Health and Disability The relationship between self-harm and alexithymia: A systematic review and meta-analysis

.

Alexithymia has also been found to be significantly higher among people with a history of self-harm (Greene, Boyes & Hasking, 2020;Norman & Borrill, 2015), defined for the purpose of this review as any act of self-injury without explicit suicidal intent. Self-harm is a major and growing public health concern (McManus, Gunnell, Cooper, Bebbington, Howard & Brugha, 2019;Pilling, Smith & Roth, 2018) and a significant risk factor for subsequent completed suicide (Carroll, Metcalfe & Gunnell, 2014;Hawton, Bergen, Cooper, Turnbull, Waters, Ness & Kapur, 2015). It has been estimated to account for over 200,000 hospital presentations per year in England alone, at a cost of £128m (Tsiachristas, Geulayov, Casey, Ness, Waters & Clements, 2020), with the number of incidences of self-harm in the community up to 10 times higher (Geulayov, Casey, McDonald, Foster, Pritchard & Wells, 2018). Estimates of the prevalence of self-harm vary because of differences in methodology and definition, but one review calculated pooled prevalence of non-suicidal self-injury as 17.2% in adolescent samples, 13.4% among young adults, and 5.5% among older adult samples (Swannell, Martin, Page, Hasking & St John, 2014). In psychiatric clinical samples, reported rates of non-suicidal self-injury range from 11 to 51% in adults and 45 to 81% in adolescents (Cipriano, Cella & Cotrufo, 2017).
People who self-harm have been found to have difficulties in emotion regulation (Wolff, Thompson, Thomas, Nesi, Bettis & Ransford, 2019) and the empirical and theoretical literature positions self-harm as a means of regulating unwelcome emotional experience (Chapman, Gratz & Brown, 2006;Klonsky, 2007;McKenzie & Gross, 2014). The association between selfharm and alexithymia may be due in part to a lack of recourse to more adaptive regulation strategies. People with high levels of alexithymia exhibit poor emotion regulation (Stasiewicz, Bradizza, Gudleski, Coffey, Schlauch & Bailey, 2012;Taylor, 2000;Venta, Hart & Sharp, 2013) and are more likely to use suppressive regulation strategies than reappraisal strategies (Swart, Kortekaas & Aleman, 2009). Using Gross (2015)'s process model, Preece, Becerra, Allan, Robinson, and Dandy (2017) have proposed that alexithymia consists of difficulties at the attention and appraisal stages of emotion regulation, driven by underdeveloped emotional schema (Lane & Schwartz, 1987) and a tendency toward avoidance of emotions (Panayiotou, Leonidou, Constantinou et al., 2015). Understanding the relationship between alexithymia and self-harm, therefore, may help to inform clinical interventions based on improving emotion regulation strategies. Originally seen as a stable trait, the evidence now suggests that alexithymia can be modified through targeted interventions (Norman, Marzano, Coulson & Oskis, 2019) which makes it a worthwhile focus of study in the context of self-harm.
A narrative review of the literature found a significant relationship between self-harm and alexithymia, particularly among women (Norman & Borrill, 2015). The relationship appeared to be driven by the alexithymia subcomponents difficulty identifying and describing feelings, rather than externallyorientated thinking. Interest in the subject continues to grow and more relevant studies have been published during the subsequent years. A recent meta-analysis found significant associations between lifetime non-suicidal self-injury (NSSI) and alexithymia (r = 0.25, Greene et al., 2020). However, the authors only included studies meeting the International Society for the Study of Self-Injury's (2018) definition of NSSI as "deliberate damage to body tissue without suicidal intent for reasons not culturally or socially sanctioned." While this approach has the advantage of definitional clarity, relevant evidence may have been missed. Historically, a range of other terms for self-harm have been used, including, but not limited to, para-suicide, deliberate self-harm, auto-destructive behavior and self-mutilation, and motivation has not always been explicitly defined as non-suicidal. This reflects evidence that the reason for self-harm is not always clear, either to the individual or to a clinician (Grandclerc, De Labrouhe, Spodenkiewicz, Lachal & Moro, 2016). Motivations may change between incidences of self-harm by the same person, or even within a single incident of self-harm (Kapur, Cooper, O'Connor & Hawton, 2013). Muehlenkamp, Claes, Havertape, and Plener (2012) found that the term used to define self-harm (deliberate self-harm versus non-suicidal self-injury) did not affect reported prevalence rates and concluded that they were measuring similar phenomena. For this reason, the current review extends the search of the literature beyond a narrow focus on NSSI to include studies that have investigated deliberate self-harm, where the motivation for the behavior is not specified. Studies that are explicitly and exclusively focused on suicide, however, are excluded, to avoid duplication of a recent meta-analysis (Hemming, Taylor, Haddock, Shaw & Pratt, 2019), which identified an effect size of r = 0.25 in the relationship between alexithymia and suicidal behavior.

Aim of the study
The aim of this study, therefore, is to synthesize the evidence concerning the relationship between self-harm and alexithymia, including its subcomponents. Self-harm is defined, for the purpose of this review, as any act of self-injury, with the exception of those which are explicitly suicidal. It is hypothesized that there will be a significant, positive relationship between self-harm and alexithymia.
A planned subgroup analyses will investigate whether the effect size of the relationship is affected by the definition of self-harm (NSSI versus a broader definition of self-harm in which motivation is not specified, and lifetime versus current self-harm). It is expected that the definition of self-harm will not be a significant moderator (Muehlenkamp et al., 2012) but that recent self-harm may be more strongly associated with alexithymia than lifetime self-harm (Greene et al., 2020). Additional subgroup analyses will test the moderating effect of gender, age and clinical versus community samples. Norman and Borrill (2015) found stronger evidence for a significant relationship between alexithymia and selfharm among women than among men. It is therefore expected that gender will be a significant moderator. As noted above, prevalence rates of both self-harm and alexithymia tend to be higher in adolescent versus adult samples, and in clinical versus community samples. Greene et al. (2020) found age, but not the sample type (clinical versus non-clinical) to be a significant moderator of the relationship between NSSI and alexithymia. This review will test these findings using a broader definition of self-harm.

Databases and search terms
The review was pre-registered with PROSPERO International prospective register of systematic reviews (CRD42018118305). Searches of six databases (PsycINFO; Medline; Web of Science; PubMed; CINAHL; and Cochrane Central Register of Controlled Trials [CENTRAL]) were conducted for the final time on 25 November 2019. Titles, abstracts and keywords were searched for alexithymia (alexithymi*) combined using the Boolean operator AND with synonyms for self-harm (suicid* OR "attempt* suicide" OR overdos* OR parasuicid* OR para-suicid* OR self-harm* OR selfharm* OR "deliberate self-harm" OR "DSH" OR self-injur* OR selfinjur* OR "nonsuicidal self-injur*" OR NSSI OR self-mutilat* OR selfmutilat* OR selfdestruct* OR selfdestruct* OR self-inflict* OR selfinflict* OR self-poison* OR selfpoison* OR self-immolat* OR selfimmolat* OR automutilat* or auto-mutilat* OR self-cut* or selfcut* OR autodestruct* or auto-destruct* OR "self-injurious behavio*" OR self-burn* OR selfburn).
• a validated measure of alexithymia must be used; • studies must include a measure of self-harm. No restriction was placed on the way in which self-harm was measured, other than to exclude any study which is explicitly and solely concerned with suicide. Studies which did not specify motivation, or which measured non-suicidal selfharm alongside (but separately from) suicide were included; • studies must report a statistical assessment of the relationship between alexithymia and self-harm, or sufficient data to allow such an assessment to be made; and • additionally, to be included in the meta-analysis, study authors needed to report or provide sufficient data to enable an effect size of the relationship between alexithymia and the presence or absence of selfharm to be calculated.
Abstracts were screened separately by two researchers and disagreements resolved through discussion.

Data extraction
Data were extracted by the lead researcher. Means and standard deviations for alexithymia, and any reported subscales, for participants with and without experience of self-harm were recorded. If these were not available, correlation statistics describing the relationship between alexithymia and self-harm were extracted. The extraction also included sample size and characteristics, and the scales used to measure alexithymia and self-harm.

Quality Assessment
The studies were checked for risk of bias using the AXIS Appraisal Tool for Cross-Sectional Studies (Downes, Brennan, Williams & Dean, 2016). The quality assessment was carried out by the lead reviewer. A second reviewer independently checked 20% of the studies and the results were compared. The tool does not provide a single, quantitative assessment of quality; rather it is designed to be used as a guide to inform interpretation of the results. It prompts the reviewer to consider, for each study, whether the aims are clear, whether the method is robust and described sufficiently to enable replication, and whether the results are complete and internally consistent. It also contains questions about ethics and conflicts of interest.

Data analysis
The meta-analyses for total alexithymia and each subscale were based on Borenstein, Hedges, Higgins, and Rothstein (2009) and calculations were made using the excel workbooks provided by Suurmond, van Rhee and Hak (2017). Because the studies reported different statistical tests, Hedges' g was used as the common effect size, with a 95% confidence interval (CI). A P value of less than 0.05 and a 95% CI that did not cross the line of no effect was interpreted as statistically significant. Effect sizes of 0.20, 0.50, and 0.80 were considered small, medium and large respectively (Cohen, 1992). For individual studies Hedges' g was derived from the mean difference in alexithymia using pooled standard deviation to account for differences in sample sizes. Where the means and standard deviations were not reported, Hedges' g was derived from the correlation statistic Pearson's r. The calculations of Hedges' g and standard errors were made using equations set out in Borenstein et al. (2009). To test whether the overall results were affected by the deriving of Hedges' g from a correlation statistic rather than the underlying means, a sensitivity analysis was conducted to test the effect of removing those studies reporting correlational data.
A random effect meta-analysis was conducted because it provides a more conservative estimate of the effect size, allowing for the fact that the effect size in samples with different characteristics (such as age or gender) might differ from the "true" effect size across the whole population (Borenstein et al., 2009). In a random effects meta-analysis, betweenstudies variance can affect statistical power (Borenstein et al., 2009) and thus, based on Jackson and Turner (2017), a minimum of five studies was set to ensure sufficient power was achieved. Heterogeneity was measured using the I 2 statistic, which describes the percentage of variation that can be attributed to differences between the studies. I 2 of less than 40% was interpreted as low heterogeneity, while I 2 of over 75% was taken to indicate considerable heterogeneity (Higgins & Green, 2011). Publication bias was checked visually using a funnel plot, and statistically using Rosenthal's fail-safe N (Rosenthal, 1979) and Begg and Muzumdar's test of bias (Begg & Mazumdar, 1994). If necessary, the trim and fill method was used to adjust for any bias (Duval & Tweedie, 2000).
Subgroup analyses were planned to examine differences in effect sizes according to: (1) adolescent (mean age ≤ 18), young adult (18-29), and adult samples (≥ 30); (2) male versus female samples; (3) clinical versus community samples; (4) lifetime versus recent self-harm; and (5) NSSI versus a broader definition of self-harm. Studies were included in the subgroup analysis if the number of participants per subgroup exceeded n = 10. Between-study variance (Ͳ 2 ) was computed separately for each subgroup, or pooled if subgroups contained fewer than five studies (Borenstein et al., 2009).
A total of 31 studies met the criteria for inclusion. Twentyseven studies provided sufficient data to be included in the metaanalysis.

Quality assessment
The studies were found to be generally good quality. There were two areas of weakness, common to the majority of studies. First, it was rare for the studies to justify whether the sample size enabled the study to be sufficiently powered. Second, only a minority of studies analyzed non-responders or missing data. This may introduce bias if the participants choosing not to respond share certain characteristics. No study was excluded from the review on grounds of quality.
Twenty studies clearly defined self-harm in a manner consistent with the International Society for the Study of Self-Injury (2018). These studies defined self-harm as NSSI in the introduction and either used a validated and accepted measure of NSSI (most commonly the Deliberate Self-harm Inventory [Gratz, 2001], used in eight studies) or were explicit that their instructions to participants had defined self-harm as without suicidal intent. A further four studies measured self-harm as well as, and as distinct from, suicide attempts, for example, through clinical assessment. It was not apparent in the remaining seven studies that the definition of self-harm given to participants excluded suicidal intent. In addition, there were differences in the type of methods included in the definition of self-harm. Some studies restricted self-harm to superficial body tissue damage (excluding, e,g., taking an overdose of pills or swallowing dangerous substances), including two studies which focused only on self-cutting (Lambert & de Man, 2007;Laukkanen, Rissanen, Tolmunen, Kylm€ a & Hintikka, 2013). Studies assessed the correlation between the frequency of selfharm and alexithymia but did not include a comparison group of participants with no selfharm (n=2) Studies reported a statistical comparison but did not report enough data to enable an effect size to be calculated (n=2)  Relationship between self-harm and alexithymia 859

Qualitative review
Scand J Psychol 61 (2020) only a subgroup of participants completed the TAS20 and it was not stated how many of these had self-harmed. Proportions have therefore been estimated using the proportion who had selfharmed in the whole sample.
Non-clinical adolescent samples. Eight studies looked at adolescent participants recruited from school settings. Seven studies reported significant and positive associations between selfharm and total alexithymia (Garisch & Wilson, 2010, 2015 (2018), did not measure total alexithymia but reported significant correlations between self-harm and the TAS20 subcomponents DIF and DDF. Garisch and Wilson (2015) was the only longitudinal study identified in this review. They reported that, not only was selfharm significantly correlated with alexithymia scores at baseline, but also that initial alexithymia scores predicted self-harm during the 5 months between baseline and follow-up.
Clinical adolescent samples. Four studies recruited adolescent participants from a clinical setting (Cerutti, Calabrese & Valastro, 2014;Lambert & de Man, 2007;L€ udtke et al., 2016). All reported a significant relationship between total alexithymia and self-harm. While two of these studies used an exclusively clinical sample (Cerutti et al., 2014;L€ udtke et al., 2016), the other two studies compared a group of adolescents with a history of self-harm, recruited in clinical settings, with a control group of adolescents with no self-harm, recruited from the community Lambert & de Man, 2007). In both cases, alexithymia (total TAS20) was significantly higher among participants with a history of self-harm.
University student samples. Of the eight studies that were based on university student samples, all reported significant and positive associations between self-harm and alexithymia (total and/or subscales) (Anderson & Crowther, 2012;Borrill, Fox, Flynn & Roger, 2009;Greene et al., 2019;Hasking & Claes, 2019;Oskis & Borrill, 2019;Paivio & McCulloch, 2004;Polk & Liss, 2007;Wester & King, 2018). Among those that reported the subcomponents of alexithymia, all found DIF to be significantly higher in those with a history of self-harm (Anderson & Crowther, 2012;Borrill et al., 2009;Greene et al., 2019;Oskis & Borrill, 2019). The findings regarding DDF were more mixed with two studies finding a significant positive association (Borrill et al., 2009;Greene et al., 2019) and one reporting a non-significant result (Oskis & Borrill, 2019), while all those which measured EOT reported a non-significant relationship with self-harm. Polk and Liss (2007) found that DIF and DDF, together with anxiety, sleep disturbance and sexual and emotional abuse were strongly correlated with a discriminant function which distinguished between a group of university students with no history of self-harm and a group of internet users who had self-harmed.
Adult clinical samples. The remaining 11 studies with adult participants drew on clinical samples. Here, the results were more varied. Of the two studies with participants with Borderline Personality Disorder (BPD), one reported that total alexithymia and all three subscales were significantly correlated with the frequency of self-harm (Mojahed, Rajabi, Khanjani & Basharpoor, 2018), while Sleuwaegen et al. (2017) found that only DDF was correlated with self-harm frequency. A further four studies focused on participants with substance dependency. Bolognini, Plancherel, Laget, St ephan, and Halfon (2003) and Evren and Evren (2005) reported significantly higher total alexithymia among those with a history of self-harm. In contrast, Oyefeso et al. (2008) compared treatment-seeking opiate addicts with and without a history of self-harm and reported a significant difference only in DIF and not in the total alexithymia score, DDF or EOT. Similarly, Verrocchio et al. (2010) found no significant relationship between self-harm and total alexithymia or any of the subscales among a group of substance dependent men.
Bedi, Muller, and Classen (2014) recruited women attending a day treatment program for survivors of abuse. They found total alexithymia to be significantly higher among those participants with a history of self-harm. These results are consistent with Zlotnick et al. (1996), which found significantly higher alexithymia in women psychiatric inpatients who had self-harmed, a high proportion of whom had suffered childhood sexual abuse. In contrast, Osuch et al. (2014) measured alexithymia and selfharm in a small (N = 32), sample of young adults with mood and/ or anxiety disorders and reported no significant differences in total alexithymia or any of the subscales between those participants with and without a history of self-harm.
Of the remaining two studies, Hsu et al., (2013) compared participants admitted to a hospital emergency room following a first-time incident of self-harm or suicide attempt (measured separately) with a control group of chronic pain outpatients. DIF was found to be significantly higher among those participants who had engaged in self-harm compared to the control group. The final study focused on adults with autism (Moseley, Gregory, Smith, Allison & Baron-Cohen, 2019). In a logistic regression model, alexithymia scores were able to differentiate significantly between people who had self-harmed within the past year and those who had never self-harmed, but not between those who had never selfharmed and those who last self-harmed over a year ago.

Quantitative analysis
Twenty-three studies with a combined sample of 8,724 were included in a meta-analysis to assess the scale of the difference in alexithymia between participants who had engaged in self-harm and people who had never self-harmed (Fig. 2). The combined studies had a medium effect size of g = 0.57 (95% CI 0.46-0.69). The overall effect was significant (Z = 10.57, p < 0.001) indicating that participants who had self-harmed had significantly higher alexithymia than participants who had never self-harmed. The heterogeneity was high (I 2 = 70.2%). A sensitivity analysis was conducted to test the effect of removing the four studies that reported the correlation between alexithymia and self-harm rather than the mean alexithymia score for those with and without a history of self-harm. Their removal made very little difference to the effect size (g = 0.58, 95% CI 0.45-0.71) although heterogeneity was reduced (I 2 = 61.6%).
The funnel plot (Fig. 3), Rosenthal's fail-safe N test, and Begg and Mazumbar's test (p = 0.206) indicated that publication bias was not a concern. Applying the trim and fill method to impute hypothetically unpublished studies made no difference to the effect size or confidence intervals.
Planned subgroup analyses of the relationship between total alexithymia and self-harm were conducted based on the demographics of the sample and the definition of self-harm (Table 2). A significantly larger effect size was observed in adolescent samples (M age ≤ 18) compared with adult samples, although when the adults samples were further subdivided into young adults and older adults the variance in effect sizes was not significant. There was a significantly larger effect size in female samples compared to male samples. The relationship between alexithymia and self-harm was significant in both male and female samples, although the effect size for men was small. The result of the gender subgroup analysis should, however, be interpreted cautiously, because it is based on only the eight studies that reported, or provided, data disaggregated by gender. The relationship between alexithymia and self-harm was significant in both clinical and community samples and the size of the effect was not significantly different between the two groups. There was no significant difference in the combined effect size of studies measuring lifetime self-harm, compared to those measuring recent self-harm. Similarly, there was no significant difference in the effect size between studies that defined self-harm as NSSI or clearly distinguished self-harm from suicide, compared with those which did not specify motivation.
Meta-analyses were conducted on the TAS20 subscales, where reported. Figures 4 and 5 illustrate that there was a significant relationship between self-harm and the subscales DIF and DDF with medium effect sizes (DIF g = 0.61, 95% CI 0.45-0.76, SE = 0.07, I 2 = 50.8%; DDF g = 0.41, 95% CI 0.29-0.53, I 2 = 25.7%, SE = 0.06). The confidence interval around the effect size for the subscale EOT crossed the line of no effect, indicating that it was not significant, and the effect size was small (g = 0.10, 95% CI À0.11-0.31, I 2 = 64.2%, SE = 0.10 Fig. 6). Begg and Mazumbar's test and Rosenthal's fail-safe N test indicated that publication bias was not a concern for DIF and DDF. However, for EOT, Rosenthal's fail-safe N test suggested that there might be publication bias, although Begg and Mazumbar's test was not significant (p = 0.225). Subgroup analyses for each subscale are set out in Table 3, although the small numbers in some of the subsets mean that these results should be interpreted with caution. The effect size in the relationship between DIF and self-harm was significantly moderated by age, with higher effect sizes observed in adolescent and young adult samples than in older adults (p = 0.021). The effect size of the relationship between DIF and self-harm was also significantly higher in community samples compared with clinical samples (p = 0.003).

DISCUSSION
The current meta-analysis found an overall medium effect size of the relationship between alexithymia and self-harm, indicating that alexithymia is significantly higher in people who have engaged in self-harm than in people who have not. The relationship is driven by the DIF and DDF subscales, while the relationship between self-harm and EOT was not significant. Specifying the motivation for self-harm as non-suicidal did not significantly alter the effect size. Similarly, there was no significant difference in the effect size between those studies that measured lifetime self-harm and those that measured recent selfharm. The effect size was significantly larger in female than in male samples and in adolescent than in adult samples.

Definition of self-harm
The effect size of the relationship between alexithymia and selfharm reported in the current review is similar to that derived by Greene et al. (2020) in their review of alexithymia and, specifically, NSSI (r = 0.25). Furthermore, the relationship between suicide and alexithymia has also been found to be of a similar effect size (r = 0.25; Hemming et al., 2019) although a separate, narrative review of alexithymia and suicidality highlighted conflicting findings between individual studies (Davey, Halberstadt, Bell & Collings, 2018    Four studies (Garisch & Wilson, 2015;Moseley et al., 2019;Sleuwaegen et al., 2017;Wester & King, 2018) reported, or provided, separate data on lifetime and recent self-harm. The data reported here include these studies in the lifetime self-harm group. When their data were included in the recent selfharm subgroup, the difference in the effect size of the relationship between self-harm and alexithymia in studies measuring lifetime versus those measuring recent self-harm remained non-significant (p = 0.131). c Between-study variance was calculated separately when there were at least five studies per subgroup and pooled where groups contained fewer than five studies (Borenstein et al., 2009). distinction in the relationship between alexithymia and the range of self-harming behaviors may indicate that, to some extent, the studies are drawing on the same population. A recent study found the prevalence of suicide attempts among people with a history of NSSI to be 40% (O'Connor, Wetherall, Cleare, Eschle, Drummond & Ferguson, 2018). Only one of the studies in the current review which focused on engagement in NSSI explicitly excluded individuals who had also attempted suicide. In addition, it may not always be possible for participants clearly to distinguish their motivation for self-harm (Grandclerc et al., 2016).

Gender
Although based only on eight studies, the result of the subgroup analysis provides strong support for a significant relationship  (2012), Sleuwaegen et al. (2017) and Wester and King (2018) also provided DIF scores for participants with recent (<12 months) self-harm. Using data from these studies relating instead to recent self-harm resulted in a slightly increased effect size of g=0.63, (95% CI 0.48 to 0.77), SE=0.07, I 2 = 42.97%.   *Sleuwaegen et al. (2017) and Wester and King (2018) provided DDF scores for parƟcipants with lifeƟme self-harm and with current self-harm. Data shown here relate to lifeƟme self-harm. Using data from these studies relaƟng instead to recent self-harm results in a slightly increased effect size of g=0.44 (95% CI 0.32 to 0.55) SE=0.05, I 2 = 19.13%. between alexithymia and self-harm in women. A small effect size was also observed in male samples. This finding can only be taken as indicative, given the small combined sample size and the fact that the majority of studies with mixed samples did not report the results by gender and therefore were excluded from this analysis. Given the continuing uncertainty about the relationship between self-harm and alexithymia in men, it would be helpful if future studies ensured sufficient sample sizes to allow the reporting of the results by gender.
It may be, however, that the finding that the relationship between alexithymia and self-harm has a larger effect size in women than in men reflects genuine gender differences. A metaanalysis has shown that men score higher on average than women on measures of alexithymia (Levant, Hall, Williams & Hasan, 2009) but are less likely than women to self-harm (Bresin & Schoenleber, 2015; see also Hawton et al., 2015). Men tend to use different methods of self-harm compared to women (Bresin & Schoenleber, 2015) and to self-harm for different reasons (Laye-Gindhu & Schonert-Reichl, 2005;Scoliers, Portzky, Madge, Hewitt, Hawton & de Wilde, 2009). For example, Rasmussen, Hawton, Philpott-Morgan, and O'Connor (2016) found that adolescent girls were more like to endorse wanting to die, and boys more likely to say that they wanted to frighten someone. The authors take this as evidence to suggest that, in adolescents at least, boys are more likely to have external motivations for selfharm than girls, which are perhaps less related to the ability to understand what it is they are feeling, as measured by the TAS20.

Age
There was also a significant difference in the effect size between adult and adolescent samples, with the size of the effect of the relationship between self-harm and alexithymia larger among adolescents than among adults. These results may be confounded by the predominance of clinical settings for adult samples, compared with a majority of community settings for the adolescent studies, a comparison of which is discussed below. Alexithymia scores tend to be higher in adolescent samples (Honkalampi, Tolmunen, Hintikka, Rissanen, Kylm€ a & Laukkanen, 2009;Oskis, Clow, Hucklebridge, Bifulco, Jacobs & Loveday, 2013). The TAS20 has been shown to be less reliable in children and young teenagers, with reliability increasing with age (Parker, Eastabrook, Keefer & Wood, 2010). It is possible that the features of alexithymia are mimicked in adolescents, who have not yet developed emotional awareness abilities, and that it is only in early adulthood that alexithymia itself can be measured as a stable personality trait. The early teenage years are also a common time for the onset of self-harm (Griffin, McMahon, McNicholas, Corcoran, Perry & Arensman, 2018;Morgan, Webb, Carr, Kontopantelis, Green & Chew-Graham, 2017). It may be that for adolescents rather more than for adults, self-harm is related to the difficulty in understanding emotions and talking about feelings. Unfortunately a systematic review of self-reported, non-suicidal reasons for self-harm found that the heterogeneity of the literature precluded any meaningful analysis of function by demographic characteristics, including age, so this remains an interesting area for future research (Edmondson, Brennan & House, 2016).

Community and clinical samples
Subgroup analysis revealed a larger effect size in those studies of community samples compared with clinical samples. The difference between the two groups was statistically significant in the subscale DIF but not total TAS20. A smaller effect size in clinical samples is unsurprising, given the evidence that levels of alexithymia are generally higher in clinical populations than in the general population (McGillivray et al., 2017). Alexithymia has been found to be higher among people with psychological disorders, such as depression (Honkalampi et al., 2000;Son et al., 2013). Among the studies in the current review, two found that depression mediated, at least partially, the relationship between  Verrocchio et al. (2010) -0.07 -0.39 0.25 9.84% 12 Wester & King (2018) 0.16 -0.10 0.42 11.10% 13 Total (95% CI) Hedges g 0.10 -0.11 0.31 Test of overall effect size Z = 1.06, p = 0.290 Heterogeneity I 2 = 64.17% Total number of parƟcipants = 2,070 *Sleuwaegen et al. (2017) and Wester and King (2018) provided EOT scores for parƟcipants with lifeƟme self-harm and with current self-harm. Data shown here relate to lifeƟme self-harm. Using data from these studies relaƟng instead to recent self-harm results in a slightly increased effect size of g=0.13 (95% CI -0.09 to 0.35), SE=0.10, I 2 = 68.90%. alexithymia and self-harm (Garisch & Wilson, 2015;Lambert & de Man, 2007). In contrast, Lee, (2016) found that alexithymia was a significant predictor of self-harm, independent of depression, and Sleuwaegen et al. (2017) observed that in their sample of BPD patients the relationship between self-harm and DDF (although not DIF or total TAS20) held, even controlling for depression. The finding of the current review, of a significant relationship between alexithymia and self-harm even in clinical settings, suggests that this is a relationship that is at least partially independent of other clinical symptoms.

Lifetime and recent self-harm
Studies that measured lifetime self-harm had a smaller combined effect size than those studies that measured current self-harm, although the difference was not statistically significant. It is hard to draw conclusions about the nature of the relationship between alexithymia and self-harm over time from these almost exclusively correlational studies. Only two studies distinguished between participants who had never self-harmed, those who last selfharmed over a year ago ("historic") and those who had self-harmed within the past year ("recent"). Anderson and Crowther (2012) found that, in their undergraduate sample, DIF scores were significantly lower among those who had never self-harmed than among those with either recent or historic self-harm. There was no significant difference in the DIF scores between participants with recent or historic self-harm. Moseley et al. (2019) reported a marginal (p = 0.53) difference between participants with recent and those with historic self-harm, but in a logistic regression alexithymia could not distinguish between participants who had never self-harmed and those with historic self-harm. If alexithymia were a stable trait, it would be expected that the relationship between alexithymia and past self-harm would be similar to that between alexithymia and recent self-harm. However, it is generally accepted that alexithymia has relative, rather than absolute, stability (Porcelli, Tulipani, Di Micco, Spedicato & Maiello, 2011) and may change over time, for example in relation to depression (Honkalampi, Hintikka, Laukkanen & Viinam€ aki, 2001) or as a result of treatment (Cameron, Ogrodniczuk & Hadjipavlou, 2014). In addition, the studies may to some degree be capturing secondary alexithymia which, like self-harm, may have developed in response to stressful life circumstances (de Vente, Kamphuis & Emmelkamp, 2006). Future research could usefully distinguish between recent and past engagement in self-harm, in order to extend understanding about the longitudinal relationship between alexithymia and self-harm.
The shortage of longitudinal studies in the review also precludes conclusions from being drawn as to whether alexithymia is a significant risk factor of self-harm (Kraemer, Kazdin, Offord, Kessler, Jensen & Kupfer, 1997). Those studies that conducted regression analyses found that alexithymia was a significant predictor of self-harm (e.g., L€ udtke et al., 2016) but this cannot be interpreted as causation. The only longitudinal study found that high alexithymia scores significantly predicted self-harm three months later in a community sample of adolescents (Garisch & Wilson, 2015). The authors conclude that alexithymia, in combination with low mood, may increase the likelihood of recourse to self-harm. However, more longitudinal Relationship between self-harm and alexithymia 871 Scand J Psychol 61 (2020) research is needed to confirm alexithymia as a risk factor. Indeed, although difficulties in emotion regulation (Buckholdt, Parra, Anestis, Lavender, Jobe-Shields, Tull & Gratz, 2015), emotion reactivity (Nock, Wedig, Holmberg & Hooley, 2008), and negative affect (Victor & Klonsky, 2014) are significant correlates of self-harm, meta-analyses have not identified them as significant risk factors of either NSSI (Fox, Franklin, Ribeiro, Kleiman, Bentley & Nock, 2015) or suicide (Franklin, Ribeiro, Fox, Bentley, Kleiman & Huang, 2017). In considering the disparity in the results between cross-sectional and longitudinal studies, Fox et al. (2015) hypothesize that emotion dysregulation, though not a significant risk factor on its own, may combine with other risk factors to increase vulnerability to self-harm and may also shed light on the function of self-harm as a means of relieving emotional distress (McKenzie & Gross, 2014). The same may be true of alexithymia, although neither of the meta-analyses cited here explicitly examined alexithymia as a risk factor (Fox et al., 2015;Franklin et al., 2017).

Model of self-harm
Many of the studies included in the review interpreted the association between alexithymia and self-harm as consistent with an affect regulation model of self-harm. According to this model, self-harm is conceptualized as a means of regulating unwanted emotional experience (Chapman et al., 2006), either to manage overwhelming emotion (Klonsky, 2007), or to feel something instead of feeling numb (Tolmunen, Rissanen, Hintikka et al., 2008). Only one of the studies in the current review analyzed the functions of self-harm in relation to alexithymia. Moseley et al. (2019) found that alexithymia was a significant predictor of participants' endorsement of NSSI as a means of regulating highenergy states, such as to relieve stress or pressure, or of communicating to others. This would appear to be consistent with the general finding of this review and Greene et al. (2020) that the relationship between self-harm and alexithymia is driven by difficulties in identifying and describing feelings which may hamper use of more adaptive regulatory strategies.

Clinical implications
Although alexithymia may be a barrier to psychological treatment (Ogrodniczuk, Piper & Joyce, 2011), there is evidence that alexithymia is modifiable if the intervention is targeted appropriately (Cameron et al., 2014). Treatments for self-harm, such as Dialectical Behavioral Therapy (Linehan, 1993), often include emotion awareness training, which may be particularly important for people with high alexithymia who struggle to understand what it is they are feeling. It has been suggested that labeling internal emotional states can act as a form of implicit emotion regulation (Torre & Lieberman, 2018). This may be the first step toward giving people the vocabulary to recognize and talk about feelings.

Limitations
Searches for the current review were limited to published data and articles published in English, which may have led to the exclusion of other relevant research. Furthermore, of the 31 studies which were identified as meeting the original search criteria, only 23 provided sufficient data to include in the meta-analysis of total alexithymia and self-harm, with a further four contributing to the analysis of the TAS20 subscales. The availability of genderspecific data was also patchy, with the result that these results require further replication. Heterogeneity between the studies was high. Although the review investigated whether specifying motivation for self-harm as non-suicidal affected the relationship between self-harm and alexithymia, there were other differences between the studies in the way in which self-harm was defined, which were too various to allow for further subgroup analyses. Definitions varied between single questions to multi-item lists, in which methods of self-harm are specified. Single question definitions have been shown to underestimate the prevalence of self-harm (Swannell et al., 2014), and therefore may not be comparable with validated measures of self-harm such as the DSHI (Gratz, 2001). In addition, while some studies used a continuous scale, taking into account the frequency of self-harm, most used a binary distinction between people who had never self-harmed, and people who had selfharmed at least once. There is evidence to suggest that frequency of self-harm is related to severity of psychological distress and that a single incident may not be comparable to habitual engagement in self-harm (Fox et al., 2015).
Alexithymia was consistently measured using the TAS20 or its predecessor the TAS26. This makes comparison between studies easier, but it relies on the scale adequately capturing the underlying trait. Other measures exist, reflecting alternative conceptions of alexithymia. For example, the Bermond-Vorst Alexithymia Questionnaire (BVAQ; Vorst & Bermond, 2001) distinguishes between a cognitive component, similar to that measured by the TAS20, and an affective component, comprising difficulties fantasizing and reduced ability to experience emotion. Both the TAS20 and the BVAQ are self-report scales, however, and, it has frequently been observed that asking people who struggle to identify their emotions to complete a questionnaire about their emotional experience is inherently problematic (Taylor, Bagby, Parker & James, 1997). It would be useful to test the findings of the current review using observer rated measures of alexithymia. Alternatively, building on evidence associating alexithymia with broader failures in interoception (awareness of bodily sensation), proxy measures, such as heartbeat detection tasks, may provide a more objective means of assessment (Brewer, Cook & Bird, 2016;Herbert, Herbert & Pollatos, 2011).

CONCLUSIONS
The current meta-analysis identified a medium effect size of the relationship between self-harm and alexithymia, particularly difficulty identifying and describing feelings, indicating that people with a history of self-harm score on average significantly higher on measures of alexithymia than people with no history of selfharm. The effect size of the relationship between self-harm and alexithymia was larger among women than men, and in adolescent than adult samples. Heterogeneity between studies was high, due perhaps to the disparity in the measures of self-harm. The results justify further investigation into why difficulty in identify and describing feelings should be associated with self-harm.

FUNDING
This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

CONFLICT OF INTEREST
The authors declare that they have no conflict of interest.

DATA AVAILABILITY STATEMENT
The data used in the meta-analysis were taken from published articles. In a few cases additional data were obtained from authors -these are listed in the results section. No new data were collected for this review study.