Effects of locality based community hospital care on independence in older people needing rehabilitation: randomised controlled trial

Protocol

Patients were eligible for the trial if they were registered with a general practitioner in the primary care trust served by the community hospital and were considered by the responsible geriatrician to be medically stable and in need of post-acute rehabilitation care. We excluded patients if they had features of medical instability (pyrexia, breathlessness at rest, history of chest pain within the previous 48 hours, or need for intravenous drugs)10; were drowsy or unconscious; required specialist rehabilitation in the stroke unit, or treatment in other departments such as surgery or coronary care; or needed new placement in a care home. Patients and their carers gave written informed consent or assent was sought from a carer or relative when patients had impaired capacity.

Baseline assessment by a research nurse included reason for admission, current input from social care services, cognitive function (abbreviated mental test score),11 activity restriction before admission (Nottingham extended activities of daily living and Barthel index),12^–14 and other outcome measures (see bmj.com).

Assignment

We stratified patients by cognitive impairment using the abbreviated mental test (categories 0-7; 8-10) and by activity restriction using the baseline Barthel index score (categories 0-13; 14-20). Randomisation to community hospital (intervention) or usual care (control) was by numbered, sealed, opaque envelopes prepared from random numbers tables, which initially used four length permuted blocks in a ratio of one intervention group to each control group. Owing to pressures on beds in the district general hospital, we changed the randomisation schedule after two months (26 patients had been recruited) to three length permuted blocks in a ratio of two intervention groups to each control group. We used short length blocks to minimise the potential for disruption to clinical services. Randomisation was carried out away from the clinical setting and, as patients were recruited from several wards, the ward staff were blinded to randomisation. The research nurse responsible for recruitment was unaware of the stratification boundaries and block length. We aimed to transfer patients randomised to the community hospital group within two days in accord with existing local practice.

Intervention group

Patients allocated to community hospital care were assessed by the multidisciplinary team and received an individual care plan designed to maximise recovery and promote independence. Home assessments before discharge are commonly arranged, and discharge is coordinated with the local social service team. The consultant visited the hospital at least twice a week and the hospital practitioner visited the hospital each weekday. Local general practitioners provided out of hours cover (see bmj.com for input from nursing and therapy teams).

Control group

Usual care consisted primarily of an extended stay in the care of the elderly ward in the local district general hospital but could include transfer to a non-locality based community hospital or to social service facilities, both of which offered slow track rehabilitation.

Outcome measures, follow up, and blinding

For the primary outcome measures we used the Nottingham extended activities of daily living scale for patients, a valid and reliable measure,12 15 and the general health questionnaire 28 for carers.16 Other outcome measures are listed in table B on bmj.com. Information on death, length of stay in hospital, and destination after discharge was collected.

Follow-up assessments were at one week after hospital discharge, and at three and six months after recruitment.

A researcher who had not been involved with recruitment visited the patients at home and completed the outcome measures. Unblinding of patient allocation was recorded and the researcher stated the perceived trial group for each patient at the final assessment. We measured agreement between perceived and actual allocation using the κ statistic.

Statistical analysis

We defined a clinically meaningful difference as a two point improvement of the Nottingham extended activities of daily living scale (0-22 scoring). We used the revised 0-66 scoring in accord with recent practice. For our sample size calculation we used a standard deviation of 5.3 recently obtained from an intermediate care study of over 300 frail older people.22 This method indicated a sample size of 400 for 1% significance and 85% power, and 240 for 5% significance and 85% power. We planned to recruit 500 patients to allow for dropouts.

Our primary analysis was an intention to treat comparison of change in scores on the Nottingham extended activities of daily living scale from baseline to six months between the intervention and control groups. We used unpaired t test and the Mann-Whitney U test for unadjusted values. We also carried out an adjusted intention to treat comparison using analysis of covariance to adjust for the baseline variables of age, sex, institutional care, mental status, and baseline Barthel index score. To allow for potential bias due to the exclusion of patients who had died, we used the Mann-Whitney U test to compare the groups for the change in Nottingham extended activities of daily living scale scores from preadmission to the six month assessment after recoding the six month scores for these patients as zero (worst outcome). The primary outcome for carers was the difference between the groups for change in general health questionnaire 28 scores between baseline and six months using the Mann-Whitney U test.

For our secondary analyses we used the Mann-Whitney U test to compare the groups for changes in the total Nottingham extended activities of daily living scale scores and general health questionnaire 28 scores between baseline and the follow-up assessments at one week and three months. We carried out similar analyses for each assessment point for the subsections of the Nottingham health profile, the total Barthel index score, and the anxiety and depression subsections of the hospital anxiety and depression scale. We used the Mann-Whitney U test to examine the differences between the groups for the carer strain index scores at the follow-up assessments at one week and three and six months. The χ² test was used to examine differences in mortality and destination after discharge between the groups. We created a dichotomous response (satisfied or not satisfied) for patient and carer satisfaction scores, and we compared the groups using the χ² test.

Participant flow and follow up

Both groups were well matched for baseline characteristics and length of hospital stay (table 1). Thirty two patients were not assessed one week after hospital discharge (29 died), 21 patients were not assessed at three months, and a further 13 patients were not assessed at six months (figure). Tables 2 and 3 provide the results of outcome measures. The outcome assessor was unblinded to treatment allocation at the six month assessment by 18 patients and carers in the treatment group and six patients and carers in the control group. The assessor correctly guessed the allocation of 67 of the remaining 128 patients (52%) at the six month assessment (eight patients had missing data). Agreement was poor (κ < 0.20).

Primary outcome analyses

The unadjusted intention to treat comparison of changes in scores on the Nottingham extended activities of daily living scale at six months showed a non-significant difference between the groups (mean difference 4.88, 95% confidence interval 0.18 to 9.59). A significant difference was suggested by the t tests, but this was largely due to the effect of an outlier. The adjusted intention to treat comparison showed a significant difference between the two groups (mean difference 5.30, 0.64 to 9.96). Although the data were reasonably normal, we did find an outlier, which once removed had no material effect on the result (mean difference 4.68, 0.27 to 9.09). The difference remained after patients who had died were assigned the worst score on the Nottingham extended activities of daily living scale.

We found no differences between the groups for emotional distress in carers, measured using the general health questionnaire.

Secondary analyses

The secondary analyses showed few differences. We found improvements in favour of the community hospital in the sleep subsection of the Nottingham health profile at one week and three months (median difference −10.5, −16.8 to 0.0 and −12.6, −22.4 to 0.0, respectively). The intervention group showed a greater increase in the depression subsection of the hospital anxiety and depression scale at one week (median difference 1.0, 0.0 to 2.0). Similar numbers of patients in the intervention and control groups died before the six month assessment (34 (23%) v 21 (27%), respectively) or were discharged to a new care home place (30/134 (22%) v 21/74 (28.4%), respectively).

Both groups had similar reported patient and carer satisfactions. At three months the intervention group showed greater satisfaction with the question “I get all the support I need from services such as meals on wheels, home helps, district nursing, etc” (odds ratio 3.43, 95% confidence interval 1.05 to 11.24).