Study design and participants
This is a single-center, randomized, double-blinded clinical trial conducted at a university hospital research department in the capital region of Denmark. Participants eligible for the trial were 18 years or above; had a diagnosis of schizophrenia, schizoaffective disorder or bipolar mood disorder (and euthymic at inclusion); were treated with at least one antipsychotic drug and had a daily use of at least one benzodiazepine or benzodiazepine-like drug for a minimum of 3 months; did not present with current violent or aggressive behavior; were not diagnosed with mental retardation, pervasive developmental disorder, dementia, hepatic impairment, terminal illness, severe somatic comorbidity, or epilepsy; were able to understand Danish; and were not allergic to any compounds in the study medication. Fertile women were only included if not pregnant or nursing and if using safe contraceptives throughout the study period.
After baseline investigations, participants were randomized to prolonged-release melatonin (PRM) 2 mg or identical placebo once daily for 24 weeks, and in parallel they were instructed to gradually reduce their daily benzodiazepine usage. Participants were instructed to ingest the study medication approximately 2 h before bedtime with a small meal.
The overall aim of the trial was to investigate if PRM can facilitate discontinuation of long-term benzodiazepine usage in chronic psychiatric patients and these results have been published elsewhere  in agreement with CONSORT guidelines. Primary and secondary outcomes are described in the published trial protocol . Here we report the results of actimetric assessments in a subset of the participants during three consecutive days and nights before and after 24 weeks of benzodiazepine tapering. The presented analyses of actimetric data were not planned in details when the trial protocol was published .
Actigraphic assessment and circadian rest-activity cycle measurement
Actigraphy does not measure sleep state and thus cannot differentiate between sleep and sedentary behavior, but it is established as a reliable instrument for evaluating sleep patterns, for studying the effects of treatments to improve sleep, and in the diagnosis of circadian rhythm disorders [2, 3, 16]. A recent review confirmed the clinical usefulness of actigraphy for objective evaluation of sleep habits and circadian rhythm disturbances in psychotic disorders .
We measured actimetry for three consecutive days and nights (72 h) to evaluate rest-activity patterns with and without melatonin and before and after benzodiazepine tapering. From the American Academy of Sleep Medicine it has been recommended to use at least three consecutive days of actigraphic recording to obtain reasonable reliability of actigraphic estimates of sleep and 24 h rhythm activity variables . The results of one night polysomnography (at baseline and follow-up), which was performed simultaneously with the first night of actigraphy, will be reported elsewhere. Subjects were asked not to change anything in day or night activities and to continue their usual circadian patterns as uninterrupted as possible while actimetric recordings were done. Recordings were performed from Tuesday to Friday or from Friday to Monday.
We used Actiwatch Spectrum (Philips Respironics) which was continuously worn on the non-dominant arm for a minimum of 72 h. The Actiwatch is a small portable device, the size of a wrist-watch, which uses an accelerometer to detect and log wrist movement (Actiwatch Clinical Implementation Guide; www.actigraphy.com).
Day and night time activity data were logged at 30 s epochs, stored within the Actiware software (version 6.0.0, Respironics, Murrysville, PA, USA), and exported as activity counts per 30 s. We did not use the automatically displayed sleep variables except for wake up time, which was used for alignment of activity counts. Data were then processed to calculate activity counts per hour and activity counts per 6 h. We analyzed activity counts with wake up time (as opposed to clock time) as reference. Wake up time was used as circadian phase marker.
Actigraphic assessed circadian rest-activity cycle parameters included the interdaily stability (IS), the intradaily variability (IV), and the relative amplitude (RA). These non-parametric circadian rhythm parameters have been recommended because they more accurately describe characteristics of disturbances in the rest-activity rhythm and because they are more sensitive to change compared with alternative statistical procedures . The IS quantifies the invariability from day to day, i.e., the predictability of the 24 h rest-activity pattern. This index will be 1 for perfect IS and thus lower values reflect higher variability between the observed days. The IV reflects the fragmentation of the rhythm, i.e., the frequency and extent of transitions between rest and activity. Thus, higher values of IV reflect higher degree of fragmentation and more frequent shifts between rest and activity. The RA is calculated from the ratio of the most active 10 h period to the least active 5 h period in the average 24 h pattern and thus higher values reflect more pronounced differences between periods of rest and activity (typically between day and night time). These non-parametric variables have been used in several studies of circadian rhythm and motor disturbances in schizophrenia [4, 5].
We used SPSS version 22 for statistical analyses. All analyses were based on complete cases for actigraphy assessment at baseline and follow-up.
Activity counts were measured in 30 s epochs, converted to 60 s epochs and entered into an excel template for calculation of the three rest-activity rhythm parameters (IS, IV, and RA) across the 72 h of actigraphic assessment at both baseline and follow-up. IS was calculated as the ratio between the variance for the average 24 h pattern around the mean and the overall variance, IV was calculated as the ratio of the mean squares of the difference between all successive hours and the mean squares around the grand mean, and RA was calculated from the most active 10 h period and the least active 5 h period in the average 24 h pattern . Changes in IS, IV, and RA between the PRM and placebo group were analyzed using the univariate general linear model with the outcome measure (24 weeks value) as the dependent variable and the baseline value, intervention group, and benzodiazepine dose reduction (from baseline to end point) as independent variables. In case of non-normal distributions, a non-parametric test (Mann–Whitney U-test) was used.
For evaluation of level of activity, the activity counts for the second recorded night and day (from midnight to midnight) at both baseline and follow-up were converted to number per hour and number per 6 h. We chose the second night and day due to the best data quality compared with the first and third night and day. Mean activity count per hour was plotted against time with actigraphy wake up time as reference for the total sample. Change in activity counts per 6 h from baseline to follow-up in the total sample as well per group was analyzed using paired samples t-test.