A typology of nonsuicidal self‐injury in a clinical sample: A latent class analysis

Abstract Nonsuicidal self‐injury(NSSI) is a behavioural concern and can present in diverse ways, varying by method, frequency, severity, function and so forth. The possible combinations of these features of NSSI produce an array of profiles that makes evaluation and management of this behaviour challenging. The aim of this study was to build upon previous work that reduces the heterogeneity of NSSI patterns by using latent class analysis (LCA) to identify a typology of NSSI. Participants consisted of 235 outpatients aged 14–35 years attending a tertiary psychiatric hospital in Singapore who had reported at least one NSSI behaviour within the last year. Eight indicators captured using the Functional Assessment of Self‐Mutilation were used in the LCA: frequency of NSSI, length of contemplation before engaging in NSSI, usage of more than three NSSI methods, suicidal ideation and four psychological functions of NSSI, that is, social‐positive, social‐negative, automatic‐positive and automatic‐negative. The LCA revealed three distinct groups: Class 1—Experimental/Mild NSSI, Class 2—Multiple functions NSSI/Low Suicide Ideation and Class 3—Multiplefunctions NSSI/Possible Suicide Ideation. Multinomial logistic regression analyses were conducted to examine the associations between class membership and sociodemographic variables as well as measures of emotion dysregulation, childhood trauma, depression and quality of life. Females were overrepresented in Class 3. In general, Class 3 had the poorest scores followed by Class 2. Our analyses suggest that different NSSI subtypes require different treatment indications. Profiling patterns of NSSI may be a potentially useful step in guiding treatment plans and strategies.


| INTRODUCTION
Nonsuicidal self-injury (NSSI) refers to the direct and deliberate destruction or alteration of bodily tissue in the absence of suicidal intent (Nock & Favazza, 2009). Common forms of NSSI include cutting, hitting, burning, scratching, biting and interfering with wound healing (Nock & Favazza, 2009). Once regarded as a symptom of borderline personality disorder, researchers and various organizations have advocated for the recognition of NSSI as an independent disorder in future editions of the Diagnostic and Statistical Manual of Mental Disorders (DSM). However, due to the poor reliability of the NSSI disorder criteria in clinical and field trials, NSSI appears instead in the section 'Conditions for Further Study' in the DSM-Fifth Edition as a distinct syndrome (Gratz, Dixon-Gordon, Chapman, & Tull, 2015;Zetterqvist, 2015). It is included as a condition 'on which future research is encouraged' , which is 'not intended for clinical use' (Plener & Fegert, 2015). This criterion has led to some agreement on how NSSI is defined and growing convergence in prevalence rates across studies. NSSI rates have been reported to be 1-4% in the adult general population, 15-35% among adolescents and young adults and 40-80% among psychiatric patients (Kerr, Muehlenkamp, & Turner, 2010).
The stereotypical self-injurer is often depicted as a young female who self-injures by cutting herself (Whitlock & Eckenrode, 2008) and does so to seek attention (Timson, Priest, & Clark-Carter, 2012). However, a number of studies have shown that males are just as likely to self-injure as females though the former are inclined towards different forms such as self-battery (Whitlock & Eckenrode, 2008). Although the motivation to influence others (e.g., elicit care and attention from significant others) may be a function of NSSI in some cases, affectregulation has been consistently shown to be the most prevalent function of self-injury. Furthermore, NSSI carried out for this purpose is often done in private and concealed from others. NSSI may also be performed in peer groups context for the purpose of sensation seeking (e.g., for excitement and exhilaration). To complicate matters, these different functions may be endorsed by the same individual (Klonsky & Muehlenkamp, 2007).
NSSI is in fact among the strongest predictors of future suicide attempts (Grandclerc et al., 2016). At the same time, only a minority of young adults who engage in NSSI actually engage in suicidal behaviour (Klonsky & Olino, 2008). Identifying individuals at risk of suicide plays a critical role in targeted prevention. Several studies have identified that more frequent engagement in NSSI, using multiple methods and using NSSI to fulfil different psychosocial functions, are associated with higher risk of suicide attempts (Andover & Gibb, 2010;Hamza & Willoughby, 2013). These findings support Joiner's interpersonal theory of suicide that NSSI and suicide are at different ends of the same continuum where greater involvement in NSSI increases an individual's acquired capability for suicide by habituating the individual to fear and pain associated with taking one's own life (Joiner, 2005).
As mentioned, NSSI can present in different ways, varying by method, frequency, severity, function and may be associated with an array of risk factors (childhood maltreatment, difficulties with emotion regulation) and psychopathology (major depressive disorder, borderline personality disorder) or the absence of these factors. The possible combinations of these factors produces a constellation of profiles of NSSI, and this multifariousness accounts for the difficulty in evaluating and managing the condition (Grandclerc et al., 2016). No current treatment for NSSI qualifies as empirically supported or efficacious (Turner, Austin, & Chapman, 2014). A better understanding of this phenomenon and an attempt to identify high-risk groups would make it more possible to optimize treatment, resource allocation and suicide prevention by identifying self-injurers who require more extensive treatment.
Reducing the heterogeneity of NSSI presentations by grouping individuals who engage in NSSI into clinically meaningful classes may aid our understanding of this condition and guide treatment plans. A small body of research has attempted to identify typologies of NSSI using a latent class analysis (LCA). LCA is a method for classifying heterogeneous individuals into homogeneous subgroups (Muthén & Muthén, 2000) based on patterns of traits and/or behaviours. One of the earliest studies was conducted by Klonsky and Olino (2008), who identified four groups of self-injurers. The groups were (1) the 'Experimental NSSI' group that experimented with NSSI on rare occasions, (2) the 'Mild NSSI' group that engaged in NSSI more frequently than the Experimental group, (3) the 'Multiple functions/Anxious NSSI' group that comprised members who used a variety of NSSI methods and endorsed several automatically and socially reinforcing functions

Key Practitioner Message
• Nonsuicidal self-injury (NSSI) is a strong predictor of suicide, though only minority of individuals who self-injure harbour suicidal intent. Patterns of NSSI can also be differentiated using indicators such as modality, frequency and psychological function.
• Three groups of individuals who self-injure were identified using latent class analysis: Class 1-Mild/Experimental, Class 2-Multiple functions NSSI/Low Suicide Ideation and Class 3-Multiple functions/Possible Suicide Ideation.
• These three classes have varying therapeutic needs. Thus, evaluating NSSI behaviour using such classification systems may be useful in guiding treatment plans and resource allocation.
of NSSI and (4) the 'Automatic-function/Suicidal NSSI' that was characterized by members who almost exclusively cut themselves for automatically reinforcing functions and had more severe clinical symptoms as well as suicidal ideation and behaviours. There have been several attempts to extend and replicate this work by including additional NSSI features (Dhingra, Boduszek, Palmer, & Shevlin, 2015) and other indicators such as childhood adversity (Vaughn, Salas-Wright, Underwood, & Gochez-Kerr, 2015) and psychiatric symptoms (Xin et al., 2016). Most of these studies were conducted with college students, and there is a dearth of research in clinical samples.
The current study was conducted in Singapore, a city state located at the tip of the Malay Peninsula in Southeast Asia. NSSI is prevalent among the youth population with a prevalence rate of 60% in a psychiatric sample (Shahwan et al., 2018) and 23% in a survey with 1,095 Singaporeans (Ho, 2019). Local government bodies have been concerned about youth suicide rates and have identified NSSI as an important risk factor (Ang, 2019). The aim of this study was to build upon previous work that had identified typologies of NSSI and examine the extent of generalizability in our local clinical population.

| Participants
Participants consisted of 235 outpatients of the Institute of Mental Health (IMH) aged between 14 and 35 years. The age range considered for our sample was based on two local definitions of youth: (1) the Children and Young Persons Act (1993), which defines a young person as 14-16 years, and (2) the National Youth Council, which defines youths as those aged 15-35 years (Singapore|Factsheets| Youthpolicy.org, n.d.). The IMH is Singapore's only tertiary-care psychiatric hospital. Data of participants who had reported at least one NSSI behaviour within the last year on the Functional Assessment of Self-Mutilation (FASM) were extracted from a larger study (Shahwan et al., 2018) for the current analysis.

| Procedure
Participants were recruited at the child and adult outpatient clinics in IMH between October 2015 and June 2016. Ethics approval was attained from the National Healthcare Group Domain Specific Review Board, and clinicians were informed of the study. Participants either volunteered by responding to posters situated in the outpatient clinics or were referred by clinicians. Patients were screened for intellectual disability as the requirements of the study required considerable insight, Primary 6 (equivalent to 6 years of schooling) reading level and the ability to retrospect. An intellectual disability diagnosis reflected in patient's medical records was a quick way to identify and exclude individuals who were most unlikely to be able to complete the survey adequately. Informed consent was obtained from participants prior to commencement of study procedures. Although a waiver of parental consent was obtained, the study was explained to parents of adolescents less than 21 years of age if they were present.
The research officers ensured that the self-administered questionnaires were completed independently by the participants.

| Measures
FASM (Lloyd-Richardson, Perrine, Dierker, & Kelley, 2007): NSSI behaviours were measured using a self-report checklist designed to assess the frequency of 11 self-harm behaviours (e.g., cutting, scratching and burning) within the last 12 months. For the purposes of the current study, frequencies of all self-harm behaviours were summed to indicate the total frequency of NSSIs. Participants were then categorized into three equal proportions based on tertiles: low (one to six times), moderate (seven to 24 times) and high (25 and above). The number of methods endorsed by the participant was also summed to obtain a total number of NSSI methods utilized. The FASM also consists of a checklist of 22 statements measuring the functions of NSSI (e.g., to avoid being with people) rated on a scale of 0 to 3 (0 = Never, 1 = Rarely, 2 = Some, 3 = Often). Previous literature has identified a four-factor structure of the FASM functions checklist: automatic-negative reinforcement (two items), automaticpositive reinforcement (three items), social-negative reinforcement (four items) and social-positive reinforcement (13 items). In the present study, the Cronbach α coefficients were 0.75, 0.65, 0.81 and 0.90, respectively. Participants who selected the response options Some or Often on any of the items corresponding to that factor were categorized as endorsing NSSI for that function (Lloyd-Richardson et al., 2007). (Gratz & Roemer, 2004): The DERS is a 36-itemself-report questionnaire that measures six aspects of emotion dysregulation: non-acceptance of emotional responses (non-acceptance; six items, α = 0.87), difficulties engaging in goal directed behaviour (goals; five items, α = 0.84), impulse control difficulties (impulse; six items, α = 0.85), lack of emotional awareness (awareness; six items; α = 0.82), limited access to emotion regulation strategies (strategies; eight items, α = 0.87) and lack of emotional clarity (clarity; five items, α = 0.80). Items are rated on a scale of 1 (almost never) to 5 (almost always), with higher DERS scores indicating greater difficulties with emotion regulation.

Difficulties in Emotion Regulation Scale (DERS)
Childhood Trauma Questionnaire (CTQ) (Bernstein & Fink, 1998): The CTQ is a 28-item retrospective self-report inventory with 25 items assessing the severity of five types of childhood trauma: emotional abuse, physical abuse, sexual abuse, emotional neglect and physical neglect. Responses were rated on a scale of 1 (never) to 5 (very often). The scores for each subscale ranges from 5 to 25, with higher scores indicating greater experiences of childhood trauma. Guidelines for the classification (none, low, moderate and severe) for each subscale were provided by the manual (Bernstein & Fink, 1998). The CTQ has demonstrated excellent psychometric rigour, with high internal consistency as whole (α = 0.97), as well as for each subscale (α = 0.81 to 0.95) (Bernstein, Ahluvalia, Pogge, & Handelsman, 1997). With regard to the present study, the Cronbach α coefficient as a whole was 0.92, whereas coefficients for the five factors were 0.84, 0.87, 0.93, 0.88 and 0.61, respectively. For the present study and to facilitate analyses and interpretation of results, scores on each subscale were further classified into three groups: none, low/moderate and severe.

| Statistical analysis
A LCA was utilized to identify subgroup heterogeneity among individuals who engaged in NSSI in the last year. In the present study, eight latent class indicators of mixed types (two ordinal and six binary) were included in the LCA. The latent class indicators are as follows: (i) frequency of NSSI (ordinal), (ii) length of contemplation before engaging in NSSI (ordinal), (iii) usage of more than three NSSI methods probabilities, a multinomial logistic regression was conducted to examine the association between sociodemographic variables (i.e., age, gender, ethnicity and education) and class membership.
Thereafter, the effect of significant sociodemographic variables (i.e., age, gender and ethnicity) was adjusted for in the following analyses. A multinomial logistic regressions model was utilized to examine the association between emotion regulation, childhood trauma and class membership. Next, a logistic regression analysis was computed to delineate the association between class membership and likelihood of having depression as measured by the PHQ-8. Two linear regression models adjusting for significant sociodemographic covariates and the PHQ-8 were also conducted in order to examine the relationship between class memberships and the physical and mental health scales of the SF-12. Statistical significance was set at the conventional level of p < 0.05, using two-sided tests. Listwise deletion was implemented in all analyses to handle missing data. All statistical analyses were conducted with Stata version 15.

| LCA and identified classes
To determine the optimal solution, the LCA began by extracting one latent class and increasing the number of extracted classes of the solution until no more could be extracted due to nonconvergence.
The results indicated that up to three latent classes solution was plau- The best fitting LCA model revealed three distinct groups. Class 1 (n = 45; 19.2% prevalence) was characterized by low probability of a past suicide attempt (0.09), low probability of using more than three forms of NSSI (0.14), having high (0.67) Table 1.

| Sociodemographic characteristics of latent classes
Sociodemographic characteristics of the latent classes are provided in Table 2. A multinomial logistic regression with class membership as the outcome was conducted to identify the sociodemographic correlates of each latent class. Results are displayed in Table 3. Results revealed that as compared with those who were below 18 years old, individuals who were 21 and above were more likely (OR = 4.74, 95% CI: 1.33-16.91, p = 0.02) to be in Class 2 than Class 1. Those of Malay ethnicity were less likely (OR = 0.26, 95% CI: 0.08-0.86, p = 0.03) to be in Class 2 than Class 1, when compared with those of Chinese ethnicity. Results also indicated that compared with males, females were more likely to be in Class 3 than in Class 1 (OR = 2.19, 95% CI: 1.01-4.78, p = 0.048). As education was not significantly associated with latent class membership, it was removed from further analyses in this paper.

| Association between class membership, emotion regulation and childhood trauma
A series of logistic and linear regression models were utilized for follow-up analyses. Means and frequencies of the scales utilized in the follow-up analyses are presented in Table 4. A multinomial logistic regression adjusting for significant sociodemographic covariates (i.e., age, gender and ethnicity), with class membership as the outcome, and the subscales of the DERS and CTQ as correlates was con-    Table 6.

| Association between perceived health, depression and class membership
T A B L E 5 Results from the multinomial logistic regression analyses between class membership, emotion regulation and childhood trauma

| DISCUSSION
Three groups of individuals who engage in NSSI were identified using LCA in this study. The three groups were differentiated by NSSI features, functions, severity and a past history of suicide attempt demonstrating the heterogeneity of self-injurers in our clinical sample.

| Three classes of self-injurers in psychiatric outpatient sample
Class 1: The first subgroup we identified consisted of 19.2% of young adults with a history of NSSI. This group was congruent to the 'Experimental/Mild NSSI' group of earlier studies (Case et al., 2019;Klonsky & Olino, 2008). Individuals in this group were characterized by low-frequency engagement in NSSI in the past year, participation in fewer methods of NSSI (compared with Class 3), low probability of a past suicide attempt and low endorsement of using NSSI to fulfil intrapersonal and interpersonal functions (compared with Classes 2 and 3) suggesting that NSSI was not part of their usual coping repertoire. Interestingly, this group had the shortest length of contemplation before committing the NSSI. The lack of forethought was also found in individuals who committed 'minor'

NSSI compared with moderate or severe injurers in Lloyd-
Richardson's study (Lloyd-Richardson et al., 2007). We propose that the lack of contemplation may be due to NSSI being of less salience to the mild/experimental group. A study by Selby, Nock, and Kranzler (2014) showed that those with stronger motivation for seeking positive reinforcement from NSSI reported more NSSI thoughts and that NSSI thoughts had a longer duration. It may be that individuals who were likely to endorse multiple functions of NSSI in the two remaining groups were contemplating the reinforcing sensations they would receive from NSSI before acting on their thoughts compared with the mild/experimental group. This group thus shares some similarities to Klonsky and Olino's (2008)

| Correlates of three classes of self-injurious behaviours
A higher distribution of participants aged 21 and above but lower distribution of participants below 18 years and who were of Malay ethnicity (compared with Chinese) were found to be in Class 2 compared with Class 1. In regard to age, NSSI is most prevalent during middle adolescence (Barrocas, Hankin, Young, & Abela, 2012 Class 3 had significantly higher rates of physical abuse falling in the severe classification compared with Class 1. Vaughn et al. (2015) in their study examining latent clusters in NSSI based on childhood adversity found that their severe abuse class evinced high levels of psychiatric/psychological issues. Our earlier study (Peh et al., 2017) had demonstrated that there was a dose-response relationship between the exposure to childhood maltreatment and self-harm frequency and that emotion dysregulation mediated this relationship.
Building on our earlier study, we observed that two particular components of difficulties in emotion regulation difficulties were associated with Class 3: lack of emotional awareness and a tendency to lose control in the face of negative emotions. This is congruent with past literature that emotion regulation difficulties are positively associated with NSSI frequency and diversity (Peterson, Chen, Karver, & Labouliere, 2019).
Putting it all together, early exposure to violent or abusive environments may disrupt healthy emotion regulation and socioemotional skills (Trickett, Negriff, Ji, & Peckins, 2011) that may predispose a child with a tendency towards experiencing high levels of negative affectivity (Marshall, Tilton-Weaver, & Stattin, 2013 often precedes NSSI and feelings of relief and calm follow rapidly after the act-supporting an emotion regulatory function of NSSI (Armey, Schatten, Haradhvala, & Miller, 2015).
Interestingly, Class 1 had higher probability of exposure to emotional abuse than Class 2. Although this relationship contradicts the trend of our other findings between abuse and NSSI, it is possible that the higher rates of emotional abuse rates are associated with other conditions or behaviours not measured in this study.

| Clinical implications
Results from the LCA are suggestive that different NSSI subgroups may have different treatment indications. Class 1 presents with selfinjurious behaviours that are experimental/mild and may benefit from treatment that monitors NSSI risk but may not require a more comprehensive treatment approach versus those in Class 3 who present with multiple risk factors and psychopathology. Class 2 demonstrates coping with negative emotions through NSSI that may maintain the behaviour, and this group may benefit from learning diverse coping strategies. Although there has yet to be a forerunner for effective treatment of repetitive self-harm, common elements across treatment models that have shown promise for individuals such as those in Class 3 who struggle with emotional awareness, impulsivity and interpersonal deficits include emotion regulation and distress tolerance skills to build greater acceptance of uncomfortable emotions (Linehan, 1993), mindfulness practice to reduce impulsivity (Dixon et al., 2019) and skills training to strengthen interpersonal bonding and family relationships such as communication training, problemsolving and/or conflict management (Hetrick, Robinson, Spittal, & Carter, 2016;Turner et al., 2014). Given that individuals are at highest risk shortly after hospital discharge, early intensive treatment may be necessary during this high-risk period to minimize this risk following discharge (Glenn, Franklin, & Nock, 2015).

| Limitations and future directions
There were a number of limitations in this study. First, the study relied on self-reported data. Although we aimed to minimize socially desirable responses through assurance of anonymity, subjective interpretation of responses could influence the responses. Next, some items such as that of the CTQ require retrospective recollection and may be subject to recall biases. Third, the study was cross-sectional in nature, precluding establishment of directionality and causality of the indicators and correlates, and the stability of the classes over time. Future studies shedding light on stability and movement across the classes over time as well as prediction of treatment outcome for each class will be valuable. Fourth, history of past suicide attempt was assessed using only a single item on the FASM. An assessment of the extent of tissue damage and suicide risk including acquired capability of suicide (i.e., lowered fear of death and increased physical pain tolerance) using a structured tool may provide a better measure of suicide risk and distinguish between those with suicide desire and those with acquired capability to provide an even finer grain distinction of individuals falling in Class 3, our largest group. Lastly, identifying latent classes entails some error because classes are not perfectly discrete. Nevertheless, our three classes have substantial overlap with that of earlier studies demonstrating generalizability of the classes across geographical locations, cultures and populations.

| CONCLUSION
Our LCA demonstrates that even in a clinical population, NSSI is not homogeneous with increasing class membership signifying increasing severity of NSSI, psychosocial skill deficits and clinical symptomatology. Clinicians working with clients with NSSI should evaluate presenting NSSI features (method and frequency), the functions served by NSSI, emotion regulation difficulties, trauma history and clinical symptoms to better understand an individual client's risk profile and assess suicide risk. Although a clear forerunner in the treatment of NSSI is presently lacking, profiling patterns of NSSI may be a potentially useful step in providing a nuanced examination among individuals with varying histories of NSSI in order to aid clinicians in their case conceptualisation and in prescribing tailored intervention.
Intervention research that shed light on treatment strategies based on NSSI subtype is recommended to improve treatment protocols and guidelines for NSSI.