et al. 2010) and shared follow up supported by electronic reminder systems (Doughty et al. 2002; Kirsh et al. 2007). Shared governance that enhanced system capacity for effective communication and collaboration (Simmons 2003) and regular interdisciplinary team meetings that enabled information exchange (Borgermans et al. 2009; Sheridan et al. 2009) were also seen as important. Successful communication channels included case conferences (Sheridan et al. 2009). Co-located GP and specialist clinics facilitated effective communication and information exchange between GPs and specialists (Simmons 2003), as well as ongoing access to specialists and shared follow up (Nocon et al. 2004).

Pragmatic, locally agreed care protocols were a key component of most of the integrated care models (Doughty et al. 2002; Simmons 2003; Nocon et al. 2004; Rea et al. 2004; Smith et al. 2004; Borgermans et al. 2009; Goderis et al. 2010; Jackson et al. 2010). The protocols included guidance for post-discharge care and review (Smith et al. 2004), shared care planning (Jackson et al. 2010), patient goal-setting and self-management (Rea et al. 2004) and structured electronic record and recall systems (Simmons 2003).

Initial and continuing education, including postgraduate training, is essential for primary care clinicians to facilitate integrated care for patients with chronic/complex chronic disease (Nocon et al. 2004; Smith et al. 2004; Kirsh et al. 2007; Borgermans et al. 2009; Sheridan et al. 2009; Jackson et al. 2010). Patient education was also identified as a core element of several care models (Doughty et al. 2002; Pearl et al. 2003; Simmons 2003; Rea et al. 2004; Kirsh et al. 2007; Borgermans et al. 2009; Jackson et al. 2010).

Most studies had improved access of care as an objective, with considerable inter-study variation in how these were achieved and what the effects were. Patient satisfaction with models of integrated care was generally high (Nocon et al. 2004). They felt their priorities and preferences were respected (Sheridan et al. 2009); they valued the geographic convenience, easier parking and better facilities, and the ‘one-stop shop’ that improved communication and gave them better access to, and continuity of, care in a friendlier, more personal service (Doughty et al. 2002; Coast et al. 2005; Salisbury et al. 2005). Patients also appreciated the reduced waiting time to their first appointment, and were satisfied with the consultation, including its duration (Coast et al. 2005; Salisbury et al. 2005). Integrated community clinics also offered them more frequent and convenient appointments with a larger pool of well-trained GPs (Coast et al. 2005). On the negative side, patients’ concerns included a lack of confidence in the skills of GPs with a special interest (GPwSIs), perceptions that the specialist was less accessible under integrated care, and some were critical of the quality of care provided (Nocon et al. 2004).

GPs expressed satisfaction with clinics staffed by GPs (Smith et al. 2004), but expressed concerns about inadequate information regarding the purpose and function of the clinic, as well as longer waiting times and suboptimal communication with specialists (Nocon et al. 2004). Referring GPs were also concerned that unnecessary referrals could potentially de-skill them, and were fearful of having no back-up and having their patients ‘poached’ (Nocon et al. 2004).

Viable funding models are essential for continuation of a program after the pilot work has been completed. Concerns around funding related to the cost of the clinic model itself, the impact of the model on existing services, and the uncertainty of future funding. One community model delivered diabetes care at half the cost of usual hospital-based outpatient care (Jackson et al. 2010). In another study, the cost of an integrated model was reported as equivalent to traditional alternatives (Doughty et al. 2002), while others found that additional costs were balanced out by social gains (Borgermans et al. 2009). Two studies reported potentially inadequate funding for the model’s specialist resources (Sheridan et al. 2009), and about sustainability of the GPwSI service without additional funding (Smith et al. 2004). Studies of costlier integrated care models attributed this to more frequent follow-up appointments (Salisbury et al. 2005), the higher cost of community-based pathology services (Coast et al. 2005), the time required for chart audits and patient home visits (Sheridan et al. 2009) or the more intense care regimen of the community model (Simmons 2003).

This review has examined the operations and effectiveness of integrated models of care at the interface of primary and secondary care – a defined subset within the many variants of primary–secondary care models. It details the limited evidence base informing the design and implementation of this particular model of care. The defined focus on models of care that involved active negotiation of the scope of care means that the number of studies identified is smaller than the other systematic reviews previously conducted.

Our findings confirm the modest impact on clinical outcomes as well as substantial impact on process outcomes of integrated care models, and the mixed costs data pertaining to them, found with other reviews of primary–secondary care integration. Importantly, no study was found that reported worse outcomes of any integrated primary–secondary care model compared with usual care, although publication bias may be a possible explanation. It appears that the modest increased costs of such interventions will have substantial impact on service utilisation, and trying to incorporate into policy and practice may be worthwhile.

Heterogeneity in outcome measurements means that no uniform conclusions could be made about ideal model types, apart from generating principles to guide model development.

We have shown that, while models can look very different on the ground, there are six operational principles underlying them. Our list of essential characteristics overlaps substantially with the characteristics identified by Ouwens et al. (2005). The two lists share three features (Multidisciplinary patient care team, Multidisciplinary clinical pathway, and Feedback, reminders and education for professionals). Ouwens et al. (2005) also identify one patient-focussed feature (self-management support and patient education), which does not feature on our list due to this research’s focus on the interface between specialist and primary care health professionals. Our review identified two new features – a need for a secure funding model and an emphasis on ease of access for patients and subsequent acceptability by patients and general practitioners.

Our review highlights the need for a secure funding model. While this is self-evident, attention to this critical feature should be paid early in the course of trialling new methods of interdisciplinary teamwork to maximise the possibility of seamless transition from pilot phase to routine practice. Economic evaluation of clinical models that demonstrate benefits in policy and practice is critical to facilitating the adoption of new clinical models by health administrators and policymakers, and should be built in to the evaluation plan at the outset.

A focus on improving the accessibility of the service to patients by placing it in the community improved efficiency in time and costs, and satisfaction in both patients and general practitioners. It appears that a patient-focussed approach to service delivery, compared with a utilitarian approach of maximising efficiency by having the services located in hospitals where the specialists are, can reap qualitative benefits and may improve attendance rates and treatment compliance.

The review has limitations. The generalisability of our findings to care settings with significant social care elements is limited by our selection criteria. By excluding studies in paediatrics, mental health and oncology from our searches a priori, we may have excluded studies which in fact did utilise the type of GP-specialist care we wished to examine. Our paper relied on studies that were conducted at the level of health service delivery and reported outcomes directly attributable to the model of care, rather than a broader health system perspective. The review included only English language articles published between January 2000 and July 2012, and it is likely that relevant unpublished reports as well as articles published in other languages exist that address our topic.

In a society in which chronic/complex chronic disease is so prevalent, effective and sustainable integrated care models have become a priority, and more research is urgently needed to identify those models that work best. In particular, further intervention studies are needed to measure the effects of primary–secondary care models, and importantly their impact on outcomes including quality of life, user satisfaction and effective resource management.

Compared with usual care, integrated primary–secondary care has limited effect on clinical outcomes, but can significantly improve service delivery measures at a modestly increased cost. Future trials of integrated models of care could consider incorporating the design elements identified in this review, which may enhance their effectiveness.

The authors declare they have no competing interests.