The authors L.K., E.L.H. and G.L. categorised the interventions independently and then by consensus. The interventions were activities; music therapy (protocol-driven); sensory interventions (all involved touch, and some included additional sensory stimulation such as light); light therapy; training paid caregivers
in person-centred care or communication skills (interventions focused on improving communication with the person with dementia and finding out what they wanted), with and without supervision; dementia care mapping; aromatherapy; training family caregivers in behavioural management therapies or cognitive–behavioural therapy (CBT); exercise; cognitive stimulation therapy; and simulated presence therapy.
Agitation level
We separated studies according to the inclusion criteria of participants in terms of level of symptoms of agitation: 1, no agitation symptom necessary for inclusion; 2, some agitation symptoms necessary for inclusion; 3, clinically significant agitation level; 4, level unspecified. We used the usual thresholds: a score above 39 on the Cohen-Mansfield Agitation Inventory (CMAI),20 and a score above 4 on the Neuropsychiatric Inventory (NPI) agitation scale,1 to denote significant agitation.
Statistical analysis
We decided a priori to meta-analyse when there were three or more RCTs investigating sufficiently homogeneous interventions using the same outcome measure, but no intervention met these criteria. To facilitate comparison across interventions and outcomes, where possible, we estimated interventions’ standardised effect sizes (SES) with 95% confidence intervals.21 In some studies the outcome was measured and reported at several time-points during the intervention. We used data from the last time-point to estimate the SES, since individual patient data were not available to incorporate repeated measures in the calculation. We also recalculated results for studies not directly comparing intervention and control groups but reporting only within-group comparisons and with one-tailed significance tests, so some of our results differ from the original analysis.