Sleep alterations in substance use disorders: a systematic review and meta-analysis

Summary

Background

Sleep disturbances are common in individuals with substance use disorders (SUDs), often persisting beyond initial abstinence and hindering recovery. However, the underlying sleep abnormalities warrant further investigation, particularly given mixed findings regarding specific substances.

Methods

This systematic review and meta-analysis aimed to identify sleep-related abnormalities associated with alcohol (AUD), benzodiazepine (BUD), cannabis (CaUD), cocaine (CoUD), methamphetamine (MUD), nicotine (NUD), and opioid (OUD) use disorders. We systematically searched Embase, PsycINFO, PubMed, Scopus, and Web of Science until November 2025, following a pre-registered protocol (PROSPERO: CRD42024531160).

Findings

We conducted a systematic review of 43 eligible publications involving approximately 7500 participants, using both objective (eg, polysomnography) and subjective (eg, Pittsburg Sleep Quality Index [PSQI]) measures. Results showed that total sleep time (TST) was reduced in AUD (−14.32, 95% CI = −16.69 to −11.96; I² = 0%), NUD (−0.33, 95% CI = −0.59 to −0.06; I² = 37%), and OUD (−38.16, 95% CI = −63.04 to −13.28; I² = 0%). Slow-wave sleep (SWS) was reduced in AUD (−3.68, 95% CI −4.99 to 2.38; I² = 73%) and CoUD (−30.69, 95% CI = −47.27 to −14.10; I² = 90%). Sleep quality, measured by the PSQI, was poorer in AUD (4.89, 95% CI = 3.01 to 6.77; I² = 98%), CoUD (0.98, 95% CI = 0.04–1.93; I² = 0%) and NUD (2.64, 95% CI = 0.41–4.88; I² = 96%). Results for CaUD could not be meta-analyzed due to scarcity of data. No study met criteria to be included for BUD or MUD.

Interpretation

These findings suggest specific relationships between specific addictive substances and sleep, highlight areas of convergence in these relationships, and indicate instances in which the same drug is related with both objective and subjective alterations. Further research is needed to explore further, at a meta-analytical level, relationships between sleep and specific substances.

Funding

National Institute on Drug Abuse.

Research in context.

Evidence before this study

Prior reviews have documented significant sleep impairment in alcohol, opioid, nicotine, and stimulant use disorders, with discrepancies between subjective and objective assessments. However, few quantitative syntheses have simultaneously examined subjective sleep scores (eg, Pittsburg Sleep Quality Index) and objective measures (eg, total sleep time) across substances. We conducted comprehensive searches in databases including PubMed, Embase, and PsycINFO (from inception to November 2025), without language restrictions, using terms related to SUDs, sleep disturbances, polysomnography, and subjective sleep tools.

Added value of this study

By pooling data from 43 eligible studies, our meta-analysis integrates both subjective and objective sleep outcomes across multiple SUDs, including alcohol, cannabis, cocaine, nicotine and opioid use disorders. This is the first synthesis quantifying both PSQI-based sleep quality and polysomnographic metrics such as slow-wave sleep and rapid eye movement (REM) sleep alterations across the mentioned different substance classes. We also identify sources of heterogeneity in measurement instruments and gaps in stimulant and benzodiazepine-related literature.

Implications of all the available evidence

The findings indicate that several sleep disturbances are present across SUDs and may serve as modifiable risk factors for relapse. Standard sleep assessment tools like the PSQI are widely deployable but partly discordant with objective data. Future research should prioritize accessible objective measures (eg, actigraphy) and consider trials of potentially beneficial interventions, such as modafinil for stimulant use concerns, or suvorexant for opioid use disorder. More clinical guidelines should integrate sleep evaluation and targeted therapies alongside traditional pharmacological and behavioral SUD treatments.

Introduction

The global burden of substance use disorders (SUDs) is rising, yet many individuals either lack adequate treatment or relapse despite care.¹ Even among people treated, outcomes are often poor, with low sustained abstinence and high relapse rates within weeks or months.² Though pharmacologic and behavioral therapies exist, they do not target a key modifiable factor: sleep disturbances. Growing evidence links untreated sleep issues to relapse and poor functioning, though sleep is rarely addressed in SUD treatment.³ This gap highlights the need to integrate sleep-focused interventions to enhance recovery. Since substances impact sleep differently,³ substance-specific assessment and management strategies are indicated.

Sleep disturbances are common in both acute and chronic substance use,⁴ including difficulties initiating or maintaining sleep, disrupted sleep architecture, and non-restorative sleep.⁴ These concerns often persist into abstinence, potentially undermining long-term recovery.⁴ The bidirectional link between substance use and sleep suggests that poor sleep may worsen substance use behaviors,⁵ while substance exposure or withdrawal may further disrupt sleep.⁵ Examples of these links include instances in which individuals with insomnia are more likely to use alcohol for its sedative effects, which quickly fade and worsen sleep, promoting increased alcohol use.⁶ Similarly, several persons who use cannabis often cite sleep aid as a motive, but cannabis withdrawal can trigger insomnia and relapse.⁵

Drug exposures may impact sleep differently, with possible effects varying by substance and across stages of use, withdrawal, or abstinence. These differences likely reflect distinct neurobiological mechanisms and receptor systems affected by each drug.^4,5,7 For instance, alcohol and opioids suppress slow-wave and rapid eye movement (REM) sleep through GABAergic and opioidergic pathways,^5,8 while stimulants such as cocaine and methamphetamine disrupt sleep-wake regulation via dopaminergic and noradrenergic activation.⁷ As a result, the timing, duration, and type of sleep disruption may differ substantially across substances and withdrawal phases. For example, early cannabis abstinence is linked to increased REM sleep, while early opioid abstinence is associated with decreased REM sleep. Cocaine use illustrates stage-specific effects: early withdrawal often brings hypersomnia and elevated REM percentage and density, while later abstinence more closely resembles chronic insomnia, marked by prolonged sleep latency, increased wake after sleep onset (WASO), and reduced sleep efficiency.^3,9

In addition to the different effects that drugs may have on sleep and heterogenous consequences in part depending on the stage of drug consumption, withdrawal, or abstinence, the literature includes uncertainties. For instance, the effects of opioids early in abstinence, and alcohol late in abstinence, remain unclear or mixed.10, 11, 12, 13 Additionally, sleep alterations associated with widely used drugs, such as benzodiazepines and methamphetamines, have not been explored with the same depth as other drugs, such as alcohol.

Furthermore, while several studies have explored polysomnography (PSG) outcomes or subjective outcomes, few have integrated both types of measurement. The paucity of studies reporting both measures leaves open important scientific questions, particularly as the available literature shows dissonance between subjective reports and polysomnographic findings. A classic example involves the “occult insomnia” found in cocaine use disorder (CoUD), in which self-reported sleep quality is seemingly normal, despite objective parameters indicating otherwise.¹⁴

Considering the above, no prior meta-analysis has quantitatively pooled data from controlled studies across different SUDs to better understand the relationships between specific SUDs and subjective and objective sleep measures. This study aims to address this gap by systematically reviewing and meta-analyzing polysomnographic and self-reported sleep outcomes across major SUDs. We hypothesized that (1) individuals with SUDs would show significantly poorer sleep continuity and architecture compared with control groups, (2) the magnitude and pattern of sleep alterations would differ by substance type, and (3) discrepancies would exist between subjective and objective sleep measures, reflecting underrecognized sleep impairment in this population.

Methods

Search strategy and selection criteria

The Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) approach was followed.¹⁵ The review protocol was registered in the International Prospective Register of Systematic Reviews (PROSPERO; CRD42024531160). The PECO strategy was elaborated for the inclusion criteria as follows: Population: human adults; Exposure: SUD [alcohol (AUD), benzodiazepine (BUD), cannabis (CaUD), cocaine (CoUD), methamphetamine (MUD), nicotine (NUD), and opioid (OUD)]; Comparators: no substance use; Outcomes: sleep measurements (subjective questionnaires, such as the Pittsburg Sleep Quality Index [PSQI]¹⁶ and objective PSG and electroencephalography [EEG] parameters, including total sleep time [TST], sleep onset latency, sleep efficiency, WASO, and the percentage of time spent in slow-wave sleep [SWS] and rapid eye movement [REM] sleep relative to TST [REM/SPT]). Interventional and observational studies published as full articles using the Roman alphabet were included, regardless of the date of publication. Studies were excluded if they were not full original articles (ie, reviews, case reports, abstracts), had a preclinical design, involved children, or investigated participants without SUDs. We also excluded studies involving participants using other substances for pain management or cancer treatment, and those with multiple concurrent illicit SUDs. Additionally, studies that did not allow clear distinction of data between controls and participants with SUDs were excluded. After initial screening, to improve generalizability or our findings, we revisited previously excluded records involving participants with co-occurring psychiatric disorders (eg, depression, anxiety, or other mental health diagnoses) to assess their eligibility. However, none met all inclusion criteria upon reassessment.

A systematic search was conducted in the electronic databases Embase, PsycINFO, PubMed, Scopus and Web of Science from inception until November 2025 to identify relevant eligible studies for this review. Identified articles were imported to Covidence (www.covidence.org) for duplicate removal and screening. Study selection and data extraction were performed independently by three reviewers (APN, HNPO and TPP). An additional reviewer was consulted for studies where the reviewers disagreed (GAA). The search strategies and exclusion criteria can be found in the Appendix (pp2-4).

Data analysis

For each included study, relevant data were extracted, effect sizes were calculated based on reported means, standard deviations (SD), and sample sizes, and if necessary, authors of original articles were contacted to provide missing data. The mean difference (MD) or standardized MD (SMD) were used to report effect sizes (Cohen's d). In cases in which such data were provided through graphs, we used a digital ruler software, National Institutes of Health ImageJ software (Bethesda, Maryland),¹⁷ to extract the means and SD. When available, sleep stage data (eg, REM%) was expressed as the percentage of total sleep period (SPT), as REM/SPT.

Heterogeneity across studies was assessed using the I² statistic and considered to be not important (I² = 0–40%), moderate (I² = 40–60%), substantial (I² = 60–90%), or considerable (I² = 90–100%).¹⁸ The results with heterogeneity above 50% were pooled using random-effects models, while fixed-effects models were used otherwise. Random-effects models were based on the DerSimonian and Laird method, accounting for both within- and between-study variance.¹⁹ The potential impact of heterogeneity was explored by examining differences in study design, participant characteristics, and sleep assessment methods. Subgroup analyses to explore potential sources of variation (eg, based on severity of SUD, time since last drug use, and demographic variables) were considered but not completed due to limitations in the data sets. Meta-regression analyses were also considered to explore moderators such as age, sex distribution, and abstinence duration; however, the limited number of studies available per outcome and substance type precluded their reliable implementation.

The revised Cochrane Risk of Bias in Non-randomized Studies-of Exposures (ROBINS-E)²⁰ was utilized to investigate the potential risk of bias in the studies analyzed. To examine publication bias, a visual evaluation of the funnel plots for each comparison was conducted. This process involved positioning each study on a graph based on its standard error (precision) and effect size.²¹ Forest plots and funnel plots were generated using Review Manager 5.

Role of the funding source

The funder had no role in the study design, data collection, analysis, interpretation, or writing of the report.

Results

Overview of included and analyzed studies

Our search retrieved 12,654 results, of which 5810 were duplicates. After the title and abstract screening, 6844 were excluded. Full text evaluation of the remaining articles yielded 43 studies that met the inclusion criteria (Fig. 1). Four of these studies could not be included in the meta-analysis because, although eligible for the review, they did not report their data in a format suitable for pooling in a Forest plot. These studies were therefore summarized in our qualitative synthesis of the results. In some cases, we contacted study authors to request additional data; however, responses to these requests did not alter the final pool of meta-analyzed studies. Ultimately, 39 studies published between 2000 and 2025 were included in the quantitative synthesis. None of the retrieved studies on benzodiazepines or methamphetamine met inclusion criteria. The methodological quality assessment of the included studies showed overall “some concerns” (Appendix p7), and visual inspection of the funnel plot shows some asymmetry, which indicates potential publication bias (Appendix p8). Sensitivity analysis accounting for these did not change results.

Fig. 1.

PRISMA study-selection flow chart.

Sample characteristics

Overall, 17 studies were published in the USA, six in China, five in Germany, three in France, two each in Egypt, India, and Israel, and one each in Canada, England, Japan, Lebanon, Poland, and Taiwan (Table 1; Appendix p5). In addition, most investigated AUD, NUD, OUD and CoUD (24, 10, four and four studies, respectively). CaUD was studied in two of the included publications. For sleep quality measurement tools, most (24 studies) used the PSQI,¹⁶ while others used other self-reported outcomes or PSG results. Approximately 7500 individuals were analyzed in this systematic review, of whom 55% were males.

Table 1.

Overview of included studies.

Author	Year	Substance	Country	Sleep measurement tool	Control (n)	Case (n)	Abstinence duration	Main results
Arbinaga22	2019	Nicotine	Spain	PSQI	24	92	NA	p < 0.001 (NUD scored higher on the PSQI)
Armitage23	2012	Alcohol	United States	PSG	16	48	3–12 weeks after their last drink	p < 0.05 (AUD presented shorter REM latency)
Asaad24	2011	Opioid	Egypt	PSG	10	33	7 days or more	ns
Brower25	2011	Alcohol	United States	PSG	10	10	1–3 months after their last drink	p < 0.05 (AUD presented shorter REM latency)
Cohen26	2019	Nicotine	Israel	PSQI	39	38	NA	ns
Cohen27	2020	Nicotine	Israel	PSQI	46	40	NA	ns
Cohrs28	2014	Nicotine	Germany	PSQI	1071	1243	NA	p = 0.016 (NUD score higher on the PSQI)
Colrain29	2009	Alcohol	United States	EEG	42	42	NA	NR
Crum30	2004	Alcohol	United States	Self-reported insomnia/hypersomnia	252	68	NA	Insomnia p = 0.0009; Hypersomnia ns
de Zambotti31	2014	Alcohol	United States	PSG and PSQI	16	14	19.8 ± 10.1 days	PSQI p < 0.001; Objective outcome mostly ns
Dong32	2025	Nicotine	China	EEG	16	19	NA	ns
Dugas33	2017	Nicotine	Canada	PSQI	453	405	NA	p < 0.05 (NUD had more PSQI scores >5)
Feige34	2007	Alcohol	Germany	PSG	23	12	3 weeks after their last drink	p = 0.003 (AUD presented shorter REM latency)
Gann35	2001	Alcohol	Germany	PSG and PSQI	30	40	20 ± 8 days after their last drink	p = 0.004 (AUD presented shorter REM latency)
Irwin10	2000	Alcohol	United States	PSG	35	32	Days since last drink: African-American 30.3 ± 20.9; Euro-American: 29.4 ± 9.6	p < 0.01 (AUD slept fewer hours)
Irwin12	2002	Alcohol	United States	PSG	32	46	Days since last drink: African-American 22.4 ± 12.0; Euro-American: 23.0 ± 12.2	ns
Irwin36	2004	Alcohol	United States	PSG	15	16	Days since last drink: 18.8 ± 8.9	ns
Irwin37	2006	Alcohol	United States	PSG	14	14	Days since last drink: 31.0 ± 28.0	ns
Irwin38	2016	Alcohol, cocaine	United States	PSG and PSQI	108	73	At least 2 weeks (Alcohol); at least 3 days (Cocaine)	p < 0.05 (AUD scored higher on the PSQI)/CoUD: ns
Jaehne39	2012	Nicotine	Germany	PSG	44	44	NA	ns
Jaehne40	2015	Nicotine	Germany	PSG and PSQI	22	11	NA	p = 0.015 (NUD presented longer REM latency)
Jakubczyk41	2019	Alcohol	Poland	AIS	110	114	NA	p < 0.01 (AUD score higher on the AIS)
Kazmi42	2022	Alcohol	United States	PSQI	14	50	NA	p < 0.001 (AUD scored higher on the PSQI)
Keen43	2022	Cannabis	United States	PSQI	244	22	NA	p = 0.01 (CaUD scored higher on the PSQI)
Lahbairi44	2022	Alcohol	France	PSQI	38	53	NA	p < 0.01 (AUD scored higher on the PSQI)
Laniepce45	2019	Alcohol	France	PSG and PSQI	20	37	NA	p < 0.0001 (AUD scored higher on the PSQI)
Laniepce46	2020	Alcohol	France	PSG and PSQI	50	54	4–21 days	Control vs Mild-AWS: ns
Li47	2021	Alcohol	China	PSQI	50	50	NA	Control vs Moderate-AWS: p < 0.01 (smokers scored higher on the PSQI)
Liu48	2019	Alcohol	China	PSQI	15	15	NA	p < 0.05 (AUD score higher on the PSQI)
Liu49	2025	Alcohol	China	PSG and PSQI	50	50	1–2 days	p < 0.05 (AUD score higher on the PSQI)
McPherson50	2021	Cannabis	United States	PSQI	170	170	NA	p = 0.026 (AUD scored higher on the PSQI)
Mehtry13	2014	Opioid	India	PSG	15	15	1 week	p = 0.028 (females with a history of chronic CaUD scored higher on the PSQI)
Morgan51	2009	Cocaine	United States	PSG	19	26	3–20 days	During cocaine abstinence, men's sleep and learning worsened compared to women.
Morgan52	2010	Cocaine	United States	PSG, PSQI	12	10	1–16 days	People with CoUD slept fewer hours
Nabhan53	2023	Nicotine	Lebanon	PSQI	262	88	NA	p = 0.042 (OUD slept fewer hours)
Nour54	2023	Opioid	Egypt	PSG	20	20	NA	p < 0.001 (NUD scored higher on the PSQI)
Ogeil55	2015	Nicotine	Australia	PSQI	117	49	NA	p = 0.039 (OUD slept fewer hours)
Pace-Schott56	2005	Cocaine	United States	PSG	5	5	NA	CoUD had poorer sleep continuity and architecture compared with control group
Singh57	2018	Alcohol	India	PSG	20	20	3–4 weeks	p = 0.002 (NUD scored higher on the PSQI)
Soundararajan58	2021	Alcohol	United States	PSQI	389	497	NA	p = 0.009 (AUD slept fewer hours)
Voinescu59	2014	Alcohol	Romania	PSQI	167	54	NA	p < 0.001 (AUD scored higher on the PSQI)
Xiao60	2010	Opioid	China	PSG, PSQI	20	20	2–10 days	OUD had poorer sleep continuity and architecture compared with control group
Yuan61	2022	Alcohol	China	PSQI	96	151	NA	p = 0.001 (AUD scored higher on the PSQI)

Note: AIS, Athens Insomnia Scale; AUD, people with alcohol use disorder; AWS, Alcohol withdrawal syndrome; CaUD, people with cannabis use disorder; CoUD, people with cocaine use disorder; EEG, Electroencephalography; ESS, Epworth Sleepiness Scale; NA, not applicable; NR, not reported; ns, no significant differences; NUD, people with nicotine use disorder; OUD, people with opioid use disorder; PSG, Polysomnography; PSQI, Pittsburg Sleep Quality Index; REM, rapid eye movement sleep.

Results of meta-analyses

Alcohol

In the publications studying the association between AUD and sleep, 1612 controls (861 males) and 1558 cases (1154 males) were included, with an overall mean age of 43 years.^10,12,23,25,29, 30, 31^,34, 35, 36, 37, 38^,41,42,44, 45, 46, 47, 48, 49^,57, 58, 59^,61 Some assessments were performed with different periods of abstinence, ranging from two to 12 weeks. Of the 24 studies investigating AUD, 19 were performed in inpatient settings.^{10,12,23,25,29,}34, 35, 36, 37, 38^,41,42,44, 45, 46^,49,57,58,61 Additionally, for the subjective analyses, the PSQI total scores were used in 12 studies,^31,35,38,42,44, 45, 46, 47^,49,58,59,61 and the Athens Insomnia Scale was used in one.⁴¹ For objective analyses, PSG was used in 15 studies,^10,12,23,25,34, 35, 36, 37, 38^{,45,46,48,49,57} while EEG was used in two.^29,32 Subgroup analyses were not feasible due to similarities across studies in participant characteristics (eg, age range, substance use pattern and severity). For alcohol, age of alcohol use onset ranged from 17.9 to 21.9 years old, the duration of AUD ranged from 12.1 to 21.4 years, and sleep measurements happened mostly 2–4 weeks after last alcohol use. Participants with AUD had, on average, PSQI scores 4.89 points higher than controls, indicating poorer subjective sleep quality (Fig. 2A). Regarding objective (ie, PSG) outcomes, we pooled TST, sleep efficiency, REM latency (REMLT), REM/SPT, SWS, and WASO. We found that TST was significantly shorter in AUD (MD: −14.32, 95% CI = −16.69 to −11.96) compared to controls (Fig. 2B). Similarly, sleep efficiency was lower (MD: −3.15, 95% CI = −4.70 to −1.60), REMLT was shorter (MD: −13.62, 95% CI = −15.80 to −11.44), REM/SPT was higher (MD: 2.70, 95% CI = 1.47–3.93), and SWS was lower (MD: −3.68, 95% CI −4.99 to 2.38) in AUD compared to controls. No significant difference between groups was found for WASO (SMD: 0.24, 95% CI = −0.31 to 0.78) (Figure S2; Appendix p10). Clinical correlations were mentioned in three studies,^42,44,49 which found that poorer sleep—reflected by higher subjective PSQI scores and reduced objective TST—was positively correlated with higher anxiety and/or depression scores.

Fig. 2.

Sleep outcomes in people who use alcohol compared with controls: (A) Pittsburg Sleep Quality Index (PSQI); (B) Total sleep time (TST). SD, standard deviation; IV, inverse variance; CI, confidence interval; df, degrees of freedom.

Cannabis

For CaUD, only two studies met inclusion criteria.^43,50 They were published in 2021⁵⁰ and 2022,⁴³ involving 414 controls (165 males) and 192 cases (120 males), with a mean age of 23 years. One study did not distinguish between actively using and abstinent participants and found that females with a history of chronic cannabis use scored higher on the PSQI than the other groups (p = 0.028).⁵⁰ The second study, including participants actively using, found worse sleep for CaUD.⁴³ Additionally, when pooling findings from both studies, people with CaUD had poorer sleep (MD: 1.33, 95% CI = 0.32–2.35) (Fig. 3).

Fig. 3.

Pittsburg Sleep Quality Index (PSQI), in people who use cannabis compared with controls. SD, standard deviation; IV, inverse variance; CI, confidence interval; df, degrees of freedom.

Cocaine

Four studies investigated CoUD, published in between 2005⁵⁶ and 2016.³⁸ Overall, 144 controls (69 males) and 73 cases (92 males) were included, with a mean age of 37.7 years. Assessments occurred after at least three days of abstinence. Pooled results from these studies demonstrated significantly worse PSQI scores in CoUD (MD = 0.98, 95% CI = 0.04–1.93; Fig. 4A), and significantly reduced SWS (MD = −30.69, 95% CI = −47.27 to −14.10; Figure S3; Appendix p11). No significant differences were found for TST (even after sensitivity analysis, excluding two studies in which other substances were used,^51,56 Fig. 4B), REM, REM latency, or sleep onset latency (Figure S3; Appendix p11). Clinical correlations were not reported.

Fig. 4.

Sleep outcomes in people who use cocaine compared with controls: (A) Pittsburg Sleep Quality Index (PSQI); (B) Total sleep time (TST), sensitivity analysis. SD, standard deviation; IV, inverse variance; CI, confidence interval; df, degrees of freedom.

Nicotine

Studies investigating NUD involved 2054 controls (1187 males) and 1918 cases (894 males), with a mean age of 27 years. Of the ten included studies,^22,26, 27, 28^{,32,33,39,40,53,55} eight were performed in an outpatient setting.^22,26, 27, 28^,32,33,53,55 All used the PSQI to evaluate sleep subjectively, and some could not be meta-analyzed due to the absence of important data, such as SDs. Additionally, three studies used PSG,^22,39,40 and one used EEG,³² to measure sleep outcomes objectively. Subgroup analyses exploring potential differences between active use in comparison to early or late abstinence were not possible because studies involved adults actively smoking, and they did not have meaningful representation of early or late abstinence. Additionally, nicotine use characteristics were reported heterogeneously (eg, some studies used Fagerström Test for Nicotine Dependence [FTND] scores, while others reported cigarettes per day or other measures), preventing consistent pooling of data across studies. Additionally, the FTND scores were similar across studies, ranging from 5.3 to 6.4. In the meta-analysis, PSQI scores did not differ between groups (MD: 1.72, 95% CI = −0.06 to 3.51; Figure S4; Appendix p12), but after sensitivity analysis excluding two studies in which participants co-used alcohol,^26,27 higher mean PSQI score (ie, worse sleep quality) was found for NUD (MD: 2.64, 95% CI = 0.41–4.88) (Fig. 5A). Regarding objective measures, a significant difference was observed in TST (SMD: −0.33, 95% CI = −0.59 to −0.06) (Fig. 5B), but not in sleep efficiency (MD: 0.22, 95% CI = −6.00 to 6.44), REMLT (MD: 0.26, 95% CI = −10.41 to 10.92), REM/SPT (MD: 0.01, 95% CI = −1.66 to 1.69) or WASO (MD: 0.25, 95% CI = −0.40 to 0.87) between people with NUD and controls (Appendix p12). People who relapsed to nicotine use showed increased REMLT and further increases during abstinence.⁴⁰ Lastly, poor sleep quality and elevated daytime sleepiness were significantly associated with psychological distress.⁵⁵

Fig. 5.

Sleep outcomes in people who use nicotine compared with controls: (A) Pittsburg Sleep Quality Index (PSQI), sensitivity analysis; (B) Total sleep time (TST). SD, standard deviation; IV, inverse variance; CI, confidence interval; df, degrees of freedom.

Opioids

Four included studies investigated OUD. Overall, 65 controls (65 males) and 88 cases (88 males) were included, with ages ranging from 18 to 45 in one study and a mean age of 30 years in the other three. All studies included people with OUD using heroin, two tramadol,^24,54 one also included codeine,¹³ and another one included methadone.⁶⁰ Subjective measures were used in one of the studies,⁶⁰ showing significantly higher scores for people with OUD (ie, worse sleep) than in controls. Subgroup analyses were not possible due to the homogeneity of the participants included (all studies including participants after 2–10 days of abstinence) and substance use characteristics reported (all studies involving participants with a mean of 7 years of opioid use). In three studies, measurements were performed one week or more after last opioid use.^13,24,60 OUD was associated with significantly reduced TST (MD: −60.15, 95% CI = −78.78 to −41.52) (Figure S5; Appendix p13), even after sensitivity analysis, excluding a study in which participants co-used other substances (MD: −38.16, 95% CI = −63.04 to −13.28) (Fig. 6). No significant differences in sleep efficiency (MD: −21.10, 95% CI = −44.12 to 1.92), REMLT (MD: 6.52, 95% CI = −21.19 to 34.23) or SWS (MD: 0.81, 95% CI = −6.01 to 7.64) were found (Appendix p13). The meta-analysis for the WASO outcome was not possible due to lack of data.

Fig. 6.

Total sleep time, after sensitivity analysis, in people who use opioids compared with controls. SD, standard deviation; IV, inverse variance; CI, confidence interval; df, degrees of freedom.

Discussion

This meta-analysis synthesized data from 43 studies involving over 7500 individuals across multiple countries, focusing primarily on AUD, CoUD, NUD, OUD, and CoUD, with limited data on CaUD, and no eligible studies on BUD or MUD. Sleep outcomes assessed included TST, SWS, REM sleep, sleep efficiency, and subjective sleep quality (primarily measured using the PSQI). The findings revealed that AUD and NUD were consistently associated with poorer sleep quality, including significantly reduced TST and worse PSQI scores. OUD, particularly involving heroin, was also linked to a marked reduction in TST, although effects on other sleep measures were mixed. While fentanyl was specifically included in our search strategy, it was not mentioned in the included OUD studies. CoUD was associated with poorer sleep quality and reduced SWS. Data on CaUD was too limited to be meta-analyzed. The overall risk of bias for most of the studies was considered as having “some concerns”. Some pooled estimates showed high heterogeneity, likely reflecting differences in abstinence duration, polysubstance use exclusion, and variability in measurement tools (eg, PSQI vs PSG). Differences between inpatient and outpatient samples may have also contributed to heterogeneity. Inpatients often undergo supervised abstinence and structured sleep-wake routines. However, they also require adjustment to new settings, which could also interfere with sleep. Meanwhile, outpatients often remain exposed to environmental and psychosocial stressors (eg, unstable housing, ongoing use, or anxiety) that may impact sleep.^3,23,25 These contextual differences limit generalizability and highlight the need for stratified analyses by treatment setting in future research. Additionally, certain substance–outcome combinations (eg, those depicted in Figs. 3 and 5, Figures S3–S5) were based on a small number of studies, and their findings should therefore be interpreted with caution due to limited statistical power and potential publication bias. Subgroup analyses based on variables such as abstinence duration, usage patterns, and demographic factors were planned but could not be conducted due to insufficient representation across these subgroups. These factors likely contributed to the variability observed, and findings should be interpreted cautiously, as the small number of studies per outcome limited the precision of heterogeneity estimates. Additionally, we could not evaluate the longitudinal trajectory of sleep changes because most studies provided only cross-sectional data or single time-point assessments. Although meta-regression was planned to identify potential moderators of heterogeneity, such as age, abstinence duration, and measurement instrument, it was not feasible due to the small number of studies available per substance–outcome combination. Nonetheless, this meta-analysis offers a quantitative synthesis that expands on previous narrative reviews^3,62,63 and provides novel insights into potential adverse effects of substance use on sleep health and the methodological gaps that should be addressed in future research.

Individuals with SUDs commonly experience sleep disturbances that are more frequent and severe than in the general population,^4,5,62 often manifesting as insomnia, fragmented sleep, and reduced restorative sleep stages.^3,64 The prevalence and nature of these concerns vary by substance, with alcohol, cocaine, and opioid use disorders typically showing more pronounced disruptions than nicotine use disorder or other psychiatric conditions.^49,63,65,66 Both subjective and objective assessments corroborate these alterations, although their relationships differ across substances and measures.^3,14 While findings in AUD appeared more consistent across subjective and objective measures, showing reduced TST, lower efficiency, and altered architecture, this likely reflects the larger number of included studies. In contrast, CoUD findings, based on a smaller number of studies, showed discrepancies between perceived and measured sleep quality,¹⁴ emphasizing the need for additional research. Such differences underscore the importance of evaluating both subjective and objective dimensions. Subjective tools, such as the PSQI, Insomnia Severity Index, Epworth Sleepiness Scale, and sleep diaries, are accessible and easily used, which may reflect the greater volume of studies using them. However, their vulnerability to recall and perception bias may obscure true sleep dysfunction.⁶⁷ PSG remains the gold standard for assessing sleep architecture with minimal bias, but its high cost and logistical demands limit broader implementation. In this context, actigraphy may offer a practical alternative: although less detailed,⁶² it shows strong agreement with PSG in estimating TST and sleep efficiency in SUD populations.⁶⁸ Expanding its use could increase the availability of objective data and help clarify discrepancies between subjective perception and actual sleep patterns.

The neurobiology of sleep disturbances in SUDs is complex, involving disruptions across neurotransmitter systems, reward pathways, and circadian regulation.⁷ SUDs may affect key neurobiological systems and measures, such as locus coeruleus-norepinephrine (LC-NE), ventral tegmental area (VTA), serotonin, glutamate, gamma-aminobutyric acid (GABA), synaptic density, and dopamine, which have been implicated in plasticity and homeostasis involved with sleep-wake control.^4,5,7,69 In the VTA, glutamatergic neurons promote wakefulness via projections to the lateral hypothalamus and nucleus accumbens, while GABAergic neurons induce non-REM sleep and inhibit arousal circuits.⁸ Substance use may influence glutamate or GABA and produce excitatory/inhibitory imbalances, disrupting sleep continuity and depth.⁸ Substance-specific effects may further complicate such processes: OUD and withdrawal, for instance, involve hyperarousal and insomnia, driven largely by LC-NE dysregulation.⁷ This connection is bidirectional as sleep disturbances may worsen SUDs, while SUDs may exacerbate sleep problems, creating a vicious cycle. Addressing sleep in SUD treatment is thus critical, as improving sleep may not only support recovery but also disrupt this cycle and reduce relapse risk.⁵

Few studies reported on the relationship between SUD and sleep (eg, whether worse sleep related to relapse or more severe addiction). Nonetheless, among the studies that examined some clinical associations, poorer sleep quality was consistently linked to heightened depressive and anxiety symptoms, even in patients with no diagnosis of mood or anxiety disorders, particularly in AUD, NUD, and OUD.^24,42,44,55 In AUD, poorer sleep correlated with anxiety and depression,⁴² while NUD relapse was tied to prolonged REMLT.⁴⁰ OUD studies found reduced SWS in people with depression,²⁴ and sleep impairments (such as shortened TST, shortened REMLT, and deficiencies in SWS) during detoxification often predicted relapse more strongly than age or mood.^5,64 Clinically, individuals with SUDs frequently experience insomnia during use or withdrawal,⁷⁰ sometimes self-medicating with sedatives that worsen sleep long-term.⁷¹ These findings suggest that sleep disturbances may function as both core symptoms and modifiable relapse factors, potentially perpetuating affective dysregulation through shared neurobiological pathways between sleep and mood, underscoring the need for integrated sleep interventions tailored to each substance's impact on sleep architecture.⁷

Emerging evidence suggests pharmacological approaches for sleep disturbances in SUDs should be substance-specific, targeting distinct neurobiological alterations. For example, in people with CoUD, modafinil has been shown to improve sleep efficiency and daytime functioning.⁷² CoUD also exhibits a biphasic withdrawal pattern, with early withdrawal being characterized by REM rebound and transient TST increases and late withdrawal by progressive REM suppression and sleep fragmentation, suggesting a window for targeted interventions like melatonin or GABAergic agents to stabilize circadian rhythms.⁶⁶ Conversely, OUD shows TST reduction and hyperarousal, consistent with locus coeruleus hyperactivity, in which orexin antagonists (eg, suvorexant) may counter withdrawal-related insomnia.⁷³ While benzodiazepines could theoretically address OUD-related hyperarousal, their addiction risk and potential effects on sleep architecture preclude routine use,⁷⁴ important considerations given our exclusion of benzodiazepine studies due to insufficient controlled data. Beyond pharmacological options, non-pharmacological treatments, such as cognitive behavioral therapy for insomnia (CBT-I), sleep hygiene psychoeducation, and chronotherapy, have shown promise in improving sleep quality and reducing relapse vulnerability in SUD populations.^4,62,74 In particular, CBT-I may complement pharmacologic approaches by addressing maladaptive cognitions, irregular sleep–wake patterns, and conditioned arousal commonly observed during abstinence.^5,70 Behavioral sleep interventions are especially relevant for NUD, where subjective complaints predominate.^3,74 Collectively, these substance-specific and multimodal strategies highlight the need for individualized approaches that address both neurobiological and behavioral contributors to sleep disturbance in SUDs.^7,63

The meta-analysis has some limitations. Many studies identified during screening lacked control groups, reducing the number of eligible papers. To minimize confounders, we excluded studies where patients received pharmacological treatment for insomnia, a common but methodologically important constraint that distinguishes our work from prior reviews. While we initially aimed to include benzodiazepines and methamphetamine, insufficient data with appropriate controls precluded their inclusion, and this highlights a gap in the SUD-related sleep literature. Subgroup analyses (eg, by substance use patterns or demographics) were also unfeasible due to limited comparable data. Notably, sex-specific gaps persist: only one eligible study examined sex effects,⁵⁰ despite evidence that women with SUDs (eg, CaUD) often experience more severe sleep disturbances and neurobiological differences than men.⁷⁵ Another limitation is that subjective questionnaires such as the PSQI assess sleep patterns over the preceding month, which may not capture rapid fluctuations across substance use, withdrawal, and abstinence phases. Similarly, the timing of PSG assessments differed from questionnaire administration, potentially contributing to variability in observed associations between subjective and objective sleep measures. Future work could consider using more than one subjective questionnaire, on multiple days, in addition to prioritizing sex-balanced samples and mechanistic studies to address existing gaps.

Building on this work, important next steps include: mechanistic studies examining how anxiety and depression relate to sleep-SUD relationships, given preliminary correlations in AUD and NUD^42,55; sex-stratified analyses to clarify differences in sleep disturbance severity and treatment responses; circadian rhythm investigations, using actigraphy to capture rest–activity cycles in SUDs, as proposed by recent work⁶⁵; and inclusive pharmacotherapy trials targeting understudied substances (eg, fentanyl) with circadian-aligned interventions (eg, timed light therapy or melatonin agonists). Actigraphy could also bridge gaps where polysomnography is impractical, particularly in community-based or polysubstance-using populations. These approaches would address current limitations while advancing translational insights into sleep as a modifiable target in SUD recovery.

This is the first systematic review and meta-analysis on the association between SUD and sleep assessing both subjective and objective sleep outcomes, including only studies with control groups. Based on our findings, it is evident that sleep disturbances are prevalent among individuals with SUDs, with both subjective and objective measures suggesting disruptions across AUD, NUD and OUD. Interestingly, although actively sought, no studies on BUD or MUD met inclusion criteria. Potential discrepancies between subjective and objective measures, particularly in SUDs like CoUD and OUD, suggest that each assessment tool may capture distinct aspects of sleep. Moreover, few included studies described relationships between sleep disturbances and clinical outcomes. Together, these findings position sleep as a modifiable target for improving SUD treatment outcomes.

Contributors

GAA and HNPO conceptualized the scope and content of this review. AMPN, HNPO, and TPP performed the initial search. GAA, HNPO, and TPP accessed and verified the data. AMPN, GAA, HNPO, and TPP drafted the first version of the manuscript. AMPN, GAA, HNPO, MNP, PTM, RSMJ, and TPP contributed to drafting the final version. GAA, MNP, and PTM critically revised the manuscript. All authors reviewed and approved the final version for publication.

Data sharing statement

Data will be provided upon reasonable request.

Declaration of interests

The authors declare no conflicts of interest. MNP discloses that he has consulted for and advised Boehringer Ingelheim and Neurofinity; been involved in a patent application with Yale University and Novartis; received research support from the Mohegan Sun Casino and the Connecticut Council on Problem Gambling; consulted for or advised legal, non-profit, healthcare and gambling entities on issues related to impulse control, internet use and addictive behaviors; performed grant reviews; edited journals/journal sections; given academic lectures in grand rounds, CME events, and other clinical/scientific venues; and generated books or chapters for publishers of mental health texts. GAA discloses that he has received research support from the National Institutes of Health (R01 DA052454–03, R33DA053592, R01DA060231-01); received consulting fees from the Center for Psychedelic Drug Research and Education at Ohio State University; and received honoraria for his role as Editor of the Journal Current Addictions Report and for presentations given at Michigan State University and the University of Pennsylvania. The other authors do not report any disclosures.