Recruitment and assignment

Researchers recruited patients from four general hospitals and six community hospitals if they were aged 80 or over, admitted as an emergency, intended to be discharged to their own home or warden controlled accommodation, and prescribed two or more drugs on discharge. Exclusion criteria were dialysis treatment and participation in an intensive discharge service on one site. We randomised patients to receive home based medication review or usual care. We used third party telephone randomisation based on a computer generated sequence in blocks of varying length. Randomisation was stratified by abbreviated mental test score (score >8v<8)12 and whether the patient was living alone. We obtained written informed consent from all participants.

Pharmacists could participate if they held a postgraduate qualification in pharmacy practice or had recent continuing professional development in therapeutics. All pharmacists participated in a two day training course, including lectures on adverse drug reactions, prescribing in elderly people, improving concordance, and communication skills.

The intervention

Initial referral to a review pharmacist included a copy of the patient's discharge letter. Pharmacists arranged home visits at times when they could meet patients and carers. Pharmacists assessed patients' ability to self medicate and drug adherence, and they completed a standardised visit form. Where appropriate, they educated the patient and carer, removed out of date drugs, reported possible drug reactions or interactions to the general practitioner, and reported the need for a compliance aid to the local pharmacist. Where a compliance aid was recommended, this was provided within the trial and a filling fee was paid to the local pharmacist. One follow up visit occurred at six to eight weeks after recruitment to reinforce the original advice.

Masking and the control group

Because of the nature of the intervention, no “placebo” could be provided. Participants were told after randomisation which group they were in. Those in the control group received “usual care.” It is possible that a small number of patients in both groups may have had their medication reviewed during the follow up period by their general practitioner or community pharmacist.

Outcome data and analysis

The primary outcome was total number of emergency admissions to hospital over six months. Secondary outcomes included deaths, admissions to residential homes and nursing homes, and self assessed quality of life measured using the EQ-5D.13^–15 Utility scores, as measured by the EQ-5D, can vary from 1 (perfect health) to -0.59 (worst imaginable health state). Patients also rated their health on a visual analogue scale from 100 (perfect health) to 0 (worst imaginable health).

We collected data on emergency admissions from hospital episode statistics. The Office for National Statistics provided mortality data. In addition, the project coordinator contacted all patients by telephone at three months and six months to collect data on admissions to nursing homes and residential homes and to maximise the response to mailed quality of life questionnaires. In addition, we collected data from practices containing more than 10 trial patients on home visits by general practitioners, attendance at general practices, and items prescribed over the six month follow up.

We used Poisson regression to compare the number of admissions between groups. We used survival analysis to compare mortality between the two groups by using the Cox proportional hazard ratio. In both analyses, we adjusted for the two stratification variables (living alone and confusion). We analysed the proportions admitted to nursing homes or residential homes by using the χ²test. We assumed that changes in quality of life followed a normal distribution and analysed them by using the two sample t test. We compared home visits by general practitioners and attendance at general practices by using Poisson regression, entering practices into the model as a random effect.

We analysed data according to randomisation group, irrespective of whether or not the patients received the intervention as planned (the intention to treat principle). We used Stata version 8.0 with statistical significance at the 5% level.

Sample size calculation

Local admission data showed a mean of 0.8 readmissions per patient within six months of discharge. A previous randomised controlled trial suggested that readmissions could be reduced by 25% in six months.11 We aimed to show a more conservative reduction of 20%. Sample size calculations based on a normal approximation to the Poisson distribution indicated that we needed to recruit 850 patients to have 90% power to show this reduction at the 5% significance level.

Participant flow and follow up

We invited 1399 patients to participate after screening them for eligibility between October 2000 and December 2002; 872 (62%) patients agreed and were randomised (1). We excluded 17 recruited patients after randomisation. Reasons for exclusion were elective baseline admission (n = 5), discharge to nursing or residential home (5), death before discharge (4), previously recruited (2), and taking fewer than two drugs (1). 1 shows that the two groups were very similar at baseline. Twenty patients withdrew from the trial, and six moved from the study area. Primary outcome data were thus available for 829 (97%) patients.

View larger version:

• In a new window

• Download as PowerPoint Slide

Fig 1

Flowchart showing progress of patients through the HOMER trial

Review pharmacists and intervention visits

We recruited 22 review pharmacists. Of 429 patients in the intervention group, 362 received first visits. Reasons for not visiting were visit not wanted (46 patients), pharmacist unavailable (11 patients), and patient unavailable due to death or early readmission (10 patients). Review pharmacists carried out a mean of 17 (range 1-37) first visits. More than 90% of first visits occurred within two weeks of recruitment (mean 7.2 days), and visits lasted a mean (SD) of 61 (23) minutes. Second visits were conducted for 297 patients and took a mean (SD) of 42 (19) minutes. Visits generated a total of 933 recommendations or comments to general practitioners (2.58/visited patient); 120 of these referred to possible drug reactions or interactions in 81 patients (22% of visited patients).

Compliance aids and pharmacists' view of intervention

Review pharmacists recommended compliance aids in 39 patients (11% of those receiving first visits). Pharmacists were asked after their second visit to record whether they believed the visits had been useful. For 216 (73%) patients, pharmacists felt the visits were definitely or probably useful; for 81 (27%) patients, pharmacists felt the visits were unlikely to be useful or not at all useful, generally when patients were found to be coping very well.

Number of hospital readmissions

A total of 178 emergency readmissions occurred in the control group and 234 in the intervention group (table 2). The Poisson model indicated a 30% greater rate of readmission in the intervention group (rate ratio = 1.30, 95% confidence interval 1.07 to 1.58; P = 0.009).

Secondary outcomes

Mortality data were available for 829 (97%) patients. Fewer deaths occurred in the intervention group (49 v 63). 2 shows the Kaplan-Meier survival graph. The hazard ratio for the intervention group compared with the control group was 0.75 (0.52 to 1.10; P = 0.14). Data on residential or nursing home admissions were available on fewer patients, as these were collected by telephone (585, 68%). 3 shows that fewer control patients than intervention patients were admitted to residential or nursing homes, but again these differences were not statistically significant.

View larger version:

• In a new window

• Download as PowerPoint Slide

Fig 2

Survival analysis over six month follow up period for patients in the HOMER trial

Quality of life data

Change in utility scores could be calculated for 308/380 (81%) surviving intervention patients and 284/362 (78%) surviving control patients. Both groups' scores decreased over the six month follow up period, but the changes were not significantly different between the groups (table 4). Scores on the visual analogue health scale also fell; the difference of 4.1 (95% confidence interval 0.15 to 8.09) units in favour of the control group was statistically significant (P = 0.042).

Primary care data

We included 165 patients from 12 practices in this analysis (84 intervention, 81 control). General practitioners carried out 204 home visits in the intervention group and 125 in the control group, a difference of 43% (rate ratio = 1.43, 1.14 to 1.80; P = 0.002). No statistically significant differences occurred between the groups in attendance at general practices or prescription items received.

Validity of trial

This trial was large, involving more than 850 patients. The entry criteria ensured a broad sample of elderly people discharged from hospital, which, together with the relatively high participation rate and the large number of pharmacists involved, means that the generalisability of these results should be high. Follow up of the main outcome was good—only 3% of participants withdrew or were lost to follow up. Hospital admission data were provided by downloads from hospital episode statistics, which are unlikely to have introduced bias. Quality of life data were provided by almost 80% of patients. Slightly more intervention patients than controls provided these at six months (81% v 78%), which could have introduced a bias. At three months, however, when response was almost equal, no between group differences were apparent on either quality of life measure. Overall, the internal validity of this study seems high, although it should be noted that follow up was only for six months. Longer term benefits might have emerged had follow up been longer. However, we can envisage no theoretical grounds for a delayed benefit emerging long after the intervention had ceased.

These results seem counterintuitive. The reviews were intended to help patients to understand their drugs better, reduce adverse drug reactions,3 and help patients to adhere to their drug regimens. We did not measure any of these intermediate outcomes directly. However, we hypothesised that the intervention would lead to a reduction in readmissions through these effects and seemed unlikely to cause harm. The pharmacists providing the interventions considered that in more than 70% of cases their interventions had probably or definitely been of value. Thus, we are left to explain how these interventions increased hospital admissions.

Possible explanations

Given the high internal validity of this study, its results are unlikely to be explained by bias or confounding. We cannot exclude the possibility of a type I error (chance). If, however, we consider the findings to be causally related, three possible explanations should be considered. The first is that pharmacists did help patients to understand their conditions better. This could have led patients to recognise warning signs earlier and promoted better help seeking behaviour, leading to more hospital admissions. This positive view is weakly supported by the non-significant decrease in deaths observed. Indeed, by preventing deaths of frailer patients our intervention may have increased admissions and general practitioner home visits—so-called “competing outcomes.”

Two less favourable interpretations are possible, however. Previous studies have shown that interventions of this type by pharmacists tend to increase adherence to prescribed drugs.16^–19 Our patients were prescribed large numbers of daily drugs (mean = 5.9/day). By encouraging better adherence, our pharmacists may have precipitated iatrogenic illness that previously had been avoided. Given the very elderly group studied, we did not collect data on adherence as we did not wish to burden patients with extra questionnaires beyond our quality of life assessment. This means that we cannot be sure that adherence improved. Indeed, another study using pharmacists to improve discharge from hospital found no effect on adherence.20

Finally, by visiting our patients at home and spending reasonably long periods of time there, we may simply have added to the complexity of their care. This may have increased anxiety and confusion or dependence on health services. The intervention group's scores on the visual analogue health scale fell more markedly than those of the control group. This suggests that they viewed their overall health as having worsened and may support a view that our intervention made patients focus more on their problems.

Evidence from other trials

Since this trial started, three large UK studies of community based medication review in elderly people have been published.20^–22 Two showed non-significant decreases in admissions,20 21 whereas the other showed a non-significant increase in admissions.22 These results, in combination with ours, indicate that it cannot be assumed that community based medication review necessarily reduces admissions, despite evidence from overseas.11 Although our finding on mortality is potentially encouraging, UK results on this outcome are equivocal, with mixed, non-significant results.20^–22

Evidence in the field of medication review is growing rapidly. More positive results seem to have resulted from interventions focused on diseases such as heart failure.10 Alternative models to home visiting also exist, including reviews within a general practice surgery.22 Such a review has the advantage of access to full patient records, potentially allowing a more thorough clinical medication review than was possible within this study. In addition, it allows pharmacists to build up a close working relationship with general practitioners, which is vital if the pharmacists' recommendations are to be followed.

Conclusions

Our trial suggests that home based medication review for older people recently discharged from hospital increased, rather than decreased, hospital admissions. It also seemed to worsen patients' quality of life compared with controls. The exact mechanism for this result is not apparent. Patients may have adhered better to their drugs, with a resultant increase in side effects or drug interactions. Alternatively, our intervention may have provoked better understanding and help seeking behaviour. Either way, a growing body of evidence suggests that further research is necessary to elucidate the most effective form and detailed effects of medication review. The recommendation in the national service framework for older people that this should be widely introduced in primary care seems to lack a clear evidence base.7

Amendment

This is Version 2 of the paper. In this version, the statistical power is given as 90% [rather than 80% as in the previous version] and the trial ISRCTN is included [omitted from the previous version].