Hospital at Home admission avoidance with comprehensive geriatric assessment to maintain living at home for people aged 65 years and over: a RCT

Article history

The research reported in this issue of the journal was funded by the HSDR programme or one of its preceding programmes as project number 12/209/66. The contractual start date was in July 2014. The final report began editorial review in February 2020 and was accepted for publication in October 2020. The authors have been wholly responsible for all data collection, analysis and interpretation, and for writing up their work. The HSDR editors and production house have tried to ensure the accuracy of the authors’ report and would like to thank the reviewers for their constructive comments on the final report document. However, they do not accept liability for damages or losses arising from material published in this report.

Copyright statement

Copyright © 2022 Shepperd et al. This work was produced by Shepperd et al. under the terms of a commissioning contract issued by the Secretary of State for Health and Social Care. This is an Open Access publication distributed under the terms of the Creative Commons Attribution CC BY 4.0 licence, which permits unrestricted use, distribution, reproduction and adaption in any medium and for any purpose provided that it is properly attributed. See: https://creativecommons.org/licenses/by/4.0/. For attribution the title, original author(s), the publication source – NIHR Journals Library, and the DOI of the publication must be cited.

2022 Shepperd et al.

Evidence

In recent decades, the focus of health care for older people has been on comprehensive geriatric assessment (CGA), defined as a multidisciplinary process to determine the older person’s medical, functional, psychological and social needs that leads to a co-ordinated plan for the delivery of health care. Organising acute hospital care for older people along these lines increases the likelihood that patients will be living in their own homes after 3 and 12 months’ follow-up, and CGA is now viewed as the gold standard for hospital-based health care delivered to older people. 2

It is possible that implementing CGA in an older person’s home, instead of in an acute hospital setting, will lead to a greater improvement in health outcomes at a lower cost. Despite service innovations seeking to expand the application of CGA as hospitals deal with the rise in emergency admissions, it is not known how CGA works in community settings that seek to provide an alternative to admission to hospital [e.g. admission avoidance hospital at home (HAH)]. Furthermore, the evidence on cost is uncertain, and older people’s and family caregivers’ participation and their contribution to the processes of CGA for their longer-term health-care needs have not been explored. Prior to this randomised trial, the main source of evidence regarding the effectiveness and cost-effectiveness of admission avoidance HAH was a meta-analysis published in a Cochrane review. 3 However, because of the small number of small trials included in this meta-analysis, the evidence of effectiveness and cost-effectiveness was uncertain. 3 Key questions to be answered include ‘Is it clinically effective and cost-effective to deliver CGA in an admission avoidance HAH setting (as opposed to delivering CGA in an inpatient setting)? and ‘Does this have an impact on older people and their caregivers?’.

Hypothesis

Consistent with the concept of healthy ageing, we hypothesised that older people who received geriatrician-led admission avoidance HAH with CGA might experience less of a decline in functional and cognitive capacity and maintain a level of independence that is difficult to achieve in a more restricted hospital environment.

Our aim was to conduct a robust evaluation, in the form of a multisite pragmatic randomised trial and process evaluation, of admission avoidance HAH services with CGA compared with admission to hospital, delivered mostly in specialised elderly care services, for which there is considerable evidence of effectiveness. 2 We report the clinical and health economic outcomes. In addition, we report the findings of a process evaluation that assessed the experiences of patient and caregivers of receiving health care in each setting, and the components and practices of organising HAH compared with those of bed-based hospital care.

Setting

Participants were recruited mainly from primary care referrals to admission avoidance HAH with CGA at the Aneurin Bevan University Health Board or a hospital-based acute assessment unit in Bradford Royal Infirmary, Bradford Teaching Hospitals NHS Foundation Trust, Guy’s and St Thomas’ NHS Foundation Trust, the Royal Devon and Exeter NHS Foundation Trust, University Hospital Monklands, St John’s Hospital, Livingston, Victoria Hospital, Kirkcaldy, Southern Health and Social Care Trust or Belfast Health and Social Care Trust. Five of the sites were based in mainly urban areas and two in a semirural area, one site covered an urban and rural area and one a mainly rural area.

Eligibility criteria

We recruited older people with frailty who required an urgent hospital admission because of an acute change in their health, such as a sudden functional deterioration, delirium or a fall, against a background of complex comorbidity.

We recruited patients who were (1) aged ≥ 65 years; (2) willing and able to give informed consent to participate in the study, or who had a relative, a friend or an independent mental capacity advocate who was involved in making a decision in the best interests of the individual if that person did not have capacity to give consent; and (3) referred to the geriatrician-led admission avoidance HAH with CGA and would otherwise require hospital admission for an acute medical event. The presence of a carer depended on the patient’s individual circumstances and was at the discretion of the clinician responsible for the patient, in accordance with current clinical practice at each site. Participants were excluded if they (1) had an acute coronary syndrome (this included myocardial infarction and unstable angina, characterised by cardiac chest pain and associated with electrocardiographic changes), (2) required an acute surgical assessment, (3) had a suspected stroke, (4) were receiving end-of-life care as part of a palliative care pathway, (5) refused HAH or were considered by clinical staff to be too high risk for home-based care (e.g. those who were physiologically unstable or at risk to themselves, or if the carer reported that home-based health care was not acceptable) or (6) were living in a residential or nursing care home setting.

Interventions

The intervention was geriatrician-led multidisciplinary admission avoidance HAH with CGA (otherwise known as hospital in the home) as an alternative to admission to hospital (also known as hospital in the home).

At the outset, during the design of the randomised trial, we established four core elements of HAH that had to be present for a site to be eligible to recruit participants:

1. geriatrician-led admission avoidance HAH

2. a multidisciplinary team (MDT)

3. health-care provision guided by the principles of CGA, which include multidisciplinary meetings and virtual ward rounds

4. direct access to elements of acute hospital-based health care, such as diagnostics and transfer to a hospital if required.

These components were considered essential to the delivery of the core function of a service that provides an alternative to inpatient hospital health care for older people, and differentiate admission avoidance HAH from the range of other services that operate across primary and secondary care. 7 The attending geriatrician had clinical responsibility in all but one site (where a primary care physician and senior nurse had clinical responsibility) and was responsible for discharging patients from the service. The MDT included nurse practitioners, who were responsible for clinical assessments, arranging investigations, documentation, discharge summaries and prescribing, physiotherapists and occupational therapists. Access was also provided to social care and mental health nurses and old-age psychiatrists. Virtual rounds were held at least daily. An existing primary care out-of-hours service provided out-of-hours health care. Between HAH visits, patients could communicate with the HAH team by telephone.

The MDT implemented treatment and management recommendations and, if required, referred patients to other services (e.g. older people’s mental health services, diagnostic services, social work, dietetics, speech and language therapy, pharmacy support and outpatient follow-up). Patients had access as usual to hospital inpatient care, general practitioners (GPs) and the primary health-care team. Intravenous drug administration and oxygen therapy were available in five sites, three sites provided intravenous administration but not oxygen therapy and one site did not provide either. Health care was provided 7 days per week, from 09.00 to early evening, admissions were restricted to Monday to Friday in all but one site, and emergency medical cover was available via the usual emergency services 24 hours per day.

At the outset, and following consultation with site principal investigators, it was anticipated that approximately 80% of those allocated to the control group (i.e. hospital) would receive their care from a geriatrician-led elderly care medicine service with CGA either in a dedicated ward or by means of input to a care plan on another type of hospital ward.

Outcome measures

Serious adverse event and adverse event reporting

We identified the following potential risks to participants: a fall (in either setting), hospital- or community-acquired infection, hospital admission (for those randomised to HAH), post-discharge hospitalisation in either group and death. We categorised an adverse event as serious if it resulted in death, was life-threatening, necessitated hospitalisation or the prolongation of existing hospitalisation, resulted in persistent or significant disability or incapacity, or was an otherwise important medical event. These categories corresponded with the expected events among this population, which include falls, pressure sores, hospital- or community-acquired infection and transfer to hospital. All serious adverse events that were related to the administration of any of the research procedures, were unexpected and were observed by the recruiting clinician or reported by the participant were recorded on the case report form (CRF) and were forwarded by the site to the trial manager after the site clinician had assessed the severity. As a minimum, the following details were recorded: description, date of onset, end date, assessment of relatedness to the intervention, other attributions/co-interventions, and action taken. The chief investigator reported serious adverse events that, in the opinion of one of the clinical leads, were ‘related’ and ‘unexpected’ in relation to the study to the Research Ethics Committee (REC) within 15 working days of becoming aware of the event.

Recruitment

We implemented a recruitment pathway that mapped to existing arrangements for referral to HAH. Eligible participants were identified from those patients who were referred by their GP to a single point of access, or who were transferred from the emergency department to an acute assessment unit and were assessed as suitable for HAH. At referral to the trial, each participant was provided with a participant information leaflet that described the research, and they were also given an opportunity to ask any questions and discuss any concerns about the research with a research nurse. Each participant had the right to withdraw from the study at any time, and reasons for withdrawal were recorded in the CRF.

Randomisation procedure and concealment of allocation

The unit of randomisation was the individual participant. We used a 2 : 1 randomisation ratio (i.e. HAH to hospital inpatient admission). We opted for this ratio to address the concern expressed by clinical leads that a 1 : 1 randomisation ratio would place unmanageable pressure on inpatient services. Randomisation was conducted by a local member of the research team using Sortition, the Oxford University’s Primary Care Clinical Trials Unit’s validated in-house online randomisation system. Telephone randomisation was used if sites did not have online access. A computer-generated randomisation sequence was used and randomisation was stratified by site, gender and known cognitive decline [measured using the Informant Questionnaire on Cognitive Decline in the Elderly (IQCODE)]. 13

Data collection

Research nurses at each site collected data on the primary outcomes from participants (and from their caregiver, if the caregiver was the designated consultee) at baseline and at 6 and 12 months, with the exception of an assessment of delirium, which was carried out 3 days, 5 days and 1 month after recruitment (Figure 1). At each site a form was completed to record whether or not death had occurred, as well as the date these data were collected from the medical records. Place of residence was recorded by the research nurses at each follow-up visit. Data were collected using a paper form or were directly entered on an electronic pro forma on OpenClinica Enterprise V.3.5 Data Management System (Waltham, MA, USA).

FIGURE 1.

Data collection during the study. a, Exact timelines varied depending on the availability of staff and interviewees.

At baseline we collected data from the patient’s clinical notes and from the clinical lead on the presenting problem that required admission to hospital, as well as demographic information (including age and education). We also collected data on the patient’s background cognitive status (using the 16-item informant-based IQCODE questionnaire),13 incident and persistent delirium (measured using the CAM8), comorbidity (measured with the Charlson Comorbidity Index14), activities of daily living (measured with the Barthel Index10), current cognitive impairment (measured with the MoCA9), health status (measured with the EQ-5D-5L11) and major health service use (see Economic analysis) during the 6 months prior to the patient’s current illness. If a participant appeared to be burdened, we collected data in two stages. We collected core data (i.e. IQCODE13 and CAM8 scores) after obtaining consent and prior to randomisation, and administered the remaining measures [i.e. the Barthel Index,10 Charlson Comorbidity Index,14 MoCA,9 EQ-5D-5L11 and the Health Resource Use Questionnaire (HRU)] soon after randomisation.

At the 6-month follow-up point, we collected data from all patients on mortality, new long-term residential care, cognitive impairment, activities of daily living, quality of life, length of stay, readmission or transfer to hospital, admission to hospital or HAH, health resource use, residential care and informal care (see Economic analysis). Twelve-month follow-up data on living at home were collected from the medical records and/or place of assessment.

Data management

We stored all paper and electronic data in a secure environment and referred to participants only by their trial number. Participants’ names appeared only on the signed consent forms, which were stored securely at each site. We followed the standard operating procedures of the University of Oxford Primary Care Clinical Trials Unit (Oxford, UK), which are compliant with the Data Protection Act15 and good clinical practice. No one outside the study team had access to either the CRFs or the database. Members of site research teams accessed identifiable data so that they could collect follow-up data. Direct access was granted to authorised representatives from the sponsor and host institution for monitoring and/or audit of the study to ensure compliance with regulations. Staff at each site entered data directly into OpenClinica or on paper versions of the CRF and questionnaires. Data from paper forms were double entered by Primary Care Clinical Trials Unit staff. Researchers from each site were asked to check the data for completeness before returning the forms to the trial team. In the case of those sites entering data directly into the database, internal validation checks were run for each data field. Any inconsistencies or missing data were highlighted as queries to be resolved by the site. The same checks were applied to paper CRFs by the data manager after data entry. Data queries were returned to sites for resolution on a regular basis.

Study oversight

The overall supervision of the trial progress was carried out by the Trial Steering Committee. An independent Data Monitoring Committee (DMC) met every 6 months to review trial progress and unblinded data, including all serious adverse events reports. The sponsor was the University of Oxford.

Statistical analysis

Data from previous trials3 and a before-and-after study of older people who received health care in an acute care of elderly unit in Scotland,16 with a length of follow-up that ranged from 6 to 12 months, informed our sample size calculation. Our proposed study effect estimate was based on a hospital event rate of 50%, with a 10 percentage point reduction in a residential setting, to 40% in the HAH group, equal to a relative risk (RR) of 0.80, which is towards the top end of the 95% confidence interval (CI) for a pooled estimate for mortality. Initially, we calculated that the sample size required to provide 90% power and based on 15% attrition for the primary outcome at 12 months would be 1552 patients. At the fifth meeting of the DMC the decision was made to amend the follow-up time for the primary outcome to 6 months, as it was agreed that it was more likely that any effect would be detected prior to 12 months in the population recruited. When we observed a lower attrition of 6%, we revised the sample size to 1055 patients, reduced the power to 83% and reduced the significance level (two-sided) for the primary outcome to 0.05 to reflect a reduced rate of recruitment towards the end of the trial. This was agreed by the Trial Steering Committee and the DMC, which had oversight of the study, and was approved by the National Institute for Health Research (NIHR). It was also agreed to reduce data collection at 12 months for the primary outcome because of concern about the burden of data completion on the study population, who were old, and because changes in the secondary outcomes were also more likely to occur at 6 months.

The original plan was to analyse data collected at both 6 and 12 months in the same model. However, it was later decided to perform separate analyses for 6- and 12-month data, based on the view that 6-month outcome data could be reported before the completion of 12 months’ follow-up. However, the 6-month analysis was delayed because of the data cleaning process. For this reason, we carried out the analysis in accordance with the statistical analysis plan, and the full model, which incorporated both time points, was analysed in a sensitivity analysis.

The primary analysis population was defined as all participants for whom data were available, and according to the group to which participants were randomly allocated regardless of deviation from protocol. For the primary outcome of living at home at 6 months, and other binary outcomes (i.e. long-term residential care, mortality, readmission or transfer to hospital, cognitive impairment and delirium), we used a generalised linear mixed-effects model (robust Poisson model with log-link function) with unstructured covariance matrix. A linear mixed-effects model was used for activities of daily living, measured by the Barthel Index. The models adjusted for intervention arm, gender and IQCODE score as fixed effects, and site as a random effect. The models for cognitive impairment, delirium and activities of daily living were also adjusted for the corresponding baseline score as a fixed effect. 17 Individual logistic regressions were performed for each baseline covariate to obtain the p-value for the association of missingness with the primary outcome. Missing IQCODE scores at baseline (n = 10) were imputed using the mean IQCODE at baseline. A fully inclusive multiple imputation was conducted with gender, age, education, place of baseline assessment, whether or not consent was signed by ‘consultee’ and other factors expected to be related to the main outcome as covariates, and the primary outcome reanalysed. Analysis was carried out using Stata SE^® version 16.0 (StataCorp LP, College Station, TX, USA).

We planned one subgroup analysis of the effect of setting (home vs. hospital) on the incidence of delirium18 in people who were cognitively impaired (defined as having a MoCA score of < 26). 19 Owing to the small number of participants with delirium, assessed by the CAM,18 six individual log-Poisson generalised linear mixed models with robust standard errors were fitted to the data, one at each of the three time points for one subgroup with a MoCA score of < 26 and another subgroup with a MoCA score of ≥ 26. The models included site as a random effect.

Sensitivity analyses

Four pre-planned sensitivity analyses were conducted with respect to living at home to explore the sensitivity of the results to different assumptions: (1) missing data for long-term residential care and/or death status were replaced with either living at home and/or alive, or not living at home and/or dead; (2) missing data were imputed using multiple imputation; (3) the model was adjusted for factors that predicted that data were missing (e.g. education level, place of assessment, presenting problem) for the outcome of living at home; and (4) both the 6- and 12-month outcomes of living at home were analysed in the same model. The model included an additional fixed effect for the interaction between intervention arm and time point so that possible differences in treatment effect could be assessed at each time point.

Economic analysis

Parts of this section have been reproduced from Singh et al. 20 © The Author(s) 2021. Published by Oxford University Press on behalf of the British Geriatrics Society. All rights reserved. For permissions, please email: journals.permissions@oup.com. This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (https://creativecommons.org/licenses/by-nc/4.0/), which permits non-commercial re-use, distribution, and reproduction in any medium, provided the original work is properly cited. For commercial re-use, please contact journals.permissions@oup.com.

We estimated the resource use, costs and health outcomes in each trial arm up to the 6-month follow-up point on an intention-to-treat basis. Costs were estimated taking the NHS and Personal Social Services (PSS) perspective, as well as the wider societal perspective, which also included the cost of informal care. Following the National Institute for Health and Care Excellence’s (NICE’s) recent recommendation, we converted EQ-5D-5L responses at baseline and 6 months to utilities using a crosswalk algorithm developed by EuroQol. 21 Quality-adjusted life-years (QALYs) were then calculated by using the exact days between the two EQ-5D-5L measurement points to estimate the area under the curve per patient. As the follow-up period was 6 months, the maximum within-trial QALY gain per patient was 0.5. We also measured health outcomes in terms of life-years living at home (LYLAHs), a proxy measure of independence and well-being in an older population, which we had used previously in an incremental cost-effectiveness analysis of individual patient data from the randomised trials that contributed to a Cochrane systematic review of the effectiveness of CGA. 2 LYLAHs were calculated by subtracting the sum of the total nights in hospital and/or in residential care over 6 months from the follow-up time (183 days or 6 months) if a participant was alive, or from randomisation to death if the patient died during follow-up.

Type and source of resource use

We collected information on health and social service resources used, and on the number of hours of informal care (i.e. unpaid help) that participants received from family and friends as a result of their health problems (Table 1). Site research nurses completed the HRU and CRF by obtaining details from participants’ medical records, and from participants and their caregivers. We also checked adverse event trial records for data on hospital and residential care admissions. Most resource use data were derived from the HRU, with the exception of data on length of stay in residential care, hospital and HAH length of stay, and readmissions or transfers to hospital, which were obtained from the medical records and entered on the CRFs or a data query form. Extreme values of all data were checked against data sources. We replaced missing values for resource use with zero if individuals had filled out any question in the HRU, assuming that the missing response meant zero use. If a patient did not have a single response on the HRU, the missing responses were treated as such.

Intervention costs in the analysis included the initial length of stay in hospital or HAH after randomisation, plus the length of stay associated with the initial assessment if the participant had been recruited from a hospital assessment unit. When resource use was collected daily or weekly over a period of 6 months, we multiplied by 183 or 26, respectively.

Valuation of resource use

Unit costs were obtained from secondary sources for each health and social care service. 18,22–25 Appendix 2 provides a list of the unit costs and their sources. Unit costs were inflated to 2017/18 prices, where necessary, using the hospital care and health services inflation index. The unit costs of hospital inpatient care were calculated by using the weighted average of all elective and non-elective hospital admissions relevant to the trial population (e.g. admissions to neonatal units were excluded), obtained from NHS Reference Costs 2017–2018. 24 Non-elective admissions were divided in to short and long stays using the length of stay per participant available in the trial data. Respite at home was not costed, as 99.5% of patients reported zero use of this service at baseline and 6 months, and we could not identify a reliable unit cost for this type of service. Volunteer work and NHS 24 (in Scotland) unit costs were not costed. Sitting service costs were set at a notional £5, as these services either are free or have a small charge. There is no defined unit cost for luncheon clubs, but a number of community centres have stated costs ranging from £2 to £6 and so we allocated an average cost of £4 per attendance. 17,26,27 The amount of health or social service used by each patient was then multiplied by the relevant unit cost.

The cost per bed-day of admission to HAH was calculated by dividing each site’s annual total spent budget in 2017/18 for HAH by the total number of bed-days (i.e. number of patients multiplied by the average length of stay per patient) in the same year.

Cost perspective

Following NICE guidelines for health technology appraisal, we estimated the costs per participant from an NHS and PSS perspective, which included costs of HAH, primary and community care, outpatient visits, hospitalisation, ambulance transportation and PSS (e.g. home care and meals on wheels). 28 Results from a societal perspective, in which the productivity cost of unpaid caregivers was added to all other costs, were reported separately.

Cost-effectiveness analysis

As the time horizon for the cost-effectiveness analysis was from baseline to 6 months, discounting of costs and outcomes was not applied. We performed an incremental analysis to assess the differences in mean costs, mean QALYs and mean LYLAHs between the two treatment groups, and estimated the incremental cost-effectiveness ratio (ICER) in terms of cost per QALY and cost per LYLAH using:

For the main analysis, we used the observed data to estimate differences in means and conducted t-tests to test for statistically significant differences. The main economic analysis is based on complete cases. The ICERs were reported from an NHS and PSS perspective, and separately from a societal perspective. Uncertainty associated with the ICER was assessed using non-parametric bootstrapping with replacement. For each of the 5000 bootstrapped samples, we estimated the means and mean between-group differences (with 95% CIs using the percentile method) in costs, QALYs and LYLAHs, and plotted ICERs on cost-effectiveness planes. Cost-effectiveness acceptability curves were displayed to show the probability that HAH was cost-effective at different levels of willingness to pay for a QALY and LYLAH gained.

Sensitivity analysis

We conducted two sensitivity analyses. First, we used linear mixed-effects regression models fitted on complete cases to estimate the differences in mean costs, QALYs and LYLAHs between the two treatment groups, after adjusting for baseline gender, known cognitive decline, baseline utilities and pre-randomisation costs as fixed effects, and site as a random effect. Pre-randomisation costs were included in the regression to estimate the differences in mean costs, and baseline utilities were included in the regression to estimate differences in mean QALYs. 29 Bootstrapped ICERs were then produced using the same method as in the main analysis. Second, we performed multiple imputation to address the impact of missing data for costs, EQ-5D-5L utilities and LYLAHs on the estimated ICERs. We first imputed missing utilities and costs at baseline using an unconditional mean estimate, as performing multiple imputation by treatment group can increase the covariate imbalance between control and treatment groups at baseline. 30,31 We also used imputed hospital length of stay at baseline using an unconditional mean estimate for the purpose of imputing LYLAHs. We then performed multiple imputation using chained equations to impute missing observations in utilities, costs at 6 months and LYLAHs based on utilities, costs and hospital length of stay at baseline, respectively, as well as gender and age. The multiple imputation process was partitioned by treatment group and 20 imputed data sets were generated, following standard practice, which suggests generating a number of imputed data sets equal to the percentage of missingness. 32 The imputed data sets were then used in the bootstrapping process similar to the main analysis.

Process evaluation

We followed Medical Research Council guidance in designing the process evaluation. 33,34 Our intention was to undertake the process evaluation with sufficient flexibility to allow us to identify and address additional questions that arose during the study, recognising that local and broader aspects of these domains were likely to become more salient as the research developed. 35,36 We aimed to identify the factors that facilitated the implementation of HAH, how the risk of functional and cognitive decline was managed in a home setting, how ongoing coping and social support was maintained, and how these factors compared with acute health care delivered in hospital.

The objectives of the process evaluation were to:

• explore the components, practices and experiences of HAH and hospital inpatient settings

• assess the contextual factors in terms of the organisational features and local policy across the sites and how these might affect implementation of the intervention and the randomised trial

• identify unanticipated consequences and aspects of the trial that were not necessarily captured quantitatively.

Methodology

Our qualitative research methodology was driven by the aim of understanding participants’ perspectives and the meanings they attached to their experiences. We combined periods of fieldwork at sites with phenomenological inquiry through interviews, using open-ended questions to yield narrative data and capture the complexity of views. Figure 2 illustrates how the process evaluation was conducted alongside the randomised trial.

FIGURE 2.

The timeline for the process evaluation conducted alongside the randomised trial. a, Exact timelines varied according to the availability of staff and interviewees.

Methods

Parts of this section have been reproduced from Mäkalä et al. 50 © The Author(s) 2020. Published by Oxford University Press on behalf of the British Geriatrics Society. This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (http://creativecommons.org/licenses/by-nc/4.0/), which permits non-commercial re-use, distribution, and reproduction in any medium, provided the original work is properly cited. For commercial re-use, please contact journals.permissions@oup.com.

We followed the Medical Research Council guidance for process evaluations. 34 We collected data on key process variables from all nine sites throughout the trial by site visits, telephone calls with the site principal investigators, conference calls with the research nurse or co-ordinator from all sites [which were organised by the trial manager (ACB)] and e-mail correspondence to obtain a detailed understanding of how the services operated in each site.

The process evaluation was located in three sites that recruited participants to the randomised trial, two NHS trusts in England and one NHS trust in Scotland. We were limited to conducting the qualitative study at three sites because funding for the process evaluation was limited to 18 months. The three selected sites (Lanarkshire, Bradford and London) represented different geographical areas, service compositions, populations and organisational arrangements that might influence the delivery and effectiveness of the intervention.

We used the five domains of the Consolidated Framework for Implementation Research:37,38 (1) outer setting (i.e. factors external to the organisation), (2) inner setting (i.e. characteristics of the organisation), (3) intervention characteristics, (4) characteristics of the individuals involved in the intervention and (5) processes of implementation (Table 2). In addition, we used a logic model of CGA delivered in hospital (Figure 3) as an initial guide to setting out the relationship between resources, activities and intended results. We used normalisation process theory (NPT) to guide our analysis and interpretation of the data from interviews with older people and carers. 39,40

FIGURE 3.

Adapted logic model of CGA. ADL, activities of daily living.

Data generation

Components of the process evaluation included document reviews of websites, operational plans, patient information leaflets, service evaluations, audit reports and presentations by the services, non-participant observations of the HAH and hospital services (including MDT meetings, discussions with staff who delivered the services and the research team) and patient and carer interviews. Field notes were taken during the observations, discussions and interviews and were de-identified. These activities were undertaken by the qualitative researcher (PM) at each of the three process evaluation sites.

Interviews with older people and their caregivers

We conducted semistructured interviews with a purposive sample of older people and their caregivers who were recruited to the randomised trial. These were conducted between June 2017 and July 2018 so that relationships with each site could be established. We invited people with and people without family caregivers or formal carers and, when available, we invited family caregivers to participate in joint interviews if patients agreed to this.

Field visits

We visited the sites to hold discussions with a range of multidisciplinary staff, managers, commissioners and representatives of primary care and social services to find out how the services were organised, the scope of HAH, the organisation of inpatient hospital care and the interaction with related services that delivered health and social care. We reviewed policy and guidance documents referred to by staff, service assessment documents, discharge templates, protocols or other documentation considered important to the services. In addition, we observed MDT meetings and virtual ward round meetings at each process evaluation site. We used a framework to record our field notes to ensure that we collected information on the local environment, ways of working, roles, interactions, relationships, activities, documents, methods of communication and other aspects that facilitated teamwork, clinical processes and older person and caregiver interactions.

Health-care professionals

We held discussions with a range of health-care professionals to ensure that there was diversity of roles and experience, and we aimed to obtain ‘information-rich’ discussions. These discussions were approved by relevant service managers and were not recorded. The qualitative researcher explained the purpose of the discussions and gained verbal consent from each professional before commencing. We also used snowball sampling: managers or clinical leads identified clinicians and support staff who had suitable experience of the service and for whom a brief ‘release’ from workload activities had been agreed so that they could participate in a discussion with the qualitative researcher. Discussions were undertaken individually or in small groups, depending on staff time constraints and availability. We continued discussions with staff throughout the study or until we had obtained credible explanations that helped to address the research questions. 47 These discussions built on those that we had during the development of the research proposal.

Multidisciplinary staff invited to discussion at each of the three sites included managers and clinical leads for services participating in the randomised trial, allied health professionals across a range of seniorities, staff nurses, ward sisters and matrons, health-care assistants, rehabilitation support workers, consultant geriatricians, junior doctors, physician assistants, primary care representatives (e.g. GPs, district nurses, community matrons), pharmacists, social workers, administrative staff, and health-care commissioners in England and representatives of Health and Social Care Partnerships in Scotland. We took account of variation in staffing between teams, including, where appropriate, pharmacists or social workers. We anticipated that team composition would be dynamic throughout the study.

Data management and analysis

Audio data were fully transcribed by a professional agency and were checked by the researcher for accuracy. We allocated pseudonyms to all participants and de-identified the transcripts. All personally identifiable data were stored in password-protected files in a secure environment. Signed consent forms for the qualitative interviews were securely stored at each site. Field notes from the site visits and observations from interviews in hospital and home settings were also de-identified. We used a spreadsheet to log all ‘raw’ data generated to detail progress and to highlight potential gaps in the evaluation. Data were managed in NVivo 11 (QSR International, Warrington, UK) to aid organising the large number and different types of data, and to facilitate analysis.

Analysis of contextual data

We produced a narrative, descriptive account of the organisation of services at each site studied, using data collected from observations and field notes that were recorded during site visits, a review of formal documents that related to the organisation and delivery of health care, and notes from discussions with a range of health-care professionals. We focused on factors guided by the Consolidated Framework for Implementation Research domains: intervention characteristics, outer setting (i.e. external to the organisation), inner setting (i.e. within the organisation), individuals involved and the processes of implementation. The evaluation of trial implementation explored the sustainability of activities and strategies used at the interface between service delivery (in varied clinical settings) and the randomised trial protocol requirements.

Analysis of qualitative interviews with patients and caregivers

We followed emergent design principles of qualitative research, which involved revisions as the research progressed. Our initial plan was to follow a grounded theory approach; however, this was adapted when we recognised that a more deductive approach was required to sensitise us to the dynamic health-care processes and the key areas that were relevant to the process evaluation. 40 We used a framework analysis to facilitate a comparative analysis of content to organise data into themes and guide the synthesis. 48 Specifically, the initial framework comprised elements from the logic model for CGA that described the inputs and activities intended to bring about the desired outcome of living at home to identify the essential features of health care in each setting of hospital or home. 2 We coded the data using NVivo, expanded through an inductive exploration of data that was not accommodated by the framework. 12

To understand the broader context of the implementation of health care in the two settings, and to assess how people individually and collectively work towards maintaining health, we used four interlinked concepts from NPT as a sensitising device to examine patients’ and caregivers’ participation in their health care. 39,49 The four concepts were (1) sense-making work (i.e. understanding what is happening), (2) relational work (i.e. interpersonal aspects of determining and meeting needs), (3) enacting work (i.e. undertaking and co-ordinating collective tasks) and (4) appraising work (i.e. reflecting on change and ongoing processes of adjustment). We developed an analytic framework, starting with the four NPT concepts, to identify the work required of patients and caregivers in relation to CGA-guided HAH and in hospital. In the second stage, we expanded the analysis by identifying factors that influenced patients’ and families’ capacity to undertake this work. We summarised findings narratively and included extracts of qualitative data to illustrate our interpretations and selected extracts across all three sites to ensure a balance. The qualitative researcher (PM) led the analysis, with SS reviewing the transcripts and GE, RS and DJS assisting with interpretation of the qualitative data.

Screening and recruitment

Sites maintained screening logs throughout the recruitment phase of the trial (i.e. between 9 February 2015 and 18 June 2018). A total of 4805 participants were screened for eligibility: 2169 (45%) were not eligible, 1581 (33%) were eligible but declined to participate and 1055 (22%) were recruited to the trial using a 2 : 1 allocation ratio [HAH, n = 700 (66.4%); inpatient hospital, n = 355 (33.7%)] (Table 3). The first participant was recruited on 14 March 2015 and the final participant was recruited on 18 June 2018 (Figure 4).

FIGURE 4.

Monthly and cumulative recruitment.

Factors affecting recruitment

Discussions with the research nurses and site principal investigators identified a range of factors that contributed to successful recruitment and referral to HAH. These included a strong link between hospital-based consultant geriatricians and those leading the delivery of the HAH services and hospital doctors’ experience of HAH. Three of the sites that had a robust referral system to HAH had established an in-reach function, usually by a matron or an acute care of the elderly nurse who would review all acute medical admissions. One site received referrals from a dedicated hospital ward that had established a system of rapid clinical assessment and ‘discharge to assess’ with onward referral to HAH. Older people who had been assessed in an acute health-care setting for > 24 hours were not eligible for recruitment, and this was a problem if there were delays (such as transport difficulties), if sites had a lengthy assessment process or if additional hospital-based investigations had been arranged. The main reason for older people declining to participate in the study was a strong preference by them or their families for HAH or, less often, a preference for hospital care. The Consolidated Standards of Reporting Trials (CONSORT) flow diagram is shown in Figure 5.

FIGURE 5.

A CONSORT flow diagram of trial participants. a, Data not available to include in the analysis. Reproduced with permission from Shepperd S, Butler C, Cradduck-Bamford A, Ellis G, Gray A, Hemsley A, et al. Is comprehensive geriatric assessment admission avoidance hospital at home an alternative to hospital admission for older people? A randomised trial [published online ahead of print April 20 2021]. Annals of Internal Medicine. 2021. 6 https://doi.org/10.7326/M20-5688. © American College of Physicians.

Reproduced with permission from from Shepperd S, Butler C, Cradduck-Bamford A, Ellis G, Gray A, Hemsley A, et al. Is comprehensive geriatric assessment admission avoidance hospital at home an alternative to hospital admission for older people? A randomised trial [published online ahead of print April 20 2021]. Annals of Internal Medicine. 2021. 6 https://doi.org/10.7326/M20-5688. © American College of Physicians.

‘Living at home’ at 6 months

There was no evidence of a difference in ‘living at home’ (i.e. not being dead and living at home, or the inverse of death or new long-term residential care) at 6 months. The adjusted RR for HAH compared with hospital was 1.05 (95% CI 0.95 to 1.15; p = 0.36). The number and percentage of participants living at home at 6 months are presented in Table 7. The results of the analysis for each of the outcomes that contributed to ‘living at home’ (i.e. mortality and new long-term residential care) are reported in Tables 9 and 10.

‘Living at home’ at 12 months

The adjusted RR for HAH compared with hospital for ‘living at home’ at 12-month follow-up was 0.99 (95% CI 0.89 to 1.10; p = 0.80). The frequency and percentage of participants living at home at 12 months, by randomised arm, are presented in Table 8.

Mortality at 6 and 12 months

There was no evidence of a difference between the HAH and hospital groups in the risk of mortality at 6 months (adjusted RR 0.98, 95% CI 0.65 to 1.47; p = 0.92) or at 12 months (adjusted RR 1.14, 95% CI 0.80 to 1.62; p = 0.47). The frequency and percentage of participants alive and dead at 3-day, 5-day, 1-month, 6-month and 12-month follow-up, and the adjusted RRs with 95% CIs, are presented in Table 9.

New long-term residential care at 6 and 12 months

There was a significant difference between the HAH and hospital groups in the risk of new long-term residential care at 6 months. The adjusted RR at 6 months was 0.58 (95% CI 0.45 to 0.76; p < 0.001) and at 12 months was 0.61 (95% CI 0.46 to 0.82; p < 0.001). The frequency and percentage of participants requiring new long-term residential care at the 6- and 12-month follow-ups, and the adjusted RRs with 95% CIs, are presented in Table 10.

Any residential care at 1-, 6- and 12-month follow-up and the adjusted relative risks (post hoc analysis)

There was no evidence of a difference in the risk of short- or long-term residential care at 1 month (adjusted RR 0.67, 95% CI 0.34 to 1.31; p = 0.24). At the 6- and 12-month follow-ups there was a significant reduction for those allocated to HAH (adjusted RR 0.64, 95% CI 0.54 to 0.76, p < 0.001; and adjusted RR 0.63, 95% CI 0.53 to 0.76; p < 0.001, respectively). The frequency and percentage of participants admitted to any residential care at 1, 6 and 12 months, and the adjusted RRs are presented in Table 11.

Delirium

The adjusted RR for HAH compared with hospital for the presence of delirium at 3 days was 1.21 (95% CI 0.54 to 2.29; p = 0.76), at 5 days was 0.93 (95% CI 0.34 to 2.47; p = 0.87) and at 1 month was 0.38 (95% CI 0.19 to 0.76; p = 0.006). The number and percentage of participants with delirium at baseline, 3 days, 5 days and 1 month, by randomised arm, and the adjusted RRs with 95% CIs are presented in Table 12. The proportion of participants with a clinical diagnosis of delirium at baseline was smaller in the HAH group [HAH 19/687 (2.8%); hospital 14/345 (4.3%)].

The mean [standard deviation (SD)] MoCA score in the group with delirium, measured with the CAM, at each follow-up time is reported in Table 13. On average, this group had a lower mean MoCA score than the study population at baseline [mean 20.5 (SD 6.7), range 0–31; n = 790].

Cognitive impairment measured with the Montreal Cognitive Assessment at 6 months

There was no evidence of a difference in the risk of cognitive impairment (defined using the MoCA) at 6 months (adjusted RR 1.06, 95% CI 0.93 to 1.21; p = 0.36). The frequency and percentage of participants who were cognitively impaired at baseline and at 6-month follow-up, and the adjusted RR with 95% CI, are presented in Table 14.

Activities of daily living (Barthel Index) at 6 months

The adjusted mean difference for activities of daily living (Barthel Index) at 6 months was 0.24 (95% CI –0.33 to 0.80; p = 0.411). The mean scores at baseline and at 6 months and the adjusted mean difference are presented in Table 15.

The adjusted mean difference at 6 months was 0.0002 (95% CI –0.1452 to 0.1455; p = 0.998) (Table 16).

Readmission or transfer to hospital at 1 month and 6 months

There was a significant difference in the risk of readmission or transfer to hospital at 1-month follow-up (adjusted RR 1.32, 95% CI 1.06 to 1.64; p = 0.012). There was no evidence of a difference at 6-month follow-up (adjusted RR 0.95, 95% CI 0.86 to 1.06; p = 0.40). The frequency and percentage of participants requiring readmission or transfer to hospital at each time point, and the adjusted RRs, are presented in Table 17.

The length of hospital and HAH stay immediately following randomisation and at 1 month and 6 months are presented in Chapter 4.

Subgroup analysis

As only a small number of participants in the subgroup with a MoCA score of < 26 presented with delirium (measured by the CAM), we did not calculate a p-value for the subgroup interaction. We have presented the results as frequencies and adjusted RRs in Table 18.

Sensitivity analyses

Three baseline factors were identified as having an association with a non-response to the primary outcome: (1) education level (p = 0.008), (2) place of assessment (p = 0.001) and (3) presenting problems (p = 0.001). These variables were included as covariates in the model used in the primary analysis. The frequency and percentage of participants living at home at 6 months, by randomised arm, and the RRs adjusted for the baseline factors that predicted missingness of the primary outcome are presented in Table 19. There was a very small non-significant change in the values of the RRs and 95% CIs, which did not change the interpretation of the primary outcome (adjusted RR 1.04, 95% CI 0.94 to 1.16; p = 0.40).

Imputing missing data for the primary outcome, ‘living at home’

The frequency and percentages of participants living at home at 6 months, and the adjusted RR when imputing missing data for the primary outcome as ‘living at home and alive’, are presented in Table 20. The values of the RR and 95% CIs did not change from the findings of the primary outcome (adjusted RR 1.04, 95% CI 0.94 to 1.15; p = 0.47).

This analysis was repeated by imputing missing values as ‘not living in long-term residential care and/or alive, or living in long-term residential care/or dead’ as the dependent variable. The frequency and percentages of participants living at home at 6 months, and the adjusted RR between the two treatment groups, are presented in Table 21. The values of the RR and 95% CIs did not change from the findings of the primary outcome (adjusted RR 1.04, 95% CI 0.94 to 1.15; p = 0.47).

In addition, we assessed whether using multiple imputation would change the findings of the primary outcome ‘living at home’ but found no evidence of this (adjusted RR 1.05, 95% CI 0.96 to 1.15; p = 0.29) (Table 22).

Reanalysis of the primary outcome, ‘living at home’, with both 6- and 12-month data

The primary outcome was analysed at each time point separately. As part of the sensitivity analysis, the primary outcome was reanalysed with both outcomes, ‘living at home’ at 6 months and ‘living at home’ at 12 months, but this did not change the findings. The number and percentage of participants living at home at 6 and 12 months, and the adjusted RR between the two treatment groups at each time point, are presented in Table 23.

Patient feedback questionnaire at 1 month

Participants were asked to complete a patient feedback questionnaire following the 1-month follow-up assessment. The responses to each question in this questionnaire are reported in Appendix 3. Overall, the responses to questions about the length of time waiting for care to start, staff receiving information about the patient’s condition, aims of care, how to contact staff, involvement in decisions and discussions with health-care staff about further health or social care services were in favour of HAH.

Results of the main economic analysis

The main economic analysis was based on complete cases, and excluded 124 patients in the HAH group and 71 patients in the control group because information on costs, EQ-5D-5L utilities or LYLAHs was incomplete for them. Table 24 presents descriptive statistics on patient resource use from baseline to the end of the initial episode of health care at the 1- and 6-month follow-up (i.e. including 1-month follow-up data). Patients randomised to the HAH group received, on average, 7.2 (SD 5.6) days of HAH care and spent 1.4 (SD 4.8) days in hospital either immediately before HAH care or as a result of treatment crossover. Patients randomised to the hospital group spent, on average, 4.9 (SD 7.6) days in hospital, and received, on average, 3.8 (SD 7.1) days of HAH treatment, mainly due to treatment crossover. As this analysis follows an intention-to-treat design, all patients who crossed over are analysed according to the group to which they were randomised. By the 6-month follow-up point, total days in hospital had increased to an average of 9.5 in the HAH group and 10.6 in the hospital group, a non-significant difference of 1.1 day (95% CI –3.795 to 1.677 days) in favour of the HAH group. Similarly small but non-significant differences were observed in subsequent HAH and residential care, favouring the HAH group, and in home care, favouring the hospital group. Informal care received was also slightly less in the HAH group, at 595 hours over the 6 months, compared with 658 hours in the hospital group (a non-significant difference of 63 hours, 95% CI –224.610 to 99.097 hours).

Table 25 presents resource use costs and health outcomes from baseline to 1-month follow-up and from baseline to 6-month follow-up. Over the 6 months after randomisation, total NHS and PSS costs, including the HAH intervention costs (i.e. cost of initial admissions to HAH), averaged to £13,975 in the HAH group and £16,521 in the hospital group, showing a significant mean difference of –£2547 (95% CI –£5059 to –£34; p = 0.047) in favour of the HAH group, mainly because of lower admissions to hospital and residential care. When a cost for informal care was added to the NHS and PSS costs (i.e. societal perspective), the mean difference in costs between the two treatment groups increased from –£2547 to –£3017 (95% CI –£5765 to –£269; p = 0.032) in favour of the HAH group.

There was no evidence of significant differences between the two treatment groups in either QALYs or LYLAHs over the 6 months from randomisation. Further details of the resource use and costs in this complete-case analysis are presented in Appendices 5 and 6 (see Appendices 7 and 8 for further details of available cases). Descriptive statistics on resource use by HAH and hospital groups are presented in Appendix 9, and health outcomes at baseline and 6 months are presented in Appendix 10.

Figure 6 reports the results from the NHS and PSS perspective in the form of cost-effectiveness planes, with the point estimates for differences in costs and QALYs (see Figure 6a) or LYLAHs (see Figure 6b) shown as dark-blue dots and the light-blue dots representing the cost–outcome pairs from the 5000 bootstraps. It can be seen that in both analyses the difference in costs normally falls below the x-axis, indicating that HAH is cost saving, whereas the difference in QALYs and in LYLAHs is more evenly distributed around the y-axis, particularly concerning QALYs, indicating a lack of clear evidence of any significant effect using these outcome measures. The joint distribution of differences in costs and effects (i.e. the bootstrap ellipse) falls mainly below the dotted line that represents the NICE willingness to pay threshold of £20,000 per QALY, and so these results indicate that, from the NHS and PSS perspective, the probability of the HAH intervention being cost-effective at a threshold of willingness to pay per QALY or per LYLAH of £20,000 is 97%. Appendix 11 presents similar cost-effectiveness planes from a societal perspective, in which these probabilities are further increased by approximately one percentage point, and cost-effectiveness acceptability curves that show how the probability that the intervention is cost-effective changes as the willingness-to-pay threshold is altered.

FIGURE 6.

Cost-effectiveness plane for the NHS and PSS perspective: (a) cost per QALY; and (b) cost per LYLAH.

Results of the sensitivity analyses

The results of the sensitivity analyses are shown in Table 26. Using linear mixed-effects regressions to adjust for any differences in baseline covariates, the difference in NHS and PSS costs between the HAH and hospital groups decreases from –£2547 to –£2265 (95% CI –£4279 to –£252; p = 0.028) and the cost difference from a societal perspective also decreases from –£3017 to –£2840, but remains significant (95% CI –£5495 to –£185; p = 0.036). Any differences in QALYs and LYLAHs remain small and non-significant. In this sensitivity analysis, the probability that HAH is cost-effective at the £20,000 threshold is 97% from the NHS perspective and 98% from a societal perspective.

Using multiple imputation for all missing data, the mean difference in NHS and PSS costs decreases from £2547 to £2458 (95% CI –£4977 to £61; p = 0.056). The difference in societal costs also remains significant [–£3083 (£1424), 95% CI –£5880 to –£287; p = 0.031]. Any differences in QALYs and LYLAHs remain very small and non-significant.

Section 1: the delivery of health care and how acute health care in the home represents a new way of working

In addition to interviewing patients and caregivers, we held discussions with a range of staff directly involved in organising and delivering health care (Table 29). We spent 3.5 days observing each of the three HAH workplaces and multidisciplinary meetings (doctor led and non-doctor led), 2.5 days observing an acute assessment unit at each site (including board and ward rounds) and 2.5 days on wards for older people at each site (including observing board rounds and MDT meetings).

We report the findings of the discussions with health-care professionals across three dimensions that had an impact on how health care was delivered and experienced in each setting: (1) the environment, (2) the workforce configuration and (3) processes of health-care delivery (Figure 7).

FIGURE 7.

(a) Overview of the analysis; and (b) environment.

Enabling support and interaction

Some patients valued the company provided by the ward environment:

The wards are smaller now, I mean, I remember way back, big long wards, and you could only speak to the people next to you. But it’s better now that you’ve only got four beds in a wee room, that’s much better because you can speak to those people.

Bridget, hospital

There was only four in the ward and the lady opposite actually lived just down the road for years. So, we could talk about how things used to be around here.

Bertha, hospital

However, many patients and caregivers described being disturbed by the noise on the ward:

There was one problem there and you couldn’t do anything about it, there was one particular lady who just sung night and day, but mum never moaned about that.

Iris’s daughter, hospital

Some considered that hospital care had disrupted personal support relationships, including with formal carers (arranged through social services prior to the acute episode of illness). Family caregivers described how health care at home allowed support networks to continue during the acute episode:

I just brought my stuff down here [to mum’s house] and I could work . . . I didn’t lose any days, because I just grabbed my laptop and worked out in the kitchen.

Aisla’s daughter, HAH

Workforce

The distinctive features of the HAH team, scope of practice and integration are described in Figure 8.

FIGURE 8.

Overview of workforce changes in HAH.

Workforce integration

Patients and caregivers described disruptions to routine district nursing input while in hospital:

The district nurses said, ‘you’ve been taken off our books now because you’re in hospital’, so she had to go through the doctor again.

Imogen’s caregiver, hospital

By contrast, district nurses’ longer-term involvement with those receiving HAH was considered beneficial. Staff at all sites said that practical tasks, such as intravenous medication administration, end-of-life care and leg ulcer management, were considered areas of potential synergy between district nursing and the HAH team. However, at times, they considered that stronger communication would enable more effective joint-working between the two.

Separate HAH and district nursing systems could create difficulties in co-ordinating the management of medicines, for example to avoid duplication of medicine administration. At one site, the integration of the electronic patient record system between HAH and GPs was considered to have improved communication about prescriptions and reduced potential medication errors. An electronic notification could be sent by HAH staff to the patient’s GP to alert them when HAH investigation results were available and to request a change to a patient’s prescription by the GP, if needed, along with advice on whether the change should be short or long term. At another site, where GPs did not continue prescribing during HAH episodes, and electronic systems were not integrated with primary care, a GP felt that HAH prescriptions could present a risk area, expressing concern that non-medical prescribers may have insufficient experience to deal with the complexities of prescribing in some situations.

The availability of social workers was limited across hospital and HAH services. In two sites, HAH teams and social services were based in the same location, the intention being to enhance collaborative ways of working. Lack of access to information technology systems shared between HAH and social services was a key issue at each site. The HAH team and social services at one site had implemented a ‘trusted assessor’ model for sharing HAH recommendations for care needs to avoid duplication of assessments.

Processes of health care

The process of delivering CGA-guided HAH is described in Figure 9.

FIGURE 9.

Overview of key processes in CGA-guided HAH.

Section 2: the role of older people and their caregivers in the delivery of health care for an acute health event

We report the findings from the interviews with patients and their caregivers around four interlinked concepts from NPT (Figure 10) to understand how people individually and collectively worked towards maintaining health. These were:49,51

1. sense-making work (i.e. understanding what is happening)

2. relationship work (i.e. interpersonal aspects of determining and meeting needs)

3. enacting work (i.e. undertaking and co-ordinating collective tasks)

4. appraising work (i.e. reflecting on change and ongoing processes of adjustment).

FIGURE 10.

Patients’ and family caregivers’ participation in CGA-guided acute health care across the dimensions of NPT.

Across the four dimensions we identified two overlapping forms of work: (1) negotiating with health-care professionals and (2) ‘doing’ or providing health- and personal care-related work. We also revised a logic model that had been developed for a Cochrane review of CGA in a hospital setting46 to add the context of delivering health care in the home and to help understand the perspectives of patients and their caregivers. We found no evidence that the term CGA, as a collaborative process that guided the assessment and planning of health care, was recognised by patients and caregivers.

Patients’ and caregivers’ ‘negotiating’ work

Patients’ and caregivers’ ‘doing’ work

Alongside ‘negotiating’ work, older people and caregivers mobilised personal and social resources to manage care and potential risks (i.e. ‘doing’ work).

Synthesis of findings: a possible mechanism of change

We used a logic model to combine the findings from the process evaluation to represent the inputs, activities, outputs, outcomes and impact of CGA-guided HAH from the perspectives of patients and their caregivers (Figure 11). This included factors external to the delivery of health care, and complements a logic model of geriatrician-led CGA in a hospital setting that was developed from the perspective of health-care professionals. 46

FIGURE 11.

Logic model of HAH developed from the perspectives of patients and family caregivers. A&E, accident and emergency.

Elements of CGA in a hospital setting that are considered critical to success include clinical leadership, specialty knowledge, experience and competence, MDT meetings, the tailoring of treatment plans to the individual and the involvement of patients and carers in goal-setting. Two areas differed in the context of HAH. We found that clinical leadership was more distributed across senior members of clinical teams, and that specialty knowledge and skills were shared beyond traditional disciplinary inpatient boundaries to ensure a workable allocation of staff for home visits that could be spread across a geographical area. Patients and caregivers receiving HAH valued the continued input from the primary care team, as well as from HAH staff.

Implementation of HAH depended on a range of factors associated with the older person, the health service and the health system, each of which had the potential for unintended consequences. We have summarised these as patient-, service- and system-level factors (Table 30).

Patient-level factors

Our findings show that caregivers’ capacity to provide additional practical and emotional support, and a suitable home environment, is crucial to the delivery of HAH for many older people. Figure 12 summarises the factors that moderated patients’ and caregivers’ capacity to ‘negotiate’ and ‘do’ the work required. Staff described situations when patients would be excluded from HAH, sometimes but not always because of the absence of a caregiver at home or being alone at night and there were concerns about safety. A further potential consequence is that an older person’s social network might have to adapt to manage health events and safety at home in ways that might be more significant than if the older person received hospital-based care.

FIGURE 12.

Factors facilitating and challenging patients’ and caregivers’ capacity to undertake acute health-care-related work (HAH and hospital).

A theory of change

To understand a plausible mechanism of action, and how HAH might have an impact on older people’s health and functional outcomes, we used the findings from this process evaluation to develop a possible theory of change (Figure 13). Older people and caregivers, in combination with the HAH team, play a crucial role by navigating health-care and social systems to support usual routines and helping the older person to adapt when an acute deterioration in their health is accompanied by emotional, relational and physical demands. Personal resources include prior experience, the environment and social networks. The theory of change is based on the assumptions that access is equitable, HAH staff manage clinical uncertainty and caregiver support will be available if required. Wider distribution of responsibility among team members, families and older people when a patient’s health and associated risks change underpinned the delivery of HAH to this population of older people. Our findings suggest that implementation of HAH can support adaptive capacity that is both older person and family centred.

FIGURE 13.

Comprehensive geriatric assessment-guided HAH ‘theory of change’.

Main findings

Consistent with the concept of healthy ageing,57 we hypothesised that older people who received HAH care might experience less decline in functional and cognitive capacity and maintain a level of independence that is more difficult to achieve in a more restricted hospital environment. The results from this randomised trial show no apparent difference in the primary outcome of living at home (i.e. the inverse of mortality or living in new long-term residential care) at 6-month follow-up, although with differential effects in each component of the outcome. There was no statistically significant effect on mortality at 6 months and some uncertainty at 12 months because of a lack of precision and wide CIs. There was a significant relative reduction in new long-term residential care for those allocated to HAH at 6 and 12 months’ follow-up, albeit with small numbers. We did not find a difference in the secondary outcomes of cognitive impairment, ability to carry out activities of daily living or the Charlson Comorbidity Index score for comorbidity. A significant reduction in new cases of delirium at 1-month follow-up in the group allocated to HAH is consistent with previous research,58 but with small numbers of patients affected. The rate of transfer to hospital was significantly higher in those allocated to HAH than in those allocated to hospital at 1 month, but not at 6 months.

The results of the economic evaluation showed that HAH is highly likely to be more cost-effective than hospital admission for older people who experience an acute change in health. Although it is uncertain whether or not HAH leads to QALY gains, this study found clear evidence that NHS and PSS costs were significantly lower in the HAH group, with no evidence of an increase in informal care. Our results suggest that, when combined with lower residential care costs, substituting 3 days of HAH for 3 days in hospital during the initial admission, a difference that was reduced to 1 day at 6-month follow-up, reduces total care costs with no apparent adverse effects on quality of life or informal care requirements.

Mechanism of action

There is a vast body of literature on the use of theory to develop a conceptual framework and the importance of theory-based evaluations of complex interventions,63–65 but guidance is limited on how to codify a mechanism of action once data have been collected. 66 The reality is that these ‘mechanisms may prove stubbornly hard to nail’. 67 The logic model provided a framework to inform intervention design and map out processes of care from the perspective of patients and caregivers, but provided limited guidance on identifying causal pathways. We used the interview findings to examine how HAH changed the delivery of health care and how it might produce outcomes by examining the environment, the workforce, processes of health care, and the work required by patients and caregivers. Building on these findings, we proposed mechanisms of action for HAH (Figure 14) to aid understanding of how HAH might have differed from bed-based hospital care, and proposed factors for consideration when developing similar interventions. Through sharing traditional roles, the HAH team members strengthened the team’s knowledge, skills and resources. The familiar physical and social environment of home can support activities of daily living and enables family caregivers to be involved in the patient’s alongside the HAH team, who provide an interface between acute and community services. It is possible that the patient's independence is maintained by the recovery in the familiar home setting, the relaxation of the strict hospital hours for activities (e.g. food, rest and hygiene) and the avoidance of family trips to the hospital. 68,69 These mechanisms are proposed in the context of the assumptions and supporting factors that are detailed in the theory of change for CGA-guided HAH (see Figure 13).

FIGURE 14.

Proposed interdependent mechanisms of action of HAH.

Workforce

A striking finding was that the HAH model of working altered traditional perceptions of disciplinary boundaries through extended scope training and informally through negotiations and support between team members. Extended scope and advanced practitioner roles, to some extent, resembled the role of junior doctors in hospitals, as HAH staff undertook initial ‘medical’ assessments, arranged investigations, entered medical progress notes into patient records, produced discharge summaries and, in some cases, prescribed medications. 70 It is possible that further development of this extended role might equip the NHS to meet the health needs of an older population, as working across traditional boundaries offered the potential for a more holistic model of care. By contrast, the increased specialisation of roles commented on by inpatient staff, with tasks divided up according to role and discipline, risks fragmentation of care and a potential increase in health-care costs. 70

Only one site employed junior doctors to work with the HAH team, and reported that they were a valuable addition and provided a way of increasing the number of clinicians working with the HAH team. In addition, exposing junior doctors in training to HAH might raise system awareness of where older people with acute health crises might be cared for. Extended scope working predominantly requires changes to the non-medical workforce, who do not tend to rotate between short-term posts, as junior doctors do, thereby enabling continuity in service delivery. However, some HAH staff described tensions that had arisen in fulfilling organisational expectations, including anxiety about whether or not they were operating outside their professional scope of practice, which reduced their willingness to carry out extended parts of their role following training. Without careful role design and attention to team development, extended roles might supplement rather than substitute for other staff. 71 Other concerns related to the potential for ‘dilution’ of multidisciplinary expertise when staff were extending their contributions to patient care. Challenges to new patterns of working and changing staff skills have been previously identified, along with the role of GPs, in the redesign of these types of services. 72,73

Caregivers

Parts of this text have been reproduced from Mäkelä et al. 50 This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (http://creativecommons.org/licenses/by-nc/4.0/), which permits non-commercial re-use, distribution, and reproduction in any medium, provided the original work is properly cited. For commercial re-use, please contact journals.permissions@oup.com. The text below includes minor additions and formatting changes to the original text.

Families and friends play a crucial role in supporting the delivery of health care to this population of older people in hospital, and even more so at home, particularly at night. 50 Caregivers (who are often older partners managing their own health problems) were frequently required to facilitate an episode of acute health care at home. In each setting, the relationships between older people, their support networks and health-care services had an impact on older people’s capacity to manage an acute deterioration in health. This was a problem if they were not involved in clinical assessments or decision-making, and could result in a lack of understanding and coherence in planning care. The relational resources of family, the neighbourhood and community health-care professionals in HAH, ordinarily disrupted during inpatient admissions, acted as a bridge to continuity of health care. The importance of health-care professionals’ understanding of caregivers’ challenges is widely established,74–76 yet their contribution to managing older people’s acute health care at home is not always recognised. 77,78

Generalisability

We were able to recruit a group of vulnerable older people (with a mean age of 83 years, comorbidity and mild to moderate restriction in daily activities, 72% of whom also had some cognitive impairment) from across the UK, but, nonetheless, achieved high rates of follow-up and outcome assessment. The characteristics of this trial population are similar to those of the service users described in the National Audit of Intermediate Care. 55,79 This provides some confidence in the generalisability of the results among our trial population to the population using these types of services as part of routine care.

We excluded participants who required end-of-life care, as we recognised that this group might have a strong preference for receiving health care at home. There was some variation among the HAH services included in this study; for example, all but one provided intravenous infusions. It is possible that those who were more burdened were less likely to be referred by the sites for interview and, when referred, were less likely to accept the invitation to be interviewed, as they did not want to participate in additional research activities or they lacked time. The majority of participants were referred from an acute assessment unit or an older persons’ frailty unit, with only a minority referred directly from home by their GP. Therefore, our results are more strongly related to patients who were referred after a rapid specialist assessment process in the local general hospital, a population who usually experience a sudden decline in functioning and might differ from the majority referred from primary care.

The process evaluation identified forms of work undertaken by older people and family caregivers at the time of, and beyond, an acute health event, and might inform how broader policy interventions that focus on prevention and self-management80 might provide support to this population.

Research of standard care interventions

In this study, participants were randomised to one of two health-care services (i.e. HAH vs. hospital) that were established prior to the study. Despite this, we encountered a number of barriers to streamlining the implementation of the randomised trial. Lengthy internal approval of documentation and amendments by the sites delayed recruitment, and an initial requirement to produce long and short forms of the patient information leaflets (a total of 45 pages) added to these difficulties. As anticipated, the preferences of clinical teams and the study population were a barrier to recruitment. The demand for hospital beds did, on occasion, divert participants who had been allocated to hospital to HAH. Changes to services that occurred during the study also had an impact on recruitment; for example, the opening of a redesigned emergency frailty unit that would have been a hub for recruiting participants was delayed. However, in the context of successfully recruiting and generating randomised evidence to support decision-making, these barriers created delays but did not stop recruitment.

A less burdensome regulatory framework would support the generation of randomised evidence to guide the delivery of standard care interventions that lack a robust evidence base. Considering standard care interventions as a potential risk to participants in the context of research, but with little risk outside a research setting, adds a substantial cost and unnecessary administrative burden to the conduct of research and to those participating in the research. In 2015, the Institute of Medicine81 published a summary of a workshop that reviewed research of standard care interventions. Although the focus was on ethics issues that related to study design and consent, the report described the tension between clinical practices that lack an evidence base and the proportionate risks associated with research into standard care interventions. 81 Further guidance in the UK context would support the cost-effective research of standard care interventions without compromising ethics and the safety and well-being of those who participate in research.

In memoriam

Mary Godfrey died as this report was being finalised. Throughout this research Mary provided valuable advice and insights, her generosity and willingness to challenge undoubtedly improved the quality of the research reported in this publication.

Trial Steering Committee

Professor Rowan Harwood, Professor of Geriatric Medicine, Faculty of Medicine & Health Sciences, University of Nottingham, Nottingham, UK (Trial Steering Committee chairperson).

Ms Amanda Edwards, now retired (was at the Social Care Institute for Excellence, London, UK).

Dr Mark Mullee, Independent Statistician, Southampton University, Southampton, UK.

Dr Sarah Pendlebury, Consultant Geriatrician, Nuffield Department of Clinical Neurosciences, John Radcliffe Hospital, Oxford, UK.

Professor Niro Siriwardena, Professor of Primary & Pre-Hospital Health Care, University of Lincoln, Lincoln, UK.

Dr Emma McIntosh, Health Economist, Health Economics and Health Technology Assessment (HEHTA), Glasgow, UK.

A lay representative.

Dr Michael Dunn, Ethox Centre, Nuffield Department of Population Health, University of Oxford, Oxford, UK.

Joyce Epstein, lay representative.

Mr Tim Davidson, Chief Executive, NHS Lothian, Edinburgh, UK.

Candace Imison, Director of Policy, Nuffield Trust, London, UK.

Natasha Curry, Acting Deputy Director of Policy, Nuffield Trust, London, UK.

Data Monitoring Committee

Professor Mike Clarke, Northern Ireland Methodology Hub, Centre for Public Health, Queen’s University Belfast, Belfast, UK (DMC chairperson).

Dr Isobel Reading, Director of the Research Design Service South Central and a Principal Medical Statistician Faculty of Medicine, University of Southampton, Southampton, UK.

Professor Graeme MacLennan, Centre for Healthcare Randomised Trials, University of Aberdeen, Aberdeen, UK.

Professor Stuart Parker, Emeritus Professor at Newcastle University, Newcastle upon Tyne, UK, and Trustee at Dunhill Medical Trust, London, UK.

Participating sites

Contributions of authors

Sasha Shepperd (https://orcid.org/0000-0001-6384-8322) produced the first draft of the report for comment, led subsequent revisions, reviewed the statistical analysis and was involved in the analysis of data from the process evaluation.

Andrea Cradduck-Bamford (https://orcid.org/0000-0002-5772-9666) contributed to report compilation, comments, trial logistics and organisation.

Christopher Butler (https://orcid.org/0000-0002-0102-3453) provided input and advice on trial management.

Graham Ellis (https://orcid.org/0000-0003-2982-9245) read and provided comments on the draft chapters.

Mary Godfrey (https://orcid.org/0000-0002-2408-534X) contributed to the design of the process evaluation and writing of Chapter 5.

Alastair Gray (https://orcid.org/0000-0003-0239-7278) supervised the conduct of the cost-effectiveness analysis and co-led the writing of Chapter 4.

Anthony Hemsley (https://orcid.org/0000-0002-5595-4528) read and provided comments on the draft chapters.

Pradeep Khanna (https://orcid.org/0000-0002-7193-456X) read the draft chapters.

Peter Langhorne (https://orcid.org/0000-0001-8185-2659) provided project management advice, and read and provided comments on the draft chapters.

Petra Mäkelä (https://orcid.org/0000-0002-0938-1175) led the process evaluation, carried out the qualitative research activities, led the qualitative data analysis and interpretation, and led the reporting of the qualitative research methodology and findings.

Sam Mort (https://orcid.org/0000-0001-6332-4641) provided statistical analysis and commented on Chapters 2 and 3.

Scott Ramsay (https://orcid.org/0000-0001-9048-1562) read and provided comments on the draft chapters.

Rebekah Schiff (https://orcid.org/0000-0001-6687-3726) read and provided comments on the draft chapters.

Surya Singh (https://orcid.org/0000-0002-3005-8073) conducted the analysis for the economic evaluation, and contributed to the interpretation and writing-up of Chapter 4.

Susan Smith (https://orcid.org/0000-0003-0639-4242) provided comments on Chapter 2 and as data manager was responsible for producing reports of missing and inconsistent data, cleaning data and supervised data entry and form tracking.

David J Stott (https://orcid.org/0000-0002-3110-7746) was involved throughout the study with design, planning, acquisition of funding, trial management, data interpretation and provided critical comments on the draft chapters.

Apostolos Tsiachristas (https://orcid.org/0000-0002-4662-8915) oversaw the analysis of the data, and contributed to the interpretation and writing-up.

Angela Wilkinson (https://orcid.org/0000-0003-4111-0671) read and provided comments on the draft report.

Ly-Mee Yu (https://orcid.org/0000-0003-0331-7364) led the writing of the statistical analysis plan, and provided oversight and validation of the statistical analysis.

John Young (https://orcid.org/0000-0003-4085-9306) provided clinical and policy expertise, advised on the design and conduct of the study, and read and provided critical comments on the draft chapters.

Publications

Shepperd S, Cradduck-Bamford A, Butler C, Ellis G, Godfrey M, Gray A, et al. A multi-centre randomised trial to compare the effectiveness of geriatrician-led admission avoidance hospital at home versus inpatient admission. Trials 2017;18:491.

Mäkelä P, Godfrey M, Cradduck-Bamford A, Ellis G, Shepperd S. A protocol for the process evaluation of a multi-centre randomised trial to compare the effectiveness of geriatrician-led admission avoidance hospital at home versus inpatient admission. Trials 2018;19:569.

Mäkelä P, Stott D, Godfrey M, Ellis G, Schiff R, Shepperd S. The work of older people and their informal caregivers in managing an acute health event in a hospital at home or hospital inpatient setting. Age Ageing 2020;49:856–64.

Shepperd S, Butler C, Cradduck-Bamford A, Ellis G, Gray A, Hemsley A, et al. Is comprehensive geriatric assessment admission avoidance hospital at home an alternative to hospital admission for older people? A randomised trial. Ann Int Med 2021;174:889–98.

Singh S, Gray A, Shepperd S, Stott DJ, Ellis G, Hemsley A, et al. Is comprehensive geriatric assessment hospital at home a cost-effective alternative to hospital admission for older people? Age Ageing 2021;51:afab220.

Data-sharing statement

All data requests should be submitted to the corresponding author, which will be considered in line with the Nuffield Department of Population Health’s Data Access Policy [URL: www.ndph.ox.ac.uk/data-access (accessed 8 December 2020)]. Access to available anonymised data may be granted following review.

Patient data

This work uses data provided by patients and collected by the NHS as part of their care and support. Using patient data is vital to improve health and care for everyone. There is huge potential to make better use of information from people’s patient records, to understand more about disease, develop new treatments, monitor safety, and plan NHS services. Patient data should be kept safe and secure, to protect everyone’s privacy, and it’s important that there are safeguards to make sure that it is stored and used responsibly. Everyone should be able to find out about how patient data are used. #datasaveslives You can find out more about the background to this citation here: https://understandingpatientdata.org.uk/data-citation.

Disclaimers

This report presents independent research funded by the National Institute for Health Research (NIHR). The views and opinions expressed by authors in this publication are those of the authors and do not necessarily reflect those of the NHS, the NIHR, NETSCC, the HSDR programme or the Department of Health and Social Care. If there are verbatim quotations included in this publication the views and opinions expressed by the interviewees are those of the interviewees and do not necessarily reflect those of the authors, those of the NHS, the NIHR, NETSCC, the HSDR programme or the Department of Health and Social Care.

Reponses to each question in the patient feedback questionnaire by randomised arm

Topic guide for discussions with patients randomised to ‘hospital’ and family caregivers

Topic guide for discussions with patients randomised to ‘hospital at home’ and family caregivers