Compassion‐based interventions in Asian communities: A meta‐analysis of randomised controlled trials

Abstract Purpose Practising compassion increases well‐being and reduces depression, anxiety, and psychological distress among clinical and non‐clinical populations. There is a rapid increase in compassion‐based interventions within the past two decades. However, the reviews are limited to predominantly Western cultures. Therefore, this meta‐analysis aimed to evaluate the literature attempting to promote and increase compassion in Asian communities. Method Eight randomised controlled trials (RCTs) conducted between 2016 to 2021 were included in the meta‐analysis with data from 1012 participants across Thailand, Japan, China and Hong Kong. Effect sizes were calculated to test the efficacy of the compassion‐based interventions on the self‐compassion outcome. Intervention efficacy was tested by comparing the intervention groups against control groups (wait‐list control and active control groups) at pre‐ and post‐interventions. Results Significant between‐group differences in change scores were found on self‐report measures of self‐compassion with large effect sizes in interventions with wait‐list control groups (d = .86) and small effect sizes in interventions with active‐control groups (d = .19). Conclusions Although compassion‐based interventions are heterogeneous in nature and limited in scope, there is promising evidence of improving self‐compassion in Asian communities. This supports for the cross‐cultural applicability of compassion‐based interventions. However, research within the Asian context is limited and at an infancy stage, signifying the importance of conducting further compassion‐based interventions in clinical and non‐clinical groups living in the Asian communities.


I N TRODUC T ION
The concept of compassion has been widely discussed in Buddhism and other practiced religions (Germer & Siegel, 2012;Strauss et al., 2016). Compassion is commonly understood as an openness to consciously turn towards suffering, rather than away from it (Gilbert, 2014a). Whilst compassion-based meditations have formed a central part of some spiritual traditions (e.g. Buddhism: Lama & Thupten, 1995), they have also been incorporated into treatment approaches in psychotherapy (Gilbert, 2013;Neff, 2003a). Practising compassion has shown increased improvements in psychological and physiological wellbeing in clinical and non-clinical populations Gilbert, 2013;MacBeth & Gumley, 2012;Neff, 2003a). The existing literature provides evidence to support the notion that compassion cultivation and practice may have a positive impact on a range of emotional, physical and life experiences whilst reducing psychopathology (Kirby, 2016). However, the majority of compassion-based interventions are limited to predominantly Western communities, with very few being conducted with Asian populations, particularly in South Asian communities (de Zoysa et al., 2022).

Theories of compassion
Self-compassion is compassion given towards oneself. One of the earliest researchers to put forward a theory and measure of self-compassion was Neff (2003a), who defined it as 'being touched by and open to one's own suffering, not avoiding or disconnecting from it, generating the desire to alleviate one's Conclusions: Although compassion-based interventions are heterogeneous in nature and limited in scope, there is promising evidence of improving self-compassion in Asian communities. This supports for the cross-cultural applicability of compassion-based interventions. However, research within the Asian context is limited and at an infancy stage, signifying the importance of conducting further compassion-based interventions in clinical and non-clinical groups living in the Asian communities.

K E Y W O R D S
Asian, compassion, cross-cultural, efficacy, intervention

Practitioner Points
• There is an increase interest in compassion-based interventions in Western countries • Most compassion-based interventions and Western concepts of compassion are influenced by Eastern philosophies and Buddhist psychology • Many Asian cultures embrace and practice Buddhism • However, there is an evident dearth of compassion related studies and compassion-based interventions in the Asian context • It is important to conduct compassion-related studies in Asian communities, considering that compassion has been found to be at least partially determined by culture.
suffering and to heal oneself with kindness' (p. 87). Neff's interpretations, based on Buddhist teachings, emphasise that self-compassion incorporates three components with opposing negative counterparts: mindfulness/over-identification, common humanity/isolation and self-kindness/self-judgement. Neff described mindfulness as being the non-judgemental, systematic observation of thoughts and feelings as they arise without denying or suppressing them. Over-identification, on the other hand, is proposed as being caught up and swept away by the negative reactivity caused by distressing thoughts and feelings. Common humanity is acknowledging that pain is a shared-human experience, whereas isolation is the perception that one is alone in their suffering. Self-kindness is treating oneself gently and warmly, and self-judgement is treating oneself from a cold and critical perspective when faced with failure and suffering.
In contrast, Gilbert (2013), another leading researcher in the compassion field defined compassion as 'a sensitivity to suffering in self and others with a commitment to try to alleviate and prevent it' (p.94). Gilbert introduced the Social Mentality Theory (SMT), which suggests that compassion emerges from the evolution of the mammalian care-giving motivational system designed to regulate negative affect. These motivational systems are referred to as social mentalities, which evolved to overcome challenges essential for survival, such as care-seeking and caregiving (Gilbert, 2005(Gilbert, , 2014b. Compassion according to Gilbert (2014a) is captured by six essential competencies relating to sensitivity, sympathy, empathy, motivation, caring and distress tolerance that are experienced across three directional flows, namely self-compassion, compassion to others and compassion from others (Gilbert, 2009).
These interventions share certain similarities and differences, indicating the multidimensional nature of compassion (Kirby, 2016). In consideration of the similarities, all the aforementioned interventions have been influenced by the Tibetan Buddhism and incorporate some form of mindfulness practice. CFT and MSC programmes focused less on mindfulness whilst the CCT, CBCT and CEB programmes spent most of the interventions focusing on mindfulness-based training. Importantly, all interventions include a portion of psychoeducation, providing a rationale for engaging in the compassion-based training. All interventions also entail activities and tasks that participants are instructed to actively practice using specific compassion strategies. These practices are also similar in most interventions and contain techniques such as breathing exercises, facial and body expressions, building compassionate inner voices, compassionate letter writing and imagery tasks aimed at producing calm and soothing sensations by activating the parasympathetic system (soothing system). A homework component is also incorporated in all interventions. Interestingly, these interventions also demonstrate the ability to be delivered in group settings (Kirby, 2016).
In consideration of the differences, CFT is notably different from other compassion-based interventions, as it is a form of psychotherapy, whereas the other interventions are simply programmes developed to increase compassion. CFT can be tailored to meet the needs of the individual, whilst the other interventions are delivered by following the prescribed session content for each session (Kirby, 2016). All interventions, except MSC, focus on compassion as a broader experience that spreads across the self and towards others, whilst MSC only focuses on self-compassion (Neff & Germer, 2013).
It is important to note that whilst these interventions have produced encouraging results for increased well-being (Kirby, 2016), these interventions have been developed in Western countries. For instance, the most widely used approaches such as CFT by Gilbert (2014a) and MSC by Neff and Germer (2013) originated in the UK and USA, respectively. In addition, only a limited number of RCT studies (e.g. Arimitsu, 2016;Tung, 2020) have been conducted so far, to support the use of these compassion-based interventions (Kirby, 2016;Matos et al., 2017;. Furthermore, literature exploring compassion-based interventions also seem to be at an infancy stage (Kirby, 2016), with the need for more rigorous trials to explore the efficacy of compassion-based interventions across clinical and non-clinical samples from a range of diverse backgrounds. With respect to the two existing meta-analyses, only three of 27 studies in the meta-analysis by Ferrari et al. (2019) and three of 17 studies by Kirby (2017) were conducted in an Asian country. Asian studies included in these reviews were also limited to China (Lo et al., 2013;Wong & Mak, 2016), Japan (Arimitsu, 2016) and Korea (Lee & Bang, 2010) suggesting that only a few countries in the East Asian cultural context have tested compassion-based interventions. In addition, neither meta-analysis (Ferrari et al., 2019;Kirby, 2017) assessed the potential influence of culture on the efficacy of these compassionbased interventions.
It is important to note that whilst these interventions have produced encouraging results for increased well-being (Kirby, 2016), these interventions have been developed in Western countries (e.g. CFT in the UK: Gilbert, 2014a, and MSC in the USA: Neff & Germer, 2013). In addition, only a limited number of RCT studies (e.g. Arimitsu, 2016;Tung, 2020) have been conducted so far, to support the use of these compassion-based interventions (Kirby, 2016;Matos et al., 2017;. Furthermore, literature exploring compassion-based interventions also seem to be at an infancy stage (Kirby, 2016), with the need for more rigorous trials to explore the efficacy of compassion-based interventions across clinical and non-clinical samples from a range of diverse backgrounds. With respect to the two existing meta-analyses, only three of 27 studies in Ferrari et al. (2019) and three of 17 studies by Kirby (2017) were conducted in an Asian country. Asian studies included in these reviews were also limited to people in China (Lo et al., 2015;Wong & Mak, 2016), Japan (Arimitsu, 2016) and Korea (Lee & Bang, 2010) suggesting that only a few countries in the East Asian cultural context have tested compassion-based interventions. In addition, neither meta-analyses (Ferrari et al., 2019;Kirby, 2017) assessed the potential influence of culture on the efficacy of these compassion-based interventions.
Therefore, the present study aimed to explore the efficacy of the existing compassion-based interventions conducted in Asian communities. A meta-analysis was conducted with the aim of answering the question; do compassion-based interventions lead to increased levels of compassion in people living in Asian communities? A secondary aim was to investigate whether identified studies also reduced the additional outcomes of depression, anxiety, and stress.

Protocol and registration
This meta-analysis adhered to the general principles of the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA: Page et al., 2021). The protocol was prospectively registered in PROSPERO, the international prospective register of systematic reviews, under the registration number CRD42020201832.

Eligibility criteria
The primary researcher and a voluntary research assistant carried out searches independently. Studies that met the following criteria were included (a) a randomised controlled trial in which the primary focus was to purposively generate compassion or self-compassion; (b) conducted in at least one Asian country; and (c) included at least one self-report measure related to compassion or self-compassion. Studies conducted across countries in the Middle East and North Africa (MENA) region were excluded as although some Middle Eastern countries are situated in the Asian continent, they are considered as countries in the MENA region (separately from other Asian countries) and share certain cultural and religious norms that are different from other Asian cultures (Alkaiyat & Weiss, 2013;Kabasakal et al., 2012). Both clinical and non-clinical populations of all ages were included. No publication date, language or study design restrictions were applied. Eligibility criteria were based on the population, intervention, comparator, outcomes and the study type (Table 1).

Search strategy
The systematic literature search was conducted using Scopus, Medline, Web of Science, AMED, APA PsycINFO, Ovid (EMBASE) and CINAHL databases. Cochrane Library, ProQuest for Dissertations and Theses and Open-Dissertations databases were systematically searched to detect any relevant grey literature. The final search took place on the 10th of March 2022. The following search terms were developed with a research librarian: TI (compassion*) AND AB (random* control*) AND AB (trial OR interven* OR stud* OR program* OR therap* OR training) AND TX (Asia* OR East* OR "Eastern culture*" OR Japan* OR Chin* OR Vietnam* OR Malaysia* OR Singapore* OR "Hong Kong" OR Korea* OR India* OR Pakistan* OR Bangladesh* OR "Sri Lanka*"). Although various other interventions have integrated compassion (e.g. Mindfulness-Based Compassion Training: Lo, 2011) or produced increased compassion (e.g. Mindfulness-Based Cognitive Therapy: Segal et al., 2002), their primary focus is not compassion cultivation. Therefore, such interventions were excluded from the search results and only the interventions with a specific focus of compassion/self-compassion cultivation were included.

Data extraction
Data relating to the following study characteristics were extracted: Name of authors and year of publication, country, intervention name, design, and underpinning theory/model, aim of the study, target population, measures used, duration of the intervention, intervention tasks and the main findings of the study. For the meta-analyses, the means, standard deviations and sample sizes for each group at pre-and post-interventions were extracted.

Analysis strategy
Version 5.4 of the RevMan software (The Cochrane Collaboration, 2020) was used for the analyses. Cohen's (1992) guidelines of small (0.2), medium (0.5) and large (0.8) effects were used when interpreting effect sizes, represented by d. Computations were based on a weighted-average of the effect sizes using a random-effects model, as it assumes that study variations can be systematic and not only due to random error (Borenstein et al., 2009). A random-effects model is also appropriate as true effects of interventions are likely to vary depending on the sample characteristics and implementation of the intervention.
The efficacy of the interventions on compassion was compared in relation to the control groups. Control groups varied between waitlist control (WLC) and active control (AC) groups. WLC groups received no intervention, and the AC groups received a different form of intervention than the intervention groups (Kirby, 2017). In addition, a sensitivity analysis was conducted to investigate the intervention efficacy separately for the intervention groups compared with the WLC groups and intervention groups compared with the AC groups. It was expected that studies with an AC would report smaller effect sizes than studies with a WLC group, as the different interventions received by the AC groups would also influence the outcome variables (Cuijpers et al., 2016;Kirby, 2017). Another sensitivity analysis was conducted to test the intervention efficacy separately for clinical populations and non-clinical populations when compared to the control groups. Both existing meta-analyses (Ferrari et al., 2019;Kirby, 2017) included participants from clinical and non-clinical backgrounds, and therefore we wanted to see whether the compassion-based interventions would have different effects on these participant groups. Where certain studies appeared to cause large heterogeneity, further sensitivity analyses were conducted with the exclusion of those studies, to understand the heterogeneity of the studies.
In addition to testing the efficacy of the interventions on compassion, this meta-analysis also investigated secondary outcome measures such as depression, anxiety and stress. Where one study incorporated multiple measures to measure an outcome (e.g. two measures of depression), a combined outcome was computed to produce a single mean difference by computing the average mean and standard deviation from each study (Borenstein et al., 2009). Previous meta-analyses of compassion-based interventions have also followed this method (Ferrari et al., 2019;Kirby, 2017). This method of providing only one effect size per study is recommended to avoid the bias of choosing one measure over another measure and to prevent attributing greater weights to studies with multiple outcomes (Borenstein et al., 2021;Ferrari et al., 2019;Kirby, 2017).

Risk of bias within studies
Risk of bias within studies was assessed using the Cochrane risk of bias tool using Revman (Higgins et al., 2011). Critical assessments were made separately for each study for the following domains: sequence generation, allocation concealment, blinding of participants and personnel, blinding of outcome assessment, incomplete outcome data, selective outcome reporting and for other biases. Based on the information provided, each study was given a judgement of 'high risk', 'low risk' or 'unclear' risk of bias.

Systematic search results
The initial database search resulted in 266 records, including 82 duplicates which were removed. Titles and abstracts of the remaining 184 papers were screened and 159 papers were excluded, as they were not related to the search. After checking the full text of 25 papers, 16 were excluded based on the eligibility criteria. Of the final nine results, three papers reported on one study (Mak et al., 2018(Mak et al., , 2019Yip, 2018) and therefore only one of them was retained (Mak et al., 2018). Reference lists of the chosen studies and other resources were searched for any potential studies and one further study was found on ResearchGate. This resulted in eight studies with quantitative data that were included in the metaanalysis. All studies were allocated a number from 1 to 8 (see Table 2) and are referred to by their assigned number (e.g. 2, 4 ) going forward. Figure 1 details the search strategy.

Study characteristics
A total of six of the eight studies included at least one of the six compassion-based interventions that Kirby (2016) outlined. Four studies were based on MSC (Neff & Germer, 2013) 1, 4, 6, 7 , one was based on CFT (Gilbert, 2014a) 2 , and another study incorporated both MSC and CMT approaches 5 . Although not outlined in Kirby's (2016) review, the remaining two studies in the present review were based on approaches by Neff (2009), andNeff et al. (2021) with one study conducting a self-compassion writing exercise (Neff, 2009) 8 and the other looking at a new approach named the Self-Compassionate Mindstate Induction (SCMI: Neff, 2021) 3 . Four studies were delivered in person, 1, 2, 5, 7 and four were delivered online 3, 4, 6, 8 , using a group 1, 2, 3, 5, 7 or a self-delivered approach 4, 6, 8 .
Intervention duration varied from 1-8 weeks. One study did not specify the duration of the intervention 3 . The authors were contacted to ascertain the information regarding the intervention duration, but no response was obtained. Four studies included a waitlist-control group (WLC) 2, 4, 5, 7 , and four studies included an active control group (AC) 1, 3, 6, 8 . The AC groups received a form of intervention different to the compassion-based interventions given to intervention groups, whilst the WLC groups received no treatment/intervention. The type of intervention received by the AC groups varied between a standard psychotherapy 1 , a neutral writing condition 3 , cognitive behavioural therapy 6 and a control writing condition 8 . The majority of the studies also reported follow-up data, 2, 4, 5, 6, 7, 8 with follow-up periods ranging from one-to twelve-month post-intervention. Table 2 gives a summary of the study characteristics. All studies were conducted within a five-year period (2016)(2017)(2018)(2019)(2020)(2021)

| 159
COMPASSION-BASED INTERVENTIONS IN ASIAN COMMUNITIES students 2, 3, 5, 7, 8 , adults from the general population 1, 6 and pregnant women 4 . Some studies specifically recruited adults with low self-compassion 2 or symptoms of anxiety and/or depression 1, 4 at baseline.

Compassion outcomes
The primary analysis was conducted to test the intervention efficacy for studies comparing the interventions with all control groups on the outcome of self-compassion. For the sensitivity analyses, separate analyses were conducted for studies with WLC groups and studies with AC groups. The effect sizes and heterogeneity statistics for self-compassion tested for the two categories can be found in Table 3. All studies, except one 3 , used the SCS (Neff, 2003a(Neff, , 2003b to measure self-compassion. Some studies used the complete scale of SCS 1, 2, 4, 5, 7 , and other studies only used the 13 positive items of the scale 6, 8 . One study used a recently developed State Self-compassion Scale-Long Form 3 (SSCS-L: Neff et al., 2021).

Compassion-Based interventions compared with all control groups
A significant medium effect size was reported for self-compassion, d = .56, k = 8, 95% CI [0.25, 0.81], p < .001 when comparing the intervention group with a control group (including WLC and AC groups) ( Figure 2). However, there was a significant amount of heterogeneity in effect sizes for self-compassion, Q(7) = 24.46, p = .0004, I 2 = 74%.

F I G U R E 1 PRISMA flow diagram of study selection [Colour figure can be viewed at wileyonlinelibrary.com]
Records removed before screening: Duplicate records removed (n = 82) Records screened (n = 184)

Included
Reports sought for retrieval (n = 1)

Identification of studies via databases and registers Identification of studies via other methods
Records identified from: Websites (n = 1) Citation searching (n = 0) When determining what may have caused the significant statistical heterogeneity, it was clear that three papers 4, 6, 8 (Guo et al., 2020;Mak et al., 2018;Wong & Mak, 2016) differed from the rest of the studies. Guo et al. (2020) was focused on a different target group (pregnant women in second or third trimester) and had a different intervention aim (prevention of post-partum depression). The paper by Mak et al. (2018) differed as it was the only mobile app-based intervention. Wong and Mak's (2016) paper also differed from the other papers as this study only used a writing exercise whereas all the other papers had multiple tasks (e.g. imagery exercises, breathing practices, etc.). To account for these differences, a sensitivity analysis was conducted with the omission of these papers (Figure 3). When these papers were removed, the results indicated a significant large effect size, d = .80, k = 5, 95% CI [0.48, 1.12], p < .001 with a non-significant heterogeneity, Q(4) = 6.87, p = .14, I 2 = 42%. This implies that the remaining studies shared similar intervention characteristics.

Compassion-based interventions compared with waitlist control groups
The sensitivity analysis resulted in a significant large effect size for self-compassion, d = .86, k = 4, 95% CI [0.46, 1.25], p < .001 when comparing the intervention group with a WLC group 2, 4, 5, 7 (Figure 4). There was a significant amount of heterogeneity in effect sizes for self-compassion, Q(3) = 8.95, p = .03, The results of the analysis comparing compassion-based interventions with WLC groups may also have been influenced by the differing target group and intervention aim of Guo et al., 2020. Therefore, to test whether this study may have affected the overall heterogeneity, a further sensitivity analysis was conducted with the omission of data of this study ( Figure 5). Results indicated that there was a large effect size, d = 1.07, k = 3, 95% CI [0.75, 1.38], p < .001, with zero heterogeneity, Q(2) = 0.07, p = .97, I 2 = 0%.

Compassion-based interventions on clinical populations
This sensitivity analysis was conducted on three papers 1, 2, 4 that included participants with a DSM-V diagnosis of Major Depressive Disorder, low compassion scores (scored below the average score of 17.35 on the SCS scale) and antenatal depressive or anxiety symptoms, respectively. When looking at the effectiveness of compassion interventions on these clinical populations (Figure 7), the results indicated a medium effect for self-compassion, d = .51, k = 3, 95% CI [.31, .72], p < .001. There was no heterogeneity of variance in the effect sizes for self-compassion, Q(2) = 1.88, p = .039, I 2 = 0%.

Secondary outcomes
Depression Four papers were included in the analysis assessing intervention efficacy on depression 1, 2, 5, 8 . One additional paper also investigated depression 4 (Guo et al., 2020) but did not report these results separately so could not be included in the analysis. There was no significant intervention effect on depression, d = − .31, k = 4, 95% CI [−.84, .24], p = .28, but a significant large heterogeneity between the studies, Q(3) = 10.15, p = .02, I 2 = 70% (Figure 10). A sensitivity analysis omitting the studies presumed to cause heterogeneity was not conducted, as there were only four papers reporting results for depression.

Anxiety
Four papers reported the intervention efficacy on anxiety 1, 2, 3, 5 . Guo et al. (2020) 4 also investigated anxiety, but the study could not be included in the analysis as results were not reported separately. There was no significant intervention effect on anxiety, d = − 1.07, k = 4, 95% CI [−3.45, 1.31], p = .38. In addition, there was a significant large heterogeneity between the studies, Q(3) = 142.96, p < .001, I 2 = 98% ( Figure 11). A sensitivity analysis omitting the studies presumed to cause heterogeneity was not conducted, as there were only four papers reporting results for anxiety.

Risk of bias within studies
The risk of bias evaluation is displayed in Figure 13. Overall, the summary figure of risk of bias indicated a low risk of bias across studies (as indicated in the grey area). However, several studies did not report the method of randomisation 2, 3, 5, 8 , performance bias 5, 7 or detection bias 1, 2, 4, 6, 7 . Whilst all

F I G U R E 1 3 Risk of bias graph across studies
studies discussed the attrition rates and possible reasons for participant dropouts, they indicated a low risk of bias for reporting bias, selection bias, and other sources of bias.

Risk of bias across studies
Due to the limited number of studies included in the present meta-analysis, this study was not powered to assess publication bias across studies (Borenstein et al., 2021).

Compassion-based interventions on self-compassion
This is the first meta-analysis to explore the efficacy and cross-cultural applicability of compassionbased interventions in Asian populations. This paper incorporated eight RCT studies, which gathered data from 1012 participants from Thailand, Japan and China (including Hong Kong) to answer the research question, 'can compassion-based interventions increase compassion in people living in Asian communities?'. The secondary aim was to investigate whether these eight studies also reduced the additional outcomes of depression, anxiety, and stress.
In relation to the first aim, results indicated that the compassion-based interventions were effective in increasing self-compassion among people in the Asian cultural context. However, several analyses displayed significantly large heterogeneity, reducing the confidence in the findings of the interventions included. Three studies were relatively different from the other studies across several characteristics such as study focus (Guo et al., 2020), intervention type (Mak et al., 2018) and intervention tasks (Wong & Mak, 2016). It was proposed that these studies may have contributed to the overall large variability, and when sensitivity analyses were carried out removing these studies from the analyses, the heterogeneity of studies became non-significant, suggesting that the remaining studies had consistent findings.
Similarly, when sensitivity analyses were conducted separately for studies with WLC groups and AC groups, significant effects were reported. As predicted, effect sizes of studies including AC groups were lower when compared to studies with WLC groups (Cuijpers et al., 2016;Kirby, 2017). This implies that the AC interventions may have also increased self-compassion in participants to some extent (Kirby, 2017). This raises the question whether the AC interventions also incorporated compassion-enhancing tasks, or whether engaging in any intervention increases self-compassion in general (Bishop et al., 2015).
Analyses were also conducted to investigate whether the impact of compassion-based interventions varied between clinical and non-clinical Asian populations. Results indicated that the compassion-based interventions were similarly effective in both participant groups and when the three papers causing variability were omitted during sensitivity analyses (Guo et al., 2020;Mak et al., 2018;Wong & Mak, 2016), the heterogeneity of studies was non-significant. This supports the view that compassion-based interventions are effective when used with a wide range of participants across the Asian context (Arimitsu, 2016). However, it is important to acknowledge that these findings were based on a limited number of eight papers altogether with only three papers conducted in clinical groups. In addition, these papers were conducted across only three countries from the Southeast Asia (Thailand) and East Asia (China, Japan). Consequently, the findings of this metaanalysis are not sufficient to draw conclusions for people in all Asian cultural contexts, and therefore future research needs be conducted in various Asian settings and participant groups (clinical, nonclinical, student populations, public, etc.). Kirby (2017) concluded in their meta-analysis that there is a lack of clarity in relation to what is the most appropriate measure of self-compassion. This lack of clarity is still evident in the current metaanalysis, as most of the studies included used the SCS to measure self-compassion, with some of them acknowledging the criticisms of the scale (e.g. Arimitsu, 2016;Huang et al., 2021). The Compassionate Engagement and Action Scales (CEAS: Gilbert et al., 2017) was developed to measure all three flows of compassion (self, to/from) whilst also addressing issues surrounding the SCS. Therefore, it seems fair to propose that the CEAS may be a more appropriate measure of compassion. However, none of the papers included in this meta-analysis used the CEAS, although other recent research using this measure in Asian countries such as Japan (Asano et al., 2020) and Sri Lanka (Kariyawasam et al., 2021(Kariyawasam et al., , 2022 have emphasised the advantages of using this measure (Asano et al., 2020). Thus, it would be useful for future research to investigate the effectiveness of compassion-based interventions using the CEAS in Asian communities to further understand the interplay between the three flows of compassion and well-being.

Compassion-based interventions on secondary outcomes
In addition to the primary outcome of self-compassion, this study also explored whether compassionbased interventions improved on the secondary outcome measures of depression, anxiety and stress. There was no significant impact of the compassion-based interventions on depression and anxiety. This contradicts with previous meta-analyses, which concluded that compassion-based interventions were effective in reducing depression and anxiety (Kirby 2017;Ferrari et al., 2019). The present meta-analysis indicated a large heterogeneity of study results on depression and anxiety constructs. In addition, although five studies tested intervention efficacy, one study (Guo et al., 2020) did not report on the results and therefore, could not be included in this analysis. On the other hand, there was a significant small effect of the interventions on stress whilst there was no significant heterogeneity in the study results. Although the individual studies reporting these results indicated that the interventions did in fact improve depression, anxiety and stress, only a few studies of the eight papers investigated these constructs and the large variability in the previously mentioned studies may have resulted in the non-significant impact on depression and anxiety, and the small significant effect on stress. Therefore, it is difficult to determine the impact of compassion-based interventions on these outcomes and further studies need to be explored to draw conclusions.

Strengths and limitations
A strength of this meta-analysis is that it only included papers with a specific focus on compassion cultivation and excluded interventions that prioritised other elements such as mindfulness (e.g. Mindfulness Based Stress Reduction, Mindfulness Based Cognitive Therapy). One paper used in this meta-analysis, however (Huang et al., 2021), based their intervention on both MSC (developed by Neff & Germer, 2013) and CMT (developed by Gilbert, 2009) approaches. It was therefore not possible to differentiate whether self-compassion increased due to the mindfulness element, the compassion element or both, when assessing the efficacy of compassion interventions based on this integrated approach (Kirby 2017).
Overall, this review highlights that the lack of research exploring the effectiveness of compassionbased interventions in Eastern cultures, as there were only three Asian countries that have researched this area so far. This was surprising given that many Asian countries such as Sri Lanka are familiar with compassion practice due to the significant Buddhist influence in such cultures (Kariyawasam et al., 2021(Kariyawasam et al., , 2022. Whilst the rigorous inclusion criteria helped to choose studies with higher methodological quality, this was also a limitation of this meta-analysis. This was because it lead to the exclusion of several studies which were either not RCTs despite being compassion-based interventions to promote compassion (e.g. Finlay-Jones et al., 2018;Noh & Cho, 2020;Yeung et al., 2022) or did not include compassion measures to assess the intervention efficacy (e.g. Lo et al., 2015).
Another limitation this meta-analysis discovered was that despite a wide search strategy, all the included studies only assessed self-compassion, disregarding the other two flows, namely compassion to others, and compassion from others . Ferrari et al. (2019) also narrowed their search to self-compassion-based interventions only, when conducting their meta-analysis of compassion-based interventions. Gilbert et al. (2017) argued that compassion is not only felt for the self but also towards and from others. Studies have discussed how these flows interact with one another (Rashid et al., 2021) and are also linked with increased well-being (Asano et al., 2020;Gilbert et al., 2017).
In addition, findings of the studies included in this review were mostly limited to non-clinical populations (e.g. Arimitsu, 2016;Guan et al., 2021), indicating the need for further research among both clinical and non-clinical populations in more Asian countries. The majority of the studies (five papers) collected data from university students, which also brings into question the generalisability of the findings to the larger context. Despite the common use of small underpowered sample sizes in many evidence-based interventions (Kirby 2017), this review noted that the papers generally included a small sample size, which also limits the generalisability of the findings. The limited number of RCTs meant that a funnel plot was not suitable to assess the risk of publication bias (Higgins et al., 2011).
Similar to Kirby (2017), this meta-analysis did not find studies that assessed compassion using heart rate variability and other bodily measures that would have increased the understanding of intervention efficacy (Luo et al., 2018;Tian et al., 2020) at a physiological level. Furthermore, the RCTs included a range of self-reported levels of depression, anxiety and stress measures, which may be particularly problematic due to the stigma of mental illness in Asian cultures (Wong & Mak, 2016). It seems fair to propose that future research should focus on using physiological measures (Finlay-Jones et al., 2018) in addition to self-report measures to help build a comprehensive understanding of the efficacy of compassion-based approaches.
The findings of this meta-analysis should be interpreted with caution when considering the impact of compassion-based interventions in Asian communities, as this meta-analysis was limited to only eight studies which were gathered from participants across three Asian countries. Therefore, whilst this was the first attempt to conduct a meta-analysis for people in Asian communities, this study cannot be generalised to all people in the Asian cultural context.
Additionally, this meta-analysis comprised studies that varied in multiple components such as the intervention duration, content and targeted population. Thus, prospective interventions could investigate the content and structure of compassion-based interventions, to determine the most suitable intervention for their targeted populations.

Clinical implications
This review found potentially positive effects on self-compassion when using compassion-based interventions in Asian populations. Both online (e.g. Mak et al., 2018;Wong & Mak, 2016) and in-person approaches (e.g. Huang et al., 2021;Tung, 2020) were found to be effective in increasing self-compassion and well-being and in reducing negative affect in Asian communities. Given that many people from Asian backgrounds do not seek help for their emotional well-being due to high shame and criticism, stigma and other help-seeking barriers in their societies (Kariyawasam et al., 2022;Mak et al., 2019), it seems fair to propose that online interventions such as the Self-Compassion App (Beaumont & Irons, 2021) maybe more appropriate for people in Asian communities. This would also reduce the need and cost of training clinicians or other specialist staff to deliver compassion-based approaches in cultures where there are high levels of poverty and limited funding for mental health clinics. Online approaches are found to be more interactive, cost-effective, quick, scalable and convenient (Chi, 2013;Mak et al., 2018). In addition, with the rapid increase of mental health complications due to the recent COVID-19 pandemic (Pfefferbaum & North, 2020), and with the new working from home environment, online interventions for facilitating well-being would be particularly convenient, timely and effective. Thus, the applicability and efficacy of online compassion-based compassion approaches are avenues of future research.

CONC LUSION
This meta-analysis explored the efficacy of compassion-based interventions in Asian populations to increase levels of compassion. The results suggested that compassion interventions increased self-compassion in clinical and non-clinical samples, providing evidence for the trans-diagnostic (Anuwatgasem et al., 2020) and cross-cultural application of these approaches (Tung, 2020). Selfcompassion increased with significant effect sizes, in studies with WLC groups when compared to studies with AC groups, indicating that active-control conditions may have also increased self-compassion. Prospective studies are encouraged to develop interventions, carefully selecting appropriate measures and assessing physiological changes to obtain outcomes that are more informed. Although there were several limitations including the limited number of studies and sample sizes, this meta-analysis encourages the use of compassion-based interventions to promote compassion and well-being in Asian communities.

AC K NOW L E DGE M E N T S
We would like to thank the two volunteer research assistants, namely Viktoria Venkatess and Kayleigh Schurer for their support in conducting the systematic search. his study received no external funding.

C ON F L IC T OF I N T ER E S T
The authors declare that there are no conflicts of interest.

DATA AVA I L A BI L I T Y S TAT E M E N T
Data sharing is not applicable to this article as no data sets were generated or analysed during the current study.