with the authors, the trials were not really randomized because a pairing procedure was performed by the investigators concurrently with their decision about inclusion: in particular, allocation was clearly not concealed. Furthermore, the fact that many outcome measures are reported without a priori establishment of a primary outcome measure (e.g. in the form of a calculated combined score) makes the possibility of selective reporting a serious drawback in study quality. Data are not reported in a sufficient degree of detail to allow us to construct combined scores and perform calculations on these, nor were the authors able to supply the additional data this would require. Scherder et al. (Scherder 1995b) have also published reports on a third very similar trial in which the intervention was TENS alone. Smallwood 2001: Randomized controlled trial assessing the longterm effect on behaviour disturbance of a combination of aromatherapy massage compared with two control interventions: aromatherapy with conversation, and massage alone. Twenty-one district general hospital inpatients with a diagnosis of dementia were randomly allocated to the three arms, seven patients in each. Interventions were given twice a week for four weeks. Behaviour disturbance was assessed by sampled video recording by evaluators blinded to treatment allocation. The investigators report a trend towards a greater reduction in behavioural disturbance in the aromatherapy plus massage arm than in the aromatherapy plus conversation arm, but this reached statistical significance only when analysis was restricted to a particular subset of data, namely measurements made during one particular time period during the day. This study was excluded from the Cochrane review of ’Aromatherapy for dementia’ (Thorgrimsen 2003) because of the unmotivated use of the interaction with time of day in data analysis (which again raises issues of selective reporting). Furthermore, one dropout is reported but it is not stated how this was treated in the data analysis, and no details are given of the type of massage offered, or the duration of each treatment. The investigators did not respond to requests for additional information. Bowles 2002: CCT assessing the effect on cognitive function, dementia-related behaviour and resistance to nursing care of massage with an essential oil blend compared with control massage with plain base cream not containing the essential oil blend. Fifty-six elderly residents in a care facility who had been diagnosed with moderate to severe dementia (using the MMSE) were divided into groups matched for mobility, sex, and type and severity of demenMassage and touch for dementia (Review) 6 Copyright © 2008 The Cochrane Collaboration. Published by John Wiley & Sons, Ltd. tia. Interventions were given in a crossover fashion, with group A receiving the treatment intervention for four weeks while group B received the control; the two groups then received the alternate intervention for another four weeks. Cognitive function was assessed the day before the residents commenced essential oil treatment and the day after they finished receiving the treatment (i.e. a 28-day separation) using the MMSE (Folstein 1975). Dementia-related behaviour was assessed using some measures described in the Cohen-Mansfield Agitation Inventory (Cohen-Mansfield 1986). The frequency and severity of other behaviours such as confusion were measured via a bespoke scale (0 to 3 where 0 = no occurrence of the behaviour and 3 = behaviour occurring more than five times or continuously during the nursing shift). Resistance to nursing care was measured on a scale of 1 to 3 where 1 = person requiring verbal redirection away from the problem behaviour and 3 = person requiring the physical intervention of two nurses (ie extremely difficult to redirect). The trialists observed an increase in behaviours classified as resistance to nursing care in the massage plus essential oil blend group in contrast to the massage alone group. The report suggests this unexpected finding be reinterpreted as an improvement in alertness. However, the trial could not be included in this review, nor the results used because of the lack of randomization of participants. Risk of bias in included studies As discussed in the descriptions of excluded and included studies above, the two included studies appeared methodologically sound, except that in Eaton 1986 the randomization procedure is not described and it is not possible to determine whether allocation was concealed. Furthermore, both of these trials were very small (only 17 and 21 patients in the active treatment groups). Effects of interventions The included studies provide evidence within two limited subfields of the overall question of effectiveness of massage and touch in dementia. One author (Remington 2002) found a decrease in agitated behaviour (rated using the modified CMAI scale) which was greater in the group receiving hand massage than in the group receiving no treatment. This treatment effect was consistently found compared with baseline for measurements taken during treatment, immediately after treatment, and one hour after treatment, and it was practically identical between the three groups receiving treatment (hand massage, calming music or both). The magnitude of this treatment effect (about