Gender differences in the influence of app-based mindfulness meditation on emotion regulation: a randomised controlled trial

Abstract

This study investigated whether gender moderates the effects of online mindfulness-based interventions (MBIs) on mental health outcomes in non-clinical populations. Specifically, we explored gender-related patterns in emotion regulation and cognitive flexibility. Data were obtained from a randomised controlled trial employing a crossover design in 2022 and 2023. Three hundred Japanese workers employed for at least 20 h weekly were randomly allocated to either an intervention or a waitlist group. The participants engaged in a four-week, app-based MBI involving daily guided practices, including brief breathing and body-scan meditations, with either loving-kindness or open-monitoring components. Perceived stress, anger, psychological flexibility, self-esteem, and emotion regulation were assessed at baseline and immediately after the four-week intervention or waitlist period. Interaction analyses between groups and gender revealed significant effects for cognitive flexibility (p = 0.023, partial η² = 0.018) and angry reaction (p = 0.007, partial η² = 0.025), although these findings did not survive correction for multiple testing. Exploratory simple slope analyses showed women showed greater cognitive flexibility (p = 0.034, Cohen’s d = 0.36) and reduced angry reactions (p < 0.001, Cohen’s d = −0.56), whereas men showed no significant changes in either variable (p = 0.202, Cohen’s d = -0.29 and p = 0.350, Cohen’s d = 0.22, respectively). This study tentatively suggests that brief online MBIs in preventive contexts may be associated with greater improvements in cognitive flexibility, with exploratory patterns indicating possible gender-related heterogeneity. However, these gender-related patterns should be interpreted cautiously. Considering individual differences, including gender, may help guide future preregistered personalisation research of digital psychological interventions.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-026-46317-z.

Introduction

Working adults are particularly vulnerable to stress and related mental health problems. Work-related stress has been shown to significantly impair mental health and impose substantial societal and economic costs¹. A growing body of evidence suggests that work-related stress can precipitate clinically diagnosable depression and anxiety even in previously healthy workers, and that supporting workers in coping with stress may help prevent the onset of clinically significant mental disorders². Consistent with this preventive perspective, a meta-analysis found that individuals who received preventive interventions were 19% less likely to develop a depressive disorder within one year compared to those who did not³.

Online interventions offer several advantages for preventive approaches, including broad accessibility, high acceptability, ease of integration into daily life, and flexibility to access the intervention at any time⁴. Mendelson and Eaton⁵ suggested that mindfulness-based interventions (MBIs) are promising preventive interventions. Mindfulness is commonly defined as the awareness that emerges through paying attention, on purpose, in the present moment, and non-judgmentally, to the unfolding of experience moment by moment⁶. However, preventive online interventions have minimal effects on most outcomes, such as depression, anxiety, well-being, and mindfulness^7,8. These small effect sizes may be attributed to a floor effect, whereby non-clinical populations are likely to have lower baseline scores on psychological symptoms, leading to less room for improvement than clinical populations⁷. Several meta-analyses have called for further research to confirm the benefits of preventive online MBIs and identify their moderators^3,7. Despite the importance of moderators, few randomised controlled trials (RCTs) have specifically investigated them in online MBIs⁹.

Gender is a promising moderator, with several studies having demonstrated the relationship between gender and mental health. Women are approximately twice as likely as men to experience major depressive disorders and most anxiety disorders^10,11, while men have a higher risk of antisocial behaviour and substance use than women^12,13. These gender differences suggest that women have a higher risk of internalising symptoms, while men have a higher risk of externalising symptoms. These differences are influenced by various biological and cultural factors¹⁴. Moreover, previous research has noted gender differences in the uptake and outcomes of MBIs. Carlson¹⁵ highlighted that MBIs are predominantly adopted by women, particularly those from Western and higher socioeconomic backgrounds, suggesting that gender may influence engagement and outcomes in MBIs. Although few empirical studies have investigated gender as a moderator in online MBIs in preventive contexts, several studies have suggested that MBIs have a greater effect on perceived stress and emotional outcomes in women than in men^9,16–19. Research highlighting the significant effects of MBIs in women has also reported improvements in mindfulness, self-compassion, and coping skills^17–19. Kang et al.¹⁷ noted that interventions that modify affect and coping strategies, including MBIs, may impact men and women differently. Previous studies have shown differences in coping strategies between genders: men tend to externalise their distress, whereas women tend to internalise their distress^20,21. These differences in coping strategies may mediate differences in psychological symptoms^21,22.

This study focuses on emotion regulation as a coping strategy. Emotion regulation refers to the processes that influence which emotions arise, when they occur, and how they are experienced or expressed²³. Gender differences have been observed in emotion regulation strategies and their relationship with psychopathology²⁴. It has been reported that women use a wider range of emotion regulation strategies more frequently than men, including rumination, reappraisal, problem-solving, acceptance, distraction, and seeking social support or religion, whereas men tend to use alcohol to cope more often than women²⁴. Additionally, increased rumination partially accounts for greater depression and anxiety rates in women than in men²⁴. It is also likely that men engage in a range of activities to regulate their emotions, but they do not recognise them as emotion regulation strategies²⁴.

Improving emotion regulation is a promising outcome of MBIs^25–27. A meta-analysis reported that MBIs have a moderate effect on emotion regulation ( = 0.20²⁸. MBIs are hypothesised to improve emotion regulation through enhanced awareness and decentring, which reduce maladaptive strategies such as rumination and promote adaptive strategies such as reappraisal²⁹. As women are more likely to engage in rumination and reappraisal, they may experience greater improvements in their emotion regulation and emotional outcomes from MBIs than men^17,24.

In addition to emotion regulation, this study included trait anger as an indicator of mental health outcome. Trait anger reflects heightened emotional reactivity and difficulties in regulating negative emotions^30,31, and has been linked to maladaptive emotion regulation strategies³². From a preventive perspective, anger represents an important manifestation of emotional dysregulation that may precede more severe internal and external problems. Accordingly, MBIs may attenuate anger by enhancing awareness and self-regulatory responses to emotional triggers³³.

Gender differences in emotion regulation may be particularly prevalent in flexible emotion-regulation choices³⁴. Emotion regulation flexibility comprises three components: sensitivity to context, availability of a diverse repertoire of regulatory strategies, and responsiveness to feedback³⁵. Goubet and Chrysikou³⁴ found that women were consistently more flexible in implementing emotion-regulation strategies. Moreover, Kashdan and Rottenberg³⁶ stated that cognitive flexibility, including emotion-regulation flexibility, spans various human abilities and that the absence of cognitive flexibility is related to many forms of psychopathology. Additionally, a cross-sectional study reported that mindfulness is closely associated with improvements in cognitive flexibility³⁷. The study suggested that MBIs could shift cognitive processes from automatic to controlled, interrupt or inhibit previously automatic responses, and enhance cognitive flexibility. This improvement provides cognitive space for identifying and modifying unhelpful or maladaptive cognitive evaluations³⁷. These mechanisms—enhanced awareness, decentring, and improved emotion regulation and cognitive flexibility—are therefore considered central to understanding individual differences, such as gender, in the effects of MBIs on mental health outcomes.

Few empirical studies have investigated the moderating effects of gender in preventive online MBIs, despite the link between gender and mental health problems. Most previous studies have focused on face-to-face or small-scale online MBIs, often targeting children and students^9,16,17,19. Additionally, prior studies that examined gender as a moderator often lacked a control group or included relatively small samples^9,18, limiting the generalisability of their findings.

This study examines gender as a potential moderator of the effects of MBIs on mental health problems in non-clinical and subclinical populations. Additionally, it focuses on gender differences in emotion regulation and cognitive flexibility, as well as the mechanisms underlying the effects of MBIs on mental health problems. This study proposes the following hypotheses (Fig. 1):

Fig. 1.

Conceptual model illustrating the hypotheses examined in this study. The figure depicts hypothesised relationships between a 4-week online mindfulness-based intervention, cognitive outcomes (emotion regulation and cognitive flexibility), and mental health outcomes (perceived stress and anger), with gender examined as a moderator of these relationships.

1. Women will show greater increases in adaptive and reductions in maladaptive emotion regulation strategies following online MBIs compared to men.

2. Women will demonstrate greater increases in cognitive flexibility following online MBIs compared to men.

3. Women will show greater reductions in perceived stress and anger following online MBIs compared to men.

Results

Baseline data

Tables 1 and 2 present the demographic characteristics, including means and standard deviations (SDs). The baseline values were summarised by group and gender. The sample that completed the baseline assessment (T₀) included 251 participants (mean age, 35.44 years [SD, 9.14 years]; women, n = 150 [60.0%]; regular employment, n = 168 [67.2%]). At baseline, the mean K6 score was 6.63 (SD = 4.79) in this non-clinical sample. Participants were employed in organisations of varying sizes, including small enterprises with fewer than 100 employees (41%), medium-sized companies with 100–299 employees (17%) or 300–999 employees (12%), and large organisations with 1,000 or more employees (30%). Regarding job roles, the majority of participants held non-managerial positions (76%), while a smaller proportion occupied supervisory or managerial roles (13%), with additional representation of specialised or professional positions (10%).

Table 1.

Demographic and baseline characteristics by group: means (standard deviations) and frequencies. Differences between groups at baseline were examined using appropriate parametric or non-parametric tests.

	Intervention group (n = 171)	Waitlist group (n = 80)	p-value	Effect size
Demographic variables
Age	34.49 (9.24)	37.45 (8.66)	0.028	-0.201 a
Gender, women/men	98 / 72	52 / 28	0.333	0.070 b
Employment status, regular / non-regular	116 / 54	52 / 28	0.716	0.032 b
Marital status, single/married	102 / 67	35 / 45	0.020	0.156 b
Educational attainment	15.79 (1.69)	15.79 (1.63)	0.691	0.005 a
Meditation experience, yes/no	48 / 122	15 / 64	0.160	0.099 b
Cognitive outcomes
CFS	46.63 (8.44)	47.00 (9.11)	0.763	-0.042
RSES	16.31 (6.86)	16.34 (6.85)	0.974	-0.004
CERQ adaptive	64.46 (8.88)	64.27 (11.06)	0.893	0.020
CERQ maladaptive	45.41 (8.11)	45.34 (9.19)	0.951	0.009
Mental health problems
PSS	28.63 (10.18)	28.50 (9.47)	0.926	0.013
Trait anger	20.01 (5.78)	19.78 (5.20)	0.756	0.041

CFS, cognitive flexibility scale; RSES, rosenberg self-esteem scale; CERQ, cognitive emotion regulation questionnaire; PSS, perceived stress scale.

Sample sizes reflect participants who completed baseline assessments. Effect sizes are reported as Cohen’s d unless otherwise indicated.

^aCliff’s delta. ^bCohen’s w.

Table 2.

Demographic and baseline characteristics by gender; means (standard deviations) and frequencies. Differences between genders at baseline were examined using appropriate parametric or non-parametric tests.

	Men (n = 100)	Women (n = 150)	p-value	Effect size
Practice days	23.67 (6.18)	22.45 (5.88)	0.035	0.195 a
Demographic variables
Age	35.89 (9.24)	35.03 (9.11)	0.423	0.104
Employment status, regular / non-regular	84 / 16	84 / 65	< 0.001	0.289 b
Marital status, single/married	55 / 44	82 / 67	1.000	0.010 b
Educational attainment	16.05 (1.94)	15.64 (1.45)	0.032	0.144 a
Meditation experience, yes/no	16 / 82	46 / 104	0.016	0.162 b
Cognitive outcomes
CFS	46.31 (8.23)	47.10 (8.94)	0.479	-0.091
RSES	16.45 (6.58)	16.26 (7.05)	0.821	0.029
CERQ Adaptive	63.95 (8.40)	64.71 (10.35)	0.537	-0.079
CERQ Maladaptive	45.04 (8.13)	45.67 (8.69)	0.566	-0.074
Mental health problems
PSS	28.16 (10.19)	28.85 (9.85)	0.606	-0.068
Trait anger	19.45 (5.56)	20.23 (5.62)	0.288	-0.139

CFS, cognitive flexibility scale; RSES, rosenberg self-esteem scale; CERQ, cognitive emotion regulation questionnaire; PSS, perceived stress scale.

Practice days refer to the number of days participants logged into the application. Effect sizes are reported as Cohen’s d unless otherwise indicated.

^aCliff’s delta. ^bCohen’s w.

In the between-group comparisons, participants in the intervention group were significantly younger (p = .028, Cliff’s delta = -0.201) and less likely to be married than those in the control group (p = .020, Cohen’s w = 0.156). In the gender-based comparisons, significant differences were found in employment status (p < .001, Cohen’s w = 0.289), educational attainment (p = .032, Cliff’s delta = 0.144), and previous experience with meditation (p = .016, Cliff’s delta = 0.162); men were more likely to be regularly employed and to have higher educational attainment, whereas women were more likely to have previous experience with meditation. No other variables differed significantly between the groups or between genders.

On average, participants practised for 22.99 days [SD, 6.01 days] during the intervention period. Men practised significantly more days than women (p = .035, Cliff’s delta = 0.195).

Group comparisons

Table 3 shows the means, SDs, and pre-post comparisons in the intervention group. Table 4 shows between-group comparisons using ANCOVA. After controlling for pre-intervention values, trait anger (F(1, 293) = 4.03, p = .046, p _adj = 0.322, partial η² = 0.014) and angry reaction (F(1, 298) = 5.35, p = .021, p _adj = 0.322, partial η² = 0.018) decreased significantly in the intervention group compared to the control group. Similarly, a marginal increase in positive reappraisal (F(1, 298) = 3.36, p = .068, p _adj = 0.322, partial η² = 0.011) and refocus on planning (F(1, 297) = 3.43, p = .065, p _adj = 0.322, partial η² = 0.011) was noted in the intervention group compared with in the control group. However, after FDR correction, these differences were no longer statistically significant (see Supplementary Tables S1 and S2).

Table 3.

Pre-post changes in the intervention group; means (standard deviations) and results of paired-samples tests.

	Intervention group (n = 229)		Pre-post comparison
	Pre	Post	Estimate	p-value	Effect size
Cognitive outcomes
CFS	46.71 (8.74)	47.44 (9.40)	2.01	0.046	0.08
RSESa	16.67 (6.94)	17.32 (6.64)	2.44	0.014	0.16
Self-competencea	8.34 (3.52)	8.73 (3.46)	2.96	0.003	0.20
Self-likinga	8.32 (3.62)	8.54 (3.45)	1.27	0.204	0.08
CERQ adaptive	64.00 (10.03)	64.38 (10.47)	0.68	0.495	0.04
CERQ maladaptive	45.17 (8.56)	44.76 (8.76)	-0.93	0.353	-0.04
Mental health problems
PSSa	28.12 (10.05)	26.13 (8.95)	-3.30	< 0.001	0.22
Trait angera	19.91 (5.95)	19.24 (5.55)	-2.31	0.021	0.16
Angry temperamenta	7.43 (2.97)	7.14 (2.89)	-2.20	0.028	0.15
Angry reactiona	9.42 (3.00)	9.11 (2.80)	-2.04	0.041	0.14

CFS, cognitive flexibility scale; RSES, rosenberg self-esteem scale; CERQ, cognitive emotion regulation questionnaire; PSS, perceived stress scale.

p values are from paired-samples t tests unless otherwise indicated. Effect sizes for paired-samples t tests are reported as Cohen’s d.

^aWilcoxon signed-rank tests were used where noted, and effect sizes are reported as r. Test statistics correspond to t-values for parametric tests and Z-values for Wilcoxon tests.

Table 4.

Comparison of changes in the intervention and control groups.

	p-value	Adjusted p-value	partial η2
Cognitive outcomes
CFS	0.363	0.717	0.003
RSES	0.985	0.985	0.000
Self-competence	0.499	0.717	0.002
Self-liking	0.515	0.717	0.001
CERQ adaptive	0.221	0.688	0.005
CERQ maladaptive	0.949	0.985	0.000
Positive reappraisal	0.068	0.322	0.011
Refocus on planning	0.065	0.322	0.011
Mental health problems
PSS	0.173	0.659	0.006
Trait anger	0.046	0.322	0.014
Angry temperament	0.447	0.717	0.002
Angry reaction	0.021	0.322	0.018

CFS, cognitive flexibility scale; RSES, rosenberg self-esteem scale; CERQ, cognitive emotion regulation questionnaire; PSS, perceived stress scale.

Adjusted p values were calculated using the false discovery rate (FDR) procedure.

To verify the robustness of these findings, a supplementary ANCOVA was conducted with additional covariates (age and marital status), as these variables differed significantly between groups at baseline. The results remained consistent.

Interaction effects between gender and intervention groups

To test Hypotheses 1–3, we examined whether there was a group (intervention, control) × gender (men, women) interaction effect in post-intervention values when controlling for pre-intervention values (Table S3).

Hypothesis 1

Gender differences in changes in emotion regulation strategies.

For Hypothesis 1, we evaluated group × gender interaction effects on emotion regulation strategies (Table S3). A marginal interaction was observed for refocus on planning (F(1, 295) = 2.82, p = .088, partial η² = 0.009); however, this effect did not remain statistically significant after FDR correction (p _adj = 0.333). The interaction did not remain marginal after controlling for the additional covariates and pre-intervention values (F(1, 285) = 1.97, p = .162, partial η² = 0.007). Exploratory simple slope analysis suggested that women in the intervention group showed significantly higher post-intervention values than those in the control group when controlling for pre-intervention values (t(295) = 2.58, p = .010, Cohen’s d = 0.42, Fig. 2), whereas no significant difference was observed among men (t(295) = -0.25, p = .803, Cohen’s d = -0.06). No other emotion regulation strategies showed significant or marginal group × gender interactions.

Fig. 2.

Interaction plot for refocus on planning.

Hypothesis 2

Gender differences in changes in cognitive flexibility.

For Hypothesis 2, we evaluated group × gender interaction effects on cognitive flexibility and related outcomes (Table S3). A significant interaction was detected for cognitive flexibility (F(1, 288) = 5.25, p = .023, partial η² = 0.018); however, this effect did not remain statistically significant after FDR correction (p _adj = 0.216). This interaction remained significant after controlling for the additional covariates (age, employment status, marital status, educational attainment, and meditation experience) and pre-intervention values (F(1, 277) = 4.04, p = .046, partial η² = 0.014). Exploratory simple slope analysis suggested that women in the intervention group exhibited significantly higher post-intervention values than those in the control group, controlling for pre-intervention values (t(288) = 2.13, p = .034, Cohen’s d = 0.36, Fig. 3), whereas no significant difference was observed among men (t(288) = -1.28, p = .202, Cohen’s d = -0.29).

Fig. 3.

Interaction plot for cognitive flexibility.

A marginal interaction was detected for self-esteem (F(1, 292) = 3.01, p = .084, partial η² = 0.010) and self-competence (F(1, 295) = 3.78, p = .053, partial η² = 0.013); however, these effects did not remain statistically significant after FDR correction (self-esteem, p _adj = 0.333; self-competence, p _adj = 0.333). Self-competence remained marginal after controlling for the additional covariates and pre-intervention values (F(1, 284) = 2.74, p = .099, partial η² = 0.010). Exploratory simple slope analysis suggested that women in the intervention group showed marginally higher post-intervention self-competence values than those in the control group, controlling for pre-intervention values (t(295) = 1.75, p = .082, Cohen’s d = 0.29, Fig. 4), whereas no significant difference was observed among men (t(295) = -1.14, p = .257, Cohen’s d = -0.26). No simple main effect of self-esteem was observed.

Fig. 4.

Interaction plot for self-competence.

Hypothesis 3

Gender differences in reductions in perceived stress and anger.

For Hypothesis 3, we evaluated group × gender interaction effects on perceived stress and anger (Table S3). A significant interaction was detected for angry reaction (F(1, 295) = 7.45, p = .007, partial η² = 0.025); however, this effect did not remain statistically significant after FDR correction (p _adj = 0.128). This interaction remained significant after controlling for the additional covariates and pre-intervention values (F(1, 284) = 6.99, p = .009, partial η² = 0.024). Exploratory simple slope analysis suggested that women in the intervention group exhibited significantly lower post-intervention values than those in the control group when controlling for pre-intervention values (t(295) = -3.43, p < .001, Cohen’s d = -0.56, Fig. 5), whereas no significant difference was observed among men (t(295) = 0.94, p = .350, Cohen’s d = 0.22).

Fig. 5.

Interaction plot for angry reaction.

Exploratory analyses: intervention type differences

Finally, exploratory analyses were conducted to examine potential differences between the two active intervention types (mindfulness meditation vs. self-compassion meditation). While detailed results are reported in the Supplementary Materials (Tables S4-S7), no significant main effects of intervention type were observed across outcomes. These analyses indicated significant intervention type × gender interactions for positive refocusing, rumination and self-blame (positive refocusing, F(1, 216) = 4.28, p = .040, partial η² = 0.019; rumination, F(1, 220) = 4.11, p = .044, partial η² = 0.018; self-blame, F(1, 217) = 7.84, p = .006, partial η² = 0.035). After applying the FDR correction, these differences were no longer statistically significant (positive refocusing, p _adj = 0.166; rumination, p _adj = 0.166; self-blame, p _adj = 0.053).

Exploratory simple slope analyses (see Supplementary Figures S1-S3) indicated that, among men, self-compassion meditation was associated with significantly greater reductions in self-blame compared with mindfulness meditation (t(217) = -2.69, p = .008, Cohen’s d = -0.58), whereas no significant difference was observed among women (t(217) = 1.11, p = .270, Cohen’s d = 0.19). For positive refocusing and rumination, simple slope analyses did not reveal clear differences between the two intervention types within each gender.

Discussion

The present study examined the effects of a brief online MBI on mental health problems and cognitive outcomes in a non-clinical working population, with a particular focus on gender differences. Overall, after controlling for multiple testing using FDR correction, we found no robust between-group differences; any observed effects were small. In uncorrected analyses, these findings suggest that participants in the intervention group exhibited decreased trait anger and angry reaction compared to those in the waitlist control group. Participants in the intervention group also showed marginally increased positive reappraisal and refocus on planning compared to those in the waitlist control group. However, these effects did not remain statistically significant after applying the FDR correction, and the corresponding effect sizes were small, suggesting modest average effects of the intervention. The results are consistent with previous findings from preventive online interventions for non-clinical populations⁷. This pattern may be partly explained by a floor effect, as participants exhibited low baseline symptom levels, leaving limited room for improvement.

Similarly, none of the group × gender interaction effects survived FDR correction, and the corresponding interaction effect sizes were small or negligible (partial η² < 0.02). Uncorrected interaction analyses suggested group × gender patterns for angry reaction and cognitive flexibility. Marginal patterns were also observed for self-competence and refocus on planning. Exploratory simple slope analyses suggested that the reduction in angry reaction among women was of medium magnitude (Cohen’s d = -0.56), whereas the remaining effects were small. Considering that the interaction analyses were not preregistered, these findings should be interpreted cautiously and considered as exploratory and hypothesis-generating patterns, rather than as confirmatory evidence of gender-specific intervention effects. Accordingly, the main contribution of this study lies in its methodological rigor and exploratory pattern detection rather than in demonstrating robust gender differences.

Before discussing potential mechanisms, interpreting the present findings requires distinguishing between state- and trait-like constructs. Measures such as perceived stress are considered relatively state-sensitive and may fluctuate in response to short-term contextual factors. Therefore, observed reductions in perceived stress should be interpreted as short-term improvements rather than enduring changes in stress vulnerability. In contrast, trait-related measures, such as trait anger, cognitive flexibility, and self-esteem, are commonly conceptualised as relatively stable individual differences. However, some prior studies have shown that these indicators can be responsive to MBIs^33,38,39. Accordingly, the observed pre–post changes should not be interpreted as evidence of enduring trait modification, but rather as short-term shifts in trait-like tendencies over the four-week intervention period.

The current study tentatively suggests exploratory group × gender patterns in cognitive flexibility, refocus on planning, and self-competence. However, because these interaction effects were small and did not survive FDR correction, they should not be interpreted as confirmatory evidence of gender-specific intervention effects. In this interpretive context, these exploratory patterns are conceptually consistent with a systematic review of MBIs that reported preliminary evidence for cognitive flexibility as a mechanism underlying the effects of MBIs on clinical and non-clinical cognitive outcomes³⁸. Cognitive flexibility encompasses three key aspects: (a) awareness that options and alternatives are available in any given situation, (b) willingness to be flexible and adapt to the situation, and (c) self-efficacy in being flexible³⁵. Accordingly, cognitive flexibility may overlap with elements such as refocus on planning and self-competence.

Refocus on planning—a cognitive-focused dimension of problem-focused coping—refers to thinking about the steps to take and ways to handle a negative event⁴⁰. Previous research found a positive correlation between cognitive flexibility and refocus on planning (r = .528)⁴¹. Self-competence is the sense that one is confident, capable, and efficacious⁴², and is closely associated with self-efficacy^43,44. This may be relevant to cognitive flexibility because even if people are aware of choices of behaviour in a given situation and are willing to be flexible, they still need to believe that they are self-efficacious in bringing about the desired behaviour⁴⁵.

The present findings tentatively suggest exploratory patterns that may indicate reductions in angry reactions following a brief online MBI. However, because the interaction effects were small and did not survive FDR correction, these findings should also be interpreted cautiously as hypothesis-generating rather than confirmatory evidence of gender-specific effects. These exploratory patterns are conceptually consistent with previous studies indicating that MBIs may be effective in mitigating angry reactions^33,46. Angry reaction refers to the frequency at which angry feelings are experienced in situations that involve frustration and/or negative evaluations⁴⁷. Anger has been hypothesised to perceptually and conceptually narrow the cognitive scope, making it challenging for angry individuals to see alternative ways of addressing the problem⁴⁸. One review reported that anger and related problems have been linked to psychological inflexibility and that enhancing cognitive flexibility may help reduce anger⁴⁹. A hypothesis is that online MBIs may lead to improved cognitive flexibility and control over anger response. The possibility should be evaluated directly in preregistered, adequately powered studies.

Additionally, exploratory analyses comparing mindfulness meditation and self-compassion meditation did not reveal significant main effects of intervention type across outcomes, consistent with the analytic decision to combine the two practices for the primary analyses. However, exploratory intervention type × gender interactions were observed for selected cognitive emotion regulation strategies. In particular, simple slope analyses suggested that, among men, self-compassion meditation was associated with greater reductions in self-blame than mindfulness meditation. These findings should be interpreted with caution, as the analyses were exploratory, and the observed effects were limited to specific outcomes. Collectively, these findings may tentatively suggest that different components of online MBIs may differentially engage emotion regulation processes across individuals, a possibility that warrants preregistered, adequately powered testing to draw conclusions about tailoring.

Before discussing possible mechanisms, it is important to acknowledge that evidence for gender differences in MBI outcomes remains inconclusive. Some meta-analyses have failed to find significant moderating effects^50,51, although methodological and statistical limitations have been noted⁵². In addition, some RCTs have reported greater improvements among men in specific domains, such as emotion suppression following MBSR⁵³, while others have found no significant differences in effectiveness between men and women in short-term online MBIs⁵⁴.

Within this context, these results tentatively suggest the possibility of gender-related heterogeneity in response to brief online MBIs. However, given the exploratory nature of the analyses, the absence of robustness after multiple-testing correction, and small effect sizes, this should be treated as hypothesis-generating rather than evidence that MBIs are more effective for women than for men. Speculatively, one possible explanation is that women implement emotion-regulation strategies more frequently and flexibly³⁴, and online MBIs may encourage this tendency. Smail-Crevier et al.⁵⁵ reported that women were more likely to use the internet for mental health-related information and demonstrated greater acceptance of e-mental health programmes for work stress than men.

Another possible explanation is that men may not accurately respond to self-reported emotion-regulation questionnaires because much of their emotion regulation may operate automatically and unconsciously²⁴. In the present study, perceived stress decreased from pre- to post-intervention in both men and women, with no interaction effect by gender. Nolen-Hoeksema²⁴ pointed out that men use various activities to regulate their emotions but do not label them as emotion-regulation strategies. The inability of men to respond accurately to questions on emotion regulation did not necessarily mean that the intervention was ineffective. Therefore, appropriate questions are required to reveal some of men’s typical emotion-regulation strategies²⁴.

A key strength of the present study was the scientific rigour based on its RCT methodology, large sample size, and high implementation rate. While several studies have examined gender as a moderator in MBIs, most have lacked a control group^9,18. Kang et al.¹⁷ conducted an RCT, but exploratory follow-up pairwise comparisons were required to investigate the potential influence of gender because the study found no interaction effects. Additionally, the implementation rate (23.17 out of 28 days, 82.8%) was considerably higher than that reported in previous studies. For example, dropout rates from studies using applications for depressive symptoms have been reported to be nearly 50% when accounting for biases⁵⁶. This study’s large sample size and high implementation rate increased the precision of effect estimates and enabled exploratory tests of interaction terms; nevertheless, the observed interaction effects were small and did not remain significant after FDR correction.

This study had several limitations. First, the target population comprised mainly Japanese workers who were relatively familiar with computers, and the study population was relatively healthy because of its preventive context. Therefore, the generalisability of this study’s results to various groups, such as the unemployed and patients with mental illnesses, remains unclear.

Second, the current study showed only small effect sizes, and these effects did not remain statistically significant after applying the FDR correction. Possible reasons for this are the short intervention duration of four weeks and the ceiling effect of targeting non-clinical workers. A four-week MBI has been found to have an effect size equivalent to an eight-week intervention and shows better results than the control group⁵⁷. However, because this study adopted an online intervention, four weeks might not have been sufficient to show a large effect. Furthermore, the absence of an active control condition limits causal inference regarding mindfulness-specific effects, leaving open whether the observed gender-related patterns reflect mechanisms specific to mindfulness practice or non-specific intervention components (e.g., expectancy, attention, or engagement with a digital tool).

Third, this study employed a crossover RCT design to enhance the statistical power of the results. As psychological intervention studies generally cannot eliminate the carryover effects of interventions through a washout period, the current study employed a one-way crossover design similar to that seen in most psychological intervention studies. We confirmed the parallel design analyses in this study and observed uncorrected group × gender interaction patterns in cognitive flexibility (F(1, 218) = 4.30, p = .039), angry reaction (F(1, 225) = 4.55, p = .034), self-competence (F(1, 224) = 3.74, p = .055), and self-esteem (F(1, 221) = 2.96, p = .086). The other indicators also exhibited similar trends. However, these sensitivity analyses were exploratory and uncorrected for multiple testing, and the corresponding interaction effects were small; therefore, they should not be interpreted as confirmatory evidence of gender moderation.

Fourth, only two out of 19 uncorrected group × gender interaction effects reached statistical significance. These observed interaction effects were small and did not remain significant after FDR correction. This study was a secondary analysis of data from a preregistered RCT, and the present hypotheses regarding gender moderation were not preregistered. Therefore, the possibility of Type I error inflation or data-driven findings cannot be ruled out. Importantly, the present study does not provide confirmatory evidence for gender moderation; rather, these results should be interpreted as exploratory, hypothesis-generating signals that require preregistered replication in adequately powered independent samples. Future studies with preregistered hypotheses and independent replication samples are warranted to test these moderation hypotheses and evaluate the robustness of any observed patterns.

Gender-related patterns in responses to online MBIs might also be interpreted through the extreme male brain theory of autism⁵⁸, which posits that men are more inclined to ‘systemise’, whereas women are more predisposed to ‘empathise’. In this context, it may be beneficial to introduce modules with tailored elements to enhance the effectiveness of psychological interventions for men in future hypothesis-driven research. Several studies have suggested that MBIs are particularly effective for women with pronounced internalising symptoms, whereas behavioural activation tends to be more effective for men^59,60. Moreover, men may exhibit a greater preference for AI-driven apps⁶¹. Future research could test whether adding behavioural activation and AI to online interventions improves efficacy for certain subgroups, including men.

The present study does not provide confirmatory evidence that gender moderates MBI effectiveness. Rather, it identifies exploratory, hypothesis-generating patterns of possible gender-related heterogeneity that warrant preregistered replication in adequately powered samples. As Baron-Cohen⁵⁸ stated, not all men have a male brain type, and not all women have a female brain type. Although gender differences exist, their explanatory power for diversity is relatively limited. As such, this work has tentative implications, suggesting that it may be important to consider individuality, including gender, when designing psychological interventions to maximise their effectiveness—a possibility that warrants further testing in future hypothesis-driven personalisation research. In the future, various moderators and intervention methods should be explored to tailor preventive interventions based on individual differences.

Methods

Study design

An RCT with a crossover design was conducted in Japan, featuring three arms: app-based brief-guided mindfulness meditation, self-compassion meditation, and a waitlist control. For the primary analyses, the two meditations were combined into a single arm to evaluate the effects of meditation. This study was approved by the Ethics Committee of the University of Tokyo (22–326). This study was registered with the University Hospital Medical Information Network Clinical Trials Registry on 10/11/2022 (Registration No. UMIN000049466). The study protocol has been published⁶². The manuscript reports exploratory secondary analyses that were not pre-specified in the protocol. This study was conducted in accordance with the Declaration of Helsinki.

Participants

The required sample size was estimated a priori using G*Power 3.1 software, based on the expected group-by-time interaction effect. Following previous research⁶³, a small effect size (Cohen’s f = 0.15) and a correlation among repeated measures of 0.5 were assumed. With a power of 0.80 and α = 0.05, 37 participants were required for each group. Considering an expected attrition rate of approximately 40% based on similar intervention studies^64,65, a total of 200 participants (about 67 per group) were planned to be recruited. Detailed procedures are provided in the study protocol⁵⁴. A total of 375 workers were recruited from various websites in Japan via open calls during two recruitment waves: November–December 2022 and June 2023, due to insufficient enrollment in the first phase. All post-test assessments were completed by September 2023. The inclusion criteria were employees who (1) worked more than 20 h per week, either full- or part-time; (2) were between the ages of 18 and 54; and (3) owned an iPhone (the app used in this study could only be used on iOS). The exclusion criteria were those who (1) were on a leave of absence, (2) were business owners or students, (3) were currently diagnosed with a mental disorder, and (4) scored ≥ 13 on the Kessler Psychological Distress Scale-6 (K6;^66,67. We targeted non-clinical participants and those with subthreshold depressive symptoms to focus on prevention; individuals with clinically depressed symptoms were excluded.

Procedure

The participants completed a web form with a screening survey (K6). We screened 375 workers according to the inclusion criteria, resulting in 300 participants. Informed consent was obtained from all individual participants included in the study. After providing informed consent via a web form, the participants were randomly allocated to the mindfulness meditation group (n = 101), self-compassion meditation group (n = 100), or the waitlist group (n = 99). Randomisation was conducted by the first author using a computer-generated blocked randomisation scheme with a block size of 15. The two meditations were combined into an intervention group (n = 201). This study was an open-label RCT, as participants engaged in the meditation independently and blinding the allocation was not feasible.

At the baseline assessment (T₀), 251 participants completed the demographic variables and all outcome questionnaires. After a four-week intervention or waitlist period, 234 participants completed the same outcome measures at the post-test assessment (T₁). Later, participants in the waitlist group practised the intervention for four weeks. After the intervention period, 71 participants in the waitlist group completed the same outcome measures at the post-test assessment (T₂). No intermediate assessments were conducted. For the primary analyses, changes during the intervention periods were defined as T₀–T₁ for the initial intervention group and T₁–T₂ for the waitlist group after crossover. These intervention-period changes were combined and compared with changes observed in the waitlist group during the non-intervention period (T₀–T₁). The number of participants at each stage is shown in Fig. 6.

Fig. 6.

Intervention flowchart.

Participants in the intervention groups practised guided mindfulness-based meditation using a smartphone application once daily for four weeks (for full content, see the study protocol)⁶². The guided meditations contained in each module were as follows: a breathing meditation (7 min) during Week 1, a short body scan (6 min) during Week 2, a loving-kindness meditation (12 min; self-compassion meditation group) or a breath, sound, and body meditation (12 min; mindfulness meditation group) during Weeks 3 and 4. Mindfulness meditation consists of focused attention (FA) meditation and open monitoring (OM) meditation⁶⁸. The first two practices correspond to FA meditation, while the breath, sound, and body meditation corresponds to OM meditation. Compassion is considered to be cultivated through MBI, and loving-kindness meditation explicitly trains this capacity; it may be optionally incorporated into MBIs^69,70. For the primary analyses, the two meditation types were combined into a single intervention arm to examine the overall effectiveness of a brief, app-based MBI in a preventive context. Potential differences between mindfulness meditation and self-compassion meditation were examined in exploratory analyses, which are reported in the Supplementary Materials. The application recorded the number of logins.

The waitlist control group received no intervention during the waitlist period. No restrictions were placed on participants regarding other treatments or care. No concomitant care was reported.

Measurements

Demographic variables

Participants’ age, gender, employment status, marital status, educational attainment, and previous meditation experience, including mindfulness, were recorded using a questionnaire. All variables collected simultaneously are reported in the study protocol⁶².

Cognitive flexibility

The Cognitive Flexibility Scale (CFS) was used to measure cognitive flexibility, which assesses: (a) awareness that options and alternatives are available in any given situation; (b) willingness to be flexible and adapt to the situation; and (c) self-efficacy in being flexible⁷¹. The CFS is commonly used to assess relatively stable individual differences⁷². The CFS is a 12-item self-reported questionnaire. Total scores range from 12 to 72, with higher scores indicating greater cognitive flexibility. The reliability and validity of the Japanese version have been previously confirmed⁷³. The internal consistency reliability of the CFS in this sample was α = 0.86.

Self-esteem

Self-esteem was measured using the Rosenberg Self-Esteem Scale (RSES)⁷⁴, comprising 10 items. Mimura and Griffiths⁷⁵ confirmed the reliability of the Japanese version. Several studies have suggested that this scale has two factors: self-competence and self-liking⁴⁴. Self-competence is the sense that one is confident, capable, and efficacious, and self-liking is the sense that one is a good person, socially relevant, and contributes to group harmony⁴². Self-esteem is generally conceptualised as a relatively stable, trait-like characteristic across the life span⁷⁶. In this sample, the internal consistency reliabilities for global self-esteem, self-competence, and self-liking were α = 0.92, 0.86, and 0.85, respectively.

Emotion regulation

The Cognitive Emotion Regulation Questionnaire (CERQ) is a 36-item scale measuring cognitive-emotion-regulation strategies to cope with the experience of threatening or stressful events or situations⁷⁷. It evaluates nine facets of cognitive-emotion-regulation strategies: positive reappraisal, putting into perspective, acceptance, positive refocusing, refocus on planning, rumination or focusing on thoughts, self-blame, other blame, and catastrophising. The first five strategies are considered adaptive, and the latter four are considered maladaptive⁴⁰. The CERQ is designed to capture relatively habitual patterns of cognitive emotion regulation and has demonstrated both temporal stability and transitions over time⁷⁸. A validated Japanese version of this questionnaire was used in the current study⁷⁹.

Perceived stress

The Perceived Stress Scale-14 (PSS-14) is a self-reported questionnaire that measures the degree of stress an individual has perceived in life within the past month^80,81. Sumi^82,83 translated the Japanese version of the PSS-14 and confirmed its reliability and validity. In this sample, the internal consistency reliability was α = 0.89.

Trait anger

The State-Trait Anger Expression Inventory-2 (STAXI-2) is a self-reported questionnaire that measures the characteristics of coping with anger arousal⁴⁷. In this study, the participants rated 10 items of the trait-anger domain to indicate how often they experienced anger over time. The trait-anger domain comprised two subscales: angry reaction and angry temperament. The angry-temperament subscale suggests the tendency to feel anger regardless of the situation, while the angry-reaction subscale indicates the frequency of anger experienced during frustration. A validated Japanese version of the questionnaire was used in this study^84,85. The internal consistency reliabilities of trait anger, angry temperament, and angry reaction were α = 0.85, 0.87, and 0.75, respectively.

Data analysis

Demographic variables and baseline psychological indices were descriptively analysed. Mean and standard deviation (SD) were calculated for each group and gender. Differences between groups and between genders at baseline were verified using Welch t-tests for normally distributed variables and Wilcoxon rank-sum tests for non-normally distributed variables. Effect sizes for Welch t-tests and Wilcoxon rank-sum tests were reported as Cohen’s d and Cliff’s delta. Normality of the data was assessed using the Kolmogorov–Smirnov test. For categorical variables, baseline differences were analysed using the chi-squared test. Effect sizes for chi-squared tests were reported as Cohen’s w.

Given the waitlist crossover design, intervention effects were examined by defining intervention-period changes as T₀–T₁ for participants initially assigned to the intervention group and T₁–T₂ for participants in the waitlist group after crossover. Changes during the non-intervention period were defined as T₀–T₁ in the waitlist group. These periods were treated as parallel conditions in the primary analyses.

Changes before and after the intervention or waitlist periods were assessed using paired t-tests for variables with normally distributed differences and Wilcoxon signed-rank tests for variables with non-normally distributed differences. Effect sizes for paired t-tests were reported as Cohen’s d. For the Wilcoxon signed-rank test, r was calculated as Z / √N. For Cohen’s d, values of approximately 0.2, 0.5, and 0.8 were interpreted as small, medium, and large effects, respectively. For effect size r, values of approximately 0.1, 0.3, and 0.5 were used as conventional benchmarks⁶⁸.

Analysis of covariance (ANCOVA), after controlling for baseline values, was used to compare changes across groups. The model regressed the outcomes on group assignment, considering pre-intervention values as covariates. Multiple comparisons were controlled using the false discovery rate (FDR; Benjamini–Hochberg method), and the FDR-adjusted p-values were explicitly stated. Effect sizes for ANCOVA were reported as partial η², with cut-off values of 0.01, 0.06, and 0.14 for small, medium, and large effects, respectively⁸⁶. As a robustness check of the results, we additionally controlled for baseline variables that were significantly different between the groups.

The effects of group and the interaction between group and gender were examined using ANCOVA after controlling for pre-intervention values. Effect sizes for the interaction were reported as partial η². To verify the robustness of the results, we additionally controlled for baseline variables that were significantly different between the groups or between genders. To further interpret the significant interaction, simple effects were examined using adjusted means. Effect sizes for independent t-tests were reported as Cohen’s d.

Exploratory analyses were additionally conducted to examine potential differences between the two active intervention types (mindfulness meditation vs. self-compassion meditation). These analyses included intervention type and its interaction with gender as between-subject factors, while controlling for baseline values using ANCOVA. Given the exploratory nature of these analyses, results are reported in the Supplementary Materials.

Missing data were handled using listwise deletion. All tests were two-tailed with an alpha level of 0.05. Data were analysed using R, version 4.5.0⁸⁷, and RStudio, version 2024.12.1.563⁸⁸.

No interim analyses were conducted, and no stopping guidelines were established, as the trial was considered minimal risk. Participants were allowed to withdraw without disclosing the reason and were able to contact the research staff via email or web-based forms, if necessary. No harms or adverse events were systematically assessed, and no participants spontaneously reported any adverse events during the study period.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Koichiro Adachi (Conceptualisation, Data curation, Formal analysis, Investigation, Methodology, Project administration, Software, Visualisation, Writing—original draft), Takumu Kurosawa (Conceptualisation, Investigation, Methodology, Project administration, Writing—review & editing), Ryu Takizawa (Conceptualisation, Funding acquisition, Methodology, Project administration, Supervision, Writing—review & editing).

Data availability

The data that support the findings of this study are not publicly available due to privacy or ethical restrictions. The data are available on request from the corresponding author.

Declarations

Competing interests

The authors declare no competing interests.