Effects of mindfulness-based interventions on food craving in adults: a systematic review and meta-analysis of controlled clinical trials

Abstract

Background

Food craving (FC) is associated with a higher body mass index (BMI) and less healthy food choices. Mindfulness-based interventions (MBIs) have garnered interest as a strategy to address maladaptive eating behaviors. We conducted a systematic review and meta-analysis to evaluate the effectiveness of MBIs in reducing FC.

Methods

We included controlled trials involving adult participants. A comprehensive search of Embase, MEDLINE (via PubMed), Scopus, Web of Science, ProQuest, and CENTRAL was conducted up to July 7, 2024. The Cochrane RoB2 tool was used to assess the risk of bias in included studies, and the GRADE framework was applied to evaluate the certainty of evidence. Effect sizes were calculated using Hedges’ g with 95% confidence intervals (CIs). Results favoring MBIs over control conditions were interpreted as statistically significant positive effects.

Results

A total of 24 studies (1920 participants) were included. FC intensity was significantly reduced in the MBI group compared to controls (Hedges’ g = 0.28, 95% CI: 0.07–0.48, p = .008, n = 23), with subgroup analysis suggesting the strongest effect for the decentering strategy. The certainty of evidence was rated as low. FC frequency did not differ significantly between the MBI and control groups (g = 0.11, 95% CI: − 0.28 to 0.49, p = .59, n = 9). The MBI group exhibited significantly lower rebound intake (i.e., food consumption following the conclusion of the restrictive phase of the study) (g = 0.83, 95% CI: 0.24–1.41, p = .006, n = 5) and a non-significant reduction in overall intake (g = 0.46, 95% CI: − 0.02 to 0.93, p = .06, n = 9) relative to controls. No adverse effects were reported.

Conclusion

Preliminary findings suggest that MBIs may be effective and safe interventions for managing FC intensity but not frequency. However, additional high-quality studies are needed to strengthen the evidence base.

Prospero registration

CRD42024584334.

Supplementary Information

The online version contains supplementary material available at 10.1186/s40359-025-03307-6.

Introduction

Rationale

Food craving (FC) is defined as a strong and seemingly irresistible desire to consume a specific food [1]. FC appears to be highly prevalent, with reports indicating that up to 100% of young women and 70% of young men experience such cravings annually [2].

FC is associated with obesity, higher body mass index (BMI), unhealthy dietary choices, and maladaptive eating behaviors such as emotional and binge eating [3–6]. It has been causally linked to excessive food intake—accounting for up to 11% of overconsumption—and may influence the success of dietary interventions [7–9].

To better understand FC, it is important to consider how it is measured across studies. FC can be assessed in two contexts: momentary (state) FC, which reflects transient, situational cravings experienced in the moment, and characteristic (trait) FC, which captures more stable, habitual tendencies to experience cravings over time [10, 11]. Among the tools used for measuring food cravings are the Food Craving Questionnaire (FCQ) developed by Cepeda-Benito et al. [11], the General Food Craving Questionnaire (G-FCQ) introduced by Nijs et al. [12], the Craving Experience Questionnaire-Strength (CEQ-S) proposed by May et al. [13], the craving subscale of the Attitudes to Chocolate Questionnaire (ACQ-C) established by Benton et al. [14], the visual analog scale (VAS), and Likert scales.

Given the significant impact of food cravings on weight gain and maladaptive eating behaviors, effective strategies for managing them are essential. One such approach that has gained increasing attention is mindfulness-based interventions (MBIs). Jon Kabat-Zinn introduced mindfulness into academic discourse in 1982 [15]. This practice involves intentionally attending to one’s present-moment experiences, thoughts, and emotions with openness and non-judgment [15]. Mindfulness has demonstrated efficacy in reducing stress, emotional eating, and stress-related overconsumption. MBIs have emerged as a promising approach for addressing maladaptive eating behaviors [16, 17].

MBIs incorporate a variety of strategies to target cravings and related behaviors, including acceptance, decentering, body scanning, self-compassion, or combinations thereof. Acceptance focuses on altering an individual’s relationship with cravings rather than suppressing or resisting them, promoting acknowledgment of cravings as mere experiences [18]. Decentering involves viewing thoughts and emotions as transient mental events distinct from the self [19]. Body scan entails carefully directing focus to different regions of the body, which can deepen one’s perception of inner sensations [20]. Self-compassion is characterized by treating oneself with kindness, particularly during periods of distress [21].

A growing body of research supports the utility of MBIs in managing food-related challenges. Studies have shown that MBIs can reduce the intensity and frequency of food cravings and improve reward-driven eating behaviors [19, 22, 23]. Emerging evidence also suggests that remotely delivered MBIs—via mobile apps or audio-guided practices—can effectively reduce maladaptive eating behaviors, anxiety, and body weight [23–29]. Although most of these studies have been short-term or conducted in laboratory settings, they highlight the feasibility of remote formats, especially in the post-COVID era.

Evidence from related fields further highlights the potential of MBIs. For instance, meta-analyses on substance use disorders report reduced craving levels, lower substance use frequency, and improved emotional regulation following MBI implementation [30, 31]. MBIs have also demonstrated benefits for specific populations, such as post-bariatric surgery patients and individuals with binge eating tendencies [32, 33]. Nevertheless, to our knowledge, no prior meta-analysis has focused specifically on FC intensity or frequency, leaving a key gap in the literature.

Objectives

We aimed to conduct a meta-analysis to provide robust evidence on the effects of MBIs on FC. Additionally, we sought to compare the efficacy of distinct MBI strategies. To this end, the present study systematically investigated trials examining the impact of MBIs on the FC intensity, FC frequency, or both in adult populations, compared to control groups.

Interventions containing broader therapeutic frameworks, such as Acceptance and Commitment Therapy (ACT) and Dialectical Behavior Therapy (DBT), were excluded. This exclusion was implemented to maintain focus on core mindfulness components without the confounding effects of additional therapeutic modalities. While ACT and DBT incorporate mindfulness as a foundational element, they extend beyond conventional mindfulness approaches through several distinct therapeutic components. ACT uniquely emphasizes values clarification, and committed action toward valued life directions [34]. DBT integrates mindfulness within a broader skills training framework that includes distress tolerance, emotion regulation, and behavioral changes [35]. In contrast, conventional MBIs primarily focus on nonjudgmental present-moment awareness through meditation practices without incorporating the behavioral change strategies, values work, or interpersonal skills training characteristic of ACT and DBT. Including ACT and DBT in this study would introduce additional therapeutic variables that could obscure the specific mechanisms and outcomes attributable to pure mindfulness practice.

We opted to include self-compassion as part of MBIs based on its frequent integration into mindfulness protocols. For example, Mindfulness-Based Eating Awareness Training (MB‑EAT), a widely studied program, explicitly integrates mindfulness meditation with practices promoting self-acceptance and compassion—helping participants counteract shame and harsh self-judgment often tied to overeating and binge behaviors [36].

Methods

The Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement guided all stages of this review [37]. The review protocol was registered in PROSPERO (CRD42024584334).

Eligibility criteria

Inclusion criteria:

• Clinical trials examining the effects of MBIs on FC intensity and/or FC frequency (analyzed separately).

• Studies involving adult populations (≥ 18 years).

• Inclusion of a comparator/control group.

• Articles published in English or Persian.

The language restriction was applied because the research team is only proficient in English and Persian, ensuring accurate interpretation and data extraction.

Exclusion criteria:

• Quasi-experimental or non-experimental study designs.

• Studies employing ACT or DBT as the primary MBI strategy.

• Research involving participants under 18 years of age.

• Absence of a control group.

• No reported outcomes for FC intensity or FC frequency.

Search & selection process

We searched the Embase, MEDLINE (via PubMed), Scopus, Web of Science, and ProQuest databases, along with the Cochrane Central Register of Controlled Trials (CENTRAL), for relevant studies on July 7, 2024.

The Population, Intervention, Comparison, Outcome (PICO) framework was used to formulate the research question, defining the target population as adults, MBIs as the intervention, and FC intensity/frequency as the outcome. The search terms consisted of mindfulness, meditation, and technique names paired with combinations of the term “craving” with terms such as food, chocolate, etc. The full search strategy is provided in Additional File 1. Additionally, the reference lists of included studies and recent reviews were manually screened to ensure no relevant studies were overlooked.

Two researchers (SA and ZM) independently screened studies for eligibility. In cases of discrepancies, a third reviewer (GA) resolved disagreements through consensus.

While we did not rerun the search due to our anticipated publication timeline being within 12 months of the initial search [38], we activated email alerts for all databases and registries where this feature was available (ProQuest, Embase, PubMed) to identify newly published studies until the date the analyses concluded (December 1st, 2024).

Data collection

We extracted available publication data (first author’s last name, country, year), follow-up duration, baseline characteristics (number of cases, age, BMI, female participant ratio, history of unhealthy eating behaviors), and information related to the intervention and its results (MBI strategy, control strategy, mode of delivery, FC intensity, FC frequency, intake amount).

Two reviewers (SA and EH) independently extracted data from each study, and a third reviewer (GA) resolved any discrepancies.

Risk of bias assessment

The Cochrane Risk of Bias tool for randomized trials (RoB2; version 2) [39] was used to assess the methodological quality of the included studies. RoB2 evaluates five domains of potential bias: bias arising from the randomization process, bias due to deviations from intended interventions, bias due to missing outcome data, bias in measurement of the outcome, and bias in selection of the reported result. Each domain is rated as “low risk,” “some concerns,” or “high risk,” and an overall risk-of-bias judgment is derived accordingly. Two reviewers (SA and ZM) independently applied the tool to each study, with a third reviewer (EH) adjudicating any discrepancies.

Data synthesis

Two reviewers (ZM & GA) performed the data synthesis and analysis. Data were analyzed using Comprehensive Meta-Analysis software version 3 (CMA v3), which allows reviewers to pool data in various formats. The primary summary statistic for continuous outcome data was the standardized mean difference using Hedges’ g, as this facilitates pooling outcomes measured heterogeneously across studies [40]. We utilized available formats provided Hedges’ g could be calculated. For multiple reports, analysis of covariance (ANCOVA) results were prioritized, as they yield the most precise and least biased intervention effect estimates per the Cochrane Handbook for Systematic Reviews of Interventions [41]. A random-effects model was employed for data pooling. To enhance readability, results favoring MBIs were denoted as positive, while those favoring control groups were denoted as negative. Hedges’ g values of 0.20, 0.50, and 0.80 are conventionally interpreted as small, medium, and large effects, respectively [42].

When data were presented as graphs, the WebPlotDigitizer web service was used to extract the outcome data [43]. In instances of missing data, we attempted to obtain the necessary information by contacting the corresponding authors via e-mail. Some studies included multiple groups within either the intervention or control arms. In these cases, we applied the Cochrane formula to calculate the overall effect size for the respective arm and entered the data into the CMA to compute the overall effect size. However, in accordance with Cochrane recommendations, subgroup analysis was performed by dividing the single-group arm in half to enable comparisons with each of the other groups in the second arm [40].

We assessed heterogeneity across studies using the Q-statistic and I² values [44]. I² values of 25%, 50%, and 75% were considered indicative of low, moderate, and high heterogeneity, respectively. To explore potential sources of heterogeneity, we conducted subgroup analyses stratified by MBI strategy, control group intervention type, intervention delivery mode, and study setting (laboratory vs. field), where feasible.

Publication bias was evaluated through visual inspection of the funnel plot, in conjunction with the Egger regression test and Begg’s adjusted rank correlation test [45–47]. The trim-and-fill method was applied to derive adjusted values when publication bias was present [48].

A sequential exclusion sensitivity analysis was performed for each outcome to assess the robustness of the results. In addition, a sensitivity analysis was conducted including only studies that identified food craving as a primary outcome, in order to assess whether the main findings were driven by studies where craving was a secondary or exploratory endpoint. To examine the effect of covariates, we conducted univariate meta-regression analyses. All regression analyses included an intercept model.

Statistical significance was defined as P < .05. The overall certainty of the evidence was assessed using the GRADEpro online tool [49].

Results

Study characteristics

A total of 24 studies reported across 23 scientific papers, comprising 1,920 participants, were included in the meta-analysis [18, 19, 21, 27]50– [68]. Seventeen of the included studies explored FC as a primary outcome [18, 19, 21, 27, 50, 51, 53]55– [59, 62, 63]65– [67]. Figure 1 presents the PRISMA flow diagram outlining the selection process [37], and Supplementary Material 2 lists the excluded full-text articles along with reasons for exclusion. While these 24 studies were initially eligible, some study arms were later excluded during data extraction. Specifically, one study each by Hinojosa-Aguayo et al. [57] and Schumacher et al. [62] were excluded due to insufficient reporting. Similarly, the first and third studies reported by Papies et al. [61] were excluded because their objectives did not align with the focus of this meta-analysis. Additionally, one control arm (waitlist) in the study by Alberts et al. [18] was excluded because participants in that group were allowed to consume the craving item, unlike the other arms.

Fig. 1.

The PRISMA flow diagram

The mean follow-up duration was 9.39 days (24 studies). Thirteen studies were completed in 1 day, while 7 studies had follow-ups of 1–2 weeks and 4 studies of 4–8 weeks. Among the participants, 81.62% were female (22 studies). The mean age was 21.31 ± 0.19 years (22 studies), and the mean BMI was 25.05 ± 0.17 (14 studies). Table 1 presents the basic characteristics of the included studies.

Table 1.

Included studies

First Author, year, country	Follow-Up	Cases	Mean Age	Mean BMI	Female (%)	MBI Strategy	Control	Results Summary
Alberts, 2010, Netherlands50	7 weeks	19	51.88	31.3	89.47	acceptance + dietary treatment + exercise	dietary treatment + exercise	Lower craving intensity in MBI (F = 8.02, p = .01)
Alberts, 2012, Netherlands51	8 weeks	26	48.5	32.7	100	mindful eating	none	Greater decreases in craving intensity for MBI (F = 9.49, p < .01)
Alberts, 2013, Netherlands18	1 day	65	21.41	NR	80	acceptance	suppression/ none	Increased craving intensity for MBI (F = 8.41, p < .001)
Baquedano, 2017, Chile52	1 day	50	28.65	23.05	66	decentering	immersion	Less craving for MBI (t = 5.5, p < .001)
Brenton‑Peters, 2023, New Zealand21	1 day	53	24.72	23.56	62.26	self-compassion	neutral writing	No effect was observed
Chapman, 2018, Australia53	4 weeks	151	30.49	26.55	100	acceptance	suppression	Better results for thought suppression (d = 0.37, p = .3)
Demos, 2019, USA54	1 day	34	42.3	31.8	79.41	acceptance	immersion/ distraction/ focus on consequences	Better results for long-term consequences, followed by distraction, acceptance, and immersion (p < .05)
Devonport, 2022, multi-national55	1 week	165	33.64	NR	80	mindful eating + diary	craving diary	The main effects of the group were not significant.
Fisher, 2016, UK56	1 day	40	30	25.4	100	mindfulness	neutral listening	No significant effect on craving intensity, lower intake in MBI
Hinojosa-Aguayo, 2022, Spain57	2 weeks	44	20.15	22.24	100	decentering	neutral listening	Lower craving intensity and intake for MBI (t=–1.82, p = .04)
Hooper, 2012, USA58	1 week	54	21.37	NR	59.26	decentering	suppression/ none	Less intake in the taste test (F = 7.98, p < .005) for MBI
Karekla, 2020, Cyprus59	1 week	65	19.65	NR	NR	decentering	restructuring/ none	Less intake & craving intensity for MBI & restructuring (F = 4.08, p < .05)
Lacaille, 2014, Canada27	2 weeks	196	19.93	22.17	89.29	awareness (aw), aw + acceptance (acc), aw + decentering (dec), aw + acc + dec	distraction	Decentering had the best effects and easier to understand (F = 2.93, p = .02)
May, 2010, UK60	1 day	48	21.83	NR	81.25	breath focus body scan + decentering	suppression/ imagery diversion/ none	No effect of intervention
May, 2010, UK60	1 day	49	20.92	NR	63.27	body scan	guided imagery/none	Less craving frequency for body scan (F = 3.28, p = .015)
Papies, 2015, Netherland61	1 day	75	NR	NR	NR	decentering	relaxed observation	MBI led to healthier food choices
Schumacher, 2017, Australia62	1 day	94	20.68	NR	100	decentering	guided imagery/mind wandering	Lower craving intensity for decentering (F = 4.41, p = .003)
Schumacher, 2018, Australia63	2 weeks	127	21.96	23.36	100	decentering	guided imagery/none	Lower craving frequency, intensity (d = 1.05, p < .001) & intake for MBI
Serpell, 2019, UK64	1 day	63	25.8	23.9	100	self-compassion	self-criticism	No effect of intervention
Shireen, 2024, Canada65	1 day	86	27.19	26.08	74.10	body scan	Neutral listening	Lower craving intensity for body scan (F = 8.12, p = .01)
Tapper, 2018, UK66	1 day	101	25.38	NR	71.29	decentering	guided imagery/ mind wandering	Less craving intensity across all conditions
Taylor, 2018, USA67	2 weeks	120	20.3	25.1	77	decentering	self-selected strategy	Less sweets intake for both conditions
Torske, 2024, Germany68	1 month	87	28	24.1	50	mindfulness	health training	Less craving intensity for MBI (F = 20.60, p < .001)
Wilson, 2021, UK19	1 day	108	26.7	NR	58.33	decentering	visualization/ mind wandering	Lower craving intensity for MBI only after 2-min audio (d = 0.4, p = .4)

MBI = Mindfulness-Based Intervention

Figure Titles

Risk of bias

The risk of bias was low for one study, 14 studies raised some concerns, and nine studies exhibited a high risk of bias. Figure 2 summarizes the risk-of-bias assessments. Refer to Additional File 3 for additional details.

Fig. 2.

Summary of quality assessment

Craving intensity measures

The FCQ includes two versions: one for state cravings (FCQ-S) and one for trait cravings (FCQ-T). The FCQ-S version comprises 15 items that respondents evaluate based on their current feelings using a five-point scale ranging from “strongly disagree” to “strongly agree.” Scores are calculated by summing responses for each subscale, yielding total scores from 15 to 75. Higher scores indicate more intense current food cravings. The FCQ-T consists of 39 items, with respondents indicating how often each statement applies to them on a six-point scale, with options ranging from one (never) to six (always). Higher scores indicate more intense food cravings. A reduced version of the FCQ-T (FCQ-T-r), developed by Meule et al. [10], comprises 15 items selected from the original 39-item version. Six studies used this scale to report food craving intensity [27, 59, 61, 65, 67, 68].

The G-FCQ is a modified adaptation of the FCQ designed to assess food cravings in a more general manner. It comprises two primary components: the G-FCQ-T (Trait) and the G-FCQ-S (State). The G-FCQ-T contains 21 items, scored using the same 1–5 scale as the original FCQ. The G-FCQ-S includes 15 items adapted from the FCQ-S, with references to “one or more specific foods” in the original FCQ-S replaced by “something tasty” to evaluate general food cravings rather than cravings for specific foods. Each item is scored on a scale from 1 (lowest) to 5 (highest). This scale has been utilized in six studies [18, 21, 50, 51, 53, 56].

For the CEQ-S, participants rate either current craving or the strongest craving experienced in the recent past (e.g. last day, last week, etc.), thereby establishing it as a measure of FC state. The CEQ-S comprises 10 items evaluating craving intensity, scored on a scale from 0 (not at all) to 10 (extremely). This scale has been used in two studies [19, 66]. In both studies craving assessments were anchored to “right now” (i.e., current state).

The ACQ-C assesses trait FC through ten items that gauge the degree to which individuals identify as someone who craves chocolate, using a 100 mm VAS. This scale was used in one study [27].

Craving intensity results

Main analysis

Data from twenty-three studies (1,681 cases) were pooled, yielding a Hedges’ g of 0.28 (95% CI: 0.07–0.48, P = .008), indicating an overall lower FC intensity in the MBI group (see Fig. 3). There was significant heterogeneity among the studies (I² = 77.54%, Tau² = 0.18, prediction interval = − 0.64 to 1.19). Sensitivity analysis confirmed the robustness of the results, as the effect remained significant under all conditions. Only the first study by May et al. [60] deviated the results by 20%; excluding this study resulted in a larger effect size of g = 0.33 (95% CI: 0.14–0.52, P = .001).

Fig. 3.

Craving intensity forest plot

Subgroup analyses-MBI strategies

To determine the effects of different MBI techniques, we conducted a subgroup analysis based on the strategies employed. Within the included studies, three primary categories were identified: those employing decentering, those utilizing acceptance, and those combining multiple techniques. The subgroup analysis revealed that decentering yielded the best results (g = 0.40, P = .01, n = 10), followed by mixed strategies (g = 0.11, P = .61, n = 6) and acceptance (g = 0.02, P = .93, n = 5). Supplementary Material 4 – Figure S1 presents the forest plot.

Subgroup analyses-control strategies

To perform subgroup analyses based on the type of control condition, we categorized interventions into three conceptual groups. The positive control group included interventions that actively encouraged participants to use techniques aimed at reducing cravings, such as cognitive suppression, distraction (e.g., focusing on non-food stimuli), guided imagery, or visualization strategies designed to downregulate desire. These techniques are often used in behavioral or cognitive training studies. The negative control group comprised interventions that could potentially intensify cravings, such as immersion in craving-related imagery, mind-wandering tasks, or self-critical reflection, which may increase reactivity to internal cues. Finally, the neutral control group included passive or informational conditions, such as waitlist, no-intervention controls, or general health education that did not involve active craving regulation strategies. This classification was applied to facilitate meaningful comparison across studies with diverse control conditions.

As expected, the results revealed the largest effect size for the negative group (g = 0.63, P < .001, n = 6), followed by the neutral group (g = 0.43, P < .001, n = 11) and the positive group (g = − 0.04, P = .62, n = 14). A forest plot illustrating these findings is provided in Additional File 4 (Figure S2).

Subgroup analyses-further analyses

Further subgroup analyses revealed that the effect of MBI was stronger on trait FC than on state FC (trait: g = 0.56, P = .008, n = 6; state: g = 0.17, P = .13, n = 19). The difference between subgroups was not statistically significant (P _between =0.11). Similarly, results showed larger effects for participants with eating behavior problems (g = 0.45, P = .003, n = 11) compared to those without (g = 0.12, P = .38, n = 12), though this difference was also nonsignificant (P _between = 0.11). No significant difference emerged between in-person and remotely delivered interventions (in-person: g = 0.26; remote: g = 0.28; P _between =0.95).

Regression analyses and bias

We also conducted meta-regression analyses based on mean BMI, age, and the proportion of female participants. None of these variables showed a significant effect based on the P-values. The R² analog was 0 for all of them (Additional File 4: Table S1).

Regarding publication bias, although the funnel plot exhibited some asymmetry, Egger’s test was not significant (P = .67) (Additional File 4: Figure S3). The overall certainty of the results was estimated to be low (Additional File 4: Table S2).

Craving frequency measures

Craving frequency was primarily assessed using custom researcher-developed tools, such as craving diaries or daily self-report questionnaires tailored to each study’s design [55, 58, 59, 63]. These tools typically required participants to record the number or frequency of cravings per day or per experimental session. Two studies [19, 66] used the frequency subscale of the Craving Experience Questionnaire (CEQ-F), which consists of three items assessing the frequency of craving-related experiences. Two studies [60, 62] assessed craving-related thoughts using thought-sampling procedures, calculating the proportion of food-related cognitions during experimental tasks.

Craving frequency

Main analysis

Data from nine studies (795 cases) were pooled, yielding a Hedges’ g of 0.11 (95% CI: − 0.28 to 0.49, P = .59), indicating no significant overall effect (Fig. 4). Again, Significant heterogeneity was observed among the studies (I² = 84.56, Tau² = 0.29, prediction interval = − 1.24 to 1.45). Sensitivity analyses confirmed that results remained non-significant across all scenarios.

Fig. 4.

Craving frequency forest plot

Subgroup analyses

Although field-based studies showed marginally higher effect sizes (g = 0.26, p = .38, n = 4) compared to laboratory-based studies (g = − 0.03, P = .92, n = 5), results remained non-significant in both subgroups, and the between-group difference was also non-significant (P _between =0.48).

Regression analyses and bias

We also conducted meta-regression analyses based on the mean age and proportion of female participants. The p-value did not indicate a significant association for either variable. The R² analog was 0 for age and 0.1 for the proportion of female participants (Additional File 4: Table S1).

Regarding publication bias, while the funnel plot exhibited slight asymmetry, Egger’s test yielded non-significant results (P = .63; see Additional File 4: Figure S4).

Intake

Main analysis

Data from eight studies (671 cases) were pooled, yielding a Hedges’ g of 0.46 (95% CI: − 0.02 to 0.93, P = .06), which indicates a nearly significant reduction in intake in the MBI group (Fig. 5). Significant heterogeneity was observed among the studies (I² = 88.28, Tau² = 0.40, prediction interval = − 1.21 to 2.12). Sensitivity analysis revealed that removing the studies by Chapman et al. [53] and Taylor et al. [67] produced significant results, with these two studies contributing to a deviation of more than 20% in the overall findings. Excluding these studies yielded a significant effect size of g = 0.75 (95% CI: 0.47 to 1.03, P < .001).

Fig. 5.

Intake forest plot

Subgroup analyses

Studies have investigated intake across two distinct phases. Some studies measured intake during periods when participants were instructed to refrain from eating (abstinence period), while others assessed intake after the active phase of the study had concluded, allowing participants unrestricted access to food (rebound intake). We conducted a subgroup analysis to compare effect sizes between intake during the experimental phase (abstinence period) and post-study rebound intake. The analysis revealed a larger effect size for rebound intake (g = 0.83, P = .006, n = 5) compared to intake during the experimental phase (g = 0.26, P = .36, n = 5). However, the between-group difference was not statistically significant (P _between =0.18).

Regression analyses and bias

We conducted meta-regression analyses using mean BMI, age, and the proportion of female participants as covariates. The p-values did not indicate significant effects for any of these variables. The R² analog was 0 for age and the proportion of female participants and 0.33 for BMI (Additional File 4: Table S1).

Regarding publication bias, while the funnel plot exhibited slight asymmetry and Egger’s test was statistically significant (P = .02), the trim-and-fill method did not impute additional studies, and the pooled results remained unchanged (Additional File 4: Figure S5).

No side effects were reported for MBIs.

Discussion

General

In this systematic review and meta-analysis, we analyzed 24 studies involving 1,920 participants to evaluate the overall effect of MBI on FC and provide preliminary evidence. While most studies were associated with either some concerns or a high risk of bias, it is important to note that these issues primarily stemmed from insufficient reporting (e.g., details about the randomization process) or the open-label design of the trials.

FC intensity/frequency

We observed an overall small-to-medium effect size (g = 0.28) for the influence of MBI on FC intensity. Among the examined MBI strategies, decentering demonstrated the highest effect; however, these subgroup comparisons are exploratory, with small sample sizes and overlapping confidence intervals. The results for acceptance and mixed strategies were not statistically significant. Compared to controls exhibiting a potential negative effect on FC or a neutral effect, MBI yielded a medium-to-large effect size and a near-medium effect size, respectively. No significant difference was observed between MBI and positive controls.

MBI yielded significantly better results than the control group for trait FC but not for state FC, suggesting that MBIs may be more effective for altering long-standing craving tendencies than for immediate craving episodes. Notably, in-person interventions did not outperform remote interventions. Furthermore, the differences observed between the MBI and the control groups in terms of FC frequency were not significant. Additionally, we did not detect any significant associations between effect size and mean BMI, age, or the proportion of female participants regarding FC intensity or frequency.

While the effect size was small-to-medium for FC intensity, the clinical and behavioral relevance of this finding warrants careful consideration. The clinical significance of effect sizes must be evaluated within the specific context of the intervention and outcome measures rather than relying solely on arbitrary benchmarks [69, 70]. Research in clinical trials suggests that investigators commonly target effect sizes between 0.2 and 0.4 across various medical domains, with an average target effect size of approximately 0.3 [71]. This indicates that our observed effect size aligns with realistic expectations for clinical interventions. Small individual-level effects can translate to substantial population health benefits when implemented broadly [72, 73]. Even a “small” effect size of 0.28 could result in meaningful public health impact if mindfulness-based interventions were scaled to large populations experiencing food cravings [73], particularly given the comparable efficacy of remote MBIs and in-person formats. This accessibility and feasibility enhance the practical significance of modest effect sizes by enabling widespread implementation.

The results indicate that acceptance alone may be insufficient to manage FC effectively. Decentering, in contrast, appears to enable individuals to manage FC independently, likely due to its unique mechanism of shifting perspective. Decentering cultivates a metacognitive stance, encouraging individuals to view cravings as transient mental events rather than immutable urges. This cognitive defusion creates psychological distance, reducing the urge’s impact on behavior—consistent with theories explaining craving reduction via decreased elaboration and automaticity [57].

Although MBI demonstrated no significant advantage over positive controls, certain studies, such as May et al. [60], suggest a potential reduction in rebound effects following MBI. Notably, the comparable efficacy of remotely delivered interventions relative to in-person formats underscores MBI’s suitability for remote implementation, this is particularly relevant for increasing accessibility, especially among populations with limited healthcare access or in post-pandemic contexts. Although we did not observe a significant reduction in craving frequency, this does not negate the utility of MBIs. Rather, MBIs may not necessarily prevent cravings but may alter the individual’s reaction to them—reducing their perceived intensity and impact on behavior.

Our findings align with recent neuroimaging research demonstrating that mindfulness interventions specifically targeting food-related stimuli can reduce midbrain reward anticipation responses [74]. This neurobiological evidence supports the mechanistic understanding that decentering may work by disrupting the automatic reward processing pathways that drive craving responses. Furthermore, experimental studies have shown that decentering reduces not only subjective craving experiences but also physiological markers such as salivary responses to food cues [75], suggesting that the benefits extend beyond self-reported measures.

The relatively stronger effect of decentering compared to acceptance strategies in our exploratory subgroup analysis is consistent with theoretical models proposing that mindfulness may reduce cravings through mechanisms such as working memory load and extinction processes [7]. While acceptance-based approaches focus on tolerating uncomfortable craving experiences, decentering actively engages cognitive resources by promoting perspective-taking and metacognitive awareness, potentially providing more robust craving management tools [76].

The comparable efficacy of remotely delivered interventions relative to in-person formats in our study is in alignment with recent research on digitally-assisted mindfulness interventions, which demonstrate that online delivery maintains therapeutic efficacy while expanding reach to underserved populations [77].

Our findings are consistent with the broader literature on MBIs for addictive behaviors. Demina et al. investigated the effect of MBIs on craving reduction in substance use disorders and behavioral addictions, involving 1228 participants, and reported an overall standard mean difference of 0.70 favoring MBIs [31]. Their review also highlighted considerable heterogeneity across studies. In a separate review, Grohmann et al. demonstrated that MBIs significantly reduce binge eating severity (g = 0.39) [33]. Additional reviews have similarly concluded that MBIs may effectively mitigate symptoms of emotional eating, external eating, and binge eating [78, 79].

Intake

Although it was not statistically significant, a medium effect size suggested that MBI was more suitable for managing intake compared to the control condition. Notably, this effect was more pronounced when considering intake following the abstinence period, as MBI demonstrated a significantly large effect size in this context. This finding further underscores the greater effectiveness of MBI in managing rebound intake compared to controls.

We did not observe a significant association between effect size and mean BMI, age, or the proportion of female participants.

Grider et al., in their systematic review on the effects of mindful eating and intuitive eating on dietary intake, reported limited evidence supporting the effectiveness of these approaches in this context and highlighted the need for further high-quality research in this field [80].

Study strengths

This review has several strengths. First, it is the first meta-analysis to systematically evaluate the effects of MBIs on both FC intensity and frequency, providing a more nuanced understanding of MBI effectiveness. Second, the review was conducted in accordance with PRISMA guidelines and used validated tools for bias and certainty assessment, including Cochrane RoB 2 and GRADE, enhancing methodological rigor. Third, the inclusion of subgroup and sensitivity analyses, including intervention types and delivery formats (e.g., remote vs. in-person), allowed for a deeper exploration of effect modifiers. Lastly, the categorization of diverse control groups enabled more meaningful interpretation of MBI outcomes relative to different comparators.

Limitations & future directions

The overall quality of the results was a concern in our review. A substantial portion of studies were rated as having “some concerns” or “high risk of bias”. This elevated risk of bias contributed to downgrading the certainty of evidence in our GRADE assessments. While the overall findings suggest that MBIs may reduce craving intensity and rebound intake, these conclusions should be considered preliminary and interpreted with caution pending higher-quality trials. Future studies should provide detailed reports on randomization, allocation procedures, and blinding, as well as specify how to access their registered protocols to prevent an underestimation of study quality.

Overall, moderate to high heterogeneity was observed across studies for all outcomes and subgroup analyses. This suggests a multitude of factors influencing study results within this field and highlights challenges in reproducing findings. Possible sources include differences in intervention duration, participant characteristics (e.g., BMI, eating behavior), delivery format (in-person vs. remote), and intervention fidelity. Outcome types (e.g., trait vs. state FC) and assessment methods may also contribute. Overall, these findings indicate that the effectiveness of MBIs for food cravings may depend on methodological and population-specific factors. To generate preliminary insights into the efficacy of MBIs for managing FCs, we intentionally included studies with diverse methodologies. This approach also allowed for a comparison of various MBI strategies, control interventions, and outcomes (e.g., FC trait vs. state). Despite conducting subgroup and sensitivity analyses, the heterogeneity remained largely unexplained, suggesting that additional unmeasured factors may be contributing to the variance. This underscores the complexity of synthesizing findings across diverse MBI studies. While our meta-analytic approach allows for quantification of pooled effects, future reviews may benefit from integrating a complementary narrative synthesis to capture contextual nuances that quantitative analyses alone may overlook.

Another limitation of the current study is the short mean follow-up period of 9 days. To better understand the sustainability of MBI effects, future research should include longer follow-up assessments (e.g. three months). Investigating long-term behavior change remains a key research gap in the evaluation of MBIs for food craving and eating behaviors.

We also recommend further exploration of rebound effects, particularly in comparisons between MBI and other active control conditions that may themselves exert positive effects.

Conclusion

MBIs may be a promising and safe strategy for managing FC intensity and subsequent intake, but not FC frequency. Notably, our analysis suggests that decentering might be the most effective MBI component for reducing FC intensity. However, further high-quality studies are needed to confirm this effect with greater certainty, as the certainty of the current results was rated as “low”. Additionally, researchers should make greater efforts to establish and implement standardized definitions and guidelines for each mindfulness-based approach.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

FC

Food Craving

MBI

Mindfulness Based Interventions

BMI

Body mass index

Author contributions

SA Conceptualization, Methodology, Data curation, Writing-Original Draft. ZM Conceptualization, Formal Analysis, Writing - Review & Editing, Supervision. EH Data curation, Writing-Original Draft, Visualization. GA Formal analysis, Writing - Review & Editing, Supervision.

Funding

This work did not receive any external funding.

Data availability

The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request.

Declarations

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.

Ethics approval

This work was approved by the ethics committee of Isfahan University of Medical Sciences: IR.MUI.PHANUT.REC.1403.098.

Consent to participate

Not applicable.