Case management for at-risk elderly patients in the English integrated care pilots: observational study of staff and patient experience and secondary care utilisation.

INTRODUCTION: In 2009, the English Department of Health appointed 16 integrated care pilots which aimed to provide better integrated care. We report the quantitative results from a multi-method evaluation of six of the demonstration projects which used risk profiling tools to identify older people at risk of emergency hospital admission, combined with intensive case management for people identified as at risk. The interventions focused mainly on delivery system redesign and improved clinical information systems, two key elements of Wagner's Chronic Care Model.
METHODS: Questionnaires to staff and patients. Difference-in-differences analysis of secondary care utilisation using data on 3646 patients and 17,311 matched controls, and changes in overall secondary care utilisation.
RESULTS: Most staff thought that care for their patients had improved. More patients reported having a care plan but they found it significantly harder to see a doctor or nurse of their choice and felt less involved in decisions about their care. Case management interventions were associated with a 9% increase in emergency admissions. We found some evidence of imbalance between cases and controls which could have biased this estimate, but simulations of the possible effect of unobserved confounders showed that it was very unlikely that the sites achieved their goal of reducing emergency admissions. However, we found significant reductions of 21% and 22% in elective admissions and outpatient attendance in the six months following an intervention, and overall inpatient and outpatient costs were significantly reduced by 9% during this period. Area level analyses of whole practice populations suggested that overall outpatient attendances were significantly reduced by 5% two years after the start of the case management schemes.
CONCLUSION: Case management may result in improvements in some aspects of care and has the potential to reduce secondary care costs. However, to improve patient experience, case management approaches need to be introduced in a way which respects patients' wishes, for example the ability to see a familiar doctor or nurse.

Introduction

Healthcare systems are often ill-equipped to respond to the rapid rise in patients with multiple health problems [1-3]. Care for such people may become fragmented between different professionals and organisations, with attendant risks to quality and safety from duplication or omissions of care. This has led to widespread calls for care to be better integrated [4, 5]. Case management, a “proactive approach to care that includes case-finding, assessment, care planning and care co-ordination” [6] is a key feature of integration and is increasingly combined with use of tools to identify patients at risk of adverse outcomes [7].

In 2008, the English Department of Health invited applications from healthcare organisations offering innovative approaches to providing better integrated care following concerns that, especially for older people, care was becoming more fragmented. The proposals were intended “to achieve more personal, responsive care and better health outcomes for a local population” [8], but no blueprint was given on how integration was to be achieved. In 2009, sixteen integrated care pilots were appointed [9]. The sites took a wide range of approaches to integration which we have described in a separate report on all sixteen integrated care pilots [10]. The largest group of sites focusing on one type of intervention were the six which focused on intensive case management of elderly people at risk of emergency hospital admission. The reason for the focus on people at risk of emergency admission was because emergency admissions had been increasing and this was thought to represent a failure of care in the community as well as generating unnecessary secondary care costs. The six case management sites used a range of methods to identify older people at risk of admission including screening the elderly population with a risk profiling tool, e.g. PARR (Predicting And Reducing Readmission) or the ‘Combined Model’ [7]. People identified as at risk of admission were assigned a case manager, most often a nurse. In terms of the Chronic Care Model [3], a commonly used framework for planning quality improvement for chronic conditions, these sites focused mainly on delivery system design and improved clinical information systems, with a more limited focus on decision support and self-management support. The overarching theory behind these changes was broadly similar for all six sites, namely that better provision of primary care in the patient’s home could improve care for patients, avoid the need for specialist intervention and, in particular, avoid unscheduled or emergency admission to hospital [11, 12]. Here we report the outcome for the six case management sites, including staff reports of changes to their own work and to patient care, changes in patients’ experience, and changes in hospital utilisation and costs. A summary of the interventions is shown in the Box 1, with further details for each site in Table 1.

Box 1:

Key features of six English integrated care pilots focusing on case management of older people.

Table 1:

Details of six case management demonstration sites.

Method

Staff and patient experience

Questionnaires were sent to health and social care staff in the six sites in Summer 2010 (early in the evaluation period) and again in Spring 2011 (towards the end of the evaluation period). These included all staff closely involved in the pilot (e.g. case managers) and a random sample of staff from lists provided by the sites whose work might be impacted by the interventions (e.g. GPs, community nurses, social workers), with data analysed separately for these two groups. The survey asked about how staff thought their roles had changed from being in an ‘integrated care pilot’ and whether they thought care for their patients had improved as a result of pilot activities.

Patient questionnaires were sent out in autumn 2009 with a second questionnaire sent in autumn 2010. We only analysed responses from those patients who returned both questionnaires and who had a documented intervention which was delivered after the first questionnaire had been sent and at least two months before the second one. A sensitivity analysis was carried out on patients who reported no change in general health status between the two rounds to allow for the possibility that the natural history of decline in many of the conditions seen might influence patient responses.

We analysed both staff and patient questionnaires using conditional logistic regression allowing for clustering of respondents within sites to test for differences in responses in the two survey rounds. Further details of patient and staff selection are given in the appendix.

Secondary care utilisation: individual patient level analysis

We analysed secondary care utilisation using data from Hospital Episode Statistics [13] (HES) which include up to fifteen diagnostic and procedural codes for all outpatient attendances and inpatient admissions in National Health Service (NHS) hospitals in England. Anonymised person level identifiers in HES data were generated by the NHS Information Centre for Health and Social Care using information supplied by the sites. Then, for patients who were documented to have received an intervention, we identified up to five controls from the national HES dataset matched for age, gender, ethnicity, area-level socio-economic deprivation using the Index of Multiple Deprivation [14], hospital utilisation in the previous year, diagnoses recorded in the previous three years, and predicted risk of future hospitalisation. Patients registered at primary care practices that were part of a pilot were excluded from being controls. Across all six sites, the analyses were based on 3646 patients confirmed to have received an intervention before September 2010 and 17,311 individually matched controls. The reasons for not being able to match controls for an additional 317 patients are documented in the appendix.

A difference-in-differences analysis was conducted to compare the two groups in terms of hospital utilisation in the six months before the intervention and the six months after. This analysis was carried out for emergency admissions, elective admissions, ambulatory care sensitive admissions, and attendance at outpatients. A concern in all matched control studies is that systematic differences might exist between intervention and control groups that are unobserved and therefore cannot be balanced between groups. We did find evidence of imbalance between the groups in this study despite the comprehensive nature of the matching, and in the appendix we describe simulation analyses which we conducted to estimate the likely effect of unobserved confounders (Appendix, section 3.1.1.). We also carried out a sensitivity analysis excluding site 2 which contributed more patients than any other site: the results of this sensitivity analysis were similar to the full analysis and are not reported here.

Notional secondary care costs were estimated by applying 2008/2009 Payment by Results tariffs to HES data. Activity not covered by the tariffs was costed using National Reference Costs. If neither tariff nor National Reference Costs were available, the activity was costed as the average tariff for the specialty under which the activity was coded. Although sites provided data on the additional costs of establishing the pilots, data were not available on the additional costs of providing new services, e.g. new case managers, so we focus in this paper on secondary care costs.

Secondary care utilisation: practice level analysis

Practice level analyses were carried out for the same pre-specified groups of pilots as the individual patient analysis. We compared 117 practices (an average of 977,082 registered patients in any one year) which took part in an intervention with a random sample of half of all other practices in England. We used a longitudinal mixed effect Poisson regression model with a wide range of covariates to compare each of the two years before the pilot interventions with the two subsequent years. Whilst we could determine the exact date on which an intervention commenced for individual patients, for practices, we took the date the site entered the integrated care pilot scheme as the ‘start date’ and analysed data for the second full year of the scheme. This allowed the maximum time for interventions to have been introduced.

Staff and Patient questionnaire data were analysed using SPSS v19 and Stata v11. Patient and practice level analysis of secondary care utilisation was analysed using SAS v9.2. Further details of the methods are in the published protocol [15] and in the Appendix.

Results

Response rates

Questionnaires were sent to 276 members of staff in the first round, with response rates of 68.5% in the first round and 50.0% in the second. We did not have information to compare the characteristics of staff responding and not responding to the survey.

One thousand three hundred and eighty-five patient questionnaires were sent out in the first round with response rates in the two rounds of 65.8% and 47.7%. However, at the original time of sampling, sites could not be sure that all patients would receive an intervention and we therefore restricted our analyses to those 460 patients who returned both questionnaires and who subsequently were documented to have had an intervention delivered after the first questionnaire had been returned and at least two months before the second one was sent out. We were not able to compare the characteristics of patients responding and not responding to the survey; however, the mean age of respondents of 79.1 years (SD 8.1 years) was similar to the mean age of 79.6 (SD 11.5) for patients who were documented to have received an intervention across all sites and the gender breakdown was similar (54% female in respondents compared to 58.5% in all patients receiving an intervention).

Across all six sites, 3646 patients were confirmed to have received an intervention before September 2010 and we identified 17,311 matched controls for these.

Staff questionnaires

Staff generally reported improved team working and communication, comparing responses early on and twelve months later in the evaluation period. For example in responses to the second questionnaire from 51 respondents whose work was directly involved with their pilot, 59.1% thought that they worked more closely with other team members, 67.5% that communication had improved within their organisation, and 65.0% that communication had improved with other organisations (compared to 4.5%, 7.5% and 2.5%, respectively who reported that these had got worse). Staff directly involved in their pilot also reported that the breadth and depth of their job had increased, that they had been given greater responsibility, and that they had more interesting jobs (46.3% more interesting, 51.2% no change, 2.4% less interesting).

In responses from 138 staff members from both groups to the question in the second survey “Have you seen improvements in patient care as a result of the Integrated Care Pilot?”, 56.7% of respondents replied ‘yes’, 14.9% ‘no’, 11.9% ‘not sure’, and 16.4% that it was ‘too early to tell’. However, comparing responses to the two survey rounds, fewer staff who had direct face-to-face contact with patients ‘strongly agreed’ with the statement “I am satisfied with the quality of care I give to patients” (41.9% before, 24.2% after, odds ratio 0.35, p<0.01). Full details of staff questionnaire responses, statistical tests and responses from different staff groups are available elsewhere [16].

Patient questionnaires

Patients gave mixed responses about the care they had received. Following an intervention, they were no more likely to have had discussions with their doctor or nurse about how to deal with their health problems (87.3% in both rounds), but they were more likely to have been told that they had a care plan (30.5% vs. 22.8%, p<0.01). This increase in care plans occurred at a time when fewer patients were reporting in national surveys that they had received a care plan (data from the national GP Patient Survey shows an adjusted odds ratio or receiving a care plan of 0.95, CI 0.93–0.98 comparing surveys completed in autumn 2009 and autumn 2010). Patients in these case management sites were also more likely to report clear follow-up arrangements and to know whom to contact when discharged from hospital, and they were less likely to report having been given the wrong medicine in the preceding six months. However, they were also less likely to be able to see a doctor or nurse of their choice, felt less involved in decisions about their care, and were less likely to feel that their opinions and preferences had been taken into account (Table 2).

Table 2:

Summary of patient questionnaire responses (n=460).

*These two questions restricted to those reporting an admission in the previous six months.

We considered the possibility that these responses reflected patients’ health having deteriorated over the 12 months between survey rounds. We therefore conducted additional analyses for 319 patients who reported no change in health status between the two rounds. This showed a similar picture with a preponderance of negative changes for care from GPs and social workers, but the changes in relation to nurses were no longer statistically significant. Responses including all non-significant findings and sensitivity analyses are shown elsewhere [16].

Secondary care utilisation

Overall changes in secondary care utilisation at individual and practice levels are shown in Table 3. The individual patient level analysis shows a significant increase in emergency admissions and significant reductions in both elective admissions and outpatient attendances for intervention patients compared to controls. For the practice level analysis which includes all practice patients and not just the small proportion of registered patients who received an intervention, the only significant change is a reduction in outpatient attendance.

Table 3:

Changes in hospital utilisation for six case management.

*More detailed figures including absolute values are shown in Tables A5i and A5ii in the appendix.

The apparent increase of 9% in emergency admissions in the individual patient analysis could have been due to imperfect matching between intervention patients and controls (e.g. intervention patients being sicker) and we have some evidence that this occurred because six month mortality was greater in intervention patients than controls (8.4% intervention patients, 4.8% controls in case management sites, Appendix Table A3). We therefore simulated the effect of an unobserved confounding variable and showed that a confounder would have to be almost twice as closely correlated with the outcome as the strongest known predictor of emergency admissions in order to reverse the apparent increase in emergency admissions (Appendix, section 3.1.1.). We conclude from this that while we cannot be certain by how much the pilot interventions increased emergency admissions, it is unlikely that they reduced them.

Table A3i:

Individual patient analysis: changes in hospital utilisation comparing six months before with six months after an intervention.

In the six months following an intervention, secondary care costs for individual patients were significantly increased for emergency admissions (increased by £172 per patient, CI £44–£300, p=0.01) but reduced for elective admissions (reduced by £329 per patient, CI £234–£424, p<0.001) and outpatient attendance (reduced by £66 per patient, CI £47–£84, p<001). Combined inpatient and outpatient costs were reduced by a mean of 9% in the six months following an intervention (£223 per patient, CI £54–£391, p=0.01).

Discussion

When invited by the English Department of Health to produce innovations to integrate care more effectively, the government deliberately gave no guidance on how integration should be achieved, rather encouraging a range of diverse approaches to be developed ‘bottom up’ by those providing care. Although this produced a diverse range of interventions, a common approach adopted by pilot sites was case management of older people identified as being at risk of emergency hospital admission. In these interventions, the main integrating activities were between primary care practices and other community-based health services with a smaller number of pilots focusing on integrating primary care with secondary care or with social services. The evaluation represents one of a very small number of evaluations of case management combined with predictive risk modelling to identify patients at risk of hospital admission.

There are some clear results from the evaluation, though not all were ones that were originally intended. Staff were enthusiastic about their own involvement in the changes: their jobs became more interesting, they could observe better communication within and between organisations, and they could see patient care starting to improve. In addition, the sites documented a range of improvements in local evaluations which are not reported here, but are summarised elsewhere [10].

Patients were mixed in their views. More patients were told they had care plans and reported that their care was better organised following hospital admissions However, although patients have reported positive views about case management in previous studies [17-19], patients in this study were less likely to see a doctor or nurse of their choice and had less positive experiences of some key aspects of communication including being involved in decisions about their care. There may be a number of explanations for this, including the possibility of frail older people having to accustom themselves to new staff and the introduction of new routines of care. Most of the interventions involved appointment of new staff, so the elderly patients in this study may also have had to get used to new health professionals as well as new approaches to care. Continuity of care is important to patients, especially those with complex conditions [20], and patients included in this study appear to have experienced a reduction in continuity of care—i.e. they found it more difficult to see a familiar doctor or nurse. We speculate that the process of care planning and a more managed approach to care which was a key part of the interventions may have had the effect of ‘professionalising’ care rather than engaging patients more personally in their care. However, it is possible that patients would have settled down to the new approaches to care, and we might have found more positive results if we have been able to go back one or two years later.

Another unexpected finding was an increase in emergency admissions in the individual patient analysis relative to matched controls, especially as all six sites specifically aimed to reduce such admissions. This was despite individual staff reports of situations where emergency admissions had been avoided. Although some of the increase may have been an artefact of the matched control analysis used, we are confident from our sensitivity analyses that these sites did not achieve their aim of reducing emergency admissions beyond changes seen elsewhere. Previous studies have found some evidence that case management may reduce admissions among elderly people [21-23] although it may be more effective for a limited range of conditions such as heart failure [24, 25]. A previous evaluation of predictive risk modelling with case management in England was thought not to have reduced admissions in part because of using a case-finding model which did not identify people at sufficiently high risk of admission [26]; this led to the development of more sophisticated models such as those used in this study [7]. One possible explanation for the rise in emergency admissions may be that case management allowed pilots to be more alert to patients requiring hospital care. This would be an example of ‘supply induced demand’ where the provision of additional forms of care has the effect of identifying unmet health need—an effect which has been observed in other settings [27]. Our evaluation had no way of testing the appropriateness of the increased admissions that appear to have resulted from the case management interventions.

In contrast to the effect on emergency admissions, we found reductions in outpatient attendance and elective admissions, leading to a net reduction in combined inpatient and outpatient costs. The reduction in outpatient attendance could have been due to better coordination by community staff helping to reduce the need for unnecessary or duplicative outpatient appointments, or as part of planned changes to move services ‘closer to home’. However, the substantial reduction in elective admissions in case management sites was unexpected as this was not a key aim for any of the six sites. Over half the reduction appeared to be related to fewer admissions for people with cancer and half of these related to fewer admissions for chemotherapy even though similar proportions of patients had a recorded diagnosis of cancer in cases and controls (26.4% cases, 25.3% controls, Appendix Table A3). We are not able fully to explain this as cancer care was not a specific focus of the sites.

We found a net reduction in secondary care costs mainly because the reduction in elective admissions and outpatient attendances out-weighed the increased cost from emergency admissions. We were not provided with data to allow us to compare these costs with the additional costs of providing new services in primary care or in other sectors. Past literature does not show a consistent relationship between better coordination of care and cost reduction: a systematic review of interventions designed to improve care coordination found that they were most likely to improve health outcomes and user satisfaction (55% and 45% of studies, respectively) but costs were reduced in only 18% of studies reviewed [28, 29].

Limitations of the study

There are challenges in drawing conclusions from this study. The pilots represent a somewhat heterogeneous group of interventions, and moreover they adapted and changed during the course of the pilot period, reflecting the changing health care environment in which they were operating. Our findings may therefore reflect a ‘real-life’ deployment of case management rather than the more artificial conditions of a randomised trial, but they do not allow us to describe a single simple intervention. In the absence of randomisation, the matched control method which we used to analyse secondary care by individual patients is subject, like all observational studies, to bias from unmeasured confounders. We found some differences in outcome between cases and controls evidenced by increase mortality in the intervention group. We assumed this was due to incomplete matching between cases and controls and therefore modelled the impact of unobserved confounders. Nevertheless, we cannot completely discount the possibility that the interventions increased mortality, and note that two recent high quality randomised controlled trials of interventions designed to prevent emergency hospital admission appeared to increase mortality in the intervention groups [30, 31]. The practice level analyses in our study are more robust to unmeasured confounding variables at individual patient level, but the weakness of this analysis is that any effect of the intervention will be diluted by the inclusion of large numbers of registered patients who were not subject to any intervention. For this reason, we suggest that the observed practice level reduction in outpatient attendance may be driven by broader changes in the sites rather than directly by the individual case management interventions. We used a range of sensitivity analyses, including simulation of the effect of confounding variables to draw what we believe are conservative conclusions. Nevertheless, we cannot completely avoid the bias that is inherent in this type of study design.

The results presented in this paper show that case management may result in improvements in some aspects of care and has the potential to reduce secondary care costs. However, case management approaches need to be introduced in a way which respects patients’ wishes, for example the ability to see a familiar doctor or nurse and has the potential to produce unexpected or negative consequences. The results are also disappointing in not meeting key objectives of the sites, despite the interventions generally including many of the elements of case management identified by Ross et al. [6].

We also note that a significant minority of staff thought that a two-year pilot period was not long enough to see whether care was improving. A recent report on an experiment to integrate care found that “two years of initial development followed by one year of live working” was required to show significant change [32]. Interventions designed to integrate care need to be monitored carefully and over the long-term to fully understand their impact on both patient experience and heath outcomes.

Overall, our evaluation shows that the link between using case management to improve care integration is not guaranteed to improve outcomes, and we have no reason to think this conclusion would be different for other countries or healthcare settings. This could be because the underlying programme theory is at fault, e.g. because supply induced demand increases appropriate admissions, or because the implementation of the interventions in this study was in some way faulty, a problem which Goodwin [33], Somme and Stampa [34] argue may be the cause of some case management interventions failing to produce their expected effects. It is also possible that our demonstration sites were implementing the wrong ‘type’ of integration—for example, well publicised approaches to integrated care in the US focus on vertical integration between primary and specialist care, and this was not a prominent feature of the English models. We were not able to distinguish between these explanations in this study, though in an accompanying paper [35] we do identify barriers and facilitators to the successful delivery of integrated care in these demonstration sites and provide a ‘routemap’ as a guide to the issues which need to be addressed when attempting to improve the integration of care.

Statement of contribution made by individual authors

Martin Roland, Richard Lewis and Tom Ling were joint principal investigators on the project. They were responsible for the design and implementation of the evaluation. Adam Steventon, Gary Abel, Martin Bardsley and Xavier Chitnis were responsible for the design, data collection and matching of hospital utilisation data and for the analysis of hospital utilisation data. Annalijn Conklin and Laura Staetsky were responsible for the development, delivery and analysis of the patient and staff questionnaires and monitoring questionnaire returns from sites. John Adams was responsible for design and oversight of the statistical analysis. Laura Brereton was responsible for coordination of the study at RAND Europe including monitoring qualitative and quantitative data collection. Sarah Tunkel was responsible for coordination of the study at Ernst and Young including day-to-day liaison with sites and monitoring data collection. All authors have contributed to drafts of the paper. Martin Roland is guarantor of the paper.

Funding and disclaimer

The study was funded by the Department of Health and was carried out by RAND Europe, the University of Cambridge and Ernst and Young. The views expressed are those of the authors and not of the Department of Health and does not constitute any form of assurance, legal opinion or advice. RAND Europe, the University of Cambridge and Ernst and Young shall have no liability to any third party in respect to the contents of this article.

Ethical and governance approval

The study received ethics approval from Cambridgeshire 3 Research Ethics Committee, reference number 09/H0306/55. The National Information Governance Board confirmed that section 251 approval was not required to process patient data from Hospital Episode Statistics without consent, reference number ECC 4-15(a)/2009.

Competing interests

The funding for this research was awarded by the Department of Health through a competitive peer reviewed process under a framework contract held by Ernst and Young. The research elements of the work described were carried out by RAND Europe, the University of Cambridge and the Nuffield Trust in close collaboration with Ernst and Young. The authors have no financial relationships with any organisations that might have an interest in the submitted work, and no other relationships or activities that could appear to have influenced the submitted work other than those declared below. Richard Lewis, as an Ernst and Young partner, has declared that Ernst and Young has been paid to deliver consultancy work unconnected to integrated care in two of the pilot sites (NHS Cumbria and Northumbria Healthcare NHS Foundation Trust). Ernst and Young is a consulting firm which may at times undertake consultancy work relevant to commissioning and provision of integrated care.

Reviewers

Nick Goodwin, PhD, Senior Fellow, Health Policy, King's Fund, UK

David Perkins, PhD, Associate Professor, Director Centre for Remote Health Research Broken Hill Department of Rural Health University of Sydney, Australia

Dominique Somme, MD, PhD, Georges Pompidou European Hospital, University of Paris Descartes, Paris, France

Table A1:

Characteristics of cases and controls in six case management sites

Table A2:

Correlation of the potential omitted confounder with intervention receipt and emergency admissions that would be required to eliminate the observed relationship.

Table A3ii:

Practice based analysis: mean number of admissions per 1000 patients per year for intervention and control practices.

*ACSC=Ambulatory care sensitive condition.