Overall RAT scores
Typical scores for the participants in our study were better than those in the prior studies that employed the instrument in elderly hospitalised patients (with or without stroke) and those with rheumatoid arthritis [[13–15]; Kusu-Orkar TG, unpublished data; Gallagher C, unpublished data; Meland E, unpublished data; Yang F, unpublished data]. These patients had worse total scores (>10) possibly because elderly hospitalised patients tend to be more debilitated than community dwelling patients. Additionally, stroke is commonly associated with both cognitive and physical impairment and rheumatoid arthritis is associated with significant dexterity problems. Though our scores were closer to those of Wen [unpublished data], who reported a median (IQR) total risk scores in patients recruited in community pharmacies of 2 (2, 3), they were slightly worse. This may be due to differences in study participant recruitment. The 6 (16%) managed participants, who had significantly worse total risk scores and cognitive risk sub-scores than the other participants, may have contributed to worse overall scores in our study. In addition, in the study by Wen [unpublished data], community pharmacists made a prior judgement on the ability of the participant to handle the interview as an inclusion criterion. As such, a degree of selection bias was unavoidable as the pharmacist could have selected participants who were 'able' physically and cognitively to participate in the interview, translating into better scores. In our study, a majority (29, 78.4%) of participants were recruited as they came into the pharmacies to collect their prescriptions, skipping the need for the pharmacists' judgement prior to the interview.
Scale reliability and validity
Internal consistency was used as a measure of scale reliability. The Cronbach's alphas obtained for the 13-item scale and the physical risk sub-scale were above the indicative value of 0.70 for acceptable internal consistency . If used alone, the cognitive risk sub-scale had a value lower that 0.70, indicating that it might not be reliable. However, internal consistency is improved by removal of motivation item 1, "Do you think your medicines are necessary for your health?" This item might be considered for removal if the cognitive risk sub-scale is to be used alone. Dexterity items "Can open 48 ml amber plastic with screw cap" and "Can open 100 ml glass bottle with normal cap" that were constant in our study sample should probably be considered for removal, as they carry no discriminatory value. This is consistent with the fact that all participants in our study successfully completed these tasks.
As with one previous study [Wen PC, unpublished data], a strong negative correlation between the instrument scores and the community pharmacists' assessment shows that there is agreement between the community pharmacists' assessments and the instrument. The ROC analysis (the AUC and 95% CI obtained) indicates that the full 13-item scale has good discriminatory capability. However, the wide 95% CI for the AUC for sub-scales with the lower bounds tending towards 0.5 imply that in some patients, the sub-scales used alone may not have good discriminatory capability. We found that a cut-off (sensitivity, specificity) total risk score of ≥4 (65%, 100%) would be the best value for screening purposes. The cut-offs obtained in our study were much lower than those in earlier studies among hospitalised elderly patients [13, 14] and those with rheumatoid arthritis . These conditions present unique cognitive and dexterity challenges to patients. In addition, while we used a community pharmacist's assessment as our criterion standard, these studies utilised a Global Clinical Assessment, which is a more commonly used assessment tool in the hospital setting. Though comparable, our 13-item cut-off is slightly higher than that of Wen [unpublished data] because our participants generally had worse scores.
Differences in scores between managed patients and other patients only resulted from the cognitive risk sub-scale. Managed participants had good physical risk scores, similar to other community dwelling elderly patients, implying that they are able to manage administration tasks just as well. Inability to self-manage medicines appears to have been the result of cognitive impairment (such as forgetfulness), which has been associated with unintentional non-adherence in the elderly [18, 19]. Our results suggest that patients with poor scores on the cognitive domain of the instrument should be considered for social care support with medication management.
Only total risk score and cognitive risk sub-score were significantly worse for multi-compartment compliance aid users compared to non-users, suggesting that multi-compartment compliance aids should be considered in elderly patients with cognitive impairment, as a memory prompt, on condition that they are physically capable of self-administration. In this way, a patient who scores poorly on both physical and cognitive sub-scales would not be considered a candidate for multi-compartment compliance aid use and alternative means of medication support would be sought. However a patient who has poor scores on the cognitive sub-scale and good scores on the physical sub-scale would benefit from multi-compartment compliance aid use.
There are strengths inherent in the use of this instrument. It is a multidimensional instrument, covering several factors associated with the inability of patients to self-manage their medicines. The short time (15 to 20 minutes) it takes to conduct the assessment should be a welcome positive for practitioners as well as patients. Item 1, in the cognitive domain is based on a validated measure of cognitive function, the Abbreviated Mental Test . The physical domain is strong as all items are based on completion of tasks related to actually taking medicines. In practice, it could help identify real physical limitations that patients may have in handling their medicines for example, failure to read standard size labels or opening child resistant caps, such that tailored interventions may be designed, for example, use of larger sized labels, replacing child resistant caps with ordinary screw caps. Generally, the instrument is easy to use and does not require pharmaceutical expertise to administer effectively. With a little training, social carers could use it to target appropriate support to those in need. They could use the instrument to bring both physical and cognitive limitations to medication management to the attention of pharmacists and other healthcare workers. It can also be used at the point of discharge from the hospital to help guide decisions about the most appropriate discharge destination as well as medication management support for individual patients.
Our study was limited by a small sample size, which could have compromised our ability to make definitive statements on the psychometric properties of the instrument. In addition, the comparison of scores between managed and non-managed patients is limited by the fact that there were only 6 managed patients. Community pharmacists set up opportunist interviews and appointments. As such we did not gather the data how many they approached to make the 31 community participants; and we could only trace 6 managed participants through the WGCHCP records. And because we neither used a pre-validated instrument for the community pharmacists' assessments, nor conducted inter-rater reliability across pharmacists, further studies are needed to validate this tool against more robust instruments. We used a conservative definition of polypharmacotherapy (4 or more medicines). Our results showed that the median (IQR) number of medicines used by our patients was much higher, 8 (6-10), ranging from 4-21. Our study therefore might have been biased by lower risk patients in terms of the number of medicines taken. Although our study demonstrated an association between cognitive risk sub-scores and both multi-compartment compliance aid use and social care support with medicines management, there are no differences between patients in the physical risk domain of the tool. Therefore we are unable to determine, using the tool, exactly which patients would be candidates for either social care support or multi-compartment compliance aids. In spite of this, we found that typical scores in "managed" participants are worse than those who use multi-compartment compliance aids (6.5 versus 4.0). We can therefore discern that patients with much worse scores may be considered for social care support, as opposed to multi-compartment compliance aids.
There are limitations inherent in the tool itself. The questions concerning motivation ("do you think your medicines are necessary for your health?" and "how confident are you about taking your medicines on your own?") rely on the participant providing an honest answer. The question "what medications do you take at the moment?" assesses people's knowledge of their medicines. There are various dimensions to a patient's knowledge of their medicines, and the evidence linking knowledge of medicines to adherence and ability to self-manage medicines is at best conflicting .