Effectiveness of Case Management for 'At Risk' Patients in Primary Care: A Systematic Review and Meta-Analysis

Background An ageing population with multimorbidity is putting pressure on health systems. A popular method of managing this pressure is identification of patients in primary care ‘at-risk’ of hospitalisation, and delivering case management to improve outcomes and avoid admissions. However, the effectiveness of this model has not been subjected to rigorous quantitative synthesis. Methods and Findings We carried out a systematic review and meta-analysis of the effectiveness of case management for ‘at-risk’ patients in primary care. Six bibliographic databases were searched using terms for ‘case management’, ‘primary care’, and a methodology filter (Cochrane EPOC group). Effectiveness compared to usual care was measured across a number of relevant outcomes: Health – self-assessed health status, mortality; Cost – total cost of care, healthcare utilisation (primary and non-specialist care and secondary care separately), and; Satisfaction – patient satisfaction. We conducted secondary subgroup analyses to assess whether effectiveness was moderated by the particular model of case management, context, and study design. A total of 15,327 titles and abstracts were screened, 36 unique studies were included. Meta-analyses showed no significant differences in total cost, mortality, utilisation of primary or secondary care. A very small significant effect favouring case management was found for self-reported health status in the short-term (0.07, 95% CI 0.00 to 0.14). A small significant effect favouring case management was found for patient satisfaction in the short- (0.26, 0.16 to 0.36) and long-term (0.35, 0.04 to 0.66). Secondary subgroup analyses suggested the effectiveness of case management may be increased when delivered by a multidisciplinary team, when a social worker was involved, and when delivered in a setting rated as low in initial ‘strength’ of primary care. Conclusions This was the first meta-analytic review which examined the effects of case management on a wide range of outcomes and considered also the effects of key moderators. Current results do not support case management as an effective model, especially concerning reduction of secondary care use or total costs. We consider reasons for lack of effect and highlight key research questions for the future. Review Protocol The review protocol is available as part of the PROSPERO database (registration number: CRD42014010824).


Methods and Findings
We carried out a systematic review and meta-analysis of the effectiveness of case management for 'at-risk' patients in primary care. Six bibliographic databases were searched using terms for 'case management', 'primary care', and a methodology filter (Cochrane EPOC group). Effectiveness compared to usual care was measured across a number of relevant outcomes: Healthself-assessed health status, mortality; Costtotal cost of care, healthcare utilisation (primary and non-specialist care and secondary care separately), and; Satisfactionpatient satisfaction. We conducted secondary subgroup analyses to assess whether effectiveness was moderated by the particular model of case management, context, and study design. A total of 15,327 titles and abstracts were screened, 36 unique studies were included. Meta-analyses showed no significant differences in total cost, mortality, utilisation of primary or secondary care. A very small significant effect favouring case management was found for self-reported health status in the short-term (0.07, 95% CI 0.00 to 0.14). A small significant effect favouring case management was found for patient satisfaction in the short-(0.26, 0.16 to 0.36) and long-term (0.35, 0.04 to 0.66). Secondary subgroup analyses suggested the effectiveness of case management may be increased when delivered by a multidisciplinary team, when a social worker was involved, and when delivered in a setting rated as low in initial 'strength' of primary care.

Introduction
Many health care systems currently face significant pressures resulting from both increasing numbers of older patients with multiple long-term conditions (multimorbidity), and pressure to reduce health care budgets or provide more efficient use of current resources [1].
To relieve these pressures, many policy makers and health system planners advocate 'integrated care' [1,2].
Integrated care is a complex concept. Broadly, it is designed to "create connectivity, alignment and collaboration" [3]. A number of different methods can be used to achieve these inter-connections, and they can occur at multiple 'levels' of the health system (e.g. financing, resource management, service delivery-see Fig 1). Outcomes of effective integration of care are presumed to be better patient experience and outcomes, as well as greater efficiency [4] (i.e. patient satisfaction; health; and cost-effectiveness), therefore potentially addressing two of the major system pressures simultaneously.
A popular model of 'integrated care' at the service delivery level is 'case management' in primary care [6,7]. Case management has been defined as: "a collaborative process of assessment, planning, facilitation, care coordination, evaluation, and advocacy for options and services to meet an individual's and family's comprehensive health needs through communication and available resources to promote quality, cost-effective outcomes" [8].
Variations exist in the delivery of case management. However, there are common components [6]: • case-finding (identifying those 'at risk' who require case management, usually through prediction of high costs in the future [9,10]) • assessment of the needs of the individual patient, and care planning (individualised care plan bringing together details of patient's personal circumstances with health and social care needs, and aiming to match these needs with service provision) • care co-ordination (navigational role of case manager involving continual communication with patients, carers, professionals and services e.g. medication management, self-care support, care advocacy and negotiation; with regular review, monitoring and adaptation of the care plan) risk-identified population to be proactively case managed [16]. In the USA, a number of health insurers and health maintenance organisations offer case management to patients with longterm conditions, for example the 'Guided care' and similar programmes [6]. It is important that the provision of case management in primary care should be based on rigorous evidence. While many descriptive reviews exist examining specific types of case management in primary care (such as nurse-led case management [17]), there is no published systematic review of a range of current case management models for high risk individuals in primary care that provides a formal meta-analytic review of its effectiveness across a range of relevant outcomes. Objectives 1. To synthesise the evidence for the effectiveness of case management in primary care for 'at risk' patients 2. To explore whether the effectiveness of case management in primary care is moderated by the particular model of case management implemented, context, and study design.

Methods
The methods and results for this review are reported in line with the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines. The review protocol is available as part of the PROSPERO database (registration number: CRD42014010824).

Eligibility criteria
Studies were included in this review if they met the following criteria: • Population: Adults (18+) with long-term condition(s) (While prevalence of multimorbidity (i.e. 'complex' cases) is highest in the elderly, the absolute numbers affected are greater in those below 65 [18]) • Intervention: • Adopting methods to identify 'at-risk' patients to receive the case management, with the aim of preventing acute exacerbations of symptoms, and/or secondary care utilisation among those at higher risk • Case management, including all of the following activities: case-finding; assessment; care planning; care co-ordination; regular review, monitoring and adaptation of the care plan • Primary care/community-based management (regardless of where the case was first identified) • Comparison: usual care or no-case management • Outcome categories: Health-self-assessed health status, mortality; Cost-total cost of care, healthcare utilisation (primary and non-specialist care and secondary care separately), and; Satisfaction-patient satisfaction • Study design: Quantitative empirical research, meeting Cochrane Effective Practice and Organisation of Care (EPOC) Group study design criteria: randomised controlled trials (RCTs), non-randomised controlled trials (nRCTs), controlled before and after studies (CBA), and interrupted time series (ITS) Exclusion criteria: • Case management targeted solely at care for patients with mental health problems, although mental health conditions could be included where they were co-morbidities alongside physical long-term conditions • Hospital discharge planning (short-term management to facilitate the transition from hospital to home [19]) • Non-English language papers and grey literature

Search Strategy
Six main electronic bibliographic databases were searched for potential studies from inception until end of April 2014: MEDLINE (Ovid), EMBASE (Ovid), CINAHL, Cochrane Register of Controlled Trials (CENTRAL), Health Management Information Consortium (Ovid), and CAB Global Health (Ovid),. The search strategy used three key blocks of terms (including subject headings as well as text-words): 1) Case management2) EPOC methodology filter [20]3) Primary care filter [21]. S1 Appendix shows an example of the full search strategy for the MED-LINE database.
Results from the above searches were combined in an Endnote library, and duplicates were removed (n = 2186) prior to study selection.

Study Selection
Study selection was carried out in two stages. First, titles and abstracts of the identified studies were screened in full by the first author. A proportion of these titles and abstracts (10%) were then independently screened by a second author (kappa coefficient = 0.78). Following this initial screening, the full texts of the identified articles were retrieved, and reviewed against the inclusion/exclusion criteria. Forty percent (n = 106/266) of the full text screening was carried out by two reviewers independently. Inter-rater reliability was high (kappa coefficient = 0.81), and any disagreements (n = 7) were resolved by group discussion (resulting in 4 included, and 3 excluded). The remaining full text screening was completed by the first author alone.

Data extraction
A data extraction form was formulated using Microsoft Excel. The form was initially piloted on two randomly selected studies. The following descriptive data were extracted for included studies: • Patient: target population; total sample size (intervention/control); proportion of males; average age; average baseline number of long-term conditions; average baseline number of emergency department visits/ hospital admissions in previous year • Outcome categories: Health-self-assessed health status, mortality; Cost-total cost of care, healthcare utilisation (primary and non-specialist care and secondary care separately), and; Satisfaction-patient satisfaction • Study design: design; study duration; unit of analysis; eligibility criteria; type of control group On a separate sheet, relevant quantitative data for the meta-analysis were extracted (see quantitative analysis section below). Where adjusted and unadjusted results were both presented, the result adjusting for the most potentially confounding variables was extracted.
25 percent (n = 9 studies) of the data were extracted by two researchers working independently. The agreement was high (kappa coefficient = 0.85, across 326 data points), and the remainder of the data were extracted by the first author, and the accuracy of extraction verified by a second reviewer.

Quality Assessment
In the original protocol, we predicted having to use multiple measures of risk of bias to suit the various study types included in the eligibility criteria. However, having identified the full text articles and study designs represented, it became clear that it would be possible and preferable to use a single quality assessment tool, the EPOC risk of bias tool [32], better allowing comparison of quality across the included studies. The EPOC risk of bias tool encompasses nine standardised criteria to judge the quality of all RCTs, nRCTs, CBA and ITS studies. Each of the nine criteria is judged on a 3-point scale, corresponding to: low risk, unclear risk, and high risk. To ease comparison between studies, the total number of criteria met by each included study was also reported. Those studies at high risk of bias (fulfilling three or less criteria) were removed from the synthesis for sensitivity analysis.

Quantitative Analysis
Meta-analysis was carried out on six outcome categories related to the three main health system goals [5]. These were: Health-self-assessed health status, mortality; Cost-total cost of care, healthcare utilisation (primary and non-specialist care and secondary care separately), and; Satisfaction-patient satisfaction. Table 1 clarifies which measures were included within each of these outcome categories.
In addition to the outcomes specified in the original protocol, we also attempted to extract data related to the outcome category of 'patient safety': admissions for ambulatory care sensitive conditions [33], and polypharmacy (simple count of medications). However, none of the included studies reported these outcome measures, so results could not be synthesised.
Meta-analysis was carried out on each outcome, distinguishing between effects over the short-term (0-12 months), and longer-term (13+ months). Meta-analysis used the standardised mean difference measure, based on the mean of the case management group minus mean of the control group, divided by the pooled standard deviation [34]. When multiple measures were available for a single study within a certain outcome category, the median effect was used, as recommended in the literature [35] (e.g. for the outcome of self-assessed health status, if a measure of activities of daily living, of restricted activity days, and a measure of QALYs were all available for a given study, the effect size for each of these would be calculated, and the median standardised mean difference would represent this studies' overall effect for this outcome). We adopted Cohen's rule of thumb for interpreting effect sizes, i.e. that 0.2 indicates a small effect, 0.5 a medium, and 0.8 a large effect [36].
Heterogeneity in the outcomes was assessed using the I 2 statistic, interpreted as the percentage of total variation in the study estimates due to heterogeneity [37]. A random effects model was chosen to present the pooled effect results based on the relatively high level of heterogeneity assumed between studies evaluating a complex intervention in a variety of service contexts.
Funnel plots were performed to assess small sample bias (which may bean indicator of publication bias), but only for those outcomes drawing on 10 or more studies, as recommended [38]. Egger's test of small-study effects was additionally performed to quantify observations in the funnel plots [39].
As a complex intervention, context may be of some importance when assessing case management [40]. Subgroup analyses were performed where 10 or more studies contributed effect size data. The pre-specified variables were: • Context: strength of primary health care orientation of the health system (low versus intermediate/ high) • Type of case management: multidisciplinary team (MDT) versus single case manager; type of risk tool used (judgement versus threshold/ predictive risk modelling); inclusion of a social worker in the case management (versus absence) • Study design: RCT versus non-RCT Table 2 discusses the justifications for these choices of subgroup. Statistical significance between subgroups was judged by overlap of each subgroup's pooled effect (i.e. overlap of confidence intervals between subgroup effects indicates no significant difference) [47].
The majority of effect sizes were calculated using the Metaeasy software add-in for Microsoft Excel (version 1.0.4) [35]. The Metaeasy software allows standardisation of effect size from a variety of input parameters (dichotomous, continuous or both data types), according to eight possible methods described by the Cochrane Collaboration [48]. When multiple methods are available for a single outcome, methods are prioritised according to expected statistical precision [35]. To maximise included results, the Metaeasy effect size calculation methods were supplemented by methods developed by Lipsey & Wilson [49], with a calculator available at http:// www.campbellcollaboration.org/resources/effect_size_input.php. Effect directions were transformed so that a positive effect represented favouring case management for all outcome measures. These final effect sizes and their standard errors were then input to STATA together with relevant study information for the subgroup analysis. The final meta-analyses were then run on STATA (version 13) [50] using the metan command [51]. Funnel plots were prepared using the metafunnel command [52], and the Egger test with the metabias command [53].

Sensitivity analyses and Multiple comparisons
Two separate post-hoc sensitivity analyses were conducted in addition to the specified PROS-PERO protocol. Studies were removed from analysis if they were: 1. at high risk of bias (meeting 3 or less of the criteria for assessment of study quality) 2. set in a Veteran's health setting, where over 90% of the patients were males.
With multiple comparisons, the chances of inflating type I errors is increased [54]. We therefore used the Holm-Bonferroni adjustment [55] for multiple comparisons to identify potential false positive results.  Strength of primary care orientation: 'Case management' may be replacing some of the functions of well-co-ordinated, person-centred primary care [12]. The effects of case management may therefore be greater when it is delivered in contexts where routine primary care services are less well developed. To test this hypothesis, we stratified results by the assessed orientation to primary care of the study country's health system. The primary care orientation scores were developed by Starfield & Shi, and take into account-for each country-both characteristics of health system policy that are conducive to primary care, as well as characteristics of clinical practice [31].

Results
Multidisciplinary team versus single case manager: The hypothesis that teams are more effective than individuals at problem solving and delivering services is established across a number of diverse organisational settings [41], and teams have also been advocated in the treatment of patients with longterm conditions [42]. We tested whether case management by teams was more effective than by individuals.
Type of risk tool used: Targeting the 'correct' patients will be vital to any effective case management programme, particularly when assessed on cost and utilisation outcomes [43]. To test whether identification of the 'correct' patients was more effective when carried out by a rule-based model, we compared clinical judgement with rule-based and predictive models.
Inclusion of a social worker in case management: Collaboration between health and social services is thought to be important for effective case management [6], particularly of multimorbid patients who frequently have a complex mix of health and social care issues [44]. It also provides an additional, 'professional' level of care integration to the intervention [45], encouraging the different disciplines to work more closely together. To test the relative effectiveness of inclusion of a social worker, we therefore stratified results by this variable.

Characteristics of included studies
Unsurprisingly, since the majority aimed at an elderly population, the average age in nearly all studies was high (mean age: 75.7, range of mean age: 49.0 to 87.3). Table 3 shows the demographic characteristics of the studies. Of note, 5 (14%) provided case management to a population composed of over 90% males, carried out in veterans' settings. Inadequate information was provided across studies on baseline number of long-term conditions, and baseline utilisation of emergency and specialist services.   Table 4 summarises the potentially relevant contextual factors. Of the 36 studies included, the majority were from the USA (n = 21, 58%). When classified according to relative strength of primary care orientation, 23 studies (64%) were set in a system with low strength of primary care, and 13 (36%) in an intermediate, or high strength system. Three studies (8%) were targeted at patients with specific conditions (COPD/chronic heart failure) while the majority targeted populations more broadly on frailty, chronic illness or high utilisation (92%).
A brief qualitative description of each intervention is also provided in Table 4. Table 5 compares some of the key attributes of each intervention more directly. Many criteria highlighted as key to understanding integrated care interventions [9] were inadequately reported which limited their utility for analysis. However, the type of risk tool, whether the case management was carried out by a MDT or single case manager, and the inclusion of a social worker in the case management could be recorded for all studies. The majority of studies used a 'threshold'/ 'predictive risk modelling' risk assessment tool (n = 32, 89%), with only 4 (11%) using clinical judgement. Twenty-one studies (58%) employed MDT case management. A social worker was involved in the case management in 12 studies (33%).
The funnel plots showed a fairly even distribution of small studies, suggesting no small study bias. The one exception was for self-assessed health status, which appeared slightly skewed towards favourable results for the intervention in smaller studies. However, results of the Egger test found no statistically significant small-study effects across any of the outcomes assessed.

Subgroup analyses
The following outcome categories met the minimum criteria of 10 studies contributing to the primary analysis: mortality (short-, and long-term), self-assessed health status (short-term), utilisation of primary and non-specialist care (short-term), and utilisation of secondary care (short-, and long-term).
The results for each of the subgroup analyses are summarised in Table 7, below (the forest plots can be found in S2 Appendix).
Power to determine differences in subgroup analyses is limited, the large number of comparisons risks inflating rates of Type I error, and there may be other differences between studies that have not been taken into account in these univariate comparisons. Therefore, these results should be treated with appropriate caution. When interpreting subgroup effects, significant difference between subgroups is the important comparative factor. Importantly, no statistically significant differences were found when comparing between subgroups. However, results perhaps indicate slightly beneficial effects of delivery of case management by an MDT, with the inclusion of a social worker, and in settings with low strength of primary care. These preliminary findings may merit further investigation. Nevertheless, any significant within-subgroup effects found were extremely small by Cohen's interpretation.

Sensitivity Analysis and Multiple comparisons
Those studies at highest risk of bias reported findings in the short-term (0-12 months) for utilisation of primary and non-specialist care and utilisation of secondary care [75,86]. Studies  using Veteran participants [67,76,77,93,102], with over 90% males, reported findings in all outcomes and time-periods assessed.
After adjusting for multiple comparisons, excluding these studies showed no significant difference from the results reported above, either for the primary analysis, or between subgroup differences for the subgroup analyses. The results of the sensitivity analysis can be found in S3 Appendix.
After Holm-Bonferroni correction was applied to all results, only two of the statistically significant results held: the finding of a significant effect on patient satisfaction in the short-term (0-12 months) in the primary analysis, and the same outcome measure in the sensitivity analysis (excluding studies with Veteran participants).

Strengths and limitations
Strengths of this study include the use of PRISMA guidelines, pre-specification of subgroups, as well as the broad search strategy. Unfortunately, the broad search impaired our ability to double-screen all studies at every stage, although we did double-screen a proportion at every stage, and our inter-rater reliability was consistently good. We did not include grey literature, due to the generally lower quality of this literature [110]. We found no evidence of small study bias in our included studies. Assessing complex service-level interventions is difficult, and RCTs may be particularly problematic in the context of patients with multimorbidity [111]. We included the range of intervention study types considered by 'The Cochrane Effective Practice and Organisation of Care (EPOC) Group'.
We view the use of meta-analysis as a major strength of this piece of work, which differentiates this review from the narrative syntheses [17, 22-24, 26-28, 30]. Some argue that metaanalysis of complex service-level interventions is inappropriate, because the effects of the intervention are so dependent on context [40], and pooling the results from different contexts is not advisable. However, as shown in the introduction, case management can be defined in terms of a number of common components. In addition, we did try to account for context differences (such as strength of the primary care system), although the precise scope of the term is unclear [112], and a lack of consistent reporting limited what was possible.
Heterogeneity was high for measures of utilisation of primary and non-specialist care in both time-periods assessed, and patient satisfaction in the long-term. This high level of heterogeneity is expected in analysis of a complex intervention, which is possibly highly dependent on context. On the whole, choosing a random effects model took into account expected heterogeneity arising from comparison of a complex intervention across different settings [113]. Nevertheless, caution must be applied to uncritical interpretation of the pooled effect, due to the level of unexplained variation observed.
When we adjusted for multiple comparisons, only increased patient satisfaction in the short-term remained significant. This type of adjustment, while it reduces the risk of false positive findings (type I error), does so at the risk of inflating the number of false negative findings (type II error) [114]. As an intervention with low risk of harm to the patient, we have chosen to present the unadjusted results as the primary analyses, with the results adjusted for multiple comparisons suggesting additional caution in interpretation.
The outcome measures we chose were broadly inclusive. For example, in self-assessed health status we included activities of daily living, as well as bed days, and more typical 'health' measures, for instance QALYs. This could be a potential weakness of this study. However, we chose these broad outcome categories attempting to synthesise as much of the relevant data as possible that were reported within the selected studies. Furthermore, these measures were reported as functional outcome measures of health in the individual studies, and were therefore synthesised as such.
Utilisation and cost outcomes have a tendency to be skewed. As expected, the studies we synthesised reporting these outcomes demonstrated significant skew (i.e. the mean is smaller than twice the standard deviation), indicating that the mean reported is not a good indicator of the centre of the distribution [115]. Future primary studies should make sure these skews are reported, and that the effects of any subsequent log transformation are detailed for more precise synthesis of these outcomes. Furthermore, although costs were detailed in a number of studies, we identified only one cost-effectiveness analysis [82], and one cost-benefit analysis [84].

Interpretation of the results in the context of other studies
It is difficult to directly compare, as most previous reviews on this subject have used narrative synthesis methods [17,23,24], or used 'vote counting' to quantify the number of studies with statistically significant results in either direction [22,[26][27][28]30]. The majority of existing reviews conclude that despite theoretical benefits, in practice there is only slight evidence of benefits [22,23], particularly related to patient satisfaction [24,27], and functional health [26]. The single previous systematic review we identified which employed meta-analysis, additionally included hospital discharge planning interventions (identifying a total of eleven studies, only six of which-three in the primary care setting-were included in meta-analysis), and only used meta-analysis for a single outcome category, 'unplanned hospital admissions', similarly finding no significant effect [25]. Our results are in line with those previous reviews, with the additional benefits of updating the evidence base, quantifying the impacts (emphasising the small benefits) across a range of outcome categories, and exploring contextual variations.
Most published reviews focused on implementation have similarly identified inadequate reporting of the methods of applying case management in practice as a major limitation of the literature [17].

Implications for research
Case management has potential to impact on patient safety issues in primary care, such as coordination and communication between professionals and levels of the health system [116]. No safety outcomes were identified in the included literature, and primary care patient safety is a notoriously under-researched area [117].
As multimorbid patients are likely at most risk for co-ordination failures, and therefore potentially have most to gain from the integration of care, measures of multimorbidity must be more consistently reported [118], even if this is a simple count of mean number of chronic diseases. Ideally, however, this would give more detailed breakdown by disease type/cluster [119] as a subgroup analysis, enabling further targeting of specific interventions to specific groups of patients who are most likely to benefit [120]. Additionally, there is some evidence that the coexistence of physical and mental health problems could lead to increased management difficulties [121]. Comorbidity of conditions should be better reported in evaluations and explored in further research.
Current evidence comes from a majority of high-income, Western settings. This potential bias requires addressing with evidence from other settings, for example Asia, where the case management approach is currently evolving.

Implications for policy and practice
Given the lack of significant effects across the majority of outcome categories, should case management generally be encouraged or incentivised for the treatment of 'at-risk' patients in primary care? This review would suggest that, as currently delivered, case management should not be regarded as a primary means of reducing overall health service utilisation and that it will not reduce costs or improve health outcomes. While we have shown some statistically significant benefits, these are not focused on primary outcomes, with the largest overall effect on satisfaction, which did not meet the usual criterion for a 'medium' effect [36]. However, the current results rest on the evidence accumulated from RCTs. There are potential problems associated with this study design in the assessment of complex interventions and conditions [111], although other designs which may be better able to reflect routine delivery of case management (such as controlled before and after designs [80]) have their own problems with internal validity.
Evidence from the subgroup analyses do perhaps point to more effective ways of delivering the intervention, namely: delivery by a MDT as opposed to a single case manager, and the inclusion of a social worker. These findings agree with the wider literature which advocates the use of a multidisciplinary team to successfully manage patients with chronic disease [42], and advocates better integration of health and social care [45]. Case management may be more effective in a system where the strength of primary health care orientation is low. However, these subgroup results should be interpreted with caution, as they are exploratory univariate analyses, which should be investigated further while controlling for potential confounding factors before firm conclusions are drawn. Furthermore, the significance of these effects did not withstand adjustment for multiple comparisons.
Further understanding of factors driving the effectiveness of case management may benefit from on-going evaluation of implementation at the local level. It is important that components of implementation are reported consistently and in detail, so that these can be included in future systematic reviews and effectiveness of individual elements of the intervention can be examined.

Conclusions
Current evidence suggests case management of 'at-risk' patients in primary care is not effective beyond small improvements in patient satisfaction. Case management should not be regarded as a proven technology in the delivery of integrated care, there remains a need for further enhancement and evaluation of its effectiveness, particularly with study designs which better incorporate context, and in lower income settings. More research is needed into more effective methods of delivery (e.g. by an MDT and including a social worker), and implementation (e.g. in a health system with poor primary care orientation), which may additionally improve effectiveness. Even with these improvements, however, case management may never be as effective as it needs to be to deliver major savings through a focus on high risk groups [122]. This highlights the need for a variety of models to deal with system pressures, including integrated care at different levels of the health care system, and with more of focus on the wider population of patients [123].