The Effect of Cognitive Behavioral Therapy for Insomnia (CBT-I) on Cardiometabolic Health Biomarkers: A Systematic Review of Randomized Controlled Trials

Abstract

Objectives:

To assess the effectiveness of Cognitive Behavioral Therapy for Insomnia (CBT-I) on cardiometabolic health biomarkers.

Method:

Cochrane CENTRAL, Embase, Medline, and PsycINFO were searched, and records were screened by two independent reviewers. Inclusion criteria were adult population, delivery of CBT-I, randomized controlled trial design, ≥1 cardiometabolic health outcome, and peer-review. Hedge’s g effect sizes were calculated, and the quality of the evidence was appraised using the Cochrane Risk of Bias 2 tool.

Results:

After screening 1649 records, 15 studies were included (total N=2067). Inflammatory markers (CRP, IL-6, TNF-α), blood pressure (SBP, DBP), and glycemic regulation (HbA1c) were most frequently reported (in ≥3 studies each). HbA1c and CRP were reduced in the CBT-I group compared to the control group (in 3 studies each). Effects varied or were null for IL-6, TNF-α, SBP, and DBP. Six studies were judged as low, four as moderate, and five as high risk of bias.

Conclusion:

CBT-I was most consistently associated with improved HbA1c and CRP, which are relatively temporally stable, suggesting influences on enduring habits rather than short-term behavior changes. High risk of bias limits the interpretation of findings. Methodologically adequate studies are needed to better understand cardiometabolic effects of CBT-I.

Introduction

Inadequate sleep has become increasingly common in the U.S.(Centers for Disease Control and Prevention, 2014) Epidemiologic studies link insomnia with prevalent and incident cardiovascular disease (CVD).(Jarrin et al., 2018; Javaheri & Redline, 2017) Insomnia has also been associated with incident obesity,(Ogilvie & Patel, 2017) diabetes, and adverse diabetes outcomes.(Ogilvie & Patel, 2018)

Although cross-sectional and longitudinal observational studies relate disordered sleep to poor cardiometabolic health, experimental sleep deprivation and extension studies provide insight into mechanisms and causality. Experimental sleep deprivation (from 8 hours to 4 hours) has been associated with lower glucose clearance and greater insulin resistance.(Spiegel et al., 1999) Additionally, it has been linked to adverse immunologic responses(Spiegel et al., 2002) and increased levels of inflammatory biomarkers associated with CVD risks, such as C-reactive protein (CRP, Meier-Ewert et al., 2004) and interleukin-6 (IL-6, Haack et al., 2007). Lesser experimental sleep deprivation, from 8 hours to 6 hours, has additionally been linked to increased levels of tumor necrosis factor-α (TNF-α, Vgontzas et al., 2004). Conversely, experimental sleep extension from 6 to 10 hours has been shown to improve insulin sensitivity and reduce appetite(Killick et al., 2015). Similar findings have been demonstrated with sleep extensions by 1 to 2 hours from participants’ baseline(Leproult et al., 2015; Tasali et al., 2014). Whether sleep extension may improve other cardiometabolic biomarkers, such as resting heart rate or blood pressure (BP), remains unclear.(Baron et al., 2019; Haack et al., 2013) While well-controlled laboratory studies have demonstrated that sleep extension and deprivation influence cardiometabolic health, few studies have examined the cardiometabolic effects of a real-world sustainable intervention focused on improving sleep health habits (i.e., versus modifying sleep duration).

Cognitive Behavioral Therapy for Insomnia (CBT-I) is an adaptation of traditional Cognitive Behavioral Therapy designed specifically to treat insomnia. CBT-I focuses on improving multiple aspects of sleep health by building cognitive and behavioral skills and is considered the gold standard treatment for insomnia by the American Psychological Association and American College of Physicians.(Boness et al., 2020; Qaseem et al., 2016) Meta-analyses of randomized controlled trials (RCTs) have shown CBT-I to have a medium to large effect size in improving insomnia symptoms, onset latency, and sleep efficiency following treatment(Hertenstein et al., 2022; van der Zweerde et al., 2019) with improvements sustained over time.(van der Zweerde et al., 2019) Reviews have also demonstrated more sustained improvements in sleep in those who received CBT-I versus pharmacologic intervention.(Morin et al., 2006)

CBT-I utilizes several components meant to improve multiple dimensions of sleep health sustainably. For example, stimulus control targets the inability to fall asleep by strengthening the bed as a cue for sleep using classical conditioning. Sleep restriction therapy decreases time spent awake in bed by limiting time in bed to time spent asleep. Relaxation techniques (e.g., deep breathing, progressive muscle relaxation) help reduce bedtime arousal, and cognitive therapy aims to restructure negative sleep-related cognitions to reduce anxiety and improve mood. Sleep hygiene education is also provided, such as recommendations related to optimal bedroom conditions and caffeine intake. Though these are the common components, CBT-I may be further individualized according to patient needs.

The evidence that CBT-I is effective for sustainably improving sleep health is strong, as is the epidemiologic evidence associating poor sleep health with cardiometabolic risk. However, no systematic review has examined the effect of CBT-I on improving cardiometabolic health or related biomarkers. Thus, the current systematic review aims to assess the effectiveness of CBT-I on cardiometabolic health indicators.

Methods

Protocol Registration

The protocol for this review was registered a priori with the International Prospective Register of Systematic Reviews (PROSPERO; ID: CRD42021256981).

Eligibility Criteria

Studies were eligible for inclusion in the review if they: (1) examined an adult (≥18 years) sample; (2) administered CBT-I; (3) were an RCT; (4) reported at least one cardiometabolic health outcome in response to CBT-I; and (5) were peer-reviewed (i.e., excluded dissertations, conference abstracts). We considered CBT-I interventions described as cognitive behavioral therapy specifically for insomnia, including one or more of the components: stimulus control, sleep restriction therapy, and cognitive therapy. Sleep hygiene alone was not considered CBT-I. There were no restrictions on study year, language, or intervention delivery modality, duration, or frequency.

Information Sources and Search Strategy

The Cochrane CENTRAL, Embase, Medline, and PsycINFO databases were searched for articles through May 4^th, 2022. Search fields included title, abstract, and keywords. The following search strategy was used:

1. Cognitive Behavioral Therapy OR CBT-I OR CBT-Insomnia OR CBTI OR CBT OR (cogniti* AND behavio* AND therap*)

AND

• (2)Insomnia OR sleep

AND

• (3)Cardiometabolic OR cardiovascular OR heart OR coronary OR cardiac OR hypertension OR blood pressure OR inflam* OR diabetes OR insulin OR glucose OR obesity OR weight OR adiposity OR cholesterol OR triglyceride

Study Selection Process

Two independent reviewers screened the studies in two stages. First, references and abstracts obtained from electronic databases were imported into Rayyan web-based software(Ouzzani et al., 2016), and duplicates were removed manually. Titles and abstracts were screened. Next, reviewers read full texts of studies to assess eligibility. We also reviewed the reference lists of included studies.

Data Collection and Data Items

Two independent reviewers conducted data extraction. Data extraction forms were created a priori and included: study sample characteristics, location, intervention characteristics (e.g., delivery method; number, frequency, and duration of sessions; CBT-I components), comparison group characteristics, outcomes of interest, follow-up frequency and duration, primary author’s narrative of findings, and data for effect size calculations.

Risk of Bias Assessment

The Cochrane Risk of Bias 2 (ROB-2) tool was used to assess the risk of bias by two independent reviewers.(Sterne et al., 2019) This tool evaluates the following domains: randomization, deviations from intended interventions, missing outcome data, measurement of the outcome, and selection of the reported result. Studies were classified as being at “high risk”, “some concerns,” or “low risk” of bias in each of the domains and overall. The ROB-2 specifies an algorithm for determining ratings within each domain based on responses to individual items.

Summary Measures and Synthesis of Results

Primary studies’ effects were narratively synthesized in text and tables. Effect sizes expressed as Hedge’s g, standardized mean differences, were calculated and interpreted as 0.20 = small, 0.50 = medium, and 0.80 = large magnitude.(Cohen, 1988). Effect sizes were calculated as the control group mean at follow-up minus the CBT-I group mean at follow-up divided by the pooled standard deviation. Positive values indicate that the control group mean is greater than the CBT-I group mean. Effect sizes were calculated for each outcome and/or time point if the information was reported. Meta-analyses were not conducted due to significant clinical and methodological heterogeneity (in biomarkers, study sample characteristics, comorbidities, and time frame of follow-up measurements), nor was reporting bias across studies for the same reason. Study reporting follows PRISMA 2020 guidelines (for Checklist, see supplementary material).(Page et al., 2021)

Results

Study Selection

Database searches yielded 1,649 articles, which were assessed for eligibility, and 15 publications from 13 unique trials met inclusion criteria (see flow diagram; Figure 1).

Figure 1.

PRISMA 2020 Flow Diagram.

From: Page MJ, McKenzie JE, Bossuyt PM, Boutron I, Hoffmann TC, Mulrow CD, et al. The PRISMA 2020 statement: an updated guideline for reporting systematic reviews. BMJ 2021;372:n71. doi: 10.1136/bmj.n71

Study Characteristics

Study sample sizes ranged from 22 to 1,142, with a median of 90 participants and total N across studies=2,067(Table 1). The mean age across studies was 56 years and the mean proportion of female participants was 63%. Comparator or control groups ranged from passive conditions (e.g., waitlist control, usual care) in 6 studies(Javaheri et al., 2020; Johann et al., 2020; McGrath et al., 2017; Yang et al., 2017; Zhang et al., 2021; Zuo et al., 2020) to active conditions (e.g., education, tai chi) in 9 studies(Alshehri et al., 2020; Cain et al., 2020; Carroll et al., 2015; Chen et al., 2011; Chen et al., 2008; Dolsen et al., 2018; Irwin et al., 2015; Irwin, Olmstead, Breen, et al., 2014; Irwin, Olmstead, Carrillo, et al., 2014). Most interventions were delivered in person, either individually or in a group, but several were delivered online. Of the 12 studies that delivered in person interventions, 4 (33%) delivered the intervention individually, 7 (58%) in a group format, and 1 (8%) used a hybrid approach with individual intervention delivery plus group discussion afterward. Most studies delivered CBT-I in weekly sessions of 45-60 minutes in length (range=30 minutes to 2 hours). The overall median intervention duration was eight weeks, slightly longer than the typical CBT-I treatment protocol of 6 sessions(Morin et al., 2006). The most frequent elements of CBT-I included cognitive therapy (93%), stimulus control (87%), sleep restriction (67%), sleep hygiene education (60%), and relaxation (60%).

Table 1.

Characteristics of Studies, Interventions, and Comparison Groups

Study ID	Sample Characteristics	Target Population	Study Location	Intervention Delivery Method	Intervention Dose	CBT-I Components	Comparison Group	Baseline and Post- Intervention Total Sleep Times (mins)
Alshehri et al. (2020)	N = 28 Age, sex, race/ethnicity, education, income NR	Patients with type 2 diabetes, ages 40-75 years	Kansas City, KS	In person, individual	6 weekly, 45-min	Sleep restriction therapy, stimulus control therapy, sleep hygiene, relaxation techniques, cognitive therapy	Health education, including brief sleep hygiene, foot care, diabetes classifications, healthy diet, and physical activity.	CBT-I Pre: 424 Post: 393 Comparison Pre: 425 Post: 395 Per actigraphy CBT-I Pre: 392 Post: 429 Comparison Pre: 321 Post: 326 Per sleep diary
Cain et al. (2020)	N = 53, Age = 29.1 (5.6), 100% female White: 41.5% Black: 56.6% Other: 3.8% Hispanic: 26.4% Non-Hispanic: 73.6% Education, income NR	Pregnant women with pre-pregnancy BMI of 25 or higher	Tampa, FL	In person group and online	7 weekly, 90 min	Stimulus control, mindfulness training, relaxation techniques, and sleep hygiene	Group prenatal care - facilitated discussions on topics of pregnancy, nutrition, appropriate weight gain, labor and delivery options, lactation and contraception	NR
Carroll et al. (2015)	N = 109, Age = 66.0, 72% female White: 86% Other: 14% Education, income NR	Ages 55+	Los Angeles, CA	In person, group	16 weekly, 2 hours	Biopsychosocial model and insomnia, cognitive restructuring, stimulus control, mood enhancement	(1) Tai chi chih; (2) Sleep Seminar	NR
Chen et al. (2008)	N = 26, Age = 49.2 (12.0), 42% female Race/ethnicity, education, income NR	Patients undergoing peritoneal dialysis	Taipei, Taiwan	In person, individual	4 weekly, 1 hour	Sleep restriction, stimulus control, relaxation, cognitive therapy, education	Sleep hygiene education	NR
Chen et al. (2011)	N = 72, Age = 58 (11), 58% female Race/ethnicity, income NR Unable to read or write: 2.5% Primary school: 41.25% Secondary school: 16.25% High school: 31.25% College: 8.75%	Patients undergoing hemodialysis	Taipei, Taiwan	Video-assisted CBT during hemodialysis, group discussion, and education after hemodialysis	18 triweekly, 30 min	Sleep restriction, stimulus control, relaxation, cognitive therapy	Sleep hygiene education	CBT-I Pre: 300 Post: 348 Comparison Pre: 324 Post: 342 Per self-report on PSQI
Dolsen et al. (2018)	N = 22, Age = 36.4, 55% female Race Asian: 22.7% African American: 13.6% White: 50% Bi-racial/multi-racial: 9.1% Ethnicity Hispanic: 9.1% Non-Hispanic: 86.4% Income <$20,000: 45.5% $20k-$35k: 4.5% $35k-$50k: 22.7% $50k-$60k: 13.6% >$60,000: 4.5% Education NR	Adults with interepisode bipolar disorder	Berkeley, CA	In person, individual	8 weekly, 50-60 min	Sleep/circadian education, stimulus control, sleep restriction, regularizing sleep-wake times, cognitive therapy	Psychoeducation related to sleep, stress, diet, health, exercise, and mood in bipolar disorder.	CBT-I Pre: 396 Post: 409 Comparison Pre: 442 Post: 491 Per sleep diary
Irwin, Olmstead, Breen, et al. (2014)	N = 90, Age = 59.8, sex NR 85.6% white; other race/ethnicity categories not reported Mean years of education: 15.75 Income NR	Breast cancer survivors	Los Angeles, CA	In-person, group	12 weekly, 2 hours	Biopsychosocial model and insomnia, cognitive restructuring, stimulus control, mood enhancement	Tai chi chih	NR
Irwin, Olmstead, Carrillo, et al. (2014)	N = 123, Age = 65.55, 72% female White: 86.18% Non-white: 13.82% Hispanic: 7.32% Non-Hispanic: 92.68% Mean years of education: 15.66 Income NR	Older adults	Los Angeles, CA	In person, group	16 weekly, 2 hours	Biopsychosocial model and insomnia, cognitive restructuring, stimulus control, mood enhancement	(1) Tai chi chih; (2) Sleep Seminar	CBT-I Pre: 371 Post: 384 Comparison 1 (Tai chi chih) Pre: 379 Post: 381 Comparison 2 (Sleep seminar) Pre: 356 Post: 379 Per PSG
Irwin et al. (2015)	N = 123, Age = 65.55, 72% female White: 86.18% Non-white: 13.82% Hispanic: 7.32% Non-Hispanic: 92.68% Mean years of education: 15.66 Income NR	Older adults	Los Angeles, CA	In-person, group	16 weekly, 2 hours	Biopsychosocial model and insomnia, cognitive restructuring, stimulus control, mood enhancement	(1) Tai chi chih; (2) Sleep Seminar	(Same study as above)
Javaheri et al. (2020)	N = 34, Age = 71.6 (9.5), 26% female White: 85.3% Black: 5.9% Asian: 5.9% Less than high school: 2.9% high school graduate: 14.7% college degree: 23.5% graduate degree: 55.9% Income NR	Adults with coronary heart disease	Boston, MA and Cleveland, OH	Interactive online program	42 daily over 6 weeks, self-paced	Stimulus control, sleep restriction, sleep hygiene, cognitive therapy, relaxation	General sleep education, then an option for treatment (waitlist control)	NR
Johann et al. (2020)	N = 46, Age = 41.0 (14.5), 63% female Race/ethnicity, education, income NR	Ages 18-65 years	Freiburg, Germany	In-person, individual	8 weekly, 50 min	Sleep hygiene education, relaxation, sleep restriction, stimulus control, and cognitive therapy	Waitlist control	CBT-I Pre: 354 Post: 378 Comparison Pre: 342 Post: 372 Per sleep diary
McGrath et al. (2017)	N = 134, Age = 59, 61% female Race/ethnicity, education, income NR	Patients with elevated blood pressure (SBP=130-160, DBP < 110)	Galway, Ireland	Online, individual	6-8 weekly, self-paced	Sleep hygiene education, stimulus control, sleep restriction therapy, cognitive therapy	Standard care, including a vascular risk factor education session	CBT-I Pre: 416 Post: 420 Comparison Pre: 428 Post: 431 Per sleep diary
Yang et al. (2017)	N = 106, Age = 56.37, 63% female Mean years of education: 9.25 Race/ethnicity, income NR	Patients with hypertension	Chongqing, Southwest China	Online, individual	At least once/week for 8 weeks, length NR	Stimulus control, sleep restriction, sleep hygiene education, cognitive therapy, and relaxation therapy	Regular treatment for hypertension - sleep-blood pressure diary and questionnaires, standard psychoeducation on cardio-cerebral vascular risk factors, and medication for hypertension	CBT-I Pre: 250 Post: 303 Comparison Pre: 256 Post: 263 Per sleep diary
Zhang et al. (2021)	N = 1142, Age = 61.71, 66% female Elementary school and below: 47.46% Junior high school: 42.99% High school and above: 9.54% Race/ethnicity, income NR	Patients with type 2 diabetes	Communities of Xuzhou City, China	In person, group	5 daily in first week, then 2 over the next 6 weeks, 50-60 min	Education about the relationship between sleep and diabetes, sleep hygiene education, regularizing the sleep-wake cycle, relaxation, cognitive therapy	Usual care (all participants): Every 3 mo, general practitioners conducted a face-to-face visit in which they recorded the participant’s status; provided advice on diet, medication, and exercise; and monitored blood sugar.	NR
Zuo et al. (2020)	N = 191, Age = 62.81, 67% female Race/ethnicity, education, income NR	Patients with type 2 diabetes	Communities of Xuzhou City, China	In person, group	Daily for the first week, then weekly for 7 weeks, 40-50 min	Education about the relationship between sleep and diabetes, sleep hygiene education, sleep restriction, cognitive therapy, relaxation, behavioral activation/aerobic exercise	Usual care: Staff recorded patient's health status, provided advice on diet, medication, and exercise, studied their physiology, and monitored their glucose levels	NR

Note: BMI = Body mass index. CBT-I = Cognitive Behavioral Therapy for Insomnia. DBP = diastolic blood pressure. NR = Not reported. PSG = Polysomnography. PSQI = Pittsburgh Sleep Quality Index (Buysse et al., 1989). SBP = systolic blood pressure.

Risk of Bias in Studies

The ROB-2 scale revealed that studies varied in quality and risk of bias across multiple domains (Tables 2 and 3). In randomization, most studies were classified as having a low risk of bias due to having a concealed and random allocation sequence with few baseline characteristic imbalances between groups. Studies rated as “some concerns” in randomization lacked information about how the allocation sequence was randomized and/or concealed. In deviations from intended interventions, most studies were classified as having a low risk of bias. Five studies were considered as having high risk of bias or with some concerns due to using per-protocol analyses instead of intention-to-treat analyses or, had a possibility that deviations from the intended control group could have arisen due to participants seeking alternate intervention if assigned to the non-attention control group, instead of additional resources. In missing outcome data, most studies were deemed as having a low risk of bias, while two studies that had >15% missing data were considered to be having a high risk of bias.

Table 2.

Risk of Bias Across Different Domains, Assessed by Cochrane Risk of Bias 2 (RoB 2) Tool.

Study first author (year)	Randomization Process	Deviations from Intended Interventions	Missing Outcome Data	Measurement of the Outcome	Selection of the Reported Result	Overall Risk of Bias
Alshehri et al. (2020)	Low	Low	Low	Low	Low	Low
Cain et al. (2020)	Low	Low	High	Low	High	High
Carroll et al. (2015)	Low	Low	Low	Low	Low	Low
Chen et al. (2008)	Low	Low	Low	Low	Low	Low
Chen et al. (2011)	Low	High	Low	Low	Low	High
Dolsen et al. (2018)	Low	Low	High	Low	Some concerns	High
Irwin, Olmstead, Breen, et al. (2014)	Some concerns	Low	Low	Low	Low	Some concerns
Irwin, Olmstead, Carrillo, et al. (2014)	Low	Low	Low	Low	Low	Low
Irwin et al. (2015)	Low	Low	Low	Low	Low	Low
Javaheri et al. (2020)	Some concerns	Low	Low	Low	High	High
Johann et al. (2020)	Low	Low	Low	Low	Low	Low
McGrath et al. (2017)	Low	High	Low	Low	High	High
Yang et al. (2017)	Some concerns	Some concerns	Low	Low	Some concerns	Some concerns
Zhang et al. (2021)	Low	Some concerns	Low	Low	Some concerns	Some concerns
Zuo et al. (2020)	Low	Some concerns	Low	Low	Some concerns	Some concerns

Risk of bias assessed in relation to cardiometabolic health outcomes.

Table 3.

Expanded Information about Risk of Bias, Assessed by Cochrane Risk of Bias 2 (RoB 2) Tool.

	Alshehri et al. (2020)	Cain et al. (2020)	Carroll et al. (2015)	Chen et al. (2008)	Chen et al. (2011)	Dolsen et al. (2018)	Irwin, Olmstead, Breen, et al. (2014)	Irwin, Olmstead, Carrillo, et al. (2014)	Irwin et al. (2015)	Javaheri et al. (2020)	Johann et al. (2020)	Mc Grath et al. (2017)	Yang et al. (2017)	Zhang et al. (2021)	Zuo et al. (2020)
1. Randomization Process
1.1 Allocation sequence random?	Y	Y	Y	Y	Y	Y	NI	Y	Y	NI	Y	Y	Y	Y	Y
1.2 Allocation sequence concealed?	Y	Y	Y	Y	Y	Y	NI	Y	Y	NI	Y	Y	NI	Y	PY
1.3 Baseline imbalances suggest a problem?	N	N	N	N	N	N	N	N	N	N	N	PN	N	N	N
Result	Low	Low	Low	Low	Low	Low	Some concerns	Low	Low	Some concerns	Low	Low	Some concerns	Low	Low
2. Deviations from Intended Interventions
2.1 Participants aware of intervention?	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y
2.2 Personnel aware of intervention?	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y
2.3 If Y/PY/NI to 2.1 or 2.2, Deviations that arose because of the trial context?	PN	PN	PN	PN	PN	PN	PN	PN	PN	PN	PN	NI	NI	NI	NI
2.4 If Y/PY to 2.3, Deviations affect outcome?	N/A	N/A	N/A	N/A	N/A	N/A	N/A	N/A	N/A	N/A	N/A	N/A	N/A	N/A	N/A
2.5 If Y/PY/NI to 2.4, Deviations balanced between groups?	N/A	N/A	N/A	N/A	N/A	N/A	N/A	N/A	N/A	N/A	N/A	N/A	N/A	N/A	N/A
2.6 Appropriate analysis to estimate the effect of assignment?	Y	Y	Y	Y	N	Y	Y	Y	Y	Y	Y	N	Y	Y	N
2.7 If N/PN/NI to 2.6, Substantial impact of the failure to analyze participants in randomized groups?	N/A	N/A	N/A	N/A	PY	N/A	N/A	N/A	N/A	N/A	N/A	PY	N/A	N/A	PN
Result	Low	Low	Low	Low	High	Low	Low	Low	Low	Low	Low	High	Some concerns	Some concerns	Some concerns
3. Missing Outcome Data
3.1 Outcome data for all participants?	Y	N	Y	Y	Y	N	Y	Y	Y	Y	Y	Y	Y	Y	Y
3.2 If N/PN/NI to 3.1, Evidence that result is not biased?	N/A	N	N/A	N/A	N/A	N	N/A	N/A	N/A	N/A	N/A	N/A	N/A	N/A	N/A
3.3 If N/PN to 3.2, Missingness could depend on true value?	N/A	Y	N/A	N/A	N/A	Y	N/A	N/A	N/A	N/A	N/A	N/A	N/A	N/A	N/A
3.4 If Y/PY/NI Likely that missingness depended on true value?	N/A	PY	N/A	N/A	N/A	NI	N/A	N/A	N/A	N/A	N/A	N/A	N/A	N/A	N/A
Result	Low	High	Low	Low	Low	High	Low	Low	Low	Low	Low	Low	Low	Low	Low
4. Measurement of the Outcome
4.1 Method of measuring the outcome inappropriate?	N	N	N	N	N	N	N	N	N	N	N	N	NI	N	N
4.2 Measurement or ascertainment of outcome differ between groups?	N	N	N	N	N	N	N	N	N	N	N	N	N	N	N
4.3 If N/PN/NI to 4.1 and 4.2, Outcome assessors aware of intervention received?	NI	Y	N	NI	NI	NI	NI	N	N	NI	NI	N	NI	PY	N
4.4 If Y/PY/NI to 4.3, Could assessment have been influenced by knowledge of intervention?	PN	PN	N/A	PN	PN	PN	PN	N/A	N/A	PN	PN	N/A	PN	PN	N/A
4.5 If Y/PY/NI to 4.4, Likely that assessment was influenced by knowledge of intervention?	N/A	N/A	N/A	N/A	N/A	N/A	N/A	N/A	N/A	N/A	N/A	N/A	N/A	N/A	N/A
Result	Low	Low	Low	Low	Low	Low	Low	Low	Low	Low	Low	Low	Low	Low	Low
5. Selection of the Reported Result
5.1 Trial analyzed in accordance with a pre-specified plan?	Y	Y	Y	Y	Y	NI	Y	Y	Y	Y	Y	Y	NI	NI	NI
5.2 Result selected from multiple outcome measurements?	N	PY	PN	N	N	PN	N	PN	PN	PY	N	PY	PN	PN	PN
5.3 Result selected from multiple analyses of the data?	PN	NI	N	N	N	PN	PN	PN	PN	N	N	N	PN	PN	PN
Result	Low	High	Low	Low	Low	Some concerns	Low	Low	Low	High	Low	High	Some concerns	Some concerns	Some concerns
Overall Risk of Bias	Low	High	Low	Low	High	High	Some concerns	Low	Low	High	Low	High	Some concerns	Some concerns	Some concerns

Note: Y = Yes. PY = Probably Yes. N = No. PN = Probably No. NI = No information. N/A = Not applicable.

All studies were classified as having a low risk of bias in the measurement of the outcome. They reported appropriately measuring the outcome consistently and the assessment of the outcome was likely not influenced by knowledge of group assignment. About half of the studies were rated as having a low risk of bias for the selection of the reported results. Studies classified as high did not have a pre-specified analytic plan, or if there was one, it lacked sufficient detail, or the results presented were likely selected from multiple outcome measurements (as reported in the protocol), with no explanation for the selection of the reported result. Overall, six studies were classified as having low risk, four had some concerns and five had a high risk of bias (Tables 2 and 3).

Effectiveness of CBT-I on Cardiometabolic Biomarkers and Health Outcomes

Studies examined the effectiveness of CBT-I in relation to a variety of cardiometabolic biomarkers. Of the 15 included studies, four reported on cardiovascular indicators such as heart rate variability and BP(Javaheri et al., 2020; Johann et al., 2020; McGrath et al., 2017; Yang et al., 2017), five on metabolic indicators, including BMI and glucose regulation(Alshehri et al., 2020; Cain et al., 2020; Irwin, Olmstead, Breen, et al., 2014; Zhang et al., 2021; Zuo et al., 2020), and seven on inflammatory biomarkers.(Chen et al., 2008; Chen et al., 2011; Dolsen et al., 2018; Irwin, Olmstead, Breen, et al., 2014; Irwin, Olmstead, Carrillo, et al., 2014; Irwin et al., 2015; Johann et al., 2020) One study reported a multisystem (i.e., lipid, inflammatory, glycemic) cardiometabolic risk.(Carroll et al., 2015) Effect sizes and narrative findings are detailed in Table 4.

Table 4.

Effect of CBT-I on Cardiometabolic Health.

Study First Author (year)	Outcome(s)	Follow-Up Period and Rate	Effect Size of Post- Intervention Between Group Differencesa	Narrative Findingsb	Summary of Findingsc
			Hedge’s g (95% CI)
Alshehri et al. (2020)	HbA1c	7 weeks, 93%	HbA1c: −0.43 (−1.21, 0.35)	There were significant between-group post-intervention differences in HbA1c.	HbA1c: +
Cain et al. (2020)	BMI HOMA-IR	35-41 weeks gestation (T2), 89% 6-8 weeks postpartum (T3), 85%	BMI T2: 0.42 (−0.16, 1.00) T3: 0.57 (−0.03, 1.16) HOMA-IR: NR	The average BMI was lower among the intervention group across all assessments. The control group noted a 36% overall decrease in HOMA-IR from T1 to T3 and the intervention noted a 42% decrease in HOMA-IR.	BMI at T2: + BMI at T3: + HOMA-IR at T2 and T3: o
Carroll et al. (2015)	Multisystem Biological Risk: high-density lipoprotein, low-density lipoprotein, triglycerides, hemoglobin A1c, glucose, insulin, C-reactive protein, and fibrinogen	4 months, 100% 16 months, 96%	NR	Mixed linear model analyses found no significant treatment by time interaction (p = .30) in the prediction of multisystem biological risk.	Multisystem biological risk: o
Chen et al. (2008)	IL-6 IL-1β IL-18 TNF-α	4 weeks, 100%	NR	Postintervention cytokine levels in the CBT and control groups were similar.	IL-6: o IL-1β: o IL-18: o TNF-α: o
Chen et al. (2011)	Hs-CRP IL-1β IL-18 OxLDL	6 weeks, 100%	NR	Post-trial of high-sensitive C-reactive protein, IL-18, and oxidized low-density lipoprotein levels significantly declined with cognitive-behavioral therapy in comparison with the control group.	Hs-CRP: + IL-1β: o IL-18: + OxLDL: +
Dolsen et al. (2018)	IL-6 sTNF-R2	10 weeks, 77%	IL-6: 0.47 (−0.5, 1.43) sTNF-R2: −0.11 (−1.06, 0.84)	From pretreatment to posttreatment, CBT-I compared to PE was associated with non-significant decreases in IL-6 and sTNF-R2.	IL-6: o sTNF-R2: o
Irwin, Olmstead, Breen, et al. (2014)	BMI CRP IL-6 TNF-α	3 months, 90%	NR	There were no significant between-group changes from baseline to posttreatment in body mass index and CRP. Levels of toll-like receptor-4–activated monocyte production of IL-6 and TNF combined showed an overall reduction in TCC versus CBT-I, with similar effects for IL-6 and TNF alone.	BMI: o CRP: o IL-6: − TNF-α: −
Irwin, Olmstead, Carrillo, et al. (2014)	High CRP, defined as > 3.0 mg/L	4 months, 91% 16 months, 88%	NR	As compared to SS, CBT was associated with a reduced risk of having high CRP at 4 months. From baseline to month 16, high CRP was less likely to be found in the CBT participants as compared to TCC and SS.	CRP, 4 months: + (compared to SS) CRP, 16 months: + (compared to both TCC and SS groups)
Irwin et al. (2015)	CRP TNF-α IL-6	4 months, 91% 16 months, 88%	NR	As compared to SS active control, CBT-I reduced levels of CRP (month 4, 16) and monocyte production of proinflammatory cytokines (month 2 only). CBT and TCC showed similar low levels of CRP at month 4 but CRP levels diverged at month 16 (levels remained low in CBT but increased in TCC). As compared to CBT-I, TCC resulted in lower levels of TLR-4 activated monocyte production of IL-6 and TNF at months 4 and 16.	CRP: + for both SS and TCC IL-6: − for TCC, + for SS TNF-α: − for TCC, + for SS
Javaheri et al. (2020)	SBP DBP	6 weeks, 85%	NR	There were reductions in systolic and diastolic BP in the treatment arm compared to increases in the control arm, but these between-group differences were non-significant.	SBP: o DBP: o
Johann et al. (2020)	Ambulatory 24-hr SBP Nocturnal SBP Ambulatory 24-hr DBP Nocturnal DBP Mean 24-hr HR Nocturnal HR Mean 24-hr LF/HF Nocturnal LF/HF Mean 24-hr SDNN Nocturnal SDNN CRP NT-proBNP	8 weeks, 96%	Mean 24-hr SBP: 0.08 (−0.50, 0.66) Nocturnal SBP: 0.23 (−0.35, 0.81) Mean 24-hr DBP: 0 (−0.58, 0.58) Nocturnal DBP: 0.25 (−0.33, 0.83) Mean 24-hr HR: −0.37 (−0.96, 0.21) Nocturnal HR: −0.11 (−0.69, 0.47) Mean 24-hr LF/HF: 0 (−0.58, 0.58) Nocturnal LF/HF: 0 (−0.58, 0.58) Mean 24-hr SDNN: 0.18 (−0.40, 0.76) Nocturnal SDNN: 0.14 (−0.44, 0.72) CRP: −0.31 (−0.89, 0.27) NT-proBNP: −0.18 (−0.76, 0.40)	There was no significant group × time interaction effect for mean 24-hr or nocturnal systolic or diastolic blood pressure, mean 24-hr or nocturnal heart rate, LF/HF, SDNN, CRP levels, or NT-proBNP levels.	Mean 24-hr SBP: o Nocturnal SBP: o Mean 24-hr DBP: o Nocturnal DBP: o Mean 24-hr HR: o Nocturnal HR: o Mean 24-hr LF/HF: o Nocturnal LF/HF: o Mean 24-hr SDNN: o Nocturnal SDNN: o CRP: o NT-proBNP: o
McGrath et al. (2017)	Ambulatory 24-hour SBP Nocturnal SBP Ambulatory 24-hour DBP Nocturnal DBP	8 weeks, 95%	Mean 24-hr SBP: −0.07 (−0.42, 0.29) Nocturnal SBP: 0.04 (−0.32, 0.39) Mean 24-hr DBP: −0.23 (−0.59, 0.13) Nocturnal DBP: −0.11 (−0.47, 0.25)	The mean change in 24-hour ambulatory and nocturnal SBP and DBP over 8 weeks was not significantly different between intervention and control arms.	Mean 24-hr SBP: o Nocturnal SBP: o Mean 24-hr DBP: o Nocturnal DBP: o
Yang et al. (2017)	SBP DBP	8 weeks, 100%	SBP: 0.58 (0.19, 0.97) DBP: −0.08 (−0.46, 0.30)	SBP and DBP decreased significantly in both groups, but decrease was more significant in CBT-I group than control group.	SBP: + DBP: +
Zhang et al. (2021)	HbA1c	2 months, 97% 6 months, 95% 12 months, 90%	HbA1c: 2 mo: 0.04 (−0.08, 0.15) 6 mo: 0.14 (0.02, 0.26) 12 mo: 0.27 (0.14, 0.39)	The effect of the intervention on HbAlc values was not significant at 2 mo but was significant between the two groups at 6 mo and at 12 mo.	HbA1c: 2 months: o 6 months: + 12 months: +
Zuo et al. (2020)	HbA1c	2 months, 100% 6 months 98%	HbA1c: 2 mo: 0.05 (−0.23, 0.34) 6 mo: 0.66 (0.36, 0.95)	No difference was found in HbAlc between the two groups after 2 months. At 6 months of follow-up, the CBT group had a significant difference in HbAlc compared with the control group.	HbA1c: 2 months: o 6 months: +

Note: BMI = Body mass index. BP = Blood pressure. CBT-I = Cognitive behavioral therapy for insomnia. CRP = C-reactive protein. DBP = Diastolic blood pressure. HbA1c = Glycated hemoglobin. HOMA-IR = Homeostatic model assessment for insulin resistance. HR = Heart rate. Hs-CRP = High-sensitivity c-reactive protein. IL-1β = Interleukin 1 beta. IL-18 = Interleukin 18. IL-6 = Interleukin 6. LF/HF = Low frequency/high frequency ratio. NR = Not reported (i.e., sufficient information not reported to calculate). NT-proBNP = N-terminal pro-brain natriuretic peptide. OxLDL = Oxidized low-density lipoprotein. PE = Psychoeducation. SBP = Systolic blood pressure. SDNN = Standard deviation of NN intervals. SS = Sleep seminar. sTNF-R2 = Serum-soluble tumor necrosis factor receptor 2. TCC = Tai chi chih. TNF-α = Tumor necrosis factor alpha.

^aPositive values indicate that control group mean was higher than CBT-I group mean at follow-up. Effect sizes were calculated for each outcome and/or timepoint measured if sufficient information was reported by authors.

^bAs reported by authors.

^cFindings (as reported by authors) as follows:

+, results showed improvements in cardiometabolic health outcome in intervention group compared to control group (i.e., favoring intervention).

o, no significant differences in cardiometabolic health outcome between intervention and control groups.

−, results showed worsening in cardiometabolic health outcome in intervention group compared to control group (i.e., favoring control).

Cardiovascular Markers

Four studies assessed CBT-I effectiveness on BP over 6-8 weeks, with 2 of them using 24-hour ambulatory BP measurement, producing averages across the 24-hour day and during nocturnal periods only.(Johann et al., 2020; McGrath et al., 2017) One study found significant medium-sized improvements (g=0.58) in systolic BP in the CBT-I group compared to the control group(Yang et al. 2017), while others reported statistically null effects. There was no significant effect for mean 24-hour or nocturnal heart rate or either twoheart rate variability (HRV) measures, or cardiac N-terminal pro-brain natriuretic peptide (NT-proBNP) levels during eight weeks of follow-up.(Johann et al. 2020)

Metabolic Markers

Three studies that assessed changes in HbA1c in patients with type 2 diabetes found significant small- to medium-sized improvements with CBT-I compared with control, such that the groups randomized to CBT-I had greater decreases in HbA1c after 7 weeks (−0.5%, g=−0.43 in Alshehri et al., 2020), 6 months (−0.2%, g=0.14 in Zhang et al., 2021; −1%, g=0.66 in Zuo et al., 2020), and 12 months (−0.2%, g=0.27 in Zhang et al., 2021) of follow-up. One study reported results in the homeostatic model assessment of insulin resistance (HOMA-IR) in pregnant women, with null results, although there was a trend favoring the intervention arm (42% decrease) compared to the control (36% decrease).(Cain et al., 2020) Regarding BMI, this study found lower mean BMI in the CBT-I group versus the control group in pregnant women (−3.5 kg/m², g=0.42, small-medium effect size) and during the postpartum period (−4.3 kg/m², g=0.57, medium effect size).(Cain et al., 2020) Lastly, one study with breast cancer survivors did not find significant differences in BMI between the CBT-I and tai chi groups at 12 weeks.(Irwin, Olmstead, Breen, et al., 2014)

Inflammatory Markers

The most reported inflammatory outcomes were CRP, IL-6, and TNF-α. Within studies that reported CRP, three demonstrated significant positive effectiveness of CBT-I (Chen et al., 2011), even when compared to the active control of tai chi(Chen et al., 2011; Irwin et al., 2015; Irwin, Olmstead, Carrillo, et al., 2014) and two found null effects(Irwin, Olmstead, Breen, et al., 2014; Johann et al., 2020). For IL-6 and TNF- α, one study found improvements when CBT-I was compared to a non-active control (sleep seminar) but no improvements when compared to tai chi.(Irwin et al., 2015) Another study found a non-significant small-medium effect size for IL-6.(Dolsen et al., 2018) CBT-I also significantly improved IL-18 and oxidized low-density lipoprotein (OxLDL), compared to the control.(Chen et al. 2011). Effects on IL-1β were null, and IL-18 were mixed.(Chen et al., 2011; Chen et al., 2008)

Multisystem Markers

One study found no significant effect of CBT-I compared to control for composite multisystem biological risk score consisting of high-density lipoprotein (HDL), low-density lipoprotein (LDL), triglycerides, fibrinogen, CRP, glucose, insulin, and HbA1c with follow-up over four months (p=0.30).(Carroll et al., 2015)

Discussion

The current review is the first to evaluate CBT-I concerning cardiometabolic health outcomes across three domains: cardiovascular, glycemic regulation, and inflammation. Overall, this review suggests that CBT-I results in consistent improvements in glucose regulation (HbA1c) and more favorable levels of the inflammatory CRP biomarker. There were inconsistent effects of CBT-I on IL-6 and largely null effects on measures of BP, heart rate, heart rate variability, and TNF-α. Though fewer studies reported on BMI, IL-18, and oxLDL, the initial studies reviewed here favored CBT-I. Besides treating insomnia, CBT-I also is linked to improvements in indicators of cardiometabolic health.

In the metabolic domain, we found CBT-I to have significant beneficial effects on HbA1c in three studies among patients with type 2 diabetes. These results are consistent with a meta-analysis of behavioral or educational sleep interventions, which found decreased HbA1c with intervention, but the effect was non-significant.(Kothari et al., 2021) Of the six studies included in this prior review, three included CBT-I, and three were sleep education interventions alone, suggesting the importance of supplementing education with cognitive-behavioral components to affect HbA1c significantly. Our findings related to CBT-I and HbA1c provide a critical addition to the literature, as few experimental laboratory-based sleep studies have examined HbA1c, which reflects glucose metabolism over 2-3 months, versus insulin sensitivity or glucose tolerance, which are modifiable after just a few nights of sleep restriction.(Spiegel et al., 1999) HbA1c is also the focus of clinical guidelines related to glycemic control among persons with diabetes(American Diabetes Association, 2022), so information related to behavioral modification may be applied in clinical settings. In interpreting these findings, it is important to note that some potential influences on HbA1c were not accounted for in the reviewed studies (e.g., weight and weight loss, diet) and there is a potential for unmeasured confounding. Findings from the other metabolic biomarkers were less consistent. HOMA-IR in pregnant women did not significantly differ between CBT-I and control groups, and BMI favored CBT-I over control in one but not both studies that measured it. These findings are similar to past studies of sleep extension, which found null effects for BMI(Al Khatib et al., 2018; Haack et al., 2013) and null effects(Leproult et al., 2015) or decreases in HOMA-IR(Killick et al., 2015; So-Ngern et al., 2019). However, several previous studies were uncontrolled or non-randomized, limiting the quality of the evidence.

The duration of the CBT-I intervention and follow-up period may have resulted in null findings. The average duration of CBT-I was eight weeks with variable follow-up assessment schedules – this poses a challenge in identifying improvements in cardiometabolic health that may be affected over a lifetime. Thus, the short intervention duration may not be sufficient to result in significant effects during the assessed study period. In particular, BMI, BP, resting heart rate, and HRV may take more than six months to improve after established interventions such as increased physical activity(Jakicic et al., 2011) and diet.(Brook et al., 2013) Consistent with this review, where some studies followed these outcomes for 6-8 weeks, even interventions in sleep laboratories and sleep extension studies report null changes in BP and heart rate over short follow-up periods.(Haack et al., 2013; Kubo et al., 2011; Reynold et al., 2014) Other outcomes, such as changes in HbA1c and biomarkers like CRP and NT-proBNP, are expected to accrue over at least a 2-3 month period. Accordingly, studies with longer follow-ups (Irwin et al., 2015; Irwin, Olmstead, Carrillo, et al., 2014), rather than shorter(Irwin, Olmstead, Breen, et al., 2014; Johann et al., 2020) for CRP had significant results. Shorter temporal changes occur in inflammatory biomarkers such as oxLDL and IL-18, and thus may be more interpretable in studies with a shorter follow-up.(Chen et al. 2011) As such, it is possible that CBT-I influenced HbA1c and CRP because they are earlier, more sensitive markers, and there was insufficient time to demonstrate robust effects for other clinical markers. Additionally, sleep restriction therapy, a component of CBT-I, may lead to a reduction in total sleep time at the beginning of CBT-I in an effort to improve sleep efficiency before expanding time in bed. Of the reviewed studies, those that used sleep restriction therapy tended to have higher rates of improvement in cardiometabolic measures (6/12 or 50% of biomarkers improved) compared to those without sleep restriction therapy (6/19 or 32% of biomarkers improved). The biomarkers with the most robust improvement (HbA1c and CRP) showed improvements in studies with and without sleep restriction therapy.

Despite the robustness of the RCT design, other important considerations include residual confounding factors and study power. All but one of the included studies that examined inflammatory biomarkers were of special populations with comorbidities or demographic characteristics that may inherently increase the risk of inflammatory processes. Specifically, these studies represented survivors of breast cancer, individuals with chronic kidney disease requiring dialysis, comorbid bipolar disorder, and older adults, all conditions that present increased inflammation. While this does not explain why CRP was improved by CBT-I, it may explain some of the inconsistencies across other inflammatory biomarkers. Such differences in effects may also be related to insufficient power to detect significant effects. All studies reporting inflammation results had sample sizes of less than 125 participants, with a median of 59 participants, and prior power calculations were often not mentioned.

A remaining question relates to the pathway by which CBT-I influences cardiometabolic health. One hypothesis is that CBT-I helps to extend sleep which in turn improves cardiometabolic components. A recent systematic review examined the effect of sleep extension interventions alone on cardiometabolic health and found positive effects on measures of insulin sensitivity, appetite, and eating behaviors.(Henst et al., 2019) However, studies had relatively short follow-up periods (≤6 weeks), and the maintenance of sleep extension interventions has not been established. In the currently reviewed studies, sleep duration improved, on average, by 20 minutes in the CBT-I group and 12 minutes in the comparison group. This modest difference in improvements between groups may have inhibited the ability of studies to demonstrate between-group differences in outcomes on the scale of those seen in experimental extension or deprivation studies. However, besides only lengthening sleep, components of CBT-I may improve sleep and influence cardiometabolic health through other pathways such as stress reduction, improving symptoms of depression and anxiety, improving sleep efficiency, and enhancing regularity of sleep-wake patterns and rhythms of daytime behaviors. Intra-individual sleep variability has also been linked to worse cardiometabolic health(Zuraikat et al., 2020) and thus, regularizing the rest-wake cycle is likely a key intervention area included in CBT-I. There was also a variety in the characteristics of the CBT-I interventions included in this review. While no clear pattern emerged in these studies about the benefits of different intervention formats, future studies are warranted to test this more directly.

Although novel, our findings should be interpreted in the context of several limitations. First, five studies were rated as having a high risk of bias. Areas of concern included the lack of information regarding the concealment of randomization, the use of per-protocol instead of intention-to-treat analysis, the lack of a pre-specified analytic plan or protocol, and selective reporting of results. These issues raise questions about the validity of the findings. Second, while the ROB-2 tool aims for objectivity, some items rely on subjective assessments and/or did not apply well to behavioral trials (e.g., whether interventionists were aware of the participants’ assigned intervention group, which is necessary for CBT-I). Third, many studies had relatively small sample sizes and baseline differences in the cardiometabolic outcomes of interest. In several studies, even though CBT-I resulted in greater improvement over time compared to the control group (in analyses of group-by-time interactions), the baseline imbalances resulted in post-intervention between-group differences favoring the control group (e.g., Alshehri et al., 2020), which are potentially misleading. Additionally, there was large methodological and clinical heterogeneity (of outcomes and study populations) hindering the ability to perform meta-analysis. Therefore, we relied on reporting the quantitative synthesis of findings across studies by calculating group differences. For standardization of future research, we recommend studies to measure cardiometabolic biomarkers that are clinically relevant, relatively stable over time, and not influenced by single nights of sleep or time of day of measurement (e.g., HbA1c, CRP). If biomarkers that fluctuate over a 24-hour period are of interest, we recommend they be measured using 24-hour ambulatory measurements or measured in a similar time frame (e.g., between 8-10 AM) among all participants and at all time points.

In conclusion, the current study contributes novel information about the effectiveness of CBT-I on cardiometabolic health, though future research could further elucidate relationships. This appears to be a developing area of research, with at least two ongoing trials examining CBT-I’s effect on cardiometabolic health in heart failure(Redeker et al., 2017) and hypertension(Levenson et al., 2017) populations. We recommend future research prioritize consistency in methods and reporting (e.g., means, standard deviations of objective sleep time, and cardiometabolic biomarker data pre- and post-intervention) to allow pooling of effects. While studies suggest the benefits of CBT-I on glucose regulation, CRP, and sleep, further studies are warranted examining moderators (e.g., aging, comorbidities) and mediators (e.g., other health behaviors, daily activity rhythms) of effects over time to better understand relationships in varied populations. Lastly, despite the strong evidence for CBT-I, it remains underutilized in primary care.(Araújo et al., 2017) There is a need for dissemination and implementation research to help increase access to the treatment, especially in populations at risk for cardiometabolic disease.