Comparative effectiveness and predictors of cognitive behavioral therapy for insomnia and its components in older adults: main outcomes of a randomized dismantling trial

Abstract

Study Objectives:

To determine the relative effectiveness and predictors of cognitive therapy (CT), behavioral therapy (BT), and cognitive behavioral therapy (CBT) for insomnia in older adults.

Methods:

In a registered clinical trial (NCT02117388), 128 older adults with insomnia disorder were randomly assigned to receive CBT, BT, or CT. Insomnia Severity Index (ISI) score was the primary outcome. Sleep diaries, fatigue, beliefs about sleep, cognitive arousal, and stress were secondary outcomes. Split-plot linear mixed models assessed within- and between-subject changes in outcomes among the treatments. As a secondary analysis, we used linear regression to test predictors of insomnia symptoms improvement, including sleep diary measures, cognitive arousal, stress, beliefs about sleep, baseline ISI score, and age. Benjamini-Hochberg correction was applied.

Results:

All groups exhibited insomnia symptom reduction at posttreatment (CT: d = −2.53, P < .001; BT: d = −2.39, P < .001; CBT: d = −2.90, P < .001) and at the 6-month follow-up (CT: d = −2.68, P < .001; BT: d = −2.85, P < .001; CBT: d = −3.14, P < .001). There were no group differences in the magnitude of ISI improvement (P_adj = .63), response (P_adj > .63), or remission (ISI < 8; P_adj > .27). All groups exhibited significant improvements in secondary outcomes at posttreatment (P_adj < .05) and at the 6-month follow-up (P_adj < .05). At posttreatment, the CT and CBT groups showed greater reductions in beliefs about sleep than the BT group (F_Interaction(2,185) = 5.99, P_adj = .03), and the CBT group showed a greater time in bed reduction than the CT group (F_Interaction(2,185) = 7.05, P_adj = .01). Baseline ISI was the only treatment predictor (b = 1.95, P_adj < .001).

Conclusions:

CBT for insomnia and its components each independently result in significant improvements in self-reported insomnia symptoms, beliefs about sleep, worry, and fatigue in older adults.

Clinical Trial Registration: Registry: ClinicalTrials.gov; Name: Treatments for Insomnia: Mediators, Moderators and Quality of Life; URL: https://clinicaltrials.gov/study/NCT02117388; Identifier: NCT02117388.

Citation:

O’Hora KP, Morehouse AB, Freidman L, et al. Comparative effectiveness and predictors of cognitive behavioral therapy for insomnia and its components in older adults: main outcomes of a randomized dismantling trial. J Clin Sleep Med. 2025;21(10):1679–1695.

BRIEF SUMMARY

Current Knowledge/Study Rationale: cognitive behavioral therapy for insomnia is an effective nonpharmacological therapy for insomnia symptoms in older adults. The cognitive and behavioral therapy components were previously shown to have differential impacts on treatment outcomes in adults with insomnia disorder, but it is unknown whether this finding extends to older adults.

Study Impact: This is the first dismantling study of cognitive behavioral therapy for insomnia in older adults. The results of this study provide a first step for future research studying the mechanisms of therapeutic change for cognitive behavioral therapy for insomnia in older adults and determining whether its effectiveness is distinct from that for relatively younger adults.

INTRODUCTION

Insomnia disorder is a prevalent sleep disorder that is characterized as difficulty initiating or maintaining sleep along with daytime impairment. Insomnia symptoms increase with age; the overall prevalence of insomnia complaints in older adults is estimated to be between 30 and 50%, compared to 10–15% in the general population.^1–4 Insomnia symptoms are associated with increased medical and psychiatric comorbidities,^5,6 greater risk of depression⁷ and anxiety⁸, and decreased quality of life.⁹ Insomnia symptoms are often treated with sedative hypnotic medications, which are associated with increased risk of falls and emergency hospitalizations in older adults,^10,11 and can have detrimental side effects, including dependence and cognitive impairment.^12,13 Cognitive behavioral therapy for insomnia (CBT-I) is a safe, nonpharmacological alternative to hypnotics, which may have superior long-term benefits.^14–16

CBT-I combines 2 component interventions that target distinct factors that contribute to insomnia symptoms: behavioral therapy (BT) and cognitive therapy (CT). BT includes sleep restriction and stimulus control.^17,18 Sleep restriction therapy alters the timing of bedtime and risetime so that the resulting time in bed (TIB) matches self-reported sleep duration¹⁷ at baseline, whereas stimulus control emphasizes going to bed only when sleepy, eliminating all sleep incompatible behaviors in bed, and waking up at the same time every morning.¹⁸ CT for insomnia, like CT for other psychiatric disorders, addresses maladaptive thoughts and beliefs about sleep that may contribute to maintaining sleep difficulties, using cognitive restructuring to change these negative thoughts to more accurate alternate ones. A greater reduction in dysfunctional beliefs about sleep (DBAS) from baseline is associated with greater insomnia symptom improvement after completion of CT.¹⁹

Several trials demonstrate that CBT-I is an effective treatment for improving self-reported insomnia symptoms in older adults, including scores on the Insomnia Severity Index (ISI), and sleep diary measures including sleep onset latency (SOL), wake after sleep onset (WASO), total sleep time (TST), and sleep efficiency (SE).^20–22 Critically, CBT-I in older adults improves daytime impairments, such as mood, memory, and quality of life^21,23–25 and may even prevent worsening physical and mental health.²⁶ However, since each CBT-I component targets different mechanisms contributing to the persistence of insomnia symptoms, it is possible that one of the components could be more effective overall or more effective for certain individuals.

Understanding the relative contribution of CT and BT to the overall efficacy of CBT-I, as well as baseline factors that predict treatment outcome for each component, could improve the efficiency of treatment to make it shorter and more accessible. Improving treatment efficacy in older adults is particularly important to meet the vast need of insomnia treatment for the growing older adult populations.³ One way to achieve these goals is through a dismantling study in which a multicomponent therapy is broken down into its component parts, either in isolation or combination, to identify therapeutic mechanisms of change.²⁷ One such dismantling study of CBT-I in adults with persistent insomnia (aged 25 years and older) reported that BT resulted in faster insomnia symptom improvement, relative to CT; however, insomnia symptoms continued to improve between posttreatment and the 6-month follow-up (6FU) in the CT group, whereas insomnia symptoms worsened in the BT group.²⁸ It is unclear whether these findings extend to older adults, in whom insomnia symptoms are more prevalent and may be related to factors distinct from those impacting younger populations.

The primary aim of the present study was to determine the effectiveness of CT and BT relative to CBT and relative to each other in terms of improving insomnia symptom severity in older adults at posttreatment and at 6FU utilizing a dismantling study design. The primary outcome measure was the ISI. Secondary outcomes included sleep diary outcomes as well as functional outcomes (beliefs about sleep, fatigue, cognitive arousal, stress).¹⁹ We hypothesized that CBT would be more effective than BT and CT in improving insomnia symptoms at posttreatment and 6FU because it would address both insomnia etiologies that are independently targeted by CT and BT. Given past findings in a younger sample,²⁸ we expected BT to result in larger insomnia symptom improvement at posttreatment than CT, which would produce more sustained improvement in insomnia symptoms at 6FU, relative to BT.

To date, research has failed to identify reliable moderators of CBT-I and its components. Therefore, a secondary aim of this study was to identify general (ie, across treatment groups) and differential (ie, specific to treatment group) predictors of insomnia symptom improvement for CT, BT, and CBT in older adults. For BT, we hypothesized that longer TIB and lower SE at baseline would predict less severe insomnia symptoms at posttreatment. For CT, we hypothesized that greater dysfunctional beliefs about sleep, higher levels of cognitive arrousal (ie, worry), and stress (ie, perceived stress and lifetime stressors) would predict less severe insomnia symptoms at posttreatment. Lastly, we hypothesized that baseline insomnia symptoms, age, and baseline sleep diary measures (WASO and TST) would serve as general predictors of treatment outcome, such that worse insomnia symptoms, younger age, increased SL and WASO, and decreased TST would be associated with decreased insomnia symptoms at posttreatment across all groups.

METHODS

We report how we determined our sample size, all data exclusions, all data manipulations, and all study measures.

Participants

Participants aged 60 years or older were recruited from the local community through flyers, newspaper ads, targeted mailings, and in-person presentations about sleep at senior centers and community organizations. An initial telephone interview was conducted to screen for potential eligibility and to provide a brief description of study procedures and research goals. Interested individuals who met initial inclusion criteria were invited to the laboratory for a comprehensive screening procedure. All participants met the following inclusion criteria: (1) 60 years of age or older; (2) independent living; (3) English-speaking; (4) met criteria for Diagnostic and Statistical Manual of Mental Disorders, fourth edition, diagnosis of insomnia on the Duke Structured Interview for Sleep Disorders²⁹; (5) ISI score > 10; and (6) were not using any hypnotic medications (ie, any prescription or nonprescription medication or supplement taken with the purpose of improving sleep) in the 3 weeks prior to or during the time of the study.

Exclusion criteria included (1) Montreal Cognitive Assessment Scale³⁰ score < 20; (2) clinically significant sleep breathing disorder (apnea-hypopnea index > 15 events/h scored using Medicare criteria³¹), periodic limb movement (PLM) disorder (>15 PLMs per hour), and/or restless legs syndrome; (3) caffeine consumption > 3 cups per day, alcohol consumption > 14 drinks per week or > 4 drinks per occasion; (4) illicit substance use (by self-report, including cannabis); (5) major psychiatric diagnosis on Axis 1 as defined in the Diagnostic and Statistical Manual of Mental Disorders, fourth edition,³² except generalized anxiety disorder or dysthymia, based on the Mini International Neuropsychiatric Interview Version 5³³; (6) received CBT-I treatment in the past 12 months; and (7) acute or unstable chronic illness. Participants with acute or unstable chronic illnesses were excluded from our sample, but participants with a stable medical comorbidity were enrolled. Medical conditions were considered stable if there were no new symptoms, treatments, or new medications in the past 4 weeks and no expected changes in the future. Enrolled participants were considered to have a medical comorbidity if they indicated they currently had heart disease, pulmonary disease, gastrointestinal disease, neurological disorder, head trauma, chronic pain disorder, endocrine disorder (such as thyroid disease), metabolic disorder (such as diabetes), kidney disease, autoimmune disorder (such as lupus), cancer, human immunodeficiency virus/acquired immunodeficiency syndrome, or other chronic disorders.

This study was conducted in Palo Alto, California, at offices in the Veterans Affairs Palo Alto Health Care System according to the principles of the Declaration of Helsinki 2008. Recruitment began September 2013 and ended March 2018. Data collection for posttreatment measures ended August 2018. Follow-up data collection ended February 2019. A summary of participant recruitment and flow through study procedures is available in Figure 1. Participants provided written informed consent after the study procedures were fully explained in accordance with the ethical guidelines of the Stanford University Institutional Review Board. Participants received financial compensation for their participation in the study ($200 for completion of the screening session, $200 for completion of treatment, $200 for completion of 6FU). All study procedures were approved by the Stanford University Institutional Review Board. Due to the minimal risk posed by this study, there was no external Data Safety and Monitoring Board. The protocol directors were the monitoring entity. Criteria for discontinuation was an increase in bothersome nightmares for a week under treatment.

Figure 1. Diagram of participant flow throughout the study.

Study design

This study was a 3-arm, parallel, randomized clinical trial. Participants were screened for trial eligibility during 2 in-person screening sessions. The first screening session consisted of collecting information on medical history, psychological and sleep questionnaires/interviews, and a cognitive test. If participants met inclusion criteria, they completed a second screening session to assess for sleep disorders that could contribute to insomnia symptoms, such as obstructive sleep apnea and PLM disorder (see the list of exclusion criteria). At the second screening session, participants completed a 1-night polysomnography recording and baseline clinical measures. Polysomnography recordings were scored for apneas and hypopneas as well as PLMs by a registered polysomnographic technologist. Older adults who met the above criteria were randomly assigned into 1 of 3 treatment groups: CT, BT, or CBT. The randomization was done in a 1:1:1 ratio by study data managers, without stratification, using a random number generator. The data managers sent the treatment assignment to study coordinators, who then informed the psychologist administering the intervention. The psychologist administering the study intervention was not involved in clinical assessments, and clinical assessors were not blinded to treatment assignment since all study outcomes were self-rated. Participants were not told their treatment assignment or given any detailed information about study interventions prior to starting treatment. A control treatment arm was not included in the study design because the primary hypotheses were focused on assessing the relative differences between the active treatment conditions, and the effectiveness of CBT-I has been extensively proven in several previous trials. Treatment consisted of six 60-minute treatment sessions, once per week, that were scheduled as close to consecutive weeks as possible. Participants completed study assessments (outlined below), sleep diaries, and an overnight polysomnography sleep recording before (at the second screening session) and after completing a 6-session therapy regimen and at the 6FU visit after treatment. We chose a 6-months follow-up period to be able to directly compare our results to results of prior CBT-I dismantling studies, which included a 6FU period.²⁸

Measures

Clinical assessments were administered by trained research staff under the supervision of the study psychologist. To ensure high-quality data, assessors were required to practice administering ratings until they reached concordance with other study staff.

Insomnia symptoms

The primary outcome was self-reported insomnia symptoms measured by the ISI, a 7-item self-administered survey assessing current (past 2 weeks) nighttime and daytime symptoms of insomnia. Participants rated each item from 0–4, with higher scores representing worse insomnia symptoms. The total score ranges from 0–28. The ISI is a reliable, valid instrument in evaluating insomnia symptoms, with good internal consistency, temporal stability, and sensitivity to changes in the general adult population and older adults.^34,35 The ISI total score was used as the primary outcome measure of treatment effectiveness. We additionally examined clinically significant changes in ISI score (a reduction of at least 8 points) and remission rates (ISI total score < 8) as primary measures of clinical significance, as defined in prior CBT-I trials.²⁸

Self-reported sleep report

Secondary outcomes of subjectively reported sleep were examined using the Consensus Sleep Diary.³⁶ Sleep diaries were collected for the previous 7 days before each study visit. The sleep diary was used to calculate self-reported values for SOL (time between “lights out” and sleep onset), WASO (amount of time spent awake between sleep onset and final awakening), total TIB (amount of time in bed spent trying to sleep), TST (time between sleep onset and sleep offset minus WASO), SE (TST/TIB * 100), caffeine and alcohol use, and naps. To calculate TIB, “lights off” time was determined using question 2 (“What time did you try to go to sleep?”) on the Consensus Sleep Diary, and “lights on” time was determined by adding question 6b (“After your final awakening, how long did you spend in bed trying to sleep”) to question 6a (“What time was your final awakening?”). Each sleep diary measure was averaged across the 7 days collected at each time point.

Dysfunctional beliefs about sleep

The Dysfunctional Beliefs and Attitudes About Sleep Scale³⁷ is a self-report questionnaire examining beliefs, expectations, and attributions about several sleep-related themes using a visual analog scale from 0–10. Scores can range from 0–160, with greater values indicating stronger beliefs about sleep. This study implements the brief 16-item version of the DBAS, which has been validated in several populations with insomnia.³⁸

Cognitive arousal

The Penn State Worry Questionnaire³⁹ was used to measure current cognitive arousal. The Penn State Worry Questionnaire is a self-report questionnaire in which participants rate 16 items assessing worry on a scale from 1–5. Total scores range from 16–80, with higher scores indicating higher levels of trait worry. This questionnaire has been successful in prior relevant moderator and mediator research⁴⁰ and has been shown to have good internal consistency and good test-retest reliability.³⁹

Stress

Current levels of perceived stress was measured by the Perceived Stress Scale,⁴¹ a 14-item self-report questionnaire that evaluates the degree to which participants find their lives to be unpredictable, uncontrollable, and overloading. Scores range from 0–40 with higher scores indicating higher perceived stress. It has been shown to have good internal consistency and validity, and it is often considered the gold standard for assessing one’s stress perception, independent of environmental stressors.

Exposures to stressful life events were assessed using the Life Stressor Checklist-R.⁴² The Life Stressor Checklist-R includes a yes/no question about exposure to 30 stressful life events, including natural disaster, death of a relative, physical/sexual abuse, and other life events. For each life event endorsed, the participant rates how upsetting the event was at the time on a scale from 1–5, with higher scores indicating more upset. The sum of self-reported ratings of how a stressful event affected the participant’s life was used in the current analysis. This score ranges from 0–150, with a higher score indicating a greater impact on one’s life.

Fatigue

The Multidimensional Fatigue Inventory (MFI)⁴³ was used to assess daytime fatigue. The MFI is a 20-item self-report questionnaire assessing general fatigue, physical fatigue, mental fatigue, reduced motivation, and reduced activity. The MFI total score ranges from 20–100, with higher scores indicating more fatigue. The MFI has been shown to have good internal consistency and validity in several populations and has been used as an outcome measure in prior randomized control trials of insomnia therapy.²⁸

Treatment credibility measures

The Credibility/Expectancy Questionnaire is a scale that was used to assess participants’ expectations and thoughts on the credibility of the rationale for treatment. Items are rated on a 1–5 Likert scale. The Credibility/Expectancy Questionnaire has high validity and test-retest reliability.⁴⁴ It was administered after the first treatment session.

The Treatment Component Adherence Scale (previously described in Manber et al⁴⁵) was used to assess participants’ adherence to and perceived helpfulness of therapy guidelines. Participants rated the extent to which they followed each treatment recommendation relevant to the study arm to which they were assigned, how difficult it was to follow, and how helpful it was on a 0–3 scale, with higher scores indicating better adherence, increased difficulty, and increased helpfulness.

Adverse events

At the start of each study visit, participants were asked about recent doctor’s appointments, medical procedures, hospital visits, physical complaints, mood changes, and accidents or near misses (ie, while driving, in the workplace, or at home).

Study interventions

All interventions were administered by a licensed psychologist not blinded to treatment assignment. One psychologist administered treatment to all participants in all groups; therefore, no differences observed are attributable to different therapists. Treatment sessions were audio recorded for quality assurance. Ten percent of treatment sessions were randomly selected to be reviewed and rated using a fidelity checklist by a psychologist trained in CBT-I to assure fidelity across participants. All participants received sleep education and sleep hygiene components as described the US Department of Veterans Affairs (VA) CBT-I manual.⁴⁶ The full study protocol and intervention materials were not publicly available prior to enrollment. Besides the dismantling of treatment components, there were no adaptation to the therapy in this study. Session length was the same for all 3 treatment groups.

Behavioral therapy

BT consisted of sleep restriction therapy and stimulus control components following the VA CBT-I manual.⁴⁶ Due to repeated instances of trying and failing to initiate and maintain sleep, individuals with insomnia typically become conditioned to associate their bed and bedroom with anxiety (arousal) and wakefulness, which perpetuates insomnia symptoms. Stimulus control reverses this conditioned arousal by limiting non-sleep–related behaviors in bed.^14,17 Moreover, BT utilizes sleep restriction to target the sleep homeostat by limiting the time an individual spends in their bed (sleep opportunity) through assigned bedtimes and rise times. The general rule is to match the person’s sleep opportunity to the amount of average TST that they get (sleep ability) at baseline. This results in faster sleep onset and sleep maintenance due to high sleep drive that accumulated before bedtime.

Cognitive therapy

CT was adapted from the treatment manual used in a previous dismantling study in adults with persistent insomnia,²⁸ which was harmonized with the CT component from the VA CBT-I manual.⁴⁶ The CT treatment module is designed to meet 3 general goals: identify dysfunctional sleep cognitions, challenge their validity, and replace them with more adaptive substitutes. CT sessions were devoted to exploring the participants’ negative attributions and fears by using thought records, cognitive restructuring, scheduled worry time, behavioral experiments, and meditation to address these cognitions.

Cognitive behavioral therapy (behavioral therapy plus cognitive therapy)

The CBT-I treatment included all components of CT and BT and was delivered according to protocols and materials from the large-scale VA implementation of CBT-I.⁴⁶ The therapist integrated the 2 therapies instead of administering the 2 therapies independently.

Data management

All research data collected were de-identified by giving participant identification numbers. Once the data files were collected, they were entered into an online REDCap electronic data capture tool hosted at the Palo Alto VA Hospital^47,48 by a research assistant. After entry, the data were reviewed and verified by a second research assistant to ensure accuracy. Raw polysomnography data files were either securely sent through encrypted email or physically delivered to the Stanford Sleep Medicine Center in Redwood City to be scored for apneas and hypopneas as well as PLMs by a registered polysomnographic technologist.

Data analysis

The primary aim of this study was to determine the relative effectiveness of CBT-I and its components in improving insomnia symptoms in older adults. The primary outcome measures were the change in the ISI total score (self-reported insomnia symptoms) from baseline to posttreatment, baseline to 6FU, and posttreatment to 6FU, as well as ISI response and remission at posttreatment and 6FU. Secondary outcomes were changes in self-reported sleep measures from sleep diaries (SOL, WASO, TST, TIB, SE) and functional outcomes (beliefs about sleep, cognitive arousal, fatigue, and stress) from baseline to posttreatment and 6FU.

Our analyses follow procedures previously implemented in Harvey et al.²⁸ Distributions of all variables were visually examined for outliers. All analyses, except for the analysis of treatment predictors (see below), were performed using an intent-to-treat approach in that participants were analyzed according to their original treatment assignments regardless of treatment adherence, dropouts, or deviations from the protocols using all data available for each participant with no data imputation. To test our hypothesis that CBT-I would result in a greater improvement of insomnia symptoms than both BT and CT groups at posttreatment and that CBT and CT would both be better than BT at 6FU, we analyzed group (CT, BT, CBT) by time (pre, post, 6FU) as fixed effects in split-plot linear mixed models (SAS 9.4 PROC MIXED, Cary, NC⁴⁹) with random intercepts and empirical (“sandwich”) estimates of fixed effect standard errors. Models included age and sex as covariates. The group-by-time interaction was evaluated to determine whether the change in insomnia symptoms across time differed between the 3 groups at each time point. Post hoc analyses of simple effects were conducted to further explore interactions. Parallel analyses were conducted for the secondary outcomes.

We also assessed the ISI treatment response and remission in each group. The proportion of individuals who achieved response (ISI reduction ≥ 8) and/or remission (ISI < 8) in each group was compared using logistic mixed models with the same predictors as outlined above (SAS 9.4 PROC GLIMMIX⁵⁰).

Within each domain of outcomes (ie, insomnia severity outcomes [including overall severity, response, and remission], sleep diary outcomes, and functional outcomes), multiplicity was addressed by computing adjusted P values for the group-by-time interactions and effects of time from baseline to both follow-up time points (ie, 3 families of tests, family size for each correction reporting in table footnote) using the Benjamini-Hochberg False Discovery Rate correction.⁵¹ All other P values reported are unadjusted. Effect sizes of time changes were calculated as the difference between means divided by the root mean squared error of the mixed model.

Lastly, we conducted a secondary analysis of general and differential predictors of insomnia symptoms, response, and remission at posttreatment. Linear regression models with the outcome at posttreatment being predicted by a treatment group-by-predictor interaction were implemented in R version 4.3.1 to test the main effects and group interaction for the hypothesized predictors for the primary outcomes (ISI total score, response, remission). If there was a significant group interaction, the predictor was considered a differential predictor (ie, specific to a certain treatment group). If there was a significant main effect, then it was considered a general predictor (ie, predicts posttreatment insomnia symptoms across treatment groups). Each model included age, sex, and baseline ISI score as covariates. A Benjamini-Hochberg correction⁵¹ was applied for both the main and interaction effects to correct for multiplicity (ie, 2 families of tests, each with 30 corrections). Due to the use of standard linear regression models, the analysis of predictors did not follow the intent-to-treat approach.

Data and study materials are available upon reasonable request. Intervention manuals utilized in this study are available through previous publications.^28,46

RESULTS

Study sample

Recruitment began September 2013 and ended March 2018. Data collection for posttreatment measures ended August 2018. Follow-up data collection ended February 2019. A summary of participant recruitment and flow through study procedures is available in Figure 1. Participants (n = 128) were randomly assigned into 1 of 3 treatment groups: CT (n = 43), BT (n = 41), or CBT (n = 44). Sample demographics are summarized in Table 1. Participants were considered to have a psychiatric comorbidity if they met criteria for generalized anxiety disorder or dysthymia. Few participants in our sample had a psychiatric comorbidity (8.6%), although 53.9% indicated a medical comorbidity. There were significant differences in age, employment, and apnea-hypopnea index among the 3 treatment groups at baseline. The BT group was older and had more retired individuals compared to the other 2 groups. The CT group exhibited a higher mean apnea-hypopnea index relative to the BT and CBT groups. No serious adverse events were reported.

Table 1.

Participant characteristics at baseline.

	CBT (n = 44)				CT (n = 43)				BT (n = 41)				Total (n = 128)				Statistic
	%	n	M	SD	%	n	M	SD	%	n	M	SD	%	n	M	SD
Sex (female)	75.0	33			62.8	27			58.5	24			65.6	84			2.8, P = .25
Age (years)			67.9	5.3			68.3	5.8			71.1	7.3			69.1	6.3	3.5, P = .03
Race/ethnicity																	2.7, P = .26
White	88.6	39			83.7	36			7.8	31			82.8	106
Black	2.3	1			2.3	1			4.9	2			3.1	4
Asian	4.6	2			9.3	4			9.8	4			7.8	10
Unknown	2.3	1			2.3	1			2.4	1			2.3	3
Hispanic	2.3	1			2.3	1			7.3	3			3.9	5
Marital status																	4.4, P = .36
Single	18.2	8			4.7	2			9.8	4			10.9	14
Married/partnered	61.4	27			67.4	29			65.9	27			64.8	83
Divorced/separated/widowed	20.5	9			27.9	12			24.4	10			24.2	31
Education																	9.6, P = .05
Some college or less	11.4	5			9.8	4			22.0	9			14.3	18
College degree	40.9	18			34.2	14			53.7	22			42.9	54
Some graduate or higher	47.7	21			56.1	23			24.4	10			42.9	54
Employment status																	10.1, P = .04
Full or part time/student	38.6	17			32.6	14			19.5	8			30.5	39
Unemployed	9.1	4			2.3	1			0	0			3.9	5
Retired	52.3	23			65.1	28			80.5	33			65.6	84
ISI			16.5	3.3			15.7	3.9			15.2	3.9			15.8	3.7	1.1, P = .32
AHI			6.8	3.7			8.8	3.5			7.3	3.7			7.6	3.7	3.8, P = .03
PLMI			9.2	16.5			14.4	26.2			11.7	31.1			11.8	25.1	0.5, P = .63
Prior antidepressant use	11.4	5			11.6	5			9.8	4			10.9	14			0.1, P = .96
Medical comorbidity (any)	54.6	24			55.8	24			51.2	21			53.9	69			0.2, P = .91
Psychiatric comorbidity (any)	13.6	6			4.7	2			7.3	3			8.6	11			2.4, P = .31
Anxiety symptomsa			4.6	4.0			5.1	4.5			4.0	4.6			4.5	4.4	0.7, P = .51
Depressive symptomsb			5.9	4.9			6.2	7.0			6.2	6.1			6.1	6.0	0.0, P = .97

Subcategory percent may not add to 100 due to rounding error. ^aMeasured by the Beck Anxiety Inventory total score.^{60 b}Measured by the Beck Depression Inventory-II.⁶¹ AHI = apnea-hypopnea index, BT = behavioral therapy, CBT = cognitive behavioral therapy for insomnia, CT = cognitive therapy, ISI = Insomnia Severity Index, M = mean, PLMI = periodic limb movement index, SD = standard deviation.

Power analysis

We sought to enroll a total of 150 participants (50 per treatment group) to provide above 80% power to detect an effect size of 0.6 at an alpha level of 0.05. To adjust this power analysis for the achieved sample of 128 participants, we calculated an adjusted detectable effect size by multiplying the original detectable effect size (0.6) by the square root of 150/128. This method uses the fact that a detectable effect size is inversely proportional to the square root of a given sample size (see, eg, Lazzeroni and Ray⁵²) and does not use any values from the actual sample. According to this calculation, the final sample of 128 participants provides 80% power to detect an effect size of 0.65 at an alpha level of 0.05.

Treatment attrition and credibility

There were no differences in treatment credibility or expectancy (P > .17) between the 3 treatment groups. All 3 groups exhibited low attrition rates; 5 of the 44 (11.3%) participants randomly assigned to CBT, 5 of the 43 (11.6%) participants randomly assigned to CT, and 8 of the 41 (19.5%) participants randomly assigned to BT did not complete a posttreatment study visit.

Relative effectiveness of CBT-I on insomnia symptom severity (primary outcome)

All adjusted means, change scores, effect sizes, and results of models for ISI scores, response, and remission are reported in Table 2 and visualized in Figure 2. Summary of results of the overall models are reported in Table S1 ^{(17.7KB, pdf)} in the supplemental material. Counter to our hypotheses, there was no significant treatment-by-time interaction for insomnia symptom severity (F(2,209) = 1.3, P_adj = .63) from baseline to posttreatment. However, there was a significant effect of time (F(2,209) = 18.58, P_adj < .001), and post hoc analyses revealed that all treatment groups exhibited significantly improved insomnia symptom severity (P < .001) from baseline to posttreatment (CBT M = −9.26, d = −2.9; BT M = −7.64, d = −2.39; CT M = −8.09, d = −2.53) with no statistically significant difference among groups.

Table 2.

Adjusted means and change scores on the Insomnia Severity Index by group and time.

Time or Change	Mean (SE), by Time and Change Scores			Effect of Time	Comparison Between Groups
	CBT	CT	BT	F(2,209)a	Effect	F(2,209)b
ISI
t1 (Pre)	16.5 (0.5)	15.8 (0.6)	15.1 (0.6)		cond/t1	1.54, P = .22
t2 (Post)	7.2 (0.6)	7.7 (0.5)	7.5 (0.7)		cond/t2	0.23, P = .80
t3 (6FU)	6.4 (0.6)	7.2 (0.6)	6.0 (0.7)		cond/t3	0.96, P = .38
Change t1–t2 (ES)	−9.3*** (−2.90)	−8.1*** (−2.53)	−7.6*** (−2.39)	18.58 P < .001, Padj < .001	cond/t1–t2	1.30, P = .27, Padj = .63
Change t1–t3 (ES)	−10.0*** (−3.14)	−8.6*** (−2.68)	−9.1*** (−2.85)	23.52, P < .001, Padj < .001	cond/t1–t3	0.87, P = .42, Padj = .63
Change t2–t3 (ES)	−0.8 ns (−0.24)	−0.5 ns (−0.15)	−1.5* (−0.46)	0.90, P = .34, Padj = .34	cond/t2–t3	0.64, P = .53, Padj = .63
ISI response % (reduction of at least 8 points from baseline)
t2 (Post)	64.45 (9.52)	59.35 (9.77)	53.25 (10.68)		cond/t2	0.30, P = .74, Padj=.74
t3 (6FU)	64.38 (10.57)	60.51 (9.94)	75.14* (9.11)		cond/t3	0.59, P = .56, Padj = .63
Change t2–t3	−0.07	1.16	21.89	3.09, P = .08, Padj = .14	cond/t2–t3	0.98, P = .38, Padj = .63
ISI remission % (ISI < 8)
t2 (Post)	44.65 (9.44)	58.62 (9.32)	62.46 (9.97)		cond/t2	0.91, P = .41, Padj = .63
t3 (6FU)	73.67* (8.85)	47.85 (9.76)	78.08* (8.50)		cond/t3	2.91, P = .06, Padj = .27
Change t2–t3	29.02*	−10.77	15.62	1.42, P = .24, Padj = .30	cond/t2–t3	2.88, P = .06, Padj = .27

*P < .05, **P < .01, ***P < .0001. ^aP values for all effects of time (n = 5) were adjusted using Benjamini-Hochberg correction. ^bP values for all group-by-time interactions (n = 9) were adjusted using Benjamini-Hochberg correction. 6FU = 6-month follow-up, BT = behavioral therapy, CBT = cognitive behavioral therapy for insomnia, CT = cognitive therapy, ES = effect size, ISI = Insomnia Severity Index, ns = nonsignificant, SE = standard error.

Figure 2. Mean ± standard deviation of Insomnia Severity Index total score for the BT group (green), CT group (purple), and CBT group (orange) at each time point (pretreatment, posttreatment, 6FU).

Asterisks indicate significant within group changes from one time point to the next. *p < 0.05, **p < 0.01, **p < 0.005. 6FU = 6-month follow-up, BT = behavioral therapy, CBT = cognitive behavioral therapy, CT = cognitive therapy.

Similarly, although there was no significant treatment-by-time interaction for insomnia symptom severity from baseline to 6FU (F(2,209) = 0.87, P_adj = .63), there was a significant effect of time from baseline to 6FU (F(2,209) = 23.52, P_adj < .001). Post hoc analyses revealed that all treatment groups significantly improved insomnia symptom severity from baseline to 6FU (CBT M = −10.0, d = −3.14; BT M = −9.1, d = −2.85; CT M = −8.6, d = −2.68). Finally, there was no significant treatment-by-time interaction for insomnia symptom severity from posttreatment to 6FU (F(2,209)=0.6, P_adj = .63). There was a significant improvement in insomnia symptom severity from posttreatment to 6FU for BT (M = −1.5, d = −0.46), but not CT (M = −0.48, d = −0.15) nor CBT (M = −0.78, d = −0.24). Taken together, CBT, BT, and CT all improved insomnia symptom severity from baseline to posttreatment, and these improvements were sustained at 6FU.

Paralleling results for the continuous measures of insomnia symptom severity, there was no significant group-by-time interaction for response (F(2,209) = 0.30, P_adj = .74) or remission (F(2,209) = 0.91, P_adj =.63) at posttreatment. The proportion of responders and remitters in each group did not differ from chance at posttreatment (P > .23; Table 2). Similarly, at 6FU, there was no significant group-by-time interaction for response (F(2,209) = 0.59, P_adj =.63) nor remission (F(2,209) = 2.91, P_adj = .27).

Relative effectiveness of CBT-I on self-reported sleep reports (secondary outcomes)

All adjusted means, change scores, effect sizes, and results of models for the sleep diary variables are reported in Table 3 and visualized in Figure 3. Summary of results of the overall models are reported in Table S1 ^{(17.7KB, pdf)} . There was a significant group-by-time interaction for TIB from baseline to posttreatment (F(2,185) = 7.05, P_adj = .01) in which the CBT group (M = −62.38 minutes, d = −1.79) exhibited a significantly larger reduction in TIB relative to CT (M = −13.97 minutes, d = −0.40) group. No other group-by-time interactions survived correction for any of the 5 sleep diary outcomes from baseline to posttreatment (all P_adj > .18), baseline to 6FU (all P_adj > .37) or from posttreatment to 6FU (all P_adj > .18). However, there was a significant effect of time on all secondary sleep outcomes from baseline to posttreatment and baseline to 6FU (all P_adj < .001; Figure 3), with the exception of TIB (baseline to posttreatment: P_adj < .001, baseline to 6FU: P_adj = .14). Post hoc analyses revealed significant improvements in SOL, WASO, TST, and SE from baseline to posttreatment and baseline to 6FU, suggesting all 3 therapies improve these measures.

Table 3.

Adjusted means and change scores on the sleep diary variables according to group and time.

Time or Change	Mean (SE), by Time and Change Scores			Effect of Time	Comparison Between Groups
	CBT	CT	BT	F(2,189)a	Effect	F(2,185)b	Post Hocc
Sleep onset latency (in min)
t1 (Pre)	28.73 (3.23)	41.61 (4.87)	28.73 (4.87)		cond/t1	2.99, P = .05
t2 (Post)	13.94 (2.62)	19.33 (2.58)	13.46 (2.00)		cond/t2	1.82, P = .16
t3 (6FU)	13.13 (2.29)	20.27 (2.59)	13.94 (2.02)		cond/t3	2.55, P = .08
Change t1–t2 (ES)	−14.79*** (−1.02)	−22.28*** (−1.54)	−14.64*** (−1.01)	68.72 P < .001 Padj <.001	cond/t1–t2	1.15, P = .32, Padj = .73
Change t1–t3 (ES)	−15.60*** (−1.08)	−21.34*** (−1.48)	−14.17*** (−0.98)	63.04 P < .001 Padj < .001	cond/t1–t3	0.85, P = .43, Padj = .73
Change t2–t3 (ES)	−0.81 ns (−0.06)	0.94 ns (0.07)	0.48 ns (0.03)	0.02 P = .88 Padj = .88	cond/t2–t3	0.19, P = .83, Padj = .83
Wake after sleep onset (in min)
t1 (Pre)	67.70 (5.98)	77.48 (9.56)	65.61 (7.40)		cond/t1	0.51, P = .60
t2 (Post)	23.57 (2.84)	43.02 (6.48)	23.30 (2.84)		cond/t2	4.10, P = .02	CT > CBT
t3 (6FU)	28.73 (5.16)	46.19 (6.37)	30.75 (5.52)		cond/t3	2.54, P = .08
Change t1–t2 (ES)	−44.13*** (−1.57)	−34.46*** (−1.23)	−42.31*** (−1.50)	105.27 P < .001 Padj <.001	cond/t1–t2	0.52, P = .59, Padj = .83
Change t1–t3 (ES)	−38.97*** (−1.39)	−31.29** (−1.11)	−34.86*** (−1.24)	65.61 P < .001 Padj < .001	cond/t1–t3	0.24, P = .79, Padj = .83
Change t2–t3 (ES)	5.16 ns (0.18)	3.17 ns (0.11)	7.46 ns (0.27)	3.53 P = .06 Padj = .08	cond/t2–t3	0.19, P = .83, Padj = .83
Total sleep time (in min)
t1 (Pre)	286.24 (12.52)	298.82 (12.60)	311.84 (13.01)		cond/t1	1.25, P = .29
t2 (Post)	346.59 (9.95)	366.63 (10.74)	357.05 (13.92)		cond/t2	0.94, P = .39
t3 (6FU)	368.55 (10.70)	372.85 (11.31)	385.98 (12.35)		cond/t3	0.59, P = .55
Change t1–t2 (ES)	60.35*** (1.24)	67.81*** (1.39)	42.21** (0.86)	40.96 P < .001 Padj <.001	cond/t1–t2	0.96, P = .38, Padj = .73
Change t1–t3 (ES)	82.30*** (1.68)	74.04*** (1.52)	71.14*** (1.46)	126.56 P < .001 Padj <.001	cond/t1–t3	0.29, P = .75, Padj = .83
Change t2–t3 (ES)	21.95* (0.45)	6.22 ns (0.13)	28.93* (0.59)	12.89 P = .0004 Padj <.001	cond/t2–t3	0.82, P = .44, Padj = .73
Total time in bed (in min)
t1 (Pre)	469.80 (7.58)	487.33 (8.51)	453.37 (8.59)		cond/t1	3.70, P = .02	CT > BT
t2 (Post)	407.42 (9.43)	473.36 (9.34)	419.22 (8.76)		cond/t2	14.10, P = .00	CT > BT = CBT
t3 (6FU)	448.96 (8.21)	485.54 (8.72)	455.55 (9.30)		cond/t3	5.14, P = .01	CT > CBT
Change t1–t2 (ES)	−62.38*** (−1.79)	−13.97 ns (−0.40)	−34.14** (−0.98)	49.84 P < .001 Padj <.001	cond/t1–t2	7.05, P = .00, Padj = .01	CBT > CT
Change t1–t3 (ES)	−20.83 * (−0.60)	−1.78 ns (−0.05)	2.17 ns (0.06)	2.43 P = .12 Padj = .14	cond/t1–t3	2.33, P = .10, Padj =.37
Change t2–t3 (ES)	41.54*** (1.20)	12.18 ns (0.35)	36.32** (1.05)	31.92 P < .001 Padj < .001	cond/t2–t3	3.35, P = .03, Padj =.18
Sleep efficiency (%)
t1 (Pre)	67.32 (2.11)	66.22 (2.49)	73.21 (2.19)		cond/t1	2.68, P = .07
t2 (Post)	86.91 (1.38)	81.15 (1.73)	86.09 (2.03)		cond/t2	3.59, P = .03	CBT > CT
t3 (6FU)	85.88 (1.46)	80.84 (1.93)	86.64 (1.58)		cond/t3	3.02, P = .05
Change t1–t2 (ES)	19.58*** (2.22)	14.92*** (1.69)	12.88*** (1.46)	150.31 P < .001 Padj < .001	cond/t1–t2	2.09, P = .03, Padj = .18
Change t1–t3 (ES)	18.56*** (2.11)	14.62*** (1.66)	13.42*** (1.52)	153.26 P < .001 Padj < .001	cond/t1–t3	1.59, P = .20, Padj = .62
Change t2–t3 (ES)	−1.02 ns (−0.12)	−0.30 ns (−0.03)	0.54 ns (0.06)	0.07 P = .80 Padj = .85	cond/t2–t3	0.25, P = .77, Padj = .83

*P < .05, **P < .01, ***P < .0001. ^aP values for all effects of time (n = 15) were adjusted using Benjamini-Hochberg correction. ^bP values for all group-by-time interactions (n = 15) were adjusted using Benjamini-Hochberg correction. ^cPost-hoc comparisons adjusted using Bonferroni adjustment (n = 3). 6FU = 6-month follow-up, ES = effect size, ns = nonsignificant, SE = standard error.

Figure 3. Mean ± standard deviation of sleep diary variables for the BT group (green), CT group (purple), and CBT group (orange) at each time point (pretreatment, posttreatment, 6FU).

Asterisks alone indicate significant within group changes from one time point to the next. Bars with asterisks indicate significant group x time interactions from baseline to each follow-up time point. *p < 0.05, **p < 0.01, **p < 0.005. 6FU = 6-month follow-up, BT = behavioral therapy, CBT = cognitive behavioral therapy, CT = cognitive therapy group, SE = sleep efficiency, SOL = sleep onset latency, TIB = time in bed; TST = total sleep time, WASO = wake after sleep onset.

Relative effectiveness of CBT-I on functional outcomes (secondary outcomes)

All adjusted means, change scores, effect sizes, and results of models for the functional outcome variables are reported in Table 4 and Visualized in Figure 4. There was a significant treatment-by-time interaction for DBAS score from baseline to posttreatment (F(2,185), P_adj = .03 (See Figure 4a)) such that the CT (P < .001, M = −1.98 points, d = −2.11) and CBT (P < .001, M = −1.98 points, d = −2.11) groups showed significantly greater reductions than the BT group (P < .001, M = −1.09 points, d = −1.16). There were no other significant group-by-time interactions at any other time point (all P_adj > .30). There was a significant effect of time for all functional outcomes from baseline to posttreatment (all P_adj < .01) and baseline to 6FU (all P_adj < .01).

Table 4.

Adjusted means and change scores on the functional outcomes according to group and time.

Time or Change	Mean (SE), by Time and Change Scores			F(2,201)a	Comparison Between Groups
	CBT	CT	BT		Effect	F(2,185)b	Post Hocc
Dysfunctional beliefs about sleep
t1 (Pre)	3.7 (0.2)	5.2 (0.21)	4.8 (0.2)		cond/t1	2.42, P = .09
t2 (Post)	3.5 (0.2)	3.3 (0.20)	3.7 (0.2)		cond/t2	1.25, P = .29
t3 (6FU)	3.3 (0.2)	3.4 (0.20)	3.3 (0.2)		cond/t3	0.08, P = .92
Change t1–t2 (ES)	−2.0*** (−2.11)	−2.0*** (−2.11)	−1.1*** (−1.16)	167.7, P < .001, Padj < .001	cond/t1–t2	5.99, P = .00, Padj = .03	CBT = CT > BT
Change t1–t3 (ES)	−2.1*** (−2.26)	−1.9*** (−1.97)	−1.5*** (−1.61)	167.7, P < .001, Padj < .001	cond/t1–t3	1.45, P = .24, Padj = .37
Change t2–t3 (ES)	−0.1 ns (−0.15)	0.1 ns (0.15)	−0.4* (−0.45)	2.53, P = .11, Padj = .12	cond/t2–t3	3.00, P = .05, Padj = .30
Penn State Worry Questionnaire
t1 (Pre)	46.6 (2.2)	44.5 (2.0)	45.2 (2.3)		cond/t1	0.25, P = .78
t2 (Post)	40.7 (2.2)	38.2 (1.9)	37.7 (2.5)		cond/t2	0.55, P = .58
t3 (6FU)	44.4 (2.1)	39.3 (2.0)	39.9 (2.2)		cond/t3	1.77, P = .17
Change t1–t2 (ES)	−5.9*** (−1.01)	−6.3*** (−1.07)	−7.5*** (−1.28)	50.69, P < .001, Padj < .001	cond/t1–t2	0.19, P = .82, Padj = .93
Change t1–t3 (ES)	−2.2 ns (−0.37)	−5.2*** (−0.89)	−5.3** (−0.90)	25.91, P < .001, Padj < .001	cond/t1–t3	1.94, P = .15, Padj = .37
Change t2–t3 (ES)	3.7*** (0.64)	1.1 ns (0.18)	2.2 ns (0.37)	11.36, P < .001, Padj = .00	cond/t2–t3	1.63, P = .20, Padj = .37
Perceived Stress Scale
t1 (Pre)	18.3 (1.0)	18.6 (1.0)	18.5 (1.0)		cond/t1	0.02, P = .98
t2 (Post)	15.4 (0.9)	15.4 (0.9)	16.6 (1.2)		cond/t2	0.60, P = .55
t3 (6FU)	17.4 (1.0)	16.9 (1.2)	16.9 (1.2)		cond/t3	0.06, P = .95
Change t1–t2 (ES)	−1.6 ns (−0.43)	−3.3*** (−0.92)	−1.8 ns (−0.51)	20.16, P < .001, Padj < .001	cond/t1–t2	1.41, P = .25, Padj = .37
Change t1–t3 (ES)	−1.2 ns (−0.33)	−1.2 ns (−0.35)	−1.6* (−0.45)	7.51, P = .01, Padj = .01	cond/t1–t3	0.08, P = .93, Padj = .93
Change t2–t3 (ES)	0.4 ns (0.11)	2.0** (0.57)	0.2 ns (0.06)	3.35, P = .07, Padj = .08	cond/t2–t3	1.59, P = .21, Padj = .37
Multidimensional Fatigue Inventory
t1 (Pre)	45.8 (1.8)	44.8 (1.9)	46.2 (2.0)		cond/t1	0.12, P = .88
t2 (Post)	38.2 (2.6)	38.4 (2.1)	39.3 (3.3)		cond/t2	0.03, P = .97
t3 (6FU)	40.3 (2.2)	36.3 (2.1)	41.6 (2.4)		cond/t3	1.55, P = .22
Change t1–t2 (ES)	−7.6** (−0.84)	−6.5*** (−0.72)	−6.9* (−0.77)	18.58, P < .001, Padj < .001	cond/t1–t2	0.07, P = .93, Padj = .93
Change t1–3 (ES)	−5.5** (−0.62)	−8.5*** (−0.95)	−4.6* (−0.51)	23.52, P < .001, Padj < .001	cond/t1–t3	1.27, P = .28, Padj = .37
Change t2–t3 (ES)	2.1 ns (0.23)	−2.0 ns (−0.22)	2.4 ns (0.26)	0.90 P = .34, Padj = .34	cond/t2–t3	1.73, P = .18, Padj = .37

*P < .05, **P < .01, ***P < .0001. ^aP values for all effects of time (n = 12) were adjusted using Benjamini-Hochberg correction. ^bP values for all group-by-time interactions (n = 12) were adjusted using Benjamini-Hochberg correction. ^cPost-hoc comparisons adjusted using Bonferroni adjustment (n = 3). 6FU = 6-month follow-up, ES = effect size, ns = nonsignificant, SE = standard error.

Figure 4. Mean ± standard deviation of all functional outcomes for the BT group (green), CT group (purple), and CBT group (orange) at each time point (pretreatment, posttreatment, 6FU).

Asterisks alone indicate significant within group changes from one time point to the next. Bars with asterisks indicate significant group x time interactions from baseline to each follow-up time point. *p < 0.05, **p < 0.01, **p < 0.005. 6FU = 6-month follow-up; BT = behavioral therapy, CBT = cognitive behavioral therapy group, CT = cognitive therapy, DBAS = dysfunctional beliefs about sleep; MFI = Multidimensional Fatigue Inventory; PSS = Perceived Stress Scale; PSWQ = Penn State Worry Questionnaire.

Predictors of insomnia symptom outcomes (secondary analyses)

The results of models testing predictors of treatment outcome are summarized in Table 5. Counter to our hypotheses, there were no significant differential predictors of treatment outcome (P_adj > .750). As hypothesized, ISI score at baseline was a general predictor for treatment response at posttreatment (χ² = 45.80, b = 1.95, P_adj < .001). However, among the predictors we tested, there were no other significant general predictors of treatment outcome (P_adj > .276).

Table 5.

Summary of results from models testing general and differential predictors of insomnia symptoms at posttreatment.

Moderator	ISI Total					ISI Response					ISI Remission
	Interaction		Main Effect			Interaction		Main Effect			Interaction		Main Effect
	F	P adj a	F	b (97.5% CI)	P adj a	χ2	P adj a	χ2	b (97.5% CI)	P adj a	χ2	P adj a	χ2	B (97.5% CI)	P adj a
BT specific predictors
TIB	0.27	.90	0.48	0.07 (−0.27, 0.13)	.78	0.46	.90	0.11	0.08 (−0.42, 0.59)	.93	1.92	.78	0.25	0.10 (−0.30, 0.52)	.88
SE	0.32	.90	4.74	−0.24 (−0.46, −0.02)	.28	0.43	.90	3.10	0.55 (−0.06, 1.22)	.28	0.08	.96	2.84	0.41 (−0.06, 0.91)	.28
CT specific predictors
DBAS	0.64	.90	0.18	−0.05 (−0.26, 0.17)	.88	2.12	.78	0.00	0.00 (−0.54, 0.55)	.99	0.57	.90	0.07	0.06 (−0.37, 0.50)	.94
PSWQ	0.38	.90	0.58	−0.07 (−0.26, 0.12)	.78	1.58	.85	0.00	0.01 (−0.48, 0.48)	.99	3.98	.75	0.04	0.04 (−0.35, 0.44)	.94
PSS	0.34	.90	0.74	0.09 (−0.11, 0.28)	.73	1.96	.78	0.30	−0.15 (−0.70, 0.39)	.88	4.64	.75	1.72	−0.274 (−0.70, 0.13)	.42
Life stressor	1.49	.75	0.02	0.01 (−0.19, 0.21)	.96	1.88	.78	0.05	−0.07 (−0.65, 0.56)	.94	5.84	.75	0.20	0.097 (−0.33, 0.53)	.88
General predictors
ISI baseline	2.64	.75	2.05	0.14 (−0.05, 0.33)	.36	0.11	.96	45.80	1.95 (1.26, 2.83)	< .001	4.08	.75	2.91	−0.34 (−0.73, 0.05)	.28
Age	1.50	.75	2.87	0.18 (−0.03, 0.39)	.28	2.65	.75	0.47	−0.19 (−0.76, 0.36)	.78	3.02	.75	3.46	−0.41 (−0.86, 0.02)	.28
WASO	1.30	.75	3.16	0.20 (−0.02, 0.41)	.28	3.30	.75	2.72	−0.56 (−1.3, 0.10)	.28	1.27	.90	2.95	−0.41 (−0.94, 0.06)	.28
TST	0.25	.90	4.44	−0.21 (−0.40, −0.01)	.28	1.12	.90	1.40	0.32 (−0.20, 0.87)	.49	0.29	.93	2.62	0.35 (−0.07, 0.79)	.28

^aP values for main effects and interactions were adjusted for each comparison across predictors (n = 30) using Benjamini-Hochberg correction. BT: behavioral therapy; CI = confidence interval, CT: cognitive therapy; DBAS: dysfunctional beliefs about sleep; ISI: Insomnia Severity Index; PSS: Perceived Stress Scale; PSWQ: Penn State Worry Questionnaire; SE: sleep efficiency; TIB: time in bed; TST: total sleep time, WASO: wake after sleep onset.

DISCUSSION

The 2 goals of the present trial were to examine the relative effectiveness of CBT-I and its components (CT and BT), as well as to identify potential predictors of treatment response in older adults with insomnia disorder. With respect to the first goal, contrary to our hypothesis, there were no statistically significant differences in the insomnia symptom improvement among groups at either posttreatment or 6FU. Instead, our results suggest that all 3 therapies lead to a significant improvement of the primary outcome measure of insomnia symptoms measured by the ISI from baseline to posttreatment and baseline to 6FU. Moreover, these improvements across treatment arms also extended to self-reported sleep diary measures, as well as functional outcomes, including DBAS, worry, stress, and fatigue. However, there were only 2 parameters for which there were differential treatment effects (DBAS and TIB). With respect to the second goal, we did not find meaningful general or differential predictors of response for insomnia symptoms, except for baseline severity of insomnia symptoms.

The present finding that all 3 treatment groups led to improvements of insomnia symptoms among older adults is consistent with prior work in healthy younger adults reporting significant insomnia symptom improvement following each of the 3 treatments.²⁸ Although many studies have documented the efficacy of CBT-I and BT in older adults.^21,26,53 this study is the first to demonstrate the effectiveness of CT in improving insomnia symptoms and functional outcomes in older adults. Unlike Harvey et al²⁸ we found that the proportion of individuals exhibiting a treatment response or remission of insomnia symptoms did not differ among the 3 treatments at either time point. In contrast, the prior dismantling study by Harvey et al²⁸ reported that, at posttreatment, there were significantly more treatment responders in the CBT and BT groups relative to the CT group and significantly more remitters in the CBT group relative to the CT group. The discrepancy between these 2 studies for response and remission may suggest that there are age-related differences in the clinically relevant outcomes of CT and BT. However, it is important to note that the younger sample had slightly higher insomnia severity scores than the current sample at baseline and there were differences in the treatment implementation between the 2 studies (ie, differing number of sessions, number of therapists), which might also have contributed to the differences in binary study outcomes that utilize specific symptom severity thresholds.

Of note, the average reduction in ISI in all 3 treatment groups was greater than 7, the empirically validated threshold for clinically meaningful change³⁴; specifically, the mean reduction in ISI scores in each treatment group was at least 7.64 points at posttreatment and at least 8.58 points at 6FU. Moreover, the group differences between the improvements in ISI across the 3 therapies were all less than 1.8 points, much smaller than the minimally important difference thresholds (7 points) for noninferiority.³⁴ Therefore, taken together, these findings indicate that all 3 therapies lead to clinically meaningful improvements in insomnia symptoms in older adults and that there does not seem to be an advantage of one over another.

All 3 therapies led to significant improvement of insomnia symptoms measured by the secondary outcome measures, including SOL, WASO, TST, and SE, from baseline to posttreatment and baseline to 6FU. Importantly, these findings replicate and extend prior work in younger adults (aged 25 years and older) with persistent insomnia.²⁸ Overall, results of these 2 studies suggest that older adults are likely to experience similar benefits as younger adults across an array of insomnia-relevant measures. However, there were 2 notable exceptions. First, in the present older adult sample, there was no difference between BT and CT in improving SOL and WASO from baseline to posttreatment, as there was in younger adults.²⁸ The second difference was that in the present older adult sample, only CBT was superior to CT in improving TIB, whereas in younger adults, both therapies that contained BT (CBT and BT) were superior to CT in improving TIB at posttreatment. These differences could potentially be attributed to the longer therapy protocol (8 sessions) employed in the study of younger adults, which resulted in more time between baseline and posttreatment assessment, along with additional therapy sessions that may have facilitated differential treatment outcomes. There are also several differences between older adults and middle-aged/young adults that could contribute to these differences. For example, there are changes in sleep architecture and in processes involved in sleep regulation^54,55 as well as lifestyle differences (eg, retirement, loss of routine) that may contribute to differential sleep outcomes.⁵⁶ These differences in sleep and lifestyle factors may mean that different factors contribute to TIB in older adults and need to be addressed in future research.

In addition to insomnia symptoms, we examined the relative effectiveness of each therapy in improving functional outcomes. All 3 therapies led to significant improvements in DBAS, cognitive arousal, and fatigue. However, there was only a group-by-time interaction observed for DBAS. Given that CT aims to alter DBAS that may contribute to cognitive arousal while attempting to sleep, it is not surprising that both groups receiving CT (ie, CT and CBT) would experience a larger improvement in DBAS, relative to the group not receiving CT (BT).¹⁹ These results in older adults are consistent with the prior dismantling study in relatively younger adults.²⁸ Also consistent with the prior dismantling study, there were no significant differences in the reduction of MFI and Penn State Worry Questionnaire scores between treatment groups, suggesting that all 3 therapies are effective in improving fatigue and cognitive arousal in older adults with insomnia.

The main outcome findings suggest that CBT, BT, and CT are all effective in improving insomnia symptoms in older adults on a group level. However, it is possible that certain individuals may benefit more from 1 specific therapy, depending on individual differences. To explore this idea, we conducted a secondary analysis examining potential predictors of treatment response. Our secondary hypotheses in terms of predictors of response were largely unsupported with only ISI score at baseline significantly predicting ISI response across all 3 treatments. Specifically, a higher ISI score at baseline was associated with being more likely to achieve a clinically significant response to treatment. This is likely because having a higher baseline score means there is more room for reduction at subsequent time points (ie, regression to the mean). There were no other significant predictors of insomnia treatment response. Future research in larger samples is required to fully understand potential predictors of insomnia treatment response in older adults.

These results should be considered within the context of several limitations. First, the study did not meet its intended recruitment goals, which resulted in decreased statistical power to detect differences among treatment groups. Similarly, this study was underpowered to serve as a nonsuperiority trial; therefore, although there were no statistically significant differences among groups, we cannot conclude that all 3 treatments are equal. Further, because there is no control group, we cannot be certain that the changes in sleep observed over time are causal to the treatment. However, several randomized controlled trials have previously proven CBT-I to be effective in improving insomnia symptoms, relative to a control, in older adults,^21,57–59 and the within-group effect sizes for insomnia symptoms improvement in the current study are similar to or larger than those reported in these prior randomized controlled trials. Therefore, it is highly likely that the changes observed over time can be attributed to the therapy and not simply a passage of time. Lastly, it is unclear the degree to which these results will generalize to the general older adult population due to the strict exclusion criteria implemented in this study, the limited ethnic/racial diversity, and the high education levels in our sample. Future research should replicate these finding in a larger sample, with increased diversity.

Despite these limitations, the reported results advance our knowledge of the relative efficacy of CBT-I components in improving insomnia symptoms and functional outcomes in older adults. The results reported suggest minimal differences between the effectiveness of the 3 therapies in older adults, with a slight advantage for CBT and BT relative to CT in terms of better remission rates at 6 months. Given the different results between the current study and the previous dismantling study in relatively younger adults,²⁸ future studies should clarify whether these differences are attributable to age-related differences between the 2 populations or other protocol differences (ie, treatment length).

DISCLOSURE STATEMENT

All authors have seen and approved the manuscript. The authors report no conflicts of interest. The authors gratefully acknowledge funding from the National Institute of Mental Health (NIMHR01MH101468) and the Mental Illness Research Education and Clinical Center at Veterans Affairs Palo Alto Health Care System.

ABBREVIATIONS

6FU

6-month follow-up

BT

behavioral therapy

CBT

cognitive behavioral therapy

CBT-I

cognitive behavioral therapy for insomnia

CT

cognitive therapy

DBAS

dysfunctional beliefs about sleep

ISI

Insomnia Severity Index

MFI

Multidimensional Fatigue Inventory

PLM

periodic limb movement

PSS

Perceived Stress Scale

PSWQ

Penn State Worry Questionnaire

SE

sleep efficiency

SOL

sleep onset latency

TIB

time in bed

TST

total sleep time

VA

Veterans Affairs

WASO

wake after sleep onset