Effect of Cognitive Behavioral Therapy for Insomnia on Alcohol Treatment Outcomes Among US Veterans

This randomized clinical trial assesses the efficacy of cognitive behavioral therapy for insomnia in veterans receiving treatment for alcohol use disorder.

Key Points

Question

Is cognitive behavioral therapy for insomnia (CBT-I) feasible and efficacious early in treatment for alcohol use disorder, and if so, does it affect alcohol use?

Findings

In this randomized clinical trial involving veterans with insomnia who were in treatment for alcohol use disorder, CBT-I was associated with greater reductions in insomnia symptoms and alcohol-related problems over time than a single session of instruction about sleep hygiene. No group differences emerged for abstinence or heavy-drinking frequency.

Meaning

Long-term abstinence may not be required to derive benefit from CBT-I, which is feasible early in alcohol use disorder treatment and may reduce alcohol-related harm.

Abstract

Importance

Three of 4 adults in treatment for alcohol use disorder (AUD) report symptoms of insomnia. Yet the first-line treatment for insomnia (cognitive behavioral therapy for insomnia, CBT-I) is often delayed until abstinence is established.

Objective

To test the feasibility, acceptability, and preliminary efficacy of CBT-I among veterans early in their AUD treatment and to examine improvement in insomnia as a mechanism for improvement in alcohol use outcomes.

Design, Setting, and Participants

For this randomized clinical trial, participants were recruited through the Addictions Treatment Program at a Veterans Health Administration hospital between 2019 and 2022. Patients in treatment for AUD were eligible if they met criteria for insomnia disorder and reported alcohol use in the past 2 months at baseline. Follow-up visits occurred posttreatment and at 6 weeks.

Interventions

Participants were randomly assigned to receive 5 weekly sessions of CBT-I or a single session about sleep hygiene (control). Participants were asked to complete sleep diaries for 7 days at each assessment.

Main Outcomes and Measures

Primary outcomes included posttreatment insomnia severity (assessed using the Insomnia Severity Index) and follow-up frequency of any drinking and heavy drinking (4 drinks for women, ≥5 drinks for men; number of days via Timeline Followback) and alcohol-related problems (Short Inventory of Problems). Posttreatment insomnia severity was tested as a mediator of CBT-I effects on alcohol use outcomes at the 6-week follow-up.

Results

The study cohort included 67 veterans with a mean (SD) age of 46.3 years (11.8); 61 (91%) were male and 6 (9%) female. The CBT-I group included 32 participants, and the sleep hygiene control group 35 participants. Of those randomized, 59 (88%) provided posttreatment or follow-up data (31 CBT-I, 28 sleep hygiene). Relative to sleep hygiene, CBT-I participants reported greater decreases in insomnia severity at posttreatment (group × time interaction: −3.70; 95% CI, −6.79 to −0.61) and follow-up (−3.34; 95% CI, −6.46 to −0.23) and greater improvements in sleep efficiency (posttreatment, 8.31; 95% CI, 1.35 to 15.26; follow-up, 18.03; 95% CI, 10.46 to 25.60). They also reported greater decreases in alcohol problems at follow-up (group × time interaction: −0.84; 95% CI, −1.66 to −0.02), and this effect was mediated by posttreatment change in insomnia severity. No group differences emerged for abstinence or heavy-drinking frequency.

Conclusions and Relevance

In this randomized clinical trial, CBT-I outperformed sleep hygiene in reducing insomnia symptoms and alcohol-related problems over time but had no effect on frequency of heavy drinking. CBT-I should be considered a first-line treatment for insomnia, regardless of abstinence.

Trial Registration

ClinicalTrials.gov Identifier: NCT03806491

Introduction

Three of 4 individuals in treatment for alcohol use disorder (AUD) report symptoms of insomnia.¹ Because alcohol disrupts sleep and can maintain insomnia symptoms over time, abstinence is recommended as the first step in improving sleep in AUD.^2,3 However, following abstinence, most patients^4,5 continue to struggle with sleep disturbance, which decreases the likelihood of treatment completion^6,7 and increases risk for relapse.⁸ Indeed, approximately 40% of individuals with AUD report use of alcohol as a sleep aid.^9,10

Cognitive behavioral therapy (CBT-I) is the first-line treatment for insomnia in the general population.^11,12,13,14 It is more effective than sleep medication,^15,16,17 with most patients reporting improvements that endure over time.¹⁸ It has demonstrated efficacy in reducing insomnia among veterans^19,20,21 and individuals with AUD,^22,23,24 although trials thus far have not found corresponding reductions in rates of relapse. However, most trials have recruited individuals with well-established histories of abstinence at baseline, and no studies have assessed the effect of CBT-I on alcohol-related problems.¹⁷

This study tested the feasibility, acceptability, and preliminary efficacy of CBT-I as an adjunct to AUD treatment in a sample of veterans with insomnia. We hypothesized that CBT-I would be feasible and acceptable. We also hypothesized that, relative to a session about sleep hygiene that served as the control, CBT-I would be associated with greater reductions in insomnia severity immediately posttreatment and frequency of heavy drinking and alcohol-related problems at the 6-week follow-up.

Methods

Participants and Procedure

All procedures were approved by the institutional review board at the University of Missouri in collaboration with Harry S. Truman Memorial Veterans’ Hospital. The trial protocol is provided in Supplement 1. Veterans enrolled in the Addiction Treatment Program at this midwestern Department of Veterans Affairs (VA) hospital were recruited between August 2019 and April 2022. Patients reporting sleep problems completed a brief screening with the research coordinator, and eligible participants were scheduled for baseline assessments. All appointments took place through the VA. Participants provided written informed consent. They completed the Mini-Mental State Examination, a clinician-administered sleep assessment, Mini International Neuropsychiatric Interview,²⁵ and self-report measures (including self-identified race and ethnicity) with a trained assessor. Race and ethnicity were self-reported by participants as Black, Hispanic, multiracial, or White and groups were later combined for reporting because of small numbers. Participants were then asked to complete 7 consecutive days of sleep diaries to confirm the diagnosis of insomnia. Recruitment stopped at the end of the funding period. Reporting following guidelines from Consolidated Standards of Reporting Trials (CONSORT).

Eligibility, Randomization, and Blinding

Eligible participants (1) were enrolled in addiction treatment, (2) reported alcohol use in the past 2 months at baseline, (3) met DSM-5 criteria for moderate to severe AUD, and (4) met DSM-5 and research criteria for insomnia disorder.^26,27 Study criteria for insomnia were sleep onset latency or wake after sleep onset of longer than 30 minutes on 3 or more nights per week for 1 month or longer and scores of 10 or higher on the Insomnia Severity Index.²⁸ Exclusion criteria included contraindications for CBT-I (eg, seizures, mania), disorder requiring immediate clinical attention (eg, psychosis, suicidal intent/plan), or current engagement in behavioral insomnia treatment. Participants were not excluded for comorbid disorders or use of sleep medication but had to be stabilized with their medications for at least 6 weeks.

Participants were randomized in a parallel design to receive a single session of sleep hygiene instruction or 5 sessions of CBT-I. Random allocation (1:1 ratio) was performed by an investigator (M.B.M.) using a random number generator. Randomization was then managed by the study coordinator, who did not inform investigators or study interventionists of group assignment until baseline diaries confirmed participant eligibility. Research investigators were blind to outcomes, and follow-up assessors were blind to participant condition.

Follow-up

Data collection occurred in 3 periods of 7 days each, with 1 week of sleep diaries and self-report measures completed at baseline, posttreatment (6 weeks after baseline), and follow-up (6 weeks after the posttreatment point). Participants in both groups were assigned daily sleep diaries during the treatment phase. Participants received up to $150 as compensation (up to $40 for the baseline visit, $50 at posttreatment, and $60 at follow-up).

Most participants were enrolled in residential alcohol treatment, which lasted about 6 weeks. For these participants, baseline typically occurred within 1 to 2 weeks of intake, posttreatment occurred close to discharge (reflecting their time in residential treatment), and the 6-week follow-up captured their transition out of treatment.

Interventions

Both interventions were delivered individually in 30- to 60-minute sessions by master’s-level graduate students in clinical/counseling psychology (all trainees delivered CBT-I and sleep hygiene sessions). Clinical encounters were supervised by a licensed clinical psychologist (M.B.M.), in consultation with a psychologist board certified in behavioral sleep medicine (C.S.M.). To ensure treatment integrity,²⁹ interventionists received initial training via mock sessions, sessions were audiotaped for ongoing training and supervision, participants received a workbook of treatment materials, and adherence to treatment was reviewed each week. The protocol was modified to allow telehealth sessions because of the rurality of the population and COVID-19 risks.

CBT-I therapists followed a 5-session protocol.^30,31 Each session began with a review of sleep diaries and adherence. Content included sleep hygiene (session 1), stimulus control and sleep restriction (session 2), relaxation (session 3), cognitive techniques (session 4), and insomnia relapse prevention (session 5).

To model usual care, participants in the sleep hygiene control group reviewed a handout on sleep hygiene recommendations with a study therapist (eg, limit caffeine, bedtime routine). Participants were asked to identify and prioritize 1 or 2 recommendations over the next 5 weeks.

A VA clinician external to the treatment team coded 10 randomly selected sessions. All sessions received fidelity ratings of 100%.

Measures

Insomnia Symptoms

The Insomnia Severity Index²⁸ is a 7-item measure of satisfaction/dissatisfaction with sleep and daytime/functional impairment. Response options range from 0 to 4, with higher scores indicating more severe symptoms. A cut score of 10 or higher is optimal for detecting insomnia in community samples.²⁸ Internal consistency was adequate (α = .77).

Consistent with recommendations,³² sleep diaries assessed time in bed, time asleep, minutes to fall asleep (sleep onset latency), minutes awake at night (wake after sleep onset), time of final awakening, time out of bed, sleep quality (0, very poor, to 4, very good), and use of alcohol to help with sleep. Total sleep time was calculated by subtracting the sum of sleep onset latency and wake after sleep onset from the time elapsed between time asleep and time of final awakening. Sleep efficiency was calculated by dividing total sleep time by time in bed (0%-100%). Participants who could not complete diaries electronically were provided with paper versions. Using at least 5 diaries is recommended to reliably estimate sleep/wake times.³³ The percentage of participants who completed 5 or more diaries was 90% at baseline, 88% at posttreatment, and 89% at follow-up.

Alcohol Outcomes

Frequency of heavy drinking was assessed using the Timeline Followback.^34,35 Using a calendar marked with holidays and events, participants indicated how many standard drinks they consumed on each day in the past 6 weeks. Heavy drinking was defined as consumption of 4 drinks for women and 5 or more drinks for men. Total drinking quantity and frequency over the 42 days is reported for descriptive purposes.

Alcohol-related problems (eg, taking foolish risks, damaging relationships) in the past 6 weeks were assessed using a 13-item version of the Short Inventory of Problems (SIP).³⁶ The SIP has demonstrated reliability and validity among individuals with AUD³⁷ and in substance use treatment.³⁶ Reliability in this sample was good (α = .95).

Intensity, frequency, and duration of alcohol cravings in the past week were assessed using the 5-item Penn Alcohol Craving Scale, which has demonstrated good internal consistency and construct validity.³⁸ Reliability in this sample was good (α = .94).

Treatment Satisfaction

Participants rated acceptability of treatment using the 8-item Client Satisfaction Questionnaire (eg, “To what extent did the treatment we provided meet your needs?”).³⁹ Response options range from 1 to 4, with higher scores indicating higher satisfaction. This measure has been validated among individuals in substance use treatment⁴⁰ and demonstrated strong reliability (α = .95).

Data Screening and Analysis

Based on previous research,^20,21,23 we expected moderate to large effects on insomnia severity and moderate effects on alcohol use outcomes. Power to detect small, moderate, and large group × time interactions (α = .05) at 3 assessment points (r = 0.25) was estimated in G*Power version 3.0.10. A sample of 56 participants provided a high likelihood (91%-99%) of detecting group × time interactions of at least moderate magnitude.

Analyses were intent-to-treat (including all randomized participants) and used maximum likelihood to include all available data.

Multilevel models were specified, as assessments (baseline, posttreatment, and 6-week follow-up) were nested within person. For heavy-drinking frequency and alcohol-related problems, which followed count distributions, generalized linear models were conducted using a negative binomial distribution. Models were conducted in SAS version 9.4 (SAS Institute) using PROC MIXED or PROC GLIMMIX with restricted maximum likelihood estimation, an unstructured covariance structure, and a person-level random intercept. Separate models were conducted for each outcome. Consistent with recommendations,⁴¹ analyses focused on planned comparisons from baseline to posttreatment and baseline to follow-up. Predictors included time (treated categorically, reference = baseline), treatment group (sleep hygiene = 0, CBT-I = 1), and the group × time interaction. Pairwise comparisons were conducted to examine between-group differences and within-group change over time, using Bonferroni adjustment to control for inflation in type I error (α = .06/5 = .008). Effect sizes were calculated as the mean difference between groups at posttreatment and follow-up, divided by the pooled baseline standard deviation (Cohen d = 0.20 was small; 0.50, medium; and 0.80, large).⁴²

We conducted a longitudinal cross-lagged mediation model⁴³ to test posttreatment insomnia severity as a mediator of the association between treatment and alcohol problems at follow-up. The model, conducted in Mplus version 8.6 with full information maximum likelihood, adjusted for autoregressive and cross-lagged effects of insomnia severity at baseline and alcohol-related problems at baseline and posttreatment.⁴⁴ A negative binomial distribution with Monte Carlo integration (5000 integration points) was specified. The indirect effect was specified using model constraints, and bootstrapped 95% confidence intervals (10 000 repetitions) were calculated. In a sensitivity analysis, we specified a longitudinal latent-difference-score mediation model⁴⁵ to explicitly model whether change in insomnia severity at posttreatment, relative to baseline, mediated the association between treatment and change in alcohol problems at follow-up, relative to posttreatment. To determine if mediation effects were specific to insomnia, we also tested depression (measured using the 9-item Patient Health Questionnaire⁴⁶) as a mediator of treatment outcomes.

Results

Participant flow is depicted in the Figure. The study cohort included 67 veterans with a mean (SD) age of 46.3 years (11.8); 61 (91%) were male. The CBT-I group included 32 participants, and the sleep hygiene control group 35 participants (Table 1). Sixty-three participants (94%) completed all allocated treatment sessions. Fifty-nine (88%) completed at least 1 follow-up (97% CBT-I, 80% sleep hygiene; χ² = 4.53, P = .03). Retention for residential vs outpatient participants was 51 (88%) vs 8 (89%) for any follow-up data and 24 (86%) vs 4 (100%) for completion of CBT-I (Table 2^{46,47,48,49,50}).

Figure. Flow Diagram.

AUD indicates alcohol use disorder; CBT-I, cognitive behavioral therapy for insomnia.

Table 1. Demographic Characteristics of Participants at Baseline (N=67).

Characteristic	No. (%)
	Full sample (N = 67)	CBT-I group (n = 32)	Sleep hygiene group (n = 35)
Age, mean (SD), y	46.3 (11.8)	45.6 (11.9)	46.9 (11.9)
Sex
Male	61 (91)	29 (91)	32 (91)
Female	6 (9)	3 (9)	3 (9)
Race and ethnicity
Multiracial, Black, or Hispanica	11 (16)	6 (19)	5 (14)
White	56 (84)	26 (81)	30 (86)
Highest level of education
No collegea	14 (21)	7 (22)	7 (20)
Some college	34 (51)	16 (50)	18 (51)
College graduate	19 (28)	9 (28)	10 (29)
Employment statusb
Employed	14 (21)	8 (25)	6 (17)
Unemployed	30 (45)	14 (44)	16 (46)
Disabled	19 (28)	8 (25)	11 (31)
Housing
My own house/apartment	36 (54)	20 (63)	16 (46)
Someone else’s house/apartment	12 (18)	7 (22)	5 (14)
Transitional/institutional housing	16 (24)	5 (16)	11 (31)
On the street	3 (5)	0	3 (9)
Marital status
Never married	10 (15)	6 (19)	4 (11)
Married or living with partner	19 (28)	11 (34)	8 (23)
Divorced, separated, or widowed	38 (57)	15 (47)	23 (66)
Branch
Air Force	7 (10)	4 (13)	3 (9)
Army	43 (64)	22 (69)	21 (60)
Navy or Marinesa	17 (25)	6 (19)	11 (31)
Combat experiences, mean (SD), No.	2.8 (2.9)	2.7 (3.2)	2.9 (2.6)
Combat-related injury	56 (84)	26 (81)	30 (86)

Abbreviation: CBT-I, cognitive behavioral therapy for insomnia

^a Groups were combined for reporting to prevent identifiability because of small numbers.

^b Four participants indicated “prefer not to respond.”

Table 2. Clinical Characteristics of Participants at Baseline (N = 67).

Characteristic (instrument)	No. (%)
	Full sample (N = 67)	CBT-I group (n = 32)	Sleep hygiene group (n = 35)
Type of alcohol treatment
Residential	58 (87)	28 (88)	30 (86)
Outpatient	9 (13)	4 (13)	5 (14)
Session via telehealtha	10 (15)	7 (22)	3 (9)
Alcohol risk (AUDIT), mean (SD) scoreb	28.5 (6.4)	29.2 (5.8)	27.8 (6.9)
Abstinence
In the past 60 d, mean (SD), No. of days	25.0 (12.8)	24.1 (10.6)	25.8 (14.6)
Lasting 0-3 wk	46 (69)	23 (72)	23 (66)
Lasting ≥4 wk	21 (31)	9 (28)	12 (34)
Drinking, 42 d (TLFB), mean (SD)
Quantity, No. of drinks	263.8 (194.3)	294.8 (196.2)	235.5 (191.0)
Frequency, No. of days	18.1 (10.3)	18.4 (9.0)	17.7 (11.4)
Sleep data (participant diary), mean (SD)
Bedtimec	10:35 pm	10:40 pm	10:30 pm
Sleep onset latency, min	53.4 (42.7)	59.1 (47.9)	48.1 (37.1)
Wake after sleep onset, min	43.7 (33.0)	41.8 (25.1)	45.5 (39.3)
Waketimec	6:15 am	6:05 am	6:10 am
Total sleep time, h	6.0 (1.2)	5.7 (1.0)	6.2 (1.2)
Time in bed, h	8.6 (1.2)	8.3 (1.1)	8.8 (1.3)
Sleep quality scored	1.7 (0.6)	1.7 (0.5)	1.6 (0.6)
Substance use in past month (NSDUH)
Cigarettes	46 (69)	20 (63)	26 (74)
Smokeless tobacco	16 (24)	12 (38)	4 (11)
Cannabis	21 (31)	10 (31)	11 (31)
Cocaine	21 (31)	10 (31)	11 (31)
Nonprescription pain relievers, heroin, or methamphetaminee	12 (18)	4 (13)	8 (23)
Symptoms of depression (PHQ-9), mean (SD) scoref	11.6 (5.6)	10.9 (5.6)	12.3 (5.6)
Trauma exposure (clinical interview)	53 (79)	25 (78)	28 (80)
Symptoms of PTSD (PCL-5), mean (SD) scoreg	33.1 (21.8)	34.3 (22.3)	32.0 (21.6)
Symptoms of anxiety (GAD-7), mean (SD) scoreh	10.3 (5.4)	9.8 (5.0)	10.8 (5.8)
Nightmares (clinical interview)	58 (87)	27 (84)	31 (89)
Sleep apnea, high risk (STOP-Bang)	33 (49)	13 (41)	20 (57)
Apnea confirmed via PSG	12 (18)	5 (16)	7 (20)
Current use of sleep medicationi	50 (75)	26 (81)	24 (69)
Diphenhydramine	7 (10)	3 (9)	4 (11)
Melatonin	18 (27)	9 (28)	9 (26)
Mirtazapine	10 (15)	4 (13)	6 (17)
Trazodone	22 (33)	14 (44)	8 (23)
Gabapentin, hydroxyzine, or prazosine	12 (18)	7 (22)	5 (14)

Abbreviations: AUDIT, Alcohol Use Disorders Identification Test⁴⁷; CBT-I, cognitive behavioral therapy for insomnia; GAD-7, 7-item Generalized Anxiety Disorder scale⁴⁸; NSDUH, National Survey on Drug Use and Health; PCL-5, PTSD Checklist for DSM-5⁴⁹; PHQ-9, 9-item Patient Health Questionnaire⁴⁶; PSG, polysomnography; PTSD, posttraumatic stress disorder; STOP-Bang, sleep apnea screening instrument⁵⁰; TLFB, Timeline Followback.

^a Coded yes if any sessions were completed remotely.

^b AUDIT scores range from 0 to 40, with higher scores indicating greater consumption and risk of harm.

^c Participants reported these times in minutes in military time; they were translated to am/pm and 5-minute intervals for ease of interpretation.

^d Sleep quality was ranked on a scale from 0, very poor, to 4, very good.

^e Groups were combined for reporting to prevent identifiability because of small numbers.

^f PHQ-9 scores range from 0 to 27, with higher scores indicating greater risk of depression.

^g PCL-5 total scores range from 0 to 80, with higher scores indicating greater risk of PTSD.

^h GAD-7 scores range from 0 to 21, with higher scores indicating greater anxiety severity.

ⁱ As reported on the clinical interview or sleep diaries; 14 participants reported use of multiple medications. Only 1 participant each reported use of the following sleep medications: aripiprazole, clonazepam, cyclobenzaprine, doxazosin, Nyquil, ropinirole, temazepam.

Primary Outcomes

Descriptive and inferential statistics for all outcomes are depicted in Table 3 and Table 4. In the prediction of insomnia severity, there was a significant group × time interaction at posttreatment and follow-up. Participants in both groups reported significant decreases in insomnia severity from baseline to posttreatment and baseline to follow-up, with CBT-I participants reporting greater decreases at posttreatment and follow-up. A smaller proportion of CBT-I than sleep hygiene participants still met study criteria for insomnia at posttreatment (32% vs 59%; χ² = 4.08, P = .04) and follow-up (40% vs 72%; χ² = 3.98, P = .046).

Table 3. Descriptive Statistics for Primary and Secondary Treatment Outcomes (N = 67).

Outcome	Mean (SD)			Cohen d (95% CI)a
	Baseline	Posttreatment	6-wk Follow-up	Baseline to posttreatment	Baseline to follow-up
Primary outcomes
Insomnia severity				0.91 (0.40 to 1.41)	0.70 (0.20 to 1.19)
CBT-I group	17.97 (4.66)	5.69 (4.70)b,c	8.07 (7.13)b,c
Sleep hygiene group	19.26 (4.01)	10.96 (6.43)b	12.42 (6.34)b
Heavy-drinking frequency				−0.08 (<0.01 to 0.54)	0.04 (−0.44 to 0.52)
CBT-I group	16.97 (9.58)	2.07 (5.26)b	5.92 (11.45)
Sleep hygiene group	17.14 (11.74)	1.34 (4.49)b	6.57 (13.56)
Alcohol problems				0.36 (−0.12 to 0.85)	0.53 (0.04 to 1.02)
CBT-I group	25.44 (10.11)	4.42 (6.19)b	4.38 (7.65)b
Sleep hygiene group	22.00 (12.17)	5.12 (9.10)b	7.00 (9.03)b
Secondary outcomes
Sleep efficiency				0.61 (0.12 to 1.10)	1.43 (0.89 to 1.96)
CBT-I group	68.47 (11.91)	87.30 (7.50)b	90.62 (5.26)b,c
Sleep hygiene group	71.44 (13.32)	82.41 (11.51)b	75.31 (14.11)
Alcohol craving				0.01 (−0.25 to 0.25)	−0.07 (<0.01 to 0.46)
CBT-I group	11.44 (8.85)	7.88 (6.73)b	9.31 (7.25)
Sleep hygiene group	10.76 (7.59)	7.15 (6.25)b	9.21 (6.86)

Abbreviation: CBT-I, cognitive behavioral therapy for insomnia.

^a Cohen d interpreted as 0.20, small; 0.50, medium; and 0.80, large. A negative effect size indicates change in favor of the sleep hygiene group.

^b Significant within-group change from baseline (P ≤ .008).

^c Significant between-group difference at that time point (P ≤ .008).

Table 4. Inferential Statistics for Primary and Secondary Treatment Outcomes (N = 67).

Outcomea	Estimate (95% CI)
	Group	Posttreatment	Follow-up	Group × posttreatment interaction	Group × follow-up interaction
Primary outcome
Insomnia severity	−1.29 (−3.99 to 1.41)	−8.07 (−10.24 to −5.89)d	−6.84 (−9.08 to −4.60)b	−3.70 (−6.79 to −0.61)d	−3.34 (−6.46 to −0.23)d
Heavy-drinking frequencyb,c	−0.01 (−0.90 to 0.87)	−2.54 (−3.50 to −1.57)d	−0.91 (−1.86 to 0.03)	0.46 (−0.89 to 1.81)	−0.14 (−1.48 to 1.19)
Alcohol problemsc	0.24 (−0.34 to 0.82)	−1.89 (−2.47 to −1.31)d	−1.39 (−1.96 to −0.81)b	−0.19 (−1.01 to 0.63)	−0.84 (−1.66 to −0.02)d
Secondary outcome
Sleep efficiency	−2.97 (−8.43 to 2.48)	10.62 (5.47 to 15.77)d	4.56 (−0.89 to 10.01)	8.31 (1.35 to 15.26)d	18.03 (10.46 to 25.60)d
Alcohol craving	0.61 (−3.00 to 4.22)	−3.32 (−5.78 to −0.86)d	−2.05 (−4.58 to 0.48)	−0.66 (−4.13 to 2.80)	−0.45 (−3.97 to 3.06)

^a F values for the interaction of group × (linear) time: insomnia severity, 3.56 (P = .03); frequency of heavy drinking, 0.37 (P = .69); alcohol problems, 2.11 (P = .13); sleep efficiency, 11.36 (P < .001); and alcohol craving, 0.08 (P = .93).

^b Heavy drinking is defined as 4 drinks for women and ≥5 for men.

^c Generalized linear model conducted using a negative binomial distribution.

^d P < .05.

At posttreatment, 18 of 53 participants (34%) reported any alcohol use in the past 6 weeks (8 CBT-I and 10 sleep hygiene; χ² = 0.46, P = .50). At follow-up, 24 of 53 participants (45%) reported alcohol use (13 CBT-I and 11 sleep hygiene; χ² = 0.03, P = .87). There was no significant group × time interaction in the prediction of heavy-drinking frequency. Participants in both groups reported significant decreases in heavy-drinking frequency from baseline to posttreatment.

There was a significant group × time interaction in the prediction of alcohol-related problems at follow-up. Again, participants in both groups reported significant decreases in alcohol-related problems from baseline to posttreatment and follow-up, with CBT-I participants reporting comparable decreases at posttreatment and significantly greater decreases at follow-up.

Secondary Outcomes

There was insufficient variability in use of alcohol as a sleep aid to conduct analyses. Eight participants (12%; 4 CBT-I and 4 sleep hygiene) reported using alcohol as a sleep aid at baseline, 6 at posttreatment (13% of those who completed diaries; 4 CBT-I and 2 sleep hygiene), and 9 at follow-up (24% of those who completed diaries; 5 CBT-I and 4 sleep hygiene).

There was no significant group × time interaction in the prediction of alcohol craving. Participants in both groups reported significant decreases in alcohol craving from baseline to posttreatment.

There was a significant group × time interaction in the prediction of sleep efficiency. Participants in both groups reported significant improvements in sleep efficiency from baseline to posttreatment, with CBT-I participants also reporting significant improvements from baseline to follow-up. Again, CBT-I participants reported greater improvements at both posttreatment and follow-up.

Treatment Satisfaction, Adverse Events, and Protocol Deviations

CBT-I participants reported greater mean (SD) satisfaction (29.3 [3.1]) with insomnia treatment than those receiving sleep hygiene instruction (24.4 [4.3]; t(49) = 4.76; P < .001).

No adverse events occurred. We modified the protocol to allow remote delivery of treatment and assessments and exclude participants who lived outside the state where supervising psychologists were licensed.

Mediation Models

In the longitudinal cross-lagged mediation model, CBT-I was associated with insomnia severity at posttreatment (estimate = −4.33, 95% CI, −7.22 to −1.43; P = .003), and posttreatment insomnia severity was associated with alcohol-related problems at follow-up (estimate = 0.24, 95% CI, 0.10 to 0.37; P = .001). The indirect effect was significant (estimate = −1.02; 95% CI, −1.91 to −0.13; P = .02), indicating that posttreatment insomnia severity mediated the association between treatment group and alcohol-related problems at follow-up. Results did not differ in the latent-difference-score mediation model. CBT-I predicted change in insomnia severity at posttreatment (estimate = −4.47; 95% CI, −7.25 to −1.46; P = .003), posttreatment change in insomnia severity predicted change in alcohol-related problems at follow-up (estimate = 0.27; 95% CI, 0.07 to 0.39; P < .001), and the indirect effect was significant (estimate = −1.26; 95% CI, −2.08 to −0.20; P = .01).

Posttreatment depressive symptoms did not mediate CBT-I effects on alcohol-related problems at follow-up in the cross-lagged (estimate = −0.49; 95% CI, −1.29 to 0.30; P = .22) or latent-difference-score models (estimate = −0.62; 95% CI, −0.12 to 0.22; P = .10).

Discussion

Previous studies have consistently demonstrated that CBT-I reduces insomnia among individuals with AUD, but none has found an effect on alcohol-related outcomes.^17,22 This is the first study to demonstrate a direct effect of CBT-I on alcohol-related problems. On average, at the end of insomnia treatment, CBT-I participants reported a 68% drop in insomnia severity, while sleep hygiene participants reported a 43% reduction; and this posttreatment improvement in insomnia was associated with reductions in alcohol-related problems at follow-up. This is a significant finding because it challenges current recommendations to postpone behavioral insomnia treatment until patients have achieved 4 weeks or longer of abstinence.³ Indeed, data from this study suggest that waiting for individuals to succeed with abstinence (which may not be necessary for all individuals with AUD⁵¹) may be a missed opportunity to improve patients’ sleep and influence their experience of alcohol-related harm.

Similar to previous trials, we did not find a significant effect of CBT-I on alcohol use. This contradicts findings that (1) normalizing sleep reduced drinking in animal models⁵² and (2) pharmacological treatment of insomnia (via gabapentin) delayed alcohol relapse at 6 to 12 weeks.^53,54 It is possible that psychosocial treatments like CBT-I have different mechanisms of action than biological/pharmacological treatments and do not affect alcohol use the same way. Indeed, in both previously mentioned pharmacological studies, gabapentin had an effect on relapse in the absence of an effect on sleep.^53,54 Alternatively, studies may not have found CBT-I effects on alcohol use because of floor effects. Specifically, because only a proportion of individuals with AUD drink alcohol in the first few months of recovery, large samples (>250) may be required to document CBT-I effects on actual drinking behavior. This study included the largest number of participants to date, with more than 30 per group, but this provides insufficient power to detect small effects.

Strengths and Limitations

To our knowledge, this is the largest clinical trial of CBT-I among people with AUD to date. However, the sample was 91% male and 84% White, and 87% of participants were in residential treatment, where substance use is monitored or prohibited. This means findings may not generalize to samples with greater racial and ethnic diversity or to women, and we have limited data on generalizability to outpatient settings. Access to residential treatment is often limited and cost-prohibitive outside the VA, so replication in outpatient and civilian settings is needed. Moreover, only about 70% of participants completed both follow-ups. This level of attrition was expected for the population, and we used state-of-the-art techniques to handle missing data.⁵⁵ However, it serves as a reminder of the context in which alcohol treatment is often provided. For people struggling with housing issues, limited social support, and/or unemployment, insomnia treatment may not be the first priority. These issues may be especially pronounced for those in outpatient (vs residential) treatment. Incorporating CBT-I into existing treatment programming is expected to improve the feasibility of delivery in this population.

The trial design also had strengths and limitations. Inclusion of individuals with a wide range of comorbidities is a strength in relevance to clinical practice; however, it likely contributed to variability in outcomes. For example, individuals with posttraumatic stress disorder or sleep apnea may benefit from CBT-I, but some sleep problems (eg, nightmares, daytime sleepiness) will persist if these conditions remain untreated. Studies testing the efficacy of transdiagnostic, modular treatments⁵⁶ that allow clinicians to tailor care to the unique sleep needs of each patient are strongly encouraged. Likewise, use of a single-session sleep hygiene control makes it impossible to know if results are unique to CBT-I or due to nonspecific therapy effects (eg, therapist/treatment time).

Finally, we are limited to interpretation of the measures included. For example, self-report is the recommended method of assessment for insomnia¹²; and among individuals with AUD, subjective sleep measures are better predictors of future drinking than objective measures.⁵⁷ However, objective measures of sleep (eg, polysomnography, actigraphy) would have provided insight on other aspects of sleep that may or may not change with CBT-I. Similarly, although inclusion of alcohol-related problems as an outcome is novel, the Short Inventory of Problems has not demonstrated measurement invariance in all studies.^37,58 Research is also needed to understand how CBT-I may have an effect on alcohol problems in the absence of an effect on drinking. It is possible that CBT-I causally reduces alcohol-related harm; for example, improvements in insomnia could lead to improvements in response inhibition that reduce the problems experienced while drinking. Alternatively, CBT-I may be correlated with alcohol-related problems because the problems people experience as a result of insomnia are similar to the problems they experience from drinking (eg, depressed mood, harmed relationships). Future studies are encouraged to test this. In either case, CBT-I seems to reduce harm in this population.

Conclusions

This randomized clinical trial found that CBT-I reduced insomnia symptoms among adults in early alcohol recovery and may improve treatment outcomes by reducing alcohol-related harm. Data suggest that delaying delivery of insomnia treatment until abstinence is established is unnecessary. CBT-I should be considered the first-line treatment for insomnia in AUD, regardless of abstinence.

Supplement 1.

Trial protocol

Supplement 2.

Data sharing statement