کالیبراسیون مشترک رفتار خودآسیبی عمدی (DSH) : به سوی یک معیار اندازه گیری مشترک

Abstract The purpose of this study was to co-calibrate items from different deliberate self-harm (DSH) behavioural scales on the same measurement metric and compare cut points and item hierarchy across those scales. Participants included 568 young Australians aged 18–30 years (62% university students, 21% mental health patients, and 17% community volunteers). Six DSH scales (containing 82 items) were administered, namely, Self-Injury Questionnaire Treatment Related (SIQTR), Self-Injurious Thoughts and Behaviours Interview (SITBI), Deliberate Self-Harm Inventory (DSHI), Inventory of Statements About Self-Injury (ISAS), Self-Harm Information Form (SHIF) and Self-Harm Inventory (SHI). Data were co-calibrated onto an underlying metric using the Rasch measurement model. The resulting calibration shows that the different scales occupy different ranges on the hierarchy of DSH methods with prevalence estimates ranging from 47.7 to 77.1%. A raw score conversion table is provided to adjust prevalence rates and to equate cut points on the six scales. A Rasch-validated hierarchy of DSH behaviours is also provided to inform the development of DSH nomenclatures and assist clinical practice.

1. Introduction Deliberate self-harm (DSH) (also referred to as self-harm) is a sub-type of self-destructive behaviours (Lundh et al., 2007). DSH involves the initiation of an intentional act to cause damage to one's own body (Kreitman, 1977), the resulting harm (or risk of harm) to oneself being direct and immediate (Babiker and Arnold, 1997), the outcome non-fatal (Morgan, 1979), and with multiple intentions and motivations possibly present (Hawton and James, 2005). DSH has been described as an “etiologically heterogeneous, multiply-determined, and complex phenomenon” (Hooley, 2008, p. 157). DSH behaviours include visible damage to surface body tissues (e.g., cutting, burning) (Wilkinson and Goodyer, 2011), lack of self-care (e.g., excessive exercising to cause an injury) (Turp, 2002), and highly dangerous acts (e.g., swallowing dangerous objects) (Linehan et al., 2006). Many forms of DSH are observed across both clinical and non-clinical populations (Nock, 2010), although some behaviours are rarely reported outside of severe mental illness (e.g., amputation, using acid to burn skin) (Walsh and Rosen, 1988). DSH behaviours may also include deliberate recklessness to cause harm (Skegg, 2005). Examples are sexual risk taking (Sansone et al., 1998), reckless driving (Patton et al., 1997), and intentional over-use of drugs (Best, 2009). Although such behaviours may have complex motivations (e.g., sensation seeking) and any harm to self may be incidental (Wilkinson and Goodyer, 2011), they are known to be highly inter-related with other forms of DSH (Brown et al., 2005, Martiniuk et al., 2009 and Nada-Raja et al., 2004). In addition, recent factor-analytic studies (Latimer et al., 2009 and Vrouva et al., 2011) have provided support for a DSH construct that includes reckless behaviours, consistent with some mainstream definitions of DSH (see Silverman, 2006). There are several behaviours that may cause harm to one's body that are not DSH. For example, factitious disorder is distinguished from DSH because the harm is a means to an end (i.e., to imitate an illness and engage medical professionals in a relationship), rather than an end in itself (i.e., to cope with psychological distress) (Babiker and Arnold, 1997). As a further example, indirect self-injury (e.g., chronic alcoholism and smoking) is distinguished from DSH because the harm is temporally remote (Ross and McKay, 1979). Presently, there is no comprehensive classification system for describing DSH (Ougrin and Zundel, 2009), although several nomenclatures (comprising of definitions and terminology) have been developed (e.g., Nock, 2010 and Pattison and Kahan, 1983). Such nomenclatures can be distinguished from each other according to the dimensions of outcome, method, lethality and intent (Ougrin and Zundel, 2009). For example, DSH can be defined as a broad spectrum of non-fatal self-injury and self-poisoning acts irrespective of degree of suicide intent or type of motivation (Hawton et al., 2006). It can also be defined as a narrower set of tissue damaging behaviours performed in the absence of a desire to die (Klonsky et al., 2003). Deciding between the paradigms of DSH without suicide intent and DSH regardless of intent to guide research and clinical practice is the subject of much discussion in the literature ( Jacobson and Gould, 2007). First, it is difficult to reliably measure intent ( Ougrin and Zundel, 2009). Second, significant suicidal ideation may accompany superficial self-harm behaviours with little or no risk of a fatal outcome ( Lundh et al., 2007). Third, some highly dangerous forms of self-harm may have little or no conscious suicide intent ( Nada-Raja et al., 2004). Fourth, suicide and non-suicide related self-harm may frequently occur in the same individual ( Nock et al., 2006). In the long term, the issues surrounding the measurement of intent will most likely be resolved by empirical research and theory building (Lundh et al., 2007 and Ougrin and Zundel, 2009). In the meantime, researchers and clinicians have developed several strategies for distinguishing suicidal self-harming from non-suicidal self-harming. For example, clinicians generally assess behaviour and then clarify intent for each specific act ( Skegg, 2005). Researchers tend to orientate participants to DSH regardless of intent or to DSH without suicide intent by the instructions and item wording contained in their measurement tools (e.g., Gratz, 2001). The accepted methodology of assessing behaviours first and intent second is assisted by the large number of published DSH measures that include behavioural scales or items to identify specific methods of self-harm (Nock et al., 2008). Such scales avoid the under-reporting of DSH associated with single item measures (Nock, 2010) and yield a pattern of behaviours that may inform risk assessment and treatment protocols (Whitlock et al., 2006). The number of methods can be counted over a person's lifetime to form a total DSH score (Sansone et al., 1998), and such scores can be used to determine clinical cut-offs for increased risk of suicide, depression, anxiety and personality disorder (Klonsky and Olino, 2008 and Nock et al., 2006). The endorsement of at least one method of DSH is an accepted procedure for estimating prevalence rates of DSH (Whitlock et al., 2006). However, there is little consistency in the range of methods of DSH contained in published behavioural scales (Gratz, 2001) which inhibits the comparison of prevalence rates across studies (Heath et al., 2009 and Zlotnick et al., 1996). This is because some behaviours are less likely to be endorsed (e.g., burning) than other behaviours (e.g., scratching), possibly due to being associated with higher levels of psychological distress (Croyle and Waltz, 2007, Walsh and Rosen, 1988 and Nock, 2010). The specific behaviours included in a DSH scale therefore have the potential to influence the prevalence rate. To resolve this problem, some researchers (e.g., Gratz, 2001) have argued for the use of a consensus set of DSH behaviours so that the prevalence rates from different studies can be compared. In other areas of mental health measurement, the variation in scale content has been addressed by co-calibrating items from different scales on the same measurement metric (La Porta et al., 2011 and Smith et al., 2006). Co-calibrations can be used to produce a raw score conversion table to allow the equating of clinical cut points and for the adjustment of prevalence rates across studies. Many of these co-calibrations of mental health scales have been conducted by fit of data to the Rasch measurement model (Rasch, 1960) (see Section 2.6). Rasch analysis assumes and tests unidimensionality (Pallant and Tennant, 2007 and Streiner and Norman, 2008), and is an accepted analytic technique in mental health measurement (Barkham et al., 2011). When applied to DSH methods, the Rasch measurement model has the potential to identify clusters of behaviours that occupy different locations on the theorised latent DSH construct. For example, it may be reasonable to expect a cluster of body surface damaging behaviours that have been labelled as non-suicidal self-injury (NSSI) (Wilkinson and Goodyer, 2011). Such finding would give support to the theoretical importance of that particular definition of DSH. A further benefit of the Rasch validated item hierarchy is the potential to inform the many tentative hierarchies (based on clinical experience and/or conceptual labelling) reported in the literature (e.g., Croyle and Waltz, 2007, Skegg, 2005 and Whitlock et al., 2008). In summary of the above, a successful co-calibration has the potential to facilitate the adjustment of prevalence rates across studies, and the equating of clinical cut-points. It also has the potential to produce an empirically validated hierarchy of DSH methods. Further, there is the potential to contribute to the development of a DSH nomenclature (at least for the dimension of method) by validating the tentative hierarchies of DSH methods and by providing evidence for “more or less well defined categories” of self-harm behaviours (Ougrin and Zundel, 2009, p. 13). Finally, a Rasch calibrated hierarchy may assist clinicians to probe for a more complete account of past DSH behaviours in order to inform the risk of future DSH, consistent with recent longitudinal studies (e.g., Glenn and Klonsky, 2011). This study, therefore, aims to use the Rasch measurement model (Rasch, 1960) to co-calibrate behavioural items extracted from selected DSH behavioural scales to (i) produce a raw score conversion table to equate clinical cut points and prevalence rates across studies; and (ii) to construct a hierarchy of DSH behaviours to inform the DSH nomenclatures and clinical practice.

3. Results 3.1. Fit to Rasch model Fit to the Rasch Model is presented in Table 2. The first column in Table 2 shows the testlets included in each analysis. The next three columns show item–trait interaction test of fit (covering all items and persons), individual item fit, and individual person fit. The fifth column reports the values of PSI and CA. The sixth column reports the PCA test of unidimensionality. Initially all 82 items were considered together (Analysis 1, Table 2). Fit to the model was poor, with a high standard deviation of the summary item–person interaction statistic, and a significant summary chi-square fit statistic. On inspection, significant levels of residual correlations were observed, and these clustered within the respective scales. Consequently, the items from each scale were summated into one testlet, so giving six items (testlets), representing each scale. The second Rasch analysis (Analysis 2 in Table 2) then evaluated (as testlets) the core linking scales, the SHI-22 and SHIF-16 (as Administered to Samples 1 and 2). The overall test of fit (item–trait interaction) indicates fit to the model (χ2=28.053, d.f.=16, P=0.031), using a Bonferroni adjusted P value of 0.025 (0.05 divided by 2). The item and person fit residuals were close to the expected values of mean residual close to 0 and a standard deviation close to 1. The PCA test showed strong support for the unidimensionality (1.49% of the t-tests were significant). The PSI estimate was 0.666 and the Cronbach's Alpha was 0.827, thereby indicating a reasonable level of person separation of persons on the latent continuum, but a skewed distribution. The mean logit value of the respondents was −1.881, suggesting that the samples were at much lower levels of DSH than the average of the two instruments taken together. The third Rasch analysis (Analysis 3 in Table 2) evaluated (as testlets) the SHI-22 and SHIF-16 in combination with the ISAS-12 (as administered to Sample 1). The overall test of fit (item–trait interaction) indicates fit to the model (χ2=18.928, d.f.=12, P=0.090). The item and person fit residuals were close to the expected values of mean residual close to 0 and a standard deviation close to 1. The PCA test showed strong support for unidimensionality (1.61% of the t-tests were significant). Further, the PSI estimate was 0.774 and the Cronbach's Alpha was 0.880 indicating a good level of person separation of persons on the latent continuum. The fourth Rasch analysis (Analysis 4 in Table 2) evaluated (as testlets) the SHI-22 and SHIF-16 in combination with the SITBI-11, DSHI-16 and SIQTR-5 (as administered to Sample 2). The overall test of fit (item–trait interaction) indicates fit to the model (χ2=16.137, d.f.=12, P=0.185). The item and person fit residuals were close to the expected values of mean residual close to 0 and a standard deviation close to 1. The PCA test showed strong support for the unidimensionality (2.14% of the t-tests were significant). Further, the PSI estimate was 0.748 and the Cronbach's Alpha was 0.821 indicating a good level of person separation of persons on the latent continuum. The fifth Rasch analysis (Analysis 5 in Table 2) evaluated (as testlets) all tests together. The overall test of fit (item–trait interaction) indicates fit to the model (χ2=36.35, d.f.=32, P=0.273). The item and person fit residuals were close to the expected values of mean residual close to 0 and a standard deviation close to 1. The PSI estimate was 0.690 thus indicating an adequate level of person separation of persons on the latent continuum. Cronbach's Alpha and the PCA test were not calculated because of the structural missing cases. 3.2. Co-calibration of scales The fit of the six published tests (as testlets) to the Rasch model in Analysis 5 allowed for the co-calibration of scales on the same measurement metric, as shown in Fig. 2. With the underlying metric range set to 0–100, the SHIF-16 has the widest operational range, and the SIQTR-5 has the narrowest. Operational range of DSH behavioural scales on latent continuum. Fig. 2. Operational range of DSH behavioural scales on latent continuum. Figure options In the present study sample, the prevalence estimates of DSH (endorsement of at least one behaviour) based on the six scales ranges from 47.7% on the DSHI-16 to 77.1% on the SHI-22. The variation is due to the fact that the scales contain items representing different levels of DSH. In particular, the SHI-22 contains ‘easy’ items reflecting low levels of self harm. Thus, subjects are more likely to endorse these easy items, and so come into the ‘self harm’ classification sooner than if they had endorsed ‘harder’ items from another scale. 3.3. Raw score equating The raw score conversion table for the six scales is provided in Table 3. The values on the latent continuum for each raw score in Table 3 have been transformed to a scale from 0 to 100. This shows, for example, that a score of 1 on the SIQTR-5 is equivalent to a score of 6 on the SHI-22. That is, these two raw scores share a very similar metric value, at around 30 units. Table 3. Raw score conversions for DSH behavioural scales. Raw score ISAS-12 SHI-22 SHIF-16 SITBI-11 DSHI-16 SIQTR-5 0 13.6 1.4 6.2 0.0 19.6 16.3 1 20.6 8.6 17.6 9.4 25.6 30.1 2 26.2 14.4 26.1 17.7 30.7 36.0 3 30.7 19.0 32.4 25.1 35.1 40.7 4 34.8 23.2 37.0 32.4 39.7 46.9 5 38.7 27.1 40.4 39.3 45.0 62.7 6 42.2 30.8 43.1 45.9 50.7 7 45.1 34.3 45.4 52.0 56.7 8 47.7 37.6 47.7 57.7 63.2 9 50.1 40.8 50.1 63.1 65.0 10 52.7 43.8 52.8 69.4 66.1 11 55.9 46.8 56.2 76.5 67.1 12 60.1 49.7 60.9 67.8 13 52.6 67.3 68.7 14 55.5 75.8 69.5 15 58.5 86.5 70.6 16 61.5 100.0 71.9 17 64.7 18 68.0 19 71.4 Example of Conversion1 on SIQTR-5 converts to 2 on DSHI-16, 3 on ISAS-12 & SHIF-16, 4 on SITBI-11, and 6 on SHI-22 (all raw scores located at about 30 on the common metric). 20 75.3 21 80.3 22 86.8 Table options 3.4. Hierarchical ordering of items Fig. 3 shows the hierarchical ordering of the 82 items contained in the six scales. Where several items are found in the same location on the common metric, one is chosen for illustrative purposes. The least frequently endorsed item is ‘dropping acid on skin’, and the most frequently endorsed item is ‘picking at a wound’. The location of all 82 items on the DSH hierarchy is available on request from the authors. Hierarchical ordering of combined item set. Fig. 3. Hierarchical ordering of combined item set.