- Email: [email protected]
Hypnotics cause insomnia: evidence from clinical trials
Sleep Medicine 15 (2014) 1168–1172
Contents lists available at ScienceDirect
Sleep Medicine j o u r n a l h o m e p a g e : w w w. e l s e v i e r. c o m / l o c a t e / s l e e p
Letters to the Editor
Hypnotics cause insomnia: evidence from clinical trials To the Editor: Many epidemiologic studies have found insomnia to be associated with hypnotic drug use. It is often supposed that insomnia causes people to take hypnotics. The reciprocal that hypnotics might cause insomnia has rarely been considered. Randomized placebocontrolled clinical trials have now demonstrated that hypnotics do indeed cause insomnia. Mayer et al. reported a 6-month clinical trial of ramelteon 8 mg versus parallel placebo [1]. During subsequent placebo run-out, the placebo group had PSG-recorded TST of 383.0 min ± SE 4.0, whereas the ramelteon group slept 372.9 min ± SE 4.1. The difference in means was significant (P < 0.05, one-tailed, my computation). By focusing on rebound insomnia compared to baseline (rather than to placebo) and by focusing on sleep latency, the authors claimed that “no withdrawal symptoms or rebound insomnia were detected after ramelteon discontinuation.” I would say ramelteon withdrawal insomnia (not rebound insomnia) was observed, since at the end of treatment, the placebo group was sleeping 380.1 min and the ramelteon group 381.4 min. Thus there was hardly a ramelteon benefit from which patients could rebound. A 12-week clinical trial of eszopiclone 2 mg in older adults provided another example [2]. The abstract stated “improvements occurred within the first week and were maintained for three months, with no evidence of rebound insomnia following discontinuation.” I must disagree with the authors, who included several leaders of sleep research. Since the design of the study was randomized, double-blind, and placebo-controlled in which the efficacy endpoints were the “subjective change from baseline” (eg, in sleep latency), withdrawal insomnia would not require a sleep latency worse than baseline. We see in fig. 2 of the study that at the start of placebo run-out, sleep latency suddenly increased by about 25 min in the group withdrawn from eszopiclone, becoming several SEM greater than the sleep latency of the placebo group, and since the placebo group started with slightly higher sleep latencies at baseline, the inferiority to placebo in change scores would exceed 25 min. The change-score advantage of eszopiclone during 12 weeks of drug administration was 4.7 min superiority to placebo, but the advantage of placebo during four weeks of drug withdrawal averaged 7.0 min [3]. The placebo-only advantage in sleep latency persisted throughout 28 days of placebo-runout. The eszopiclone-withdrawn group was also inferior to the only-placebo group in subjective total sleep time change for the first three days following drug discontinuation. A subset of participants in the eszopiclone study wore wrist actigraphs. The authors stated that there would be a separate report of the objective actigraphic data, but I have not located even an abstract, so it seems up to me to interpret the objective data available at http://www.clinicaltrials.gov under NCT00386334 [3]. Actigraphically, mean improvement in TST over 12 weeks was 4.1 min greater for eszopiclone than placebo, but the advantage decreased, so that by week 12, the placebo group had actually improved 0.9 min
more. After drug withdrawal, the placebo group’s TST was 12.9 min more improved than the eszopiclone group. Objectively, the disadvantage of eszopiclone during the four-week withdrawal was over three times as great as its advantage during drug administration. Similarly, placebo was 0.7 min better than eszopiclone in sleep latency reduction over 12 weeks (13.2 min superior by week 12), and placebo was 13.5 min superior to eszopiclone in sleep latency reduction during withdrawal. In a 1-year trial of suvorexant 30–40 mg using subjective outcomes [4], the investigators proposed a drug-rebound criterion requiring exacerbation worse than the insomnia at baseline. The study suffered from failed randomization, in that the group assigned to suvorexant had worse insomnia than the placebo group at baseline (less sTST, P < 0.05 and more insomnia severity, P < 0.01, two-tailed), creating a bias toward greater improvement in the suvorexant group due to regression to the mean. Failed randomization likewise biased against the placebo group after 1-year drug withdrawal. The suvorexant-withdrawn group indeed had a higher percentage of TST rebound exacerbation than the group receiving only placebo (P = 0.057), a difference which would have been significant, correcting for the failed randomization. Examining their fig. 2, we see that the group withdrawn from suvorexant had increased subjective sleep latencies during the entire eight weeks of drug withdrawal observation, that this would be so even adjusting for the failed randomization, and that there was little evidence of insomnia remission even 8 weeks after withdrawal. This study demonstrated a lasting “return of insomnia” [their words, fig. 2] in the group withdrawn from suvorexant, leaving suvorexant-treated patients worse than the group maintained on placebo throughout. Comparing month 12 of suvorexant treatment with months 1 and 2 after withdrawal, the suvorexant-withdrawn group showed a worse drop in sTST after withdrawal and a greater increase in sleep latency (P < 0.0001 all comparisons), as well as worse insomnia intensity (P < 0.0001 and P < 0.05) compared to those only treated with placebo. I must disagree with the authors that “abrupt discontinuation of suvorexant under double-blind conditions was not associated with . . . significant withdrawal or rebound insomnia” [4]. In my opinion, these three placebo-controlled trials observed withdrawal insomnia throughout 1, 4, and 8 weeks after hypnotic discontinuation, leaving patients who had received the hypnotic worse off than those randomized to placebo. The hypnotics caused insomnia, principally by prolonging sleep latency, and the harm lasted as long as observation persisted. Further research will be needed to examine if mechanisms causing this persistent posthypnotic insomnia are more neuropharmacologic or behavioral. It is disappointing that withdrawal insomnia was caused by hypnotics of such different neuropharmacologic specificities.
Conflict of interest The ICMJE Uniform Disclosure Form for Potential Conflicts of Interest associated with this article can be viewed by clicking on the following link: http://dx.doi.org/10.1016/j.sleep.2014.08.001.
Letters to the Editor/Sleep Medicine 15 (2014) 1168–1172
References [1] Mayer G, Wang-Weigand S, Roth-Schechter B, Lehmann R, Staner C, Partinen M. Efficacy and safety of 6-month nightly ramelteon administration in adults with chronic primary insomnia. Sleep 2009;32:351–60. [2] Ancoli-Israel S, Krystal AD, McCall WV, Schaefer K, Wilson A, Claus R, et al. A 12-week, randomized, double-blind, placebo-controlled study evaluating the effect of eszopiclone 2 mg on sleep/wake function in older adults with primary and comorbid insomnia. Sleep 2010;33:225–34. [3] Sunovion. NCT003386334.; 2010 [accessed 10.09.09]. [4] Michelson D, Snyder E, Paradis E, Chengan-Liu M, Snavely DB, Hutzelmann J, et al. Safety and efficacy of suvorexant during 1-year treatment of insomnia with subsequent abrupt treatment discontinuation: a phase 3 randomised, doubleblind, placebo-controlled trial. Lancet Neurol. 2014;13:461–71. doi:10.1016/ S1474-4422(14)70053-5.
Daniel F. Kripke * Viterbi Family Sleep Center, Scripps Clinic W-207, 10666 North Torrey Pines Road, La Jolla, CA 92037, USA * Address: Viterbi Family Sleep Center, Scripps Clinic W-207, 10666 North Torrey Pines Road, La Jolla, CA 92037, USA. Tel.: +1 858 554 8845; fax: +1 858 554 8492. E-mail address: [email protected]
http://dx.doi.org/10.1016/j.sleep.2014.08.001 1389-9457/© 2014 Published by Elsevier B.V.
Response to Dr. Kripke’s letter In his letter Dr. Kripke raises the question if hypnotics can cause insomnia. At first glance it seems like a contradiction in itself that hypnotics might cause insomnia due to rebound. At second glance the idea sounds challenging, because this has never been directly addressed by studies. Thinking further about this idea makes us wonder if this thought might just serve as another argument supporting the public condemnation of hypnotics. But before drawing final conclusions, methodological aspects have to be considered. One of the methodological key issues is if results during withdrawal period should be compared to baseline or to the end of treatment data. Another is the crucial question whether the assessment of rebound should be done by means or by single patient comparisons resulting in percentages of patients with rebound. Let’s discuss these issues point by point. The contradiction: A World Health Organization (WHO) study recently reported that insomnia is the most frequent co-morbidity concerning health impact [1]. It is well known that benzodiazepines initially shorten sleep latency (SL), but do not increase sleep length or duration of slow wave sleep [2]. On the other hand, total sleep time (TST) becomes shorter and worsens after withdrawal. Nevertheless benzodiazepines do not cause insomnia; they simply do not heal insomnia and can cause dependency. There is ample evidence that insomnia decreases quality of life strongly and that many patients suffer chronically without getting help (cognitive behavioral therapy [CBT] not being paid for, hypnotics being considered to be too dangerous). Would anyone with pain be left alone because opioids are dangerous and may cause addiction? Do insomniacs have to suffer until they get the ‘right’ support, if ever? There must be a balance of pro and cons that needs to be considered, which should also be done with hypnotics. The study protocols: In most short and long term studies the primary study goals were to determine the efficacy and safety of hypnotics. Very few studies have looked at rebound after the hypnotic had been withdrawn. In those studies that looked at post treatment effects 1 day to 2 months after withdrawal, the assess-
1169
ment was mostly done at one or few time points only, either with questionnaires and/or with polysomnography. Even then an immense improvement of up to 60% (TST, SL) was seen in patients and controls. There are strong patient benefits from the study visits, study inherent patient care, and considerable day-to-day variations of the evaluation of sleep quality and other sleep parameters [3,4]. Rebound versus withdrawal insomnia: Kales et al. [5] introduced ‘drug-withdrawal insomnia’ for disturbed sleep, more frequent and intense dreams, and occasional nightmares after abrupt withdrawal of drugs (barbiturates) administered in multiple nightly doses over a long time. Withdrawal insomnia “results from both psychological factors and physiological changes” [5]. In contrast ‘rebound insomnia’ consists of “a marked worsening of sleep following the abrupt withdrawal of benzodiazepine drugs administered in only single doses nightly for short periods” indicated by a significant worsening of sleep latency, wake after sleep onset, and total wake time beyond baseline [6] i.e., an increase of at least 40% of total wake time in the mean group value [7]. Later on this definition was also used for other sleep parameters including subjective assessments and daytime performance. Roehrs et al. [8] have found a difference between pretreatment sleep and sleep during discontinuation of benzodiazepines, regardless of the type of treatment with active drug or placebo, suggesting that pill discontinuation itself might cause rebound insomnia in the sense of a placebo effect. Considering the high day-by-day variation of sleep parameters in insomnia, it seems reasonable to assess both baseline sleep and possible rebound effects over longer time intervals than several days, e.g., in the study of Hajak et al. [9] on 1507 insomnia patients the rebound rates were lower on day 1 of discontinuation than on day 31. Therefore in studies using polysomnographic evaluations the comparison of one or two nights at baseline, end of treatment, and withdrawal is highly questionable to assess rebound insomnia. Dr. Kripke re-evaluated long term studies, which have very different study designs and goals, and either use subjective sleep logs, questionnaires, polysomnography or a combination. All studies he cited had different rebound definitions. The methods show statistically significant differences of SL and TST at different time points of evaluation. The ramelteon study [10] defined rebound as a mean sleep latency during placebo run-out equal to or worse than mean sleep latency at baseline (both in diaries and PSG). The suvorexant [11] study defined rebound as “exploratory endpoints in the first 3 days of discontinuation phase”. This study is the only one that presents the rebound rate (number of patients affected by rebound) in percent. The statistics in different insomnia studies use comparison of baseline to endpoint or discontinuation mean values, which raises two problems: Presenting mean values without giving the range does not allow discrimination of good from poor responders. Secondly, the number of patients from start to end diminishes up to 50%. Therefore comparing mean values at any point of the study does not allow seeing inter-individual changes of sleep parameters. Since rebound (worsening beyond baseline levels) affects single patients, it is not appropriate to use this method to assess withdrawal or rebound insomnia, but this is what is done in nearly all studies. At least the baseline mean of patients who reached the withdrawal study phase must be given. For example: independently of exactly equal TSTs in placebo and treatment groups at study start patients are lost from baseline to end of study (around 25% in both groups of the eszopiclone study [12], 4% in the suvorexant study, and 30% in the ramelteon study). Except for the suvorexant study, we do not know if the dropouts had been good or poor responders. Therefore these data must be interpreted with caution, leaving more patients in the ongoing study with mild and moderate insomnia [12] than
Contents lists available at ScienceDirect
Sleep Medicine j o u r n a l h o m e p a g e : w w w. e l s e v i e r. c o m / l o c a t e / s l e e p
Letters to the Editor
Hypnotics cause insomnia: evidence from clinical trials To the Editor: Many epidemiologic studies have found insomnia to be associated with hypnotic drug use. It is often supposed that insomnia causes people to take hypnotics. The reciprocal that hypnotics might cause insomnia has rarely been considered. Randomized placebocontrolled clinical trials have now demonstrated that hypnotics do indeed cause insomnia. Mayer et al. reported a 6-month clinical trial of ramelteon 8 mg versus parallel placebo [1]. During subsequent placebo run-out, the placebo group had PSG-recorded TST of 383.0 min ± SE 4.0, whereas the ramelteon group slept 372.9 min ± SE 4.1. The difference in means was significant (P < 0.05, one-tailed, my computation). By focusing on rebound insomnia compared to baseline (rather than to placebo) and by focusing on sleep latency, the authors claimed that “no withdrawal symptoms or rebound insomnia were detected after ramelteon discontinuation.” I would say ramelteon withdrawal insomnia (not rebound insomnia) was observed, since at the end of treatment, the placebo group was sleeping 380.1 min and the ramelteon group 381.4 min. Thus there was hardly a ramelteon benefit from which patients could rebound. A 12-week clinical trial of eszopiclone 2 mg in older adults provided another example [2]. The abstract stated “improvements occurred within the first week and were maintained for three months, with no evidence of rebound insomnia following discontinuation.” I must disagree with the authors, who included several leaders of sleep research. Since the design of the study was randomized, double-blind, and placebo-controlled in which the efficacy endpoints were the “subjective change from baseline” (eg, in sleep latency), withdrawal insomnia would not require a sleep latency worse than baseline. We see in fig. 2 of the study that at the start of placebo run-out, sleep latency suddenly increased by about 25 min in the group withdrawn from eszopiclone, becoming several SEM greater than the sleep latency of the placebo group, and since the placebo group started with slightly higher sleep latencies at baseline, the inferiority to placebo in change scores would exceed 25 min. The change-score advantage of eszopiclone during 12 weeks of drug administration was 4.7 min superiority to placebo, but the advantage of placebo during four weeks of drug withdrawal averaged 7.0 min [3]. The placebo-only advantage in sleep latency persisted throughout 28 days of placebo-runout. The eszopiclone-withdrawn group was also inferior to the only-placebo group in subjective total sleep time change for the first three days following drug discontinuation. A subset of participants in the eszopiclone study wore wrist actigraphs. The authors stated that there would be a separate report of the objective actigraphic data, but I have not located even an abstract, so it seems up to me to interpret the objective data available at http://www.clinicaltrials.gov under NCT00386334 [3]. Actigraphically, mean improvement in TST over 12 weeks was 4.1 min greater for eszopiclone than placebo, but the advantage decreased, so that by week 12, the placebo group had actually improved 0.9 min
more. After drug withdrawal, the placebo group’s TST was 12.9 min more improved than the eszopiclone group. Objectively, the disadvantage of eszopiclone during the four-week withdrawal was over three times as great as its advantage during drug administration. Similarly, placebo was 0.7 min better than eszopiclone in sleep latency reduction over 12 weeks (13.2 min superior by week 12), and placebo was 13.5 min superior to eszopiclone in sleep latency reduction during withdrawal. In a 1-year trial of suvorexant 30–40 mg using subjective outcomes [4], the investigators proposed a drug-rebound criterion requiring exacerbation worse than the insomnia at baseline. The study suffered from failed randomization, in that the group assigned to suvorexant had worse insomnia than the placebo group at baseline (less sTST, P < 0.05 and more insomnia severity, P < 0.01, two-tailed), creating a bias toward greater improvement in the suvorexant group due to regression to the mean. Failed randomization likewise biased against the placebo group after 1-year drug withdrawal. The suvorexant-withdrawn group indeed had a higher percentage of TST rebound exacerbation than the group receiving only placebo (P = 0.057), a difference which would have been significant, correcting for the failed randomization. Examining their fig. 2, we see that the group withdrawn from suvorexant had increased subjective sleep latencies during the entire eight weeks of drug withdrawal observation, that this would be so even adjusting for the failed randomization, and that there was little evidence of insomnia remission even 8 weeks after withdrawal. This study demonstrated a lasting “return of insomnia” [their words, fig. 2] in the group withdrawn from suvorexant, leaving suvorexant-treated patients worse than the group maintained on placebo throughout. Comparing month 12 of suvorexant treatment with months 1 and 2 after withdrawal, the suvorexant-withdrawn group showed a worse drop in sTST after withdrawal and a greater increase in sleep latency (P < 0.0001 all comparisons), as well as worse insomnia intensity (P < 0.0001 and P < 0.05) compared to those only treated with placebo. I must disagree with the authors that “abrupt discontinuation of suvorexant under double-blind conditions was not associated with . . . significant withdrawal or rebound insomnia” [4]. In my opinion, these three placebo-controlled trials observed withdrawal insomnia throughout 1, 4, and 8 weeks after hypnotic discontinuation, leaving patients who had received the hypnotic worse off than those randomized to placebo. The hypnotics caused insomnia, principally by prolonging sleep latency, and the harm lasted as long as observation persisted. Further research will be needed to examine if mechanisms causing this persistent posthypnotic insomnia are more neuropharmacologic or behavioral. It is disappointing that withdrawal insomnia was caused by hypnotics of such different neuropharmacologic specificities.
Conflict of interest The ICMJE Uniform Disclosure Form for Potential Conflicts of Interest associated with this article can be viewed by clicking on the following link: http://dx.doi.org/10.1016/j.sleep.2014.08.001.
Letters to the Editor/Sleep Medicine 15 (2014) 1168–1172
References [1] Mayer G, Wang-Weigand S, Roth-Schechter B, Lehmann R, Staner C, Partinen M. Efficacy and safety of 6-month nightly ramelteon administration in adults with chronic primary insomnia. Sleep 2009;32:351–60. [2] Ancoli-Israel S, Krystal AD, McCall WV, Schaefer K, Wilson A, Claus R, et al. A 12-week, randomized, double-blind, placebo-controlled study evaluating the effect of eszopiclone 2 mg on sleep/wake function in older adults with primary and comorbid insomnia. Sleep 2010;33:225–34. [3] Sunovion. NCT003386334.
Daniel F. Kripke * Viterbi Family Sleep Center, Scripps Clinic W-207, 10666 North Torrey Pines Road, La Jolla, CA 92037, USA * Address: Viterbi Family Sleep Center, Scripps Clinic W-207, 10666 North Torrey Pines Road, La Jolla, CA 92037, USA. Tel.: +1 858 554 8845; fax: +1 858 554 8492. E-mail address: [email protected]
http://dx.doi.org/10.1016/j.sleep.2014.08.001 1389-9457/© 2014 Published by Elsevier B.V.
Response to Dr. Kripke’s letter In his letter Dr. Kripke raises the question if hypnotics can cause insomnia. At first glance it seems like a contradiction in itself that hypnotics might cause insomnia due to rebound. At second glance the idea sounds challenging, because this has never been directly addressed by studies. Thinking further about this idea makes us wonder if this thought might just serve as another argument supporting the public condemnation of hypnotics. But before drawing final conclusions, methodological aspects have to be considered. One of the methodological key issues is if results during withdrawal period should be compared to baseline or to the end of treatment data. Another is the crucial question whether the assessment of rebound should be done by means or by single patient comparisons resulting in percentages of patients with rebound. Let’s discuss these issues point by point. The contradiction: A World Health Organization (WHO) study recently reported that insomnia is the most frequent co-morbidity concerning health impact [1]. It is well known that benzodiazepines initially shorten sleep latency (SL), but do not increase sleep length or duration of slow wave sleep [2]. On the other hand, total sleep time (TST) becomes shorter and worsens after withdrawal. Nevertheless benzodiazepines do not cause insomnia; they simply do not heal insomnia and can cause dependency. There is ample evidence that insomnia decreases quality of life strongly and that many patients suffer chronically without getting help (cognitive behavioral therapy [CBT] not being paid for, hypnotics being considered to be too dangerous). Would anyone with pain be left alone because opioids are dangerous and may cause addiction? Do insomniacs have to suffer until they get the ‘right’ support, if ever? There must be a balance of pro and cons that needs to be considered, which should also be done with hypnotics. The study protocols: In most short and long term studies the primary study goals were to determine the efficacy and safety of hypnotics. Very few studies have looked at rebound after the hypnotic had been withdrawn. In those studies that looked at post treatment effects 1 day to 2 months after withdrawal, the assess-
1169
ment was mostly done at one or few time points only, either with questionnaires and/or with polysomnography. Even then an immense improvement of up to 60% (TST, SL) was seen in patients and controls. There are strong patient benefits from the study visits, study inherent patient care, and considerable day-to-day variations of the evaluation of sleep quality and other sleep parameters [3,4]. Rebound versus withdrawal insomnia: Kales et al. [5] introduced ‘drug-withdrawal insomnia’ for disturbed sleep, more frequent and intense dreams, and occasional nightmares after abrupt withdrawal of drugs (barbiturates) administered in multiple nightly doses over a long time. Withdrawal insomnia “results from both psychological factors and physiological changes” [5]. In contrast ‘rebound insomnia’ consists of “a marked worsening of sleep following the abrupt withdrawal of benzodiazepine drugs administered in only single doses nightly for short periods” indicated by a significant worsening of sleep latency, wake after sleep onset, and total wake time beyond baseline [6] i.e., an increase of at least 40% of total wake time in the mean group value [7]. Later on this definition was also used for other sleep parameters including subjective assessments and daytime performance. Roehrs et al. [8] have found a difference between pretreatment sleep and sleep during discontinuation of benzodiazepines, regardless of the type of treatment with active drug or placebo, suggesting that pill discontinuation itself might cause rebound insomnia in the sense of a placebo effect. Considering the high day-by-day variation of sleep parameters in insomnia, it seems reasonable to assess both baseline sleep and possible rebound effects over longer time intervals than several days, e.g., in the study of Hajak et al. [9] on 1507 insomnia patients the rebound rates were lower on day 1 of discontinuation than on day 31. Therefore in studies using polysomnographic evaluations the comparison of one or two nights at baseline, end of treatment, and withdrawal is highly questionable to assess rebound insomnia. Dr. Kripke re-evaluated long term studies, which have very different study designs and goals, and either use subjective sleep logs, questionnaires, polysomnography or a combination. All studies he cited had different rebound definitions. The methods show statistically significant differences of SL and TST at different time points of evaluation. The ramelteon study [10] defined rebound as a mean sleep latency during placebo run-out equal to or worse than mean sleep latency at baseline (both in diaries and PSG). The suvorexant [11] study defined rebound as “exploratory endpoints in the first 3 days of discontinuation phase”. This study is the only one that presents the rebound rate (number of patients affected by rebound) in percent. The statistics in different insomnia studies use comparison of baseline to endpoint or discontinuation mean values, which raises two problems: Presenting mean values without giving the range does not allow discrimination of good from poor responders. Secondly, the number of patients from start to end diminishes up to 50%. Therefore comparing mean values at any point of the study does not allow seeing inter-individual changes of sleep parameters. Since rebound (worsening beyond baseline levels) affects single patients, it is not appropriate to use this method to assess withdrawal or rebound insomnia, but this is what is done in nearly all studies. At least the baseline mean of patients who reached the withdrawal study phase must be given. For example: independently of exactly equal TSTs in placebo and treatment groups at study start patients are lost from baseline to end of study (around 25% in both groups of the eszopiclone study [12], 4% in the suvorexant study, and 30% in the ramelteon study). Except for the suvorexant study, we do not know if the dropouts had been good or poor responders. Therefore these data must be interpreted with caution, leaving more patients in the ongoing study with mild and moderate insomnia [12] than