Study participants were recruited from eight specialist psychogeriatric nursing homes and three private nursing homes in Melbourne, Australia, between 2009 and 2011. The following criteria were required for inclusion: (i) a rating by nursing staff of at least mild dementia on the Clinical Dementia Rating scale ; (ii) a physically agitated behaviour that occurred at least several times each day in daylight hours, at times other than during nursing care, to a degree that required staff intervention; (iii) an assessment by the nursing staff, visiting medical practitioner and/or psychiatrist that the behaviour was not due primarily to pain, physical illness, depression or psychosis; (iv) residence in the facility for at least three months, and (v) consent to study participation by the next of kin or guardian. Exclusion criteria included: (i) an acute, life-threatening illness; (ii) a variable psychotropic medication regime, and (iii) a medical condition that precluded the use of topical oils. No attempt was made to distinguish one type of dementia from another in this severely disabled group.
The trial was approved by all relevant committees including the Monash University Human Research Ethics Committee. Written consent was obtained from the next of kin or legal guardians of all participants.
Four samples of 100% pure lavender oil (Lavandula angustifolia) were submitted for electrophysiological analysis to the Department of Pharmacology and Toxicology, Otago University, New Zealand, using established methods . A sample sourced from the Ukraine out-performed others in reducing synaptic activity and membrane excitability in cultured rat embryo pyramidal cells and was therefore selected for use in this trial (Prof. George Lees, personal communication). On preliminary testing, 1 ml of 30% lavender in jojoba oil proved to be the maximum feasible volume and concentration that could be delivered topically. Jojoba oil, which had no electrophysiological activity, was used as a control.
The full protocol of this randomized, placebo-controlled, single-blind cross-over trial has been described previously . Participants were allocated randomly by the project manager using an Excel random number generator to either a lavender or control study condition with no pre-set blocking. They were later crossed over to the other condition. To capture any cumulative effects, each of the two experimental conditions comprised three exposures over a one week period with a four-day washout period between them. Treatments were administered at times when nursing staff reported that the selected physically agitated behaviour was most likely to be present, excluding times of personal nursing care. To minimise confounding by intercurrent illness and medication changes, the trial was limited to a two-week period and nursing and medical staff were asked not to alter participants’ psychotropic medications if possible.
The lavender and control oils were stored in identical vials, marked as A or B. Only a single researcher, who had no other involvement in the study, was aware of the allocation. A nursing staff member then massaged 1 ml of either the lavender or control oil into both forearms for one minute each, giving a total of 2 mls per session. Since lavender plasma levels peak after 20 minutes, and are barely detectable after 90 minutes, participants were observed for 30 minutes before and 60 minutes after application . To maintain observer blinding, nurses applying the oil wore a nose clip and research assistants, who completed the observations, applied a mixture of essential oils to their upper lip to disguise lavender’s fragrance. It was not considered practicable or desirable to attempt to blind participants, all of whom had marked cognitive impairment, to the treatment condition.
A discretely positioned research assistant recorded if the selected target behaviour was present (one point) or absent (zero points) during each minute giving a maximum score of one point per minute of observation. Results were computed as mean counts over the three 30-minute observation periods. Inter-rater reliability was high with an average Kappa of 0.98 over five training sessions. This time-sampling method has been used successfully in our previous studies .
Observers also used the Philadelphia Geriatric Center Affect Rating Scale to record the predominant type of affect (pleasure, contentment, interest, anger, sadness, anxiety/ fear) evident during every minute . The predominant affect was rated as one and all others were rated zero giving a maximum score of one point per minute of observation. Results were averaged as described above. The mean Kappa for inter-rater reliability was 0.92. For reasons of brevity, results for pleasure, contentment and interest are categorized here as positive affects and those for anger, sadness and anxiety/fear as negative affects.
Other measures included the Mini-Mental State Examination (MMSE) and the Cohen-Mansfield Agitation Inventory (CMAI) [26, 27]. To identify a suitable behaviour for use as an outcome measure, the CMAI was completed at baseline by nursing staff who rated physically agitated behaviours (e.g. pacing) on a seven-point scale ranging from “occurs never” to “occurs several times an hour” in the previous fortnight.
Sample size was calculated with respect to the primary outcome measure of physically agitated behaviour for a two-sided hypothesis test with a Type I error rate of 0.05 and power set at 90%, based on data from an earlier repeated measures study of simulated family presence versus music therapy in a similar population and setting . In that study, simulated family presence had an effect size of 0.45 relative to the control condition. However, given the mixed evidence supporting lavender, it seemed prudent to apply the smaller effect size of 0.32 for music therapy. It was estimated that 77 participants might be required to detect a difference of this magnitude. At the same time, there were 18 data points for every participant (6 applications x 3 summary behaviour and affect counts per application) which generated considerable statistical power for only a limited number of analyses.
The behaviour and affect scores were over-dispersed relative to Poisson expectations and were therefore modelled using random effects negative binomial regression, taking account of participants’ age, gender and dementia severity. The results of the analyses are shown as mean scores and incidence rate ratios for completed cases. Ratios over one signify an increase in a mean value, and ratios below one signify a decrease in that value, with respect to a reference condition. Ratios are displayed for treatment (lavender versus the control reference condition), time (post-exposure versus the pre-exposure reference condition) and treatment-time interaction.