The effect of exercise interventions on substance-use outcomes: a meta-analytic and systematic review

Abstract

Introduction:

A number of studies have been conducted to assess the efficacy of exercise-based interventions for substance use disorders. Nicotine use has been overrepresented in prior meta-analytic and systematic reviews, potentially obscuring the effects of exercise on outcomes related to other drugs. The aim of this meta-analysis and systematic review is to offer an updated account of the impact of exercise on substance-use outcomes for drugs other than nicotine.

Methods:

Eligible studies included peer reviewed articles published before August 2023 describing randomized controlled trials or clinical trials involving adults in which exercise served as the primary treatment intervention and substance use and/or craving outcomes were assessed for drugs other than nicotine. In addition to omnibus effects, meta-regression models were conducted to assess study design (within vs. between), data collection (acute vs. long-term), and outcome measure (drug craving vs. drug use) as potential moderating variables.

Results:

A total of 19 articles describing 17 unique studies were selected for analysis, including data from 1,363 individual participants. An omnibus RVE meta-analysis indicated a significant reduction in substance use outcomes following exercise intervention. The moderator analysis additionally indicated a significant effect of design type such that between-subject designs displayed smaller magnitude reductions than within-subject designs. Other potential moderators were not significant.

Conclusion:

These findings indicate that a variety of exercise-based interventions produce moderate improvements in substance use and craving for people using non-medical drugs and corroborate prior reports that exercise is effective as an adjunctive or standalone treatment for substance use disorder.

Introduction

Epidemiological studies have long reported an inverse relationship between physical activity and substance use, such that greater engagement in physical activity is associated with less drug use (Halladay et al., 2024; Korhonen et al., 2009; Ströhle et al., 2007; Terry-McElrath and O’Malley, 2011). Preclinical studies have shown a causal relationship between physical activity and lower degrees of drug intake. For instance, studies conducted in rodents have shown that wheel running (Aarde et al., 2015; Cosgrove et al., 2002; Lacy et al., 2014; Reguilón et al., 2020; Sanchez et al., 2015; Smith et al., 2008), treadmill running (Alizadeh et al., 2018; Hosseini et al., 2009), and resistance training (Smith et al., 2018; Strickland et al., 2016) reduce the intake of drugs from diverse pharmacological classes (e.g. nicotine, alcohol, stimulants, and opioids).

A number of clinical trials have been conducted to determine the effects of exercise on physical health, psychological well-being, and drug-use outcomes in substance-using populations. These studies have reported positive effects on measures of physical and mental health outcomes, independent of substance use outcomes (Dolezal et al., 2013; Flemmen et al., 2014; Liu et al., 2021; Muller and Clausen, 2015; Nygård et al., 2018; Rawson et al., 2015). The effects of exercise on nicotine use have been over-represented in these studies, and several systematic reviews and meta-analyses have described positive effects of exercise on short-term nicotine craving and smoking cessation (Chen et al., 2022; Darabseh et al., 2022; Santos et al., 2021). The effects of exercise on substance-use outcomes for other drugs are less clear; however, a meta-analysis conducted in 2014 reported generally positive findings (Wang et al., 2014).

The aim of this meta-analysis and systematic review is to provide an updated evaluation of the effects of exercise on substance-use outcomes for drugs other than nicotine. To this end, we reviewed clinical trials in which exercise served as the primary treatment intervention. In addition to determining omnibus effects, we examined study design (within vs. between), time of data collection (acute vs. long-term), and outcome measure (drug craving vs. drug use) as potential moderating variables.

Methods

Search Strategy

We conducted a search for studies using PubMed/Medline, SCOPUS, and PsycINFO. The search included terms related to exercise (“resistance train”, “aerobic”, “strength train”, “exercise”), drug use (“use disorder”, “drug use”, “substance use”, “abstinence”) and drug (“cocaine”, “heroin”, “opioid”, “amphetamine”, “methamphetamine”, “marijuana”, “cannabis”, “alcohol”). The search was limited to studies printed in English, available online, and published before August 2023.

Inclusion Criteria

Studies included in the initial search were constrained by several key inclusion criteria. Selected studies included participants who were adults (≥ 18 years old) and reported drug use. Specifically, included studies either involved participants seeking/in treatment for a substance use disorder or non-treatment seeking participants who met the criteria for substance use disorder. Though nicotine use per se was not excluded, only studies in which the outcome involved a drug other than nicotine were included. All included studies used exercise as the primary intervention, though the type of exercise was not restricted (aerobic or anaerobic), and the duration of the exercise varied across studies. Both clinical trials and randomized controlled trials were included. All included studies specifically assessed drug use or craving of a drug other than nicotine as outcome measures. Studies were excluded if the outcomes did not involve explicit use or craving outcomes; for example, studies only assessing psychological factors like depression or anxiety that may correlate with substance use were not included. Effect sizes were extracted, as possible, using raw data or interpolated data from manuscript figures. During the initial search, 23 articles were selected; 4 of these were later excluded from analysis due to lack of data. The final analysis included 19 articles describing 17 studies. The selection process is depicted in Figure 1.

Figure 1. Consort Diagram.

Data Analysis

Omnibus effect size estimates were generated using an RVE meta-regression method (Hedges et al., 2010; Tipton, 2015). This method allowed for the incorporation of dependent effect sizes into a single model (e.g., multiple measures within-person or the mixture of between- and within-subject comparisons present in most studies) without violation of assumptions of independent observations. A modified form of the RVE method was used with a small-sample size adjustment (Tipton, 2015).

First, omnibus tests were conducted with all observed effect sizes included. Meta-regression models were then conducted with moderators including design (between- versus within-subject comparison), time of assessment (acute versus long-term), and measure type (craving versus substance use indicator). Publication bias tests were then conducted on the average effect size estimates within study given that traditional publication bias measures (e.g., Egger’s plot for funnel asymmetry) have not yet been widely adopted or validated for RVE models. Separate publication bias tests were conducted for between and within-subject comparison given the robust difference observed in the omnibus test and noted heterogeneity in collapsing these measures within study. Tests were conducted in R with RVE models evaluated using the robumeta and meta packages (Fisher and Tipton, 2015; Viechtbauer, 2010).

Results

Narrative Review

A total of 19 articles describing 17 unique studies were included in the final analysis, comprising data from 1,363 individual participants (Brown et al., 2014; Buchowski et al., 2011; Colledge et al., 2017; Cutter et al., 2014; De La Garza et al., 2016; Grandjean da Costa et al., 2017; Gunillasdotter et al., 2022; Hallgren et al., 2021, 2014; He et al., 2021; Jensen et al., 2019; Roessler et al., 2017; Salem et al., 2022; Trivedi et al., 2017; Ussher et al., 2004; Wang et al., 2020, 2016; Zhang and Zhu, 2020).

Of the 19 included studies, 8 focused on stimulant-related outcomes (42.1%), 7 on alcohol (36.8%), 3 on opioids (15.8%), 2 on multiple drugs (10.5%), and 1 on cannabis (5.2%). Running and cycling were the most common form of exercise; each were used in 6 studies (31.6%). Cycling tended to be used in conjunction with acute assessments. Aerobic exercise of any form accounted for the majority of studies. 11 studies (57.9) involved long-term assessment (one month or longer), and the remaining studies assessed outcome measures in an acute time frame (42.1%), defined as a data collection period of less than two weeks. Most of the studies considered acute involve a single session of exercise. Only craving was assessed acutely and only one study involving long-term assessment measured craving. 11 studies included both between and within subject analyses (57.9%). 5 studies used a between subject design (26.3%), all of which collected data long-term. 3 studies used a within subject design (15.8%), all of which used acute assessments only. Table 1 includes a narrative summary of included studies. The treatment status of participants varied from non-treatment seeking, to treatment seeking but not abstinent, to treatment seeking and abstinent. The duration and frequency of exercise varied widely, even among studies employing the same exercise modality.

Table 1.

Summary of Included Studies

Article	N	Drug	Exercise	Exercise Duration	Exercise Frequency	Study Design	Time of Data Collection	Outcome
Colledge et al. 2017	24	Opioid	Various		2x/week		Long-Term (12 week)	Use
Ussher et al. 2004	20	Alcohol	Cycling	10 minutes	Once	Between, Within	Acute (10 minutes)	Craving
Rawson et al. 2015	135	Stimulant	Running	55 minutes	3x/week	Between	Long-Term (6 months)	Use
Salem et al. 2022		Stimulant	Running	55 minutes	3x/week	Between, Within	Long-Term (8weeks)	Craving
Jensen et al. 2019	105	Alcohol	Running	30–45 minutes	1–2 hr/week	Between, Within	Long-Term (6 months)	Use
Roessler et al. 2017	175	Alcohol	Running		2x/week	Between	Long-Term (12 months)	Use
Buchowski et al. 2011	12	Cannabis	Running	30 minutes	5x/week	Within	Acute (2 weeks)	Craving
Wang et al. 2020	60	Opioid	Cycling	20 minutes	Once	Between, Within	Acute (40 minutes)	Craving
Gunillasdotter et al. 2022	140	Alcohol	Cycling, Yoga	30 minutes	Ix/week	Between, Within	Long-Term (13weeks)	Use
Hallgren et al. 2021		Alcohol	Cycling	12 minutes	Once	Within	Acute (30 minutes)	Craving
Hallgren et al. 2014	18	Alcohol	Yoga		Daily	Between, Within	Long-Term (10weeks)	Use
Wang et al. 2016	92	Stimulant	Cycling	50 minutes	Once	Between, Within	Acute (50 minutes)	Craving
Brown et al. 2014	49	Alcohol	Aerobics		90–160 minutes/week	Between	Long-Term (6 months)	Use
Grandjean da Costa et al. 2017	14	Polydrug	Cycling	20 minutes	Once	Within	Acute (Immediate)	Craving
Trivedi et al. 2017	302	Stimulant	Walking	30–50 minutes	3x/week	Between	Long-Term (12 weeks)	Use
De La Garza et al. 2016	24	Stimulant	RunningandWalking	30 minutes	3x/week	Between, Within	Long-Term (4weeks)	Use, Craving
Cutter et al. 2014	29	Opioid, Stimulant	Active Game Play	20–25 minutes	5x/week	Between	Long-Term (8weeks)	Use
He et al. 2021	90	Stimulant	Aerobic, Resistance	1 hour	Once	Between, Within	Acute (Immediate)	Craving
Zhang&Zhu 2020	74	Stimulant	Taijiquan	50 minutes	5x/week	Between, Within	Long-Term (6 months)	Use
*Bichler et al. 2017	16	Alcohol				Between	Acute	Craving
*Roessler, 2010	38	Various				Within	Long-Term	Use
*Weinstock et al. 2014	31	Alcohol				Between, Within	Long-Term	Use
*Murphy et al. 1986	60	Alcohol				Between, Within	Long-Term	Use

Meta-Analysis

Table 2 contains meta-analysis outcomes from the omnibus analysis. The omnibus RVE meta-analysis indicated a significant and medium effect size reduction in substance use outcomes following exercise interventions, d = −0.58 [−0.83 – −0.32], p < .001. Coefficients from moderator analyses are also included in Table 2. A significant effect of design type was observed such that between-subject designs/comparisons showed a smaller magnitude reduction than within-subject designs/comparisons, b = 0.39 (0.12 – 0.66), p = .007 (i.e., effect size estimate within-subject = −0.80; between-subject = −0.41). This difference can be observed in the Figures 2 and 3 forest plot depicting between (Figure 2) and within (Figure 3) subject comparisons. Moderating effects for time of collection (p = .13) and outcome (p = .07) were not statistically significant, but acute measurement and craving outcomes tended to show larger effect sizes than long-term (e.g., clinical trials) measurement and measures of substance use.

Table 2.

Summary of Primary RVE Meta-Analysis

Measure	Effect estimates (n)	Coefficient Estimate of Robust Variance Estimation (RVE) Meta-Analysis	Cohen’s d
n = 135
Studies = 17
Effect Size
Omnibus (95% CI, I2)	135	−0.58*** (−0.83 – −0.32), I2= 88.0%	−0.58
Moderators (95% CI)
Design
Within	66	−0.80*** (−1.11 – −0.49)	−0.80
Between	69	0.39** (0.12 – 0.66)	−0.41
Time of Collection
Long-Term	74	−0.41*** (−0.60 – −0.23)	−0.41
Acute	61	−0.45 (−1.04 – 0.15)	−0.86
Outcome
Craving	67	−0.86*** (−1.39 – −0.33)	−0.86
Use Indicator	68	0.49# (−0.036 – 1.01)	−0.37

^#p < .10

^**p < .01

^***p < .001

Figure 2. Forest Plot of Between-Subject Comparisons. Data are presented with the average effect size within each study for visual purposes.

Figure 3. Forest Plot of Within-Subject Comparisons. Data are presented with the average effect size within each study for visual purposes.

Publication Bias

Estimates of publication bias were taken from average effect sizes within each study. Publication bias estimates were taken separately for within and between subject comparisons given the noted heterogeneity in estimates and difficulty in collapsing these measures within study. The between-subject meta-analysis using the average effect size estimate indicated a similar effect size as the omnibus moderator tests, d = −0.43 [−0.71 – −0.15], p = .002, k = 15. Egger’s test for funnel plot asymmetry was not statistically significant consistent with visual inspection of the funnel plot (Supplemental Figure 1). Jackknife analyses suggested minimal change in the omnibus effect size (d = −0.47 to −0.32).

The within-subject meta-analysis using the average effect size estimate indicated a similar effect size as the omnibus moderator tests, d = −0.78 [−1.10 – −0.46], p < .001, k = 12. Egger’s test for funnel plot asymmetry was not statistically significant consistent with visual inspection of the funnel plot (Supplemental Figure 2). Jackknife analyses suggested minimal change in the omnibus effect size (d = −0.86 to −0.65).

Discussion

The purpose of the present study was to provide an updated account of exercise-based interventions on substance-use outcomes. An analysis of 17 studies indicated an effect of medium size, such that exercise intervention decreased experimental metrics of substance use and craving. Notably, this analysis included studies assessing outcomes related to the use of non-medical drugs, but not studies involving nicotine. Historically, studies investigating the impact of exercise-based intervention for nicotine/tobacco use have been overrepresented in systematic reviews and meta-analyses relative to outcomes associated with other drugs (Martinez-Calderon et al., 2023; Patterson et al., 2022; Wang et al., 2014). The present findings corroborate that exercise-based intervention produces moderate improvements in substance-use outcomes in people using drugs other than nicotine.

A significant effect of study design indicates that within-subject designs result in larger improvements in outcome measures compared to between-subject designs. Although moderating effects of time of collection and outcome did not reach significance, acute measurements and craving outcomes tended to show larger effect sizes than long-term measurement and use outcomes. There are several considerations in interpreting these results. Across all included studies, only craving was assessed in an acute time frame (data collected less than two weeks from exercise intervention) and all craving measures were analyzed within-subject, though in some studies a between-subject analysis was also conducted. In the included studies, long-term assessment was more likely to occur in the context of a clinical trial, while acute assessments were frequently employed in randomized controlled trials, many involving a single session of physical activity (Grandjean da Costa et al., 2017; Hallgren et al., 2021; He et al., 2021; Ussher et al., 2004; Wang et al., 2020, 2016). As the assessment of intake is not feasible within a study involving a single session (immediate pre- and post- exercise assessment), this effect of study design may primarily reflect the malleability of craving to acute change.

The prevalence of studies and systematic reviews conducted on the therapeutic potential of physical activity for the treatment of mental health disorders, including substance-use disorders, in the last decade attests to enduring public and scientific interest in this topic. However, several researchers have raised serious concerns about the quality of this research. For example, Martinez-Calderon et al. (2023) evaluated 18 systematic reviews with 53 meta-analyses comprising 103 clinical trials published between 1998 and 2022. They note that the majority of the systematic reviews focused on nicotine/tobacco and that the reviews devoted to nicotine tended to focus on use/misuse, whereas the reviews devoted to other drugs focused on psychological symptoms. Other concerns included lack of transparency during article selection, risk of bias, and lack of specificity. Their report concludes with a call to improve the quality of systematic reviews investigating the impact of exercise on substance-use related outcomes.

The present analysis addresses several of the concerns laid out by Martinez-Calderon et al. (2023). First, our exclusion criteria at each stage of the article selection process are explicitly stated in the methods section, and articles initially selected but excluded from analysis are cited and listed in Table 1. Second, we directly assessed publication bias in the design of our analysis. Publication bias was assessed using Egger’s test for funnel plot asymmetry and metrics of publication bias were conducted separately for between- and within-subject comparisons, to affirm that no individual studies nor particular study design was driving the results of the omnibus analysis. Finally, we conducted a targeted meta-analysis with specific attention to moderators to evaluate which methodological factors may influence the conclusions of other systematic reviews. Importantly, Martinez-Calderon et al. (2023) point out that including studies and outcomes related to different substances contributes to potentially problematic heterogeneity among studies. Though the present analysis excludes studies involving nicotine, which have been previously overstated in the literature relative to other drugs, it does include a variety of drugs (alcohol, stimulants, opioids, and cannabis) in unequal proportions. A total of 19 studies that would have otherwise met inclusion criteria were excluded because they involved nicotine. Though these studies may be numerous enough to constitute an individual meta-analysis, no other drug was represented frequently enough in our search to justify conducting separate analyses.

The quality of the studies in this field has been called into question. One recent systematic review (Patterson et al., 2022) assessed the risk of bias in studies of exercise interventions for substance use-related outcomes. This assessment indicated that the factor with the most risk of bias was blinding of participants or personnel to condition. Though understandable, this nevertheless represents an important issue. Though several studies employ a control group exposed to an alternative intervention (health education, for example; He et al. 2021), none could effectively obscure the nature of intervention. This impact of functional unblinding also likely partly contributes to the larger magnitude effect on more proximal and malleable outcomes like drug craving as opposed to distal and reticent to change outcomes like drug intake (e.g., urine drug screens). Balancing this limitation is the fact that real-world exercise-based interventions are inherently unblinded and that expectations can be a therapeutic target for enhancing the positive effects of these interventions (e.g.,(Helfer et al., 2015)).

This study does not address mechanism, but previous studies provide evidence that exercise may reduce drug use through several means. Exercise reliably decreases measures of depression (Liu et al., 2025), anxiety (Ramos-Sanchez et al., 2021), and chronic pain (Ninneman et al., 2024), all of which are comorbid risk factors that drive drug use. Exercise also increases measures of self-esteem (Park et al., 2014) and self-efficacy (Gilanyi et al., 2023), both of which are associated with lower rates of drug use. In laboratory models, exercise serves as an alternative nondrug reinforcer that allocates behavior away from drug use and drug seeking (Carroll, 2021). Exercise also produces functional changes in dopaminergic (Abdullah et al., 2022), opioidergic (Pettrey et al., 2024), cannabinergic (Brellenthin and Koltyn, 2016), and glutaminergic (Lynch et al., 2013) neurotransmitter systems to reduce the acute reinforcing effects of drugs and mitigate the pathological changes associated with substance use disorder. These mechanisms are not mutually exclusive and likely operate in a complementary fashion across psychobiological endpoints to reduce maladaptive patterns of drug use.

This analysis and those predating it collectively suggest that exercise improves outcomes associated with substance use and craving for a variety of drugs from different pharmacological classes (Ashdown-Franks et al., 2020; Patterson et al., 2022; Wang et al., 2014; present study). Exercise is a low cost, accessible, and flexible treatment option for anyone who is physical able to engage. In contrast, there are significant personal and structural barriers for many people seeking pharmacological treatment or psychotherapy for substance use disorders, including time and financial commitment (Rapp et al., 2006, Farhoudian et al. 2022). If exercise is a viable treatment option, its incorporation into treatment as an intervention or adjunctive therapy may improve access for people seeking treatment for substance use disorders. Future work is needed to inform implementation/effectiveness work given the challenges and barriers of translating exercise-based interventions into real-world treatment programs (e.g., see Horrell et al., 2020 for discussion of perceived barriers). Given the potent role of non-drug activities like exercise in maintaining long-term recovery (Acuff et al., 2023, 2024), efforts to facilitate the incorporation of physical activity-based programs into standard-of-care treatment stand to greatly benefit substance use disorder care.

Public significance statement:

In recent years, public and scientific interest in exercise as a low-cost, personalized adjunctive or standalone treatment for substance use disorder has grown significantly. A number of studies have been conducted to assess the efficacy of such interventions for a variety of substances and substance-related outcomes. This targeted meta-analysis synthesizes current evidence and corroborates the positive impact of exercise intervention on craving and use of non-medical drugs.