Mindful attention as a mechanism underlying non-suicidal self-injury: evidence from a longitudinal intervention trial Authors: Jian Li, Fengmin Ni, Yangyang Tao, Ya Xie, Enze Tang, Ziyan Wu, Minlu Liang, Pei Zhang, Marcus Rodriguez, Jan Glasenapp, Chunyan Yin, Jing Long, Yuan Liu, Yan Sun, Xianliang Wu, Guangya Liu, Xiaomei Jiang, Xianglan Wang, Hua Lv, Zhuowei Zhang, Suhong Wang, Zachary Rosenthal, Chun Wang Published: 2026 ABSTRACT BackgroundNon-suicidal self-injury (NSSI) represents a significant public health issue, yet its underlying mechanisms are not fully understood. Theoretical models posit attentional deficits as key potential factors, but most supporting evidence has been cross-sectional. This study aimed to provide longitudinal evidence for the role of mindful attention, a mindfulness-related attentional process, in NSSI by leveraging data from a randomized controlled trial.MethodsA total of 196 individuals with NSSI were randomized to 13 weeks of dialectical behavior therapy skills training (DBT-ST; n = 97) or social support group therapy (SSGT; n = 99). They completed baseline and post-treatment assessments of mindful attention (Mindful Attention Awareness Scale), NSSI frequency (Ottawa Self-Injury Inventory). Given its close link with mindful attention and NSSI, emotion dysregulation (Difficulties in Emotion Regulation Scale) was also assessed.ResultsResults indicated that the DBT-ST group showed significantly greater improvements in NSSI behaviors, mindful attention, and emotion dysregulation compared to the SSGT group. Mediation analyses revealed that improvements in mindful attention did not directly mediate NSSI reduction. Instead, mindful attention contributed indirectly through improvements in emotion regulation, which in turn predicted reductions in NSSI frequency. Cross-lagged panel analyses further demonstrated that improvements in mindful attention prospectively associated with subsequent improvements in emotion regulation, underscoring mindful attention as an upstream process in the reduction of NSSI.ConclusionThese findings indicated that attention deficits played a role in NSSI and suggested a potential sequential process in which DBT-ST-related improvements in attention may support subsequent gains in emotion regulation, which are associated with reductions in NSSI.Supplementary InformationThe online version contains supplementary material available at https://doi.org/10.1186/s12888-026-08316-2. Introduction Protecting oneself from harm is a fundamental biological instinct; however, humans sometimes engage in behaviors that contradict this instinct. One of the most puzzling behaviors is non-suicidal self-injury (NSSI), defined as the deliberate destruction of one’s own body tissues without suicidal intent [1]. NSSI is most prevalent during adolescence and commonly co-occurs with various psychiatric disorders [2, 3], showing a prevalence of 14.3% among adolescents aged 10–19 years with psychiatric comorbidities [4]. Due to its high prevalence and its association with elevated suicide risk [5, 6], NSSI has become a major focus of clinical research. Yet, despite extensive efforts, the mechanisms underlying NSSI remain incompletely understood. Attentional deficits may play a critical role in NSSI [7, 8]. Achieving emotion relief is one of the important functions of NSSI [1]. Several theoretical models posit that attention redirection underlies this function: the self-inflicted physical pain redirects attention away from negative emotions, thereby facilitating emotional relief [9]. For example, according to the emotion cascade model, negative emotions tend to trigger rumination and intensify emotional responses, creating a vicious cycle. NSSI individuals disrupt this cycle by shifting their attention away through the pain resulted from self-harm, to mitigate emotional distress [10]. In contrast, healthy individuals regulate negative affect through adaptive attentional strategies, such as voluntarily shifting attention away from distressing cues or focusing on positive aspects of an experience [11]. Thus, the reliance on highly salient external stimuli (i.e., pain) to redirect attention may suggest impairments in attentional processes. Consistent with this view, emerging evidence shows that individuals with NSSI demonstrate abnormalities in attentional processes [8, 12]. For instance, researchers found that adolescents with an NSSI history were impaired at inhibiting attention to angry adult faces, compared to those without NSSI history [13]. Moreover, attentional impairments have been linked to NSSI frequency [14], underscoring their potential role in the disorder’s pathology. However, most existing evidence comes from cross-sectional studies. It remains unclear whether changes in attentional processes are involved in treatment-related reductions in NSSI. To address this question, the present study focused specifically on mindful attention, which refers to the tendency to maintain awareness of present-moment experiences. This focus is particularly relevant to NSSI, as becoming absorbed in automatic, ruminative, or maladaptive internal processes are tightly linked to NSSI [15]. Specifically, we leveraged the data acquired from an intervention study involving dialectical behavior therapy skills training (DBT-ST) [16] to achieve this goal. DBT-ST has been proven to be an important intervention in reducing NSSI, as demonstrated in multiple randomized clinical trials [17, 18]. Importantly, mindfulness is a central component of DBT-ST and is designed to cultivate present-moment awareness, attentional stability, and the ability to observe internal experiences without immediately acting on them [19]. Therefore, DBT-ST provides a useful context in which to examine whether mindful attention changes during treatment and whether such changes are associated with reductions in NSSI. If attentional deficits play central roles in NSSI, we should observe that DBT-ST leads to greater reduction in NSSI behavior compared to control group, and most importantly, this improvement should relate to enhancements in attentional abilities (i.e., mindful attention). In addition, we also tracked changes in participants’ emotion regulation. This was done for two main reasons. First, a substantial body of evidence indicates that emotion dysregulation plays a critical role in the pathology of NSSI [20]. However, most of this evidence has been derived from cross-sectional studies, and it remains unclear whether reductions in NSSI are accompanied by improvements in emotion regulation. Second, emotion regulation is closely intertwined with attention-related processes [21, 22]. The ability to flexibly shift attention is considered a key strategy for regulating negative affect, and cognitive reappraisal, a core emotion regulation strategy, also relies on attentional deployment [23]. Examining how these two interrelated processes evolve over time, particularly their temporal ordering, can provide valuable insight into the psychological processes underlying treatment-related reductions in NSSI. In summary, this study aimed to provide longitudinal evidence for the role of attentional deficits in NSSI and to characterize its temporal relationship with emotion regulation. To this end, we: (1) examined treatment effects of DBT-ST (versus social support group therapy, SSGT) on NSSI, mindful attention, and emotion regulation; (2) conducted mediation analyses to test whether improvements in mindful attention and emotion regulation accounted for reductions in NSSI behaviors; and (3) employed cross-lagged panel analysis to investigate the temporal sequence between mindful attention and emotion regulation. Notably, SSGT served as an active control, as it followed the same group procedures as DBT-ST but without skills training, instead providing peer support [24, 25]. Method Participants Participants with NSSI were enrolled in a two-arm, multi-center randomized controlled trial (RCT) designed to evaluate the effectiveness of 13-week DBT-ST. This study was a secondary analysis of the RCT registered at ClinicalTrials.gov (NCT04094623), here we only briefly described the trial design. Details were provided in the supplementary materials, and could be found in the primary article [16]. The inclusion criteria included: (a) meeting the DSM-5 diagnostic criteria for one or more mental disorders; (b) meeting the DSM-5 diagnostic criteria for NSSI, (c) 10 ~ 59 years of age, (d) be able to understand written and spoken Chinese. Participants were excluded if they met any of the following criteria: (a) a serious neurological disease; (b) a serious physical disease; (c) life-threatening suicidal plans or attempts within one month; (d) severe aggressive behaviors, mania episode, substance dependence or abuse; (e) other severe psychopathology or physical illness requiring immediate inpatient treatment; (f) pregnancy or breast-feeding. Treatment Both the DBT-ST group and the SSGT group received a 13-week psychotherapy program, meeting once weekly for 120 min per session. DBT-ST encompassed four core modules: mindfulness (1st-3rd session), distress tolerance (4th-6th session), emotion regulation (7th-10th session), and interpersonal effectiveness (11th-13th session). The program included psychoeducation, in-session practice of newly introduced skills, and weekly homework assignments to facilitate skill reinforcement. Each session began with a review of homework and a brief mindfulness exercise. The SSGT group followed the same procedures as the DBT-ST group, differing only in the absence of specific thematic content; it provided a supportive environment solely as a control group. Measurements Assessments were conducted at baseline and post-treatment. At baseline, the severity of major mental disorders (depression disorder, borderline personality disorder, and bipolar disorder) was measured. The severity of depressive symptoms was measured by Hamilton Depression Rating Scale (HAMD) [26]; the severity of borderline personality was measured by Borderline Symptom List-23 (BSL-23) [27]; and the severity of bipolar disorder was measured by Mood Disorder Questionnaire (MDQ) [28]. The frequency of NSSI behaviors within 1 month was measured using the Ottawa Self-Injury Inventory (OSI) [29]. The OSI is a self-administered questionnaire that assesses the cognitive, affective, behavioral, and environmental aspects of NSSI. In accordance with the aims of our study, only the first item of the OSI was considered to examine the frequency of NSSI behavior episodes within 1 month. Mindful attention was measured using the Mindful Attention Awareness Scale (MAAS) [30] because it assesses present-moment attention and awareness in daily life, which corresponds closely to the mindfulness-related attentional capacity targeted by DBT-ST. The MAAS consists of 15 items that emphasize open and receptive awareness of, as well as attention to, present-moment experiences. Responses are rated on a 6-point Likert scale, with higher scores indicating better mindful attention. In the current study, the Cronbach’s alpha was 0.850 at baseline and 0.916 at post-treatment. Emotion regulation was measured using the Difficulties in Emotion Regulation Scale (DERS) [31]. The DERS is a self-reported tool consisting of 36 items used to assess clinical difficulties with emotion regulation. All 36 items are scored using a 5-point Likert scale, ranging from 1 (rarely) to 5 (almost always), with higher scores indicating more difficulty in emotion regulation. In the current study, the Cronbach’s alpha for the DERS was 0.914 at baseline and 0.940 at post-treatment. The public registration listed NSSI as the primary outcome; MAAS and DERS were not individually listed as registered outcomes. However, both measures were included in the original assessment battery and statistical analysis plan of the primary RCT as secondary outcome/process-related measures and were reported in the primary effectiveness paper. The mediation and cross-lagged panel analyses reported here should therefore be interpreted as secondary mechanistic analyses of the completed RCT dataset. Statistical analyses Chi-square, and t tests were conducted to examine potential differences in sociodemographic and clinical variables between the DBT-ST and SSGT groups. According to the intention-to-treat (ITT) basis, multiple repeated-measures ANOVAs were conducted to reveal the therapeutic effect of DBT-ST, with timepoints (baseline and post-treatment) as within-subject variable and group (DBT-ST and SSGT) as between-subject variable. Following a significant interaction, simple effects analyses with bonferroni correction were performed. Partial η2 (ηp2) was used to report the effect size. These analyses were conducted using SPSS version 27.0. To assess whether changes in mindful attention and emotion dysregulation significantly mediated the association between treatment condition and NSSI change, we conducted a residualized-change chained mediation model [32]. Residualized change scores were calculated for mindful attention, emotion dysregulation, and NSSI behaviors by regressing each post-treatment score on its corresponding baseline score and saving the unstandardized residual. These residuals represent post-treatment individual differences adjusted for baseline levels and were used as the mediator and outcome variables in the mediation model. Treatment condition was entered as the independent variable, residualized changes in mindful attention and emotion dysregulation were entered as sequential mediators, and residualized change in NSSI behaviors was entered as the outcome. Indirect effects were tested using 5,000 bootstrap resamples. In addition, we conducted an Analysis of Covariance (ANCOVA) two-wave mediation analysis as a sensitivity analysis, in which baseline levels of the mediator and outcome variables were included as covariates. The residualized-change mediation model was estimated using SPSS PROCESS version 3.3. Both the cross-lagged model and the ANCOVA two-wave mediation model were analyzed using Mplus statistical software version 8.3. There were 0.5% to 21.9% missing data due to variety of reasons. Little’s Test for Missing Completely at Random (MCAR) indicated that the data were missing completely at random. Thus, missing data was addressed using single data imputation with expectation-maximization algorithm [33]. Results Descriptive statistics Table 1 presented the demographic characteristics for the DBT-ST group and the SSGT group. There were no significant differences between two groups in terms of demographic and clinical characteristics at baseline (ps > 0.1). Table 1Demographic characteristics of study participantsDBT-ST group(n = 97)SSGT group(n = 99)p c Age, mean (SD) 19.67 (5.98)19.68 (6.13)0.994 Gender (%) Male13 (13.40%)17 (17.17%)0.593 Female84 (86.60%)82 (82.82%)Primary diagnosis, No. (%) a Depression disorder42 (43.30%)42 (42.42%)0.586 Borderline personality disorder29 (29.90%)32 (32.32%) Bipolar disorder19 (19.59%)16 (16.16%) Anxiety disorder1 (1.03%)4 (4.04%) Stress disorder01 (1.01%) Obsessive-compulsive disorder1 (1.03%)2 (2.02%) Pathology gambling02 (2.02%) Schizophrenia1 (1.03%)0 Schizoaffective disorder1 (1.03%)0HAMD b33.96 ± 13.8733.30 ± 13.230.738 BSL-23 54.98 ± 22.3651.25 ± 20.890.229 MDQ 4.15 ± 2.253.86 ± 2.340.368Note. a Primary diagnosis data was incomplete for the DBT-ST group (n = 3) due to missing values. b HAMD data was incomplete for the DBT-ST group (n = 4) due to missing values. c Age, HAMD, MDQ, and BSL-23 difference between two groups was compared by t-test, gender difference between two groups was compared by chi-square test, and primary diagnosis difference between two groups were compared by fisher’s exact test. HAMD = Hamilton Depression Rating Scale; BSL-23 = Borderline Symptom List-23; MDQ = Mood Disorder Questionnaire Table 1Demographic characteristics of study participantsDBT-ST group(n = 97)SSGT group(n = 99)p c Age, mean (SD) 19.67 (5.98)19.68 (6.13)0.994 Gender (%) Male13 (13.40%)17 (17.17%)0.593 Female84 (86.60%)82 (82.82%)Primary diagnosis, No. (%) a Depression disorder42 (43.30%)42 (42.42%)0.586 Borderline personality disorder29 (29.90%)32 (32.32%) Bipolar disorder19 (19.59%)16 (16.16%) Anxiety disorder1 (1.03%)4 (4.04%) Stress disorder01 (1.01%) Obsessive-compulsive disorder1 (1.03%)2 (2.02%) Pathology gambling02 (2.02%) Schizophrenia1 (1.03%)0 Schizoaffective disorder1 (1.03%)0HAMD b33.96 ± 13.8733.30 ± 13.230.738 BSL-23 54.98 ± 22.3651.25 ± 20.890.229 MDQ 4.15 ± 2.253.86 ± 2.340.368Note. a Primary diagnosis data was incomplete for the DBT-ST group (n = 3) due to missing values. b HAMD data was incomplete for the DBT-ST group (n = 4) due to missing values. c Age, HAMD, MDQ, and BSL-23 difference between two groups was compared by t-test, gender difference between two groups was compared by chi-square test, and primary diagnosis difference between two groups were compared by fisher’s exact test. HAMD = Hamilton Depression Rating Scale; BSL-23 = Borderline Symptom List-23; MDQ = Mood Disorder Questionnaire Demographic characteristics of study participants Note. a Primary diagnosis data was incomplete for the DBT-ST group (n = 3) due to missing values. b HAMD data was incomplete for the DBT-ST group (n = 4) due to missing values. c Age, HAMD, MDQ, and BSL-23 difference between two groups was compared by t-test, gender difference between two groups was compared by chi-square test, and primary diagnosis difference between two groups were compared by fisher’s exact test. HAMD = Hamilton Depression Rating Scale; BSL-23 = Borderline Symptom List-23; MDQ = Mood Disorder Questionnaire Treatment effect of DBT-ST Repeated-measures ANOVA revealed significant timepoints × group interaction for NSSI behaviors (F = 4.218, p = 0.041, ηp2 = 0.021), mindful attention (F = 6.192, p = 0.014, ηp2 = 0.031), and emotion dysregulation (F = 11.290, p < 0.001, ηp2 = 0.055). Simple effects analyses showed that following treatment, both DBT-ST and SSGT groups exhibited significant reductions in NSSI behaviors (DBT-ST, MD = -0.728, p < 0.001; SSGT, MD = -0.433, p < 0.001). Importantly, the reduction was significantly larger in the DBT-ST group (Fig. 1A), supporting its therapeutic effect. Similarly, compared to SSGT group, DBT-ST group showed greater improvements in mindful attention (DBT-ST, MD = 6.900, p < 0.001; SSGT, MD = 2.394, p = 0.062) and emotion dysregulation (DBT-ST, MD = -16.542, p < 0.001; SSGT, MD = -7.145, p < 0.001) (see Fig. 1B-C and eTable 1). Collectively, these findings suggested that DBT-ST showed treatment effect on NSSI behaviors and other key psychopathological components related to NSSI. Furthermore, the treatment effects observed here were broadly comparable to previous research [34], suggesting that our findings were clinical meaningful, which enabled us to investigate the mechanism underlying such behavioral improvement. Fig. 1Intervention results of DBT-ST and SSGT groups on (A) NSSI behaviors, (B) mindful attention, and (C) emotion dysregulation. Note. Error bars represent ± 1 standard error of the mean. OSI, Ottawa Self-Injury Inventory; NSSI, non-suicidal self-injury; MAAS, Mindful Attention Awareness Scale; DERS, Difficulties in Emotion Regulation Scale. *p < 0.05, **p < 0.01, ***p < 0.001; ns, nonsignificant Intervention results of DBT-ST and SSGT groups on (A) NSSI behaviors, (B) mindful attention, and (C) emotion dysregulation. Note. Error bars represent ± 1 standard error of the mean. OSI, Ottawa Self-Injury Inventory; NSSI, non-suicidal self-injury; MAAS, Mindful Attention Awareness Scale; DERS, Difficulties in Emotion Regulation Scale. *p < 0.05, **p < 0.01, ***p < 0.001; ns, nonsignificant Mediation underlying treatment effect of DBT-ST The correlations among the baseline and post-treatment variables were presented in eTable 2. Except for the non-significant correlation between baseline NSSI behavior scores and post-treatment emotion regulation scores, all other variable pairs were significantly correlated. To further examine these associations while accounting for baseline levels, chained-mediation analyses were conducted using residualized change scores. Results revealed two significant indirect pathways (Fig. 2 and eTable 3). First, the path “treatment condition → residualized change in emotion dysregulation → residualized change in NSSI behaviors” showed a significant indirect effect of -0.116, 95% CI [-0.227, -0.028], accounting for 55.8% of the total indirect effect. Second, the sequential path “treatment condition → residualized change in mindful attention → residualized change in emotion dysregulation → residualized change in NSSI behaviors” also showed a significant indirect effect of -0.060, 95% CI [-0.141, -0.007], accounting for 28.8% of the total indirect effect. By contrast, the indirect path “treatment condition → residualized change in mindful attention → residualized change in NSSI behaviors” was not significant, β = -0.032, 95% CI [-0.105, 0.024]. The ANCOVA two-wave mediation model produced comparable results (eFigure 1 and eTable 4). Together, these findings suggested that DBT-ST-related reductions in NSSI were statistically accounted for, in part, by improvements in emotion dysregulation and by a sequential pathway in which improvements in mindful attention were associated with subsequent reductions in emotion dysregulation. Fig. 2Path diagram of the mediation model based on residualized change scores. Note. The standardized coefficients path estimates were represent in the figure. Dashed line indicates non-significant paths. *p < 0.05, **p < 0.01, ***p < 0.001 Path diagram of the mediation model based on residualized change scores. Note. The standardized coefficients path estimates were represent in the figure. Dashed line indicates non-significant paths. *p < 0.05, **p < 0.01, ***p < 0.001 Temporal relationship between attention deficits and emotion dysregulation To clarify the temporal ordering of attention and emotion improvements, cross-lagged panel analysis was conducted. Bivariate correlations between mindful attention and emotion dysregulation at baseline and post-treatment were reported in eTable 5, which showed significant associations between all variables in both the DBT-ST group and SSGT group. The two-wave cross-lagged panel model was just-identified, or saturated, with zero degrees of freedom; therefore, standard model fit indices indicated perfect fit and were not informative for evaluating model adequacy. For the DBT-ST group, mindful attention significantly predicted subsequent reductions in emotion dysregulation (β = -0.261, p < 0.01), whereas emotion dysregulation did not significantly predict later mindful attention (Fig. 3). This pattern suggested that, within DBT-ST, improvements in mindful attention prospectively associated with enhancements in emotion regulation rather than the reverse. Stability paths were also significant for both mindful attention (β = 0.478, p < 0.001) and emotion dysregulation (β = 0.368, p < 0.001) across baseline and post-treatment, indicating relative temporal consistency of these constructs. By contrast, the SSGT group demonstrated a bidirectional relationship: mindful attention significantly predicted later emotion dysregulation (β = -0.205, p < 0.05), while emotion dysregulation also significantly predicted later mindful attention (β = -0.343, p < 0.001). This reciprocal pattern implied that, without targeted intervention such as DBT-ST, mindful attention and emotion dysregulation may reinforce one another with no leading component. Stability effects were again significant for mindful attention (β = 0.380, p < 0.001) and emotion dysregulation (β = 0.578, p < 0.001). Together, these findings provided preliminary evidence that improvements in attentional abilities may temporally precede improvements in emotional regulation in NSSI. Fig. 3Cross-lagged analysis for mindful attention and emotion dysregulation in (A) DBT-ST group and (B) SSGT group. Note. T1 = baseline; T2 = post-treatment. *p < 0.05, **p < 0.01, ***p < 0.001 Cross-lagged analysis for mindful attention and emotion dysregulation in (A) DBT-ST group and (B) SSGT group. Note. T1 = baseline; T2 = post-treatment. *p < 0.05, **p < 0.01, ***p < 0.001 Discussion The objective of this study was to investigate the important role of attentional deficits in NSSI by providing longitudinal evidence that reductions in NSSI behaviors were accomplished through enhancement in this component. We showed that after a 13-week psychotherapy session, participants with NSSI (i.e., DBT-ST group) exhibited larger reduction in NSSI behaviors, and greater improvement in mindful attention and emotion regulation abilities, compared to control group (i.e., SSGT group). This finding laid the foundation for further investigation on the mechanism underlying such treatment effects. Mediation analyses indicated that the therapeutic effects of DBT-ST operated through enhancements in attention and emotion regulation. In particular, the effect of attention on NSSI was achieved by indirectly facilitating reductions in emotion dysregulation. More importantly, cross-lagged panel analyses revealed that improvements in attention abilities temporally preceded that in emotion regulation. These findings underscored the pivotal role of attention in NSSI, with emotion regulation acting as a mediating process, which contributed to the emerging body of knowledge on the relationship between attention, emotional regulation and NSSI. Attentional functioning is fundamental in our daily lives [35], and deficits of it are associated with a broad range of psychiatric diseases, including depression disorder [36], anxiety disorder [37], and borderline personality disorder [38]. These disorders are closely associated with NSSI [4]. Additionally, emerging evidence has shown that individuals with NSSI show impairments in attentional processes. For example, through behavioral experiments and eye-tracking technique, researchers have found that adolescents with a history of self-harm tend to pay more attention to NSSI related stimuli [39]. Such behavioral evidences are echoed with neuroimaging findings, which have found a strong association between NSSI and deficits in brain regions in frontoparietal attention network, such as dorsolateral prefrontal cortex, dorsal anterior cingulate cortex, and anterior cingulate cortex [40–42], highlighting the link between deficits in attention and NSSI. However, the aforementioned evidence mainly comes from cross-sectional studies [12, 14, 43]. Evidence from longitudinal study has been lacking. Building on the cross-sectional findings that NSSI participants were impaired in attentional processes, the current study provided more direct evidence for the attentional hypothesis of NSSI by showing that NSSI’s improvement was linked to improvement in mindful attention. The association between emotion dysregulation and NSSI is well-established, as both theoretical models and empirical evidence of NSSI suggest that individuals who engage in NSSI typically show impairments in emotion regulation [20, 44]. Our results extend this literature by providing longitudinal evidence that reductions in NSSI behaviors were accompanied by improvements in emotion regulation, thereby further supporting the emotion dysregulation hypothesis of NSSI. Notably, the direct pathway from changes in mindful attention to reductions in NSSI behaviors was not significant, whereas the sequential indirect pathway through emotion dysregulation was significant. This pattern does not necessarily indicate that mindful attention is unimportant. Rather, it suggests that, in a clinical intervention context, improvements in foundational attention-related processes may need to be translated into more proximal emotion regulation capacities before they can effectively reduce maladaptive behaviors such as NSSI. Consistent with this interpretation, subsequent cross-lagged panel analysis demonstrated that in the DBT-ST group, improvements in mindful attention prospectively associated with improvements in emotion regulation, whereas the reverse pathway was not significant. Nevertheless, because the sequential indirect effect was small in magnitude, this pathway should be interpreted cautiously as preliminary evidence for a statistically detectable process rather than as evidence for a strong clinical mechanism. Together, these findings suggest that attention may function as an upstream therapeutic process that indirectly contributes to reductions in NSSI by supporting subsequent improvements in emotion regulation, with emotion regulation serving as the more proximal mediator of behavioral change. Several caveats should be considered when interpreting this sequential pathway. First, attention and emotion regulation are intrinsically interconnected. Individuals with better emotion regulation abilities tend to experience less distraction by irrelevant stimuli [45], whereas people with better attention abilities are also the ones who show superior emotion regulation performance [46]. Thus, attention may support emotion regulation, but the two constructs should not be viewed as entirely independent processes. Second, the tool used to assess emotion regulation may here overlap partially with that of mindful attention. Specifically, we adopted MAAS and DERS to respectively quantify mindful attention and emotion regulation. Some DERS items involve awareness- or attention-related content, such as “I have no idea how I am feeling” and “When I’m upset, I have difficulty thinking about anything else” [31]. Therefore, shared method variance and partial conceptual overlap may have contributed to the observed association between mindful attention and emotion dysregulation and may have strengthened the sequential mediation pathway. However, the correlations between MAAS and DERS were moderate rather than high at both assessment points, and the correlation between their change scores was also moderate, with coefficients approximately around 0.40 (eTable 2). This pattern suggests that the two measures captured related but non-redundant constructs. Third, DBT-ST involves training in both of attention and emotion regulation [17], which may make it difficult to isolate the unique direct effect of attention on NSSI reduction. Therefore, the present findings should be interpreted as supporting a statistical process model in which improvements in attention are associated with subsequent improvements in emotion regulation, rather than as definitive evidence for a causal mechanism. Future research could employ interventions that precisely target on attention to further elucidate the connection between these two. It is also important to note that the SSGT group showed significant reductions in NSSI behaviors and emotion dysregulation, despite not showing a statistically significant improvement in mindful attention. This finding is not unexpected, as SSGT served as an active control condition. Although SSGT did not include DBT skills training, it provided structured group contact, peer support, opportunities for emotional sharing, and validation, all of which may reduce distress and contribute to improvements in emotion regulation and self-injurious behaviors. Thus, improvement in the SSGT group may reflect nonspecific therapeutic factors or support-related mechanisms. Our findings therefore should not be interpreted as suggesting that mindful attention is the only pathway through which NSSI can improve. Rather, the mediation results indicate that mindful attention may help explain the additional benefit of DBT-ST over SSGT, particularly by supporting subsequent improvements in emotion regulation. There were several limitations to note. First, the sample was predominantly female and relatively young, which may limit the generalizability of the findings. Although this demographic distribution is broadly consistent with the epidemiological profile of NSSI, future studies should examine whether the present findings extend to male patients and older adults with NSSI. Second, the study included only two measurement time points, which hindered a more accurate understanding of the change in these variables over time. Third, all scales used in this mediation analysis were self-report measures. Thus, the potential influence of expectancy effects on post-intervention self-reports cannot be excluded. Future research could utilize an objective and more precise measures to assess this variable. Overall, this study explored the relationship between attention, emotion regulation and NSSI. The results supported the view that attention played an important role in NSSI behaviors, though through a mediation by emotion regulation. The mediation and cross-lagged findings suggest a potential sequential process in which improvements in attention may support subsequent gains in emotion regulation, which in turn are associated with reductions in NSSI. These findings highlight attention as a candidate upstream therapeutic process in DBT-ST and implies that interventions for NSSI may benefit from placing greater emphasis on mindfulness- or attention-related training early in treatment, thereby strengthening attention as a foundational capacity for subsequent emotion regulation improvement. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Supplementary Material 1