Randomized, blinded, two-armed, parallel group superiority trial (Figure ).
Flow chart of the SMART trial design.
The trial has been approved by the Committee on Biomedical Research Ethics of The Capital Region in Denmark (H-1-2011-025) and the Danish Medicines Agency (EudraCT 2010-024065-46) and is registered at ClinicalTrials.gov (NCT01431092). The trial will be conducted in accordance with the latest version of the Declaration of Helsinki [26
] and the International Conference on Harmonization (ICH) - Good Clinical Practice (GCP) guidelines for clinical trials [27
]. The trial will be monitored by the GCP Unit at Copenhagen University Hospital.
Participants will be recruited from outpatient clinics under the Mental Health Services in the Capital Region of Denmark. In case of recruitment difficulties the inclusion area can be extended to Region Zealand.
• Patients diagnosed with schizophrenia or schizoaffective disorder (ICD-10 (International Classification of Diseases, 10th edition) criteria for schizophrenia (F20) or schizoaffective disorder (F25) must be fulfilled at inclusion or previously as documented by chart review; fulfillment of relevant DSM-IV-TR (Diagnostic and Statistical Manual of Mental Disorders, 4th edition, text revision) criteria will also be registered).
• Attached as outpatient to the Mental Health Services of the Capital Region of Denmark or Region Zealand (in case of recruitment difficulties).
• Treated with the same antipsychotic drug for at least 3 months before inclusion (change of dose, antipsychotic polypharmacy, and prescription/discontinuation of add-on drugs are allowed but the basic antipsychotic drug must have remained the same).
• Benzodiazepine derivative administration (at least one of: chlordiazepoxide, diazepam, clobazam, clonazepam, flunitrazepam, nitrazepam, bromazepam, alprazolam, lorazepam, lormetazepam, oxazepam, or triazolam) for at least 3 months before inclusion. Treatment with benzodiazepine-related drugs (zolpidem, zopiclone, zaleplon) is not sufficient to be included in the trial but these drugs will be included in the taper process.
• Age 18-55 years (both inclusive).
• Fertile women: Negative pregnancy test at baseline and the use of safe contraceptives (intrauterine devices or hormonal contraception) throughout the trial period and 1 day after withdrawal of trial medication. This does not apply to sterile or infertile participants, i.e. surgically sterilized or post menopausal women.
• Written informed consent.
• Currently under treatment for abuse of alcohol or drugs.
• Known aggressive or violent behavior.
• Known mental retardation, pervasive developmental disorder, or dementia.
• Epilepsy, terminal illness, severe co morbidity, or unable to understand Danish.
• Allergy to compounds in the trial medication (melatonin, lactose, starch, gelatin, and talc).
• Hepatic impairment (known diagnosis).
• Pregnancy or nursing.
• Lack of informed consent.
Experimental intervention and comparison
All trial participants are gradually tapered off their usual benzodiazepine treatment (including benzodiazepine-related drugs) following general national treatment guidelines (Institute for Rational Pharmacotherapy, IRF) [28
], i.e. 10-20% dose reduction every week or every two weeks. Gradual taper is preferred to sudden discontinuation both nationally and internationally and this approach is also supported by a Cochrane review [29
]. The largest (percentage) dose reduction should be planned during the beginning of the tapering process where it is most easily tolerated. When treated with short-acting benzodiazepine derivatives (elimination half-life 24 hours or less) the participants are offered to shift to diazepam (run-in phase). Diazepam is preferred for tapering because of its long elimination half-life (and therefore a stable plasma concentration) plus the availability of low dose tablets. If preferred by the individual participant the tapering process will be conducted with his/her usual benzodiazepine derivative. Participants driving a car will not be offered to shift to diazepam because of the long elimination half-life. Participants will be advised not to drink alcohol during the trial period. The discontinuation plan will continuously be adjusted according to the individual participant. If necessary, the discontinuation can be (temporarily) paused. There is no upper limit of the duration of staying on the same dose. The continuous contact with the participants will follow general recommendations in this area based on thorough clinical experience [30
]. However, most current recommendations stem from general practice [33
Trial medication (Circadin® 2 mg and matching placebo, respectively) begins simultaneously with the tapering process. The participants are instructed to ingest the trial medication 2 hours before bedtime (between 9 and 11 p.m.) following a light meal. The participants are randomized and given the trial medication once they are ready to begin tapering off their benzodiazepine(s), i.e. after baseline assessments and after possible shifting to long-acting benzodiazepine (if the latter is relevant). The participants are treated with the trial medication during the entire trial period including during the follow-up assessment after 6 months, irrespective of their final dose of benzodiazepine. Hereafter, the trial medication is abruptly discontinued.
Any co medication is allowed. We aim to keep the co medication constant during the trial period but any necessary changes (from a clinical point of view) are allowed and will be controlled for in the analysis if necessary. It is possible that the withdrawal process will reveal symptoms and signs (e.g. anxiety disorder, depression) where treatment with other non-dependence-producing medication is indicated (e.g. antidepressant drugs) [30
]. The participants will continue in the trial despite such medication changes.
Outcome measures and assessments
Primary outcome measure
• Benzodiazepine (including benzodiazepine related drugs) dose at 6 months follow-up.
Secondary outcome measures
• Pattern of benzodiazepine dose over time (see statistical methods).
• The fraction of participants who has completely discontinued benzodiazepines 6 months after initiating trial medication.
• Psychophysiology: Selective attention expressed as P300 amplitude [34
]. Pattern of P300 amplitude over time is analyzed.
• Neurocognitive assessment: Brief Assessment of Cognition in Schizophrenia (BACS) composite score [36
]. Pattern of BACS score over time is analyzed.
• Sleep evaluation (one night polysomnography (PSG) [37
] conducted at home): Sleep efficiency at 6 months follow-up.
To evaluate if this single night PSG is representative (irregular sleep patterns are expected), supplementary wrist actigraphy [38
] is performed for 3 consecutive days and nights.
• Subjective assessment of sleep quality: Pittsburgh Sleep Quality Index (PSQI) global score [39
] at 6 months follow-up.
• Benzodiazepine withdrawal symptoms: Benzodiazepine Withdrawal Symptom Questionnaire (BWSQ-2) [40
]. Pattern over time is analyzed.
Other variables measured to evaluate adverse events/adverse reactions of the trial medication
• Clinical laboratory tests: Fasting blood glucose, fasting lipid panel, haematology, serum chemistry (sodium, potassium, creatinine, calcium, alanine transaminase, alkaline phosphatase, and international normalized ratio). An abnormal value (outside the reference interval) will only be registered as an adverse event if it was not present at baseline and if an intervention is required to correct the value (e.g. initiating treatment, referring the participant to further examinations).
• Physical examination including blood pressure, weight, height, waist circumference, and electrocardiogram. An abnormal result will only be registered as an adverse event if it was not present at baseline and if an intervention is required to correct the abnormality.
• Adverse events (AEs), serious adverse events (SAEs), and suspected unexpected serious adverse reactions (SUSARs) are registered at 2, 4, and 6 months follow-up or whenever occurring. AEs are defined as any adverse change in health occurring during the trial reported by the participants on request. Date of occurrence and duration of each AE is registered along with a clinical evaluation of its possible relation to the trial medication. AEs are not registered after the trial period (i.e. the last visit for each participant). A SAE is defined as any untoward medical occurrence that results in death, is life-threatening, requires inpatient hospitalization or prolongation of existing hospitalization, results in persistent or significant disability/incapacity, is a congenital anomaly/birth defect, or requires intervention to prevent permanent impairment or damage. SUSARs are unexpected SAEs judged to be related to the trial medication. Preplanned hospitalizations during the trial period and symptoms undoubtedly attributable to benzodiazepine withdrawal will not be registered as AEs.
Other variables measured at baseline to characterize the participants
• Age, sex, diagnosis, age at illness onset, illness duration, and co morbidity.
• Sociodemographic characteristics (education, job, marital status, family, housing), ethnicity, and handedness.
In addition, we measure the plasma concentration of benzodiazepines at 6 months follow-up to confirm the compliance of the participants reporting complete benzodiazepine discontinuation. Any co medication is registered at baseline and at 2, 4, and 6 months follow-up.
Baseline assessments are performed after inclusion and before run-in (if relevant) and randomization. Follow-up assessments are performed 6 months after initiating withdrawal and trial medication. Several of the assessments are also performed at 2 and 4 months (See Table for summary of data collection).
Central randomization is performed by the Copenhagen Trial Unit (CTU) with computer generated, permuted randomization allocation sequence with block size unknown to the investigator. The investigator (or other research staff) will call the CTU and provide a personal pin code, participant civil registration and identification number, and the value of the stratification variable of benzodiazepine dose (low (≤15 mg diazepam equivalents) versus high (>15 mg diazepam equivalents)) at baseline. Then the randomization will be announced as a trial medication package number.
Trial participants as well as trial staff are blinded to the allocated treatment. The blinding will be maintained by using matching placebo, and an independent unit to perform the randomization and do the packaging and labeling of the trial medication. Both Circadin and placebo are encapsulated in lactose containing gelatin capsules to optimize the blinding. CTU holds the randomization code which will not be broken until all data are registered, all analyses finished, and conclusions drawn [42
]. The randomization code will only be broken during the trial period in case of emergency if the investigator decides that knowledge about the trial medication will affect the treatment of a SAE or in case of a SUSAR. Blinded data will be handed over to the CTU, which will be in charge of double data entry and which will conduct the statistical analyses blinded to the intervention.
The efficacy of the intervention with respect to benzodiazepine dose, sleep efficiency, and PSQI at 6 months follow-up is analyzed using the univariate general linear model with the outcome measure (6 months value) as the dependent variable and the indicator of intervention and the baseline value as the independent variables. If the assumptions of the model cannot be fulfilled either directly or after transformation a non-parametric method will be used.
The efficacy of the intervention with respect to the fraction of participants who completes the benzodiazepine withdrawal (follow-up dose: 0) is analyzed using a logistic regression model where logit(p) is the dependent variable, p is the probability of completing the withdrawal, and a binary intervention indicator is the independent variable [43
The effect of the intervention on the pattern of benzodiazepine dose, P300, BACS, and BWSQ-2 over time is analyzed in a model describing outcome measure as a function of time (2, 4, and 6 months after withdrawal was initiated):
Without the intervention indicator the model can be described as:
where int is the intercept, t is the independent variable time, baseline is the baseline value of the outcome measure, and a through d are the coefficients of the function. The model is expanded to include a binary indicator of intervention to test the overall effect of intervention and its main effect and interaction with time and time squared. In the analysis of the pattern of benzodiazepine dose, P300, BACS, and BWSQ-2 over time, we will apply a mixed model with repeated measures (MMRM). Using the Akaike's criterion, we will determine which co-variance structure fits the data the best: an unstructured (un), a compound symmetric (cs), or a first order autoregressive AR(1) with and without variance heterogeneity. Time is included as a continuous variable and sequential hypothesis testing will be applied [44
Missing values will not lead to bias when the MMRM is applied if data is missing at random which is not a very rigorous condition. Provided significant results are obtained using any of the above described methods, the potential influence of values not missing at random will be assessed for most outcome measures in a sensitivity analysis. Let BEST be the group where a significant and beneficial effect has been observed. Missing values will be replaced by optimistic ones in the other group and by pessimistic ones in BEST. Optimistic and pessimistic values are inferred as follows:
• Dose after 2, 4, and 6 months: Pessimistic values will be imputed as follows: Data will be reviewed from left to right setting the 2 months value, if it is missing, to equal the baseline value. If the 4 or 6 months value is missing it will be set to equal the preceding value (which may be an imputed one). When imputing optimistic values, the data will be reviewed from right to left setting the 6 months value to 0 if it is missing. A missing 4 months as well as a missing 2 months value will be imputed with the subsequent value (which may be an imputed one).
• Withdrawal completion: "Yes" is an optimistic value and "no" is a pessimistic value.
• Continuous outcome measures only assessed after 6 months: An optimistic value equals the highest value in the whole sample and a pessimistic value the lowest value, if increasing the outcome measure is considered beneficial. Vice versa if decreasing the outcome measure is considered beneficial. Using min and max of delta of the whole material pessimistic and optimistic baseline values are constructed by extrapolating from the 6 months value.
• For other analyses using the above mixed model the sensitivity analysis will include a worst case analysis following the same technique as previously published [45
We do not plan any interim analysis because we do not foresee any circumstances that should lead to closing the whole trial prematurely.
Sample size estimation
Data directly illustrating the distribution of benzodiazepine dose in patients with schizophrenia after going through a discontinuation trial is not available in the literature. In the following we apply our own unpublished data from a comprehensive chart review including registration of benzodiazepine dose in 99 consecutive outpatients diagnosed with schizophrenia.
We assume that the trial participant, after having followed a discontinuation plan for a certain period of time, either has completed the withdrawal (intake of benzodiazepines stopped) or has paused the gradual taper and continues on a reduced dose. During the taper off, we assume that the dose decay curve approximately follows an exponential function, because the discontinuation plan assumes a constant withdrawal rate. If we know the dose at baseline, the rate of withdrawal (slope), and the time of stopping the withdrawal (t), we can calculate the final dose for each participant as
We further assume that the participant must follow the discontinuation plan for 25 weeks to have completed the discontinuation. We wish to be able to detect a statistically significant difference of minimum 8 weeks between the two intervention groups with regard to average duration of adherence to the discontinuation plan with 90% probability. We furthermore assume that the participants in the control group on average will be able to follow the discontinuation plan for two months (8 weeks) while the participants in the melatonin group are assumed to be able to follow the discontinuation plan for four months (16 weeks). The slope is assumed to vary between 10 and 20% per week in both groups.
The distribution of benzodiazepine dose in the melatonin and placebo group is assumed to equal the distribution found among our previous sample of outpatients. The distribution of the final dose under the given assumptions is therefore found as follows:
For each intervention group the slope is assumed to follow an even distribution between 0.1 and 0.2. A slope following this distribution is allocated randomly to each participant. The time passing before a participant stops the withdrawal is randomly allocated to each participant as time is normally distributed with a mean of 8 weeks and a standard deviation (SD) of 2 weeks in the placebo group and a mean of 16 weeks and a SD of 2 weeks in the melatonin group. Finally, the dose at follow-up is calculated for each participant.
Table shows the mean and SD of dose at follow-up in each intervention group according to this model. The distributions are skewed and far from normal. They are normalized with a square root transformation, but the variances are still significantly different (see Table ).
Simulated data of benzodiazepine dose (mg diazepam equivalents) at follow-up (in each intervention group) to estimate the sample size (see text).
The larger of the two SDs (0.73) is used to compensate for the inaccuracy of the estimates. With alfa = 0.05, beta = 0.1, and delta = 1.51 - 0.91 = 0.6, we estimate a sample size for each group of 38. We round up to 40 per intervention group.
As evident from the statistical analysis of the secondary outcomes, there might be other explanations for a possible effect of melatonin, e.g. an accelerated rate of discontinuation (increased slope) in the melatonin group. But it seems reasonable to suggest a net effect, which is approximately equivalent to the one gained by improving the length of withdrawal with 2 months. Furthermore, the baseline doses in the experimental group may show to be somewhat higher than among our previous outpatient sample. By adding 5 to all the baseline doses and repeating the analysis, we get delta = 0.8 and SD = 0.65 which equals a sample size of 18 in each group (the non-transformed means of follow-up dose in the two groups are 3.92 and 1.39).
Consequently, a total sample size of 80 patients must be regarded as a conservative estimate.