6.2.1. Introduction

Many researchers have investigated risk factors for self-harm (Fliege et al., 2009; Gratz, 2003; Owens et al., 2002) and for suicide (McLean et al., 2008; Nock et al., 2008). However, these studies do not often distinguish risk factors for a first episode of self-harm from those risk factors for repetition of self-harm. Knowledge of these factors can provide an understanding of the characteristics of those who repeat self-harm or who go on to die by suicide. There will be an overlap between individual risk factors and risk assessment scales (see Section 6.3) which may include combinations of risk factors. Aside from traditional risk factors, those factors that may protect against repeated self-harm or suicide are also considered. Establishing causal relations between risk factors and outcome is difficult because many studies have been observational. In addition there is often a strong association between different risk factors, and measuring one may be a proxy measure for another. This section, however, is aimed at giving guidance on factors to consider in a clinical assessment and not for predicting risk.

6.2.2. Clinical review protocol

The review protocol including the review questions, information about the databases searched and the eligibility criteria used for this section of the guideline can be found in Appendix 8 (further information about the search strategy can be found in Appendix 9). Information on the review protocol can be found in Table 6.

Table 6

Clinical review protocol for the review of case identification tools.

6.2.3. Studies considered¹¹

Forty-nine prospective cohort studies (out of 6,077 references generated by the search) providing relevant clinical evidence met the eligibility criteria for this review. The GDG decided to include only prospective cohort studies for the following three reasons. First, prospective studies are less subject to selection bias and participant recall bias than retrospective studies. Second, prospective cohort studies could identify temporal relationships between risk factors and outcome that might have implications for management. The third reason was practical: to ensure that the number of studies was manageable within the timeframe of this guideline. Of the 49 studies, all were published in peer-reviewed journals. In addition, 41 studies were excluded from the analysis. Further information about both included and excluded studies can be found in Appendix 15c.

Twenty-six out of 49 prospective studies that reported effect measures such as RRs, odds ratios (ORs) or hazard ratios (together with CIs) were selected for possible meta-analysis. These are presented as clinical evidence in Section 6.2.4 (risk factors for non-fatal repetition), Section 6.2.5 (risk factors for fatal repetition) and Section 6.2.9 (risk factors for children and young people).

The process for selection of studies for meta-analysis is described below:

• A list of risk factors examined in each of the 26 studies was drawn up.
• The studies that reported the effect measure for the same risk factor were grouped together.
• For each risk factor, a meta-analysis was conducted for studies that reported the same type of effect measure together with 95% CI (for example, two studies that reported the OR of depression were pooled). A narrative synthesis was presented for those studies that could not be pooled.

Other risk factors that were not reported in a way to allow outcomes to be extracted are included in the narrative synthesis presented in Sections 6.2.6, 6.2.7 and 6.2.10. These studies either did not report 95% CIs, only reported effect measures by subgroups (for example, gender or single ethnic group), only reported p-values, or comprised either a mixture of people who self-harmed for the first time or repeatedly self-harmed (in which results were not separable).

For the section concerning children and young people, studies that recruited participants up to the age of 20 years were included; for the section concerning older adults, one study was identified that included participants aged 60 years or above. These age ranges were wider than those that might be seen in clinical services, but were used in this guideline because of the age cut-offs included in the studies.

6.2.4. Clinical evidence for risk factors for repetition of self-harm in adults (non-fatal outcome)

All studies in this section included clinical populations recruited after presenting to hospital following an index episode of self-harm. Therefore, the factors examined are those associated with a higher risk of repetition of self-harm. As mentioned in Section 6.2.3, all risk factors reviewed below are findings from prospective studies only.

The quality of evidence is presented according to following criteria:

• Study sample – is the study representative of the population of interest with regard to key characteristics, and is it sufficient to limit potential bias to results?
• Loss to follow-up – is the loss to follow-up unrelated to key characteristics, and is it sufficient to limit potential bias?
• Putative risk factor – has this been adequately measured in study participants?
• Outcome of interest – has this been adequately measured in study participants?
• Potential confounders – have the important confounds been appropriately accounted for, limiting potential for spurious association?
• Statistical analysis – has the study used an appropriate design, which limited the potential for presentation of invalid results?

Evidence from each important outcome and the overall quality of evidence are presented. The study characteristics for studies included in the meta-analysis can be found in Appendix 15c.

History of previous self-harm as a risk factor for repetition

Pooled adjusted data

A history of previous self-harm is associated with higher risk of repetition. Three studies (COLMAN2004 [Colman et al., 2004], JOHNSTON2006 [Johnston et al., 2006], MCAULIFFE2008 [McAuliffe et al., 2008]) with a total of approximately 5,000 participants were pooled in the meta-analysis and their combined adjusted OR was 2.7 (95% CI, 2.13 to 3.42).

The repetition rate for self-harm during follow-up (up to 2 years) was 25% (COLMAN2004), 11% (JOHNSTON2006) and 30% (MCAULIFFE2008).

The majority of participants in COLMAN2004 (66%), JOHNSTON2006 (55%) and MCAULIFFE2008 (59%) had a history of previous self-harm. Specifically, most participants in COLMAN2004 received a psychiatric diagnosis, half of the participants in JOHNSTON2006 had received previous psychiatric treatment and a number of participants in MCAULIFFE2008 had alcohol problems. Three studies varied in the extent to which they adjusted for current symptoms related to depression. The adjusted factors can be found in Table 7.

Table 7

History of self-harm – adjusted factors.

The follow-up period ranged from 1 to 2 years. The three studies were conducted in Canada, the UK and European countries. There was no significant heterogeneity after pooling these studies.

Attenuation of the association following adjustment was examined in the two studies (COLMAN2004, MCAULIFFE2008) that reported both unadjusted and adjusted OR. The pooled unadjusted OR was 5.86 (95% CI, 3.23 to 10.65). After adjusting for depression, age and gender, the adjusted OR decreased to 3.81 (95% CI, 1.98 to 7.35).

The quality of evidence for history of self-harm as a risk factor for repetition of self-harm is summarised in Table 8.

Table 8

History of self-harm – quality of evidence.

Pooled unadjusted data

Five studies (BILLE-BRAHE1994 [Bille-Brahe & Jessen, 1994], COLMAN2004, JOHNSSON1996 [Johnsson et al., 1996], MCAULIFFE2008, OWENS1994 [Owens et al., 1994]) provided raw data and a pooled unadjusted OR of 3.09 (95% CI, 1.99 to 4.8), and an observed heterogeneity (I² = 52%) was calculated. It is important to note that unadjusted ratios do not take confounding variables into consideration, and thus findings may result from association with another unmeasured risk factor.

Narrative synthesis

Aside from the studies reviewed above, a narrative synthesis of seven other studies (HAW2007 [Haw et al., 2007], ALLGULANDER1990 [Allgulander & Fisher, 1990], KAPUR2006 [Kapur et al., 2006], DIESERUD2000 [Dieserud et al., 2000], SIDLEY1999 [Sidley et al., 1999], VAN AAIST1992 [Van Aaist et al., 1992], PETRIE1992 [Petrie & Brook, 1992]) with a total of approximately 23,000 participants reported a history of previous self-harm as a risk factor for repetition. Of these seven studies, two studies adjusted for confounding variables (HAW2007, KAPUR2006) while the remaining five did not.

Depressive symptoms as a risk factor for repetition

Pooled adjusted data

People with depressive symptoms are associated with higher risk of repetition. Three studies (CHANDRASEKARAN2008 [Chandrasekaran & Gnanaselane, 2008], COLMAN2004, DIESERUD2003 [Dieserud et al., 2003]) with a total of approximately 700 participants were pooled and reported an adjusted OR of 2.63 (95% CI, 1.72 to 4.04).

The repetition rate during follow-up was 25% (COLMAN2004), 16% (DIESERUD2003) and 23% (CHANDRASEKARAN2008).

A lifetime psychiatric diagnosis was reported in the majority of participants in COLMAN2004 (66% major depression) and in a few participants in CHANDRASEKARAN2008 (26% depression) at baseline. The breakdown of psychiatric diagnosis was not reported in DIESERUD2003, where depressive symptoms were measured by the Beck Depression Inventory (BDI; Beck & Steer, 1987). The majority of participants in COLMAN2004 and DIESERUD2003 had a history of previous self-harm. In CHANDRASEKARAN2008, only participants who reported their index episode as their first episode of self-harm were included. The adjusted factors for each study can be found in Table 9.

Table 9

Depressive symptoms – adjusted factors.

The follow-up period ranged from 1 to 2 years. They were conducted in three different countries. There was no significant heterogeneity reported after pooling the three studies.

Attenuation following adjustment was examined in two of these studies (COLMAN2004, DIESERUD2003), which reported both unadjusted and adjusted OR. The pooled unadjusted OR was 2.98 (95% CI, 0.9 to 9.85). After adjusting for previous self-harm, age and gender, the adjusted OR decreased to 2.19 (95% CI, 1.25 to 3.81).

The quality of evidence for depressive symptoms as a risk factor for repetition of self-harm is summarised in Table 10.

Table 10

Depressive symptoms – quality of evidence.

Narrative synthesis

The narrative findings from four studies (SCOLIERS2009 [Scoliers et al., 2009], CHRISTIANSEN2007 [Christiansen & Jensen, 2007], KAPUR2006 and SIDLEY1999) with a total of approximately 12,000 participants also found that having depressive symptoms, scoring high on a scale measuring hopelessness, and the current use of antidepressants all increased the risk of repetition of self-harm.

KAPUR2006 and SCOLIERS2009 reported an unadjusted hazard ratio of 1.28 (95% CI, 1.14 to 1.44) and RR of 1.85 (1.23 to 2.78), respectively.

The majority of participants (over 50%) in most studies had a history of previous self-harm. The use of antidepressants (CHRISTIANSEN2007) as a risk factor was controlled for other confounds, while hopelessness and having other depressive symptoms (KAPUR2006, SCOLIERS2009, SIDLEY1999) was not controlled for other confounds. SIDLEY1999 reported hopelessness as a short-term predictor of repetition (within 6 months).

Psychiatric history as a risk factor for repetition

Pooled unadjusted data

Two studies (JOHNSSON1996 [Johnsson et al., 1996], OWENS1994) with a total of approximately 1000 participants reported raw data that could be used to calculate a pooled unadjusted OR. The pooled unadjusted OR was 3.46 (95% CI, 2.26 to 5.3). This showed that people with a psychiatric history might be at a higher risk of repetition, bearing in mind this had not been adjusted for confounders. Neither study specified a diagnosis. Data were collected objectively from local psychiatric services' case registers in OWENS1994, and from psychiatric hospital records in JOHNSSON1996.

Repetition rates were reported as 40% (JOHNSSON1996) and 12% (OWENS1994) respectively. Forty-eight per cent (JOHNSSON1996) and 35% (OWENS1994) of participants had a history of previous self-harm before the index admission. JOHNSSON1996 specified the diagnosis (68% with personality disorder, 35% with major depressive disorder [MDD]), while OWENS1994 reported 33% of participants had past psychiatric contact. None of the aforementioned variables was adjusted for in the pooled ratio. There might be confounding factors that limit the strength of findings.

JOHNSSON1996 conducted the study in Sweden for 5 years and OWENS1994 followed up participants for 1 year in the UK.

The quality of evidence for psychiatric history as a risk factor for repetition of self-harm is summarised in Table 11.

Table 11

Psychiatric history – quality of evidence.

Narrative synthesis

Five other studies with a total of approximately 20,000 participants reported narratively having a psychiatric history (without specifying diagnosis) as a risk factor. Three studies reported separately psychiatric treatment (JOHNSTON2006, KAPUR2006) and admission to a mental health hospital (CHRISTIANSEN2007) as a significant risk factor even after adjusting for confounding variables. Two studies (HAW2007, SIDLEY1999) reported the same but only presented unadjusted effects. In HAW2007, psychiatric diagnosis as a risk factor was only reported in participants who were admitted following their first self-harm attempt (not following subsequent episodes).

Gender as a risk factor for repetition

Pooled adjusted data

Two studies (CHEN2010 [Chen, 2010], SCOLIERS2009) with a total of 1,844 participants, reported females as being at a higher risk for repetition. The studies were pooled, resulting in an adjusted RR of 1.96 (95% CI, 1.22 to 3.15). Adjusted factors can be found in Table 12.

Table 12

Gender – adjusted factors.

The pooled unadjusted RR of the same two studies was 1.8 (95% CI, 1.2 to 2.71). CHEN2010 reported a repetition rate of 9.5% over 4 years, while SCOLIERS2009 reported an overall repetition rate of 30% over 5 years.

In SCOLIERS2009, 34% of participants had a history of previous self-harm. CHEN2010 did not report this information. The majority of participants were younger than 40 years old (SCOLIERS2009) and the mean age was 37 years (CHEN2010). Most of the participants were married and had less than 10 years of education (CHEN2010). Other important demographics such as employment or clinical variables were not reported in CHEN2010. SCOLIERS2009 reported that 61% of their participants had high anxiety scores and 46% had high depression scores at follow-up, but these factors were adjusted for in the statistics model.

Both studies followed-up participants for approximately 5 years. CHEN2010 was conducted in Taiwan and SCOLIERS2009 was conducted in Belgium.

The quality of evidence in the meta-analysis for gender as a risk factor for repetition of self-harm is summarised in Table 13.

Table 13

Gender – quality of evidence in the meta-analysis.

Pooled unadjusted data

Four studies (CHANDRASEKARAN2008, JOHNSSON1996, KRARUP1991 [Krarup et al., 1991], OWENS1994) reported raw data that could be used to calculate the unadjusted OR. However, the pooled unadjusted OR had a wide CI and thus there was no clear indication of the direction of the effect if any (OR unadjusted 1.01 [95% CI, 0.5 to 2.04]). A moderate heterogeneity was also observed (I² = 53%), which might be explained by the uncontrolled confounding variables.

The quality of evidence for pooled unadjusted OR is summarised in Table 14.

Table 14

Pooled unadjusted odds ratio – quality of evidence.

Narrative synthesis

In studies where outcomes were not extractable, ZAHL2004 (Zahl & Hawton, 2004) reported young female multiple repeaters (more than two episodes) being at a higher risk compared with repeaters who had two or fewer episodes. This finding is only applicable to females.

In contrast, CHRISTIANSEN2007 reported that being male was associated with a higher risk of repeating a suicide attempt. HEATH2008 suggested that there were no gender differences in risk of repetition, based on a college sample that had a female majority. However, the finding was unadjusted for potential confounds, which should be subject to careful interpretation.

Marital status as a risk factor for repetition

Pooled unadjusted data

Some evidence suggests that not being married, or being of single status, could be a higher risk for repetition of self-harm. Four studies (BILLE-BRAHE1994, CHANDRASEKARAN2008, JOHNSSON1996, OWENS1994) with a total of approximately 1,700 participants reported raw data that could be used to calculate a pooled unadjusted OR. The finding was not significant, with an unadjusted OR of 1.36 (95% CI, 0.85 to 2.16), and was subject to heterogeneity (I² = 63%). Both BILLE-BRAHE1994 and OWENS1994 reported not being married as a risk factor. Nevertheless, the reported statistics in both studies were limited because they were unadjusted.

The quality of evidence for marital status as a risk factor for repetition of self-harm is summarised in Table 15.

Table 15

Marital status – quality of evidence.

Narrative synthesis

Three other studies (JOHNSTON2006, KAPUR2006, DIESERUD2000) with a total of approximately 15,000 participants narratively reported not being married as a risk factor. Findings from KAPUR2006 and DIESERUD2000 were unadjusted for confounds and therefore of limited conclusiveness. JOHNSTON2006 reported an adjusted OR of 1.39 (95% CI, 1.09 to 1.76), which was adjusted for previous self-harm, psychiatric treatment, employment status and ethnicity.

Suicide intent as a risk factor for repetition

Pooled unadjusted data

Two studies (OWENS1994, DIESERUD2003) were pooled to report an unadjusted OR of 0.9 (95% CI, 0.32 to 2.52), providing no conclusive evidence for suicide intent as a risk factor for repetition of self-harm; there was also substantial heterogeneity when pooling the studies (I² = 78%). Suicide intent was defined as ‘suicide threat’ or ‘leaving note’ in OWENS1994; a cut-off score on Beck's Suicide Intent Scale (SIS) was used in DIESERUD2003.

The quality of evidence for suicide risk as a risk factor for repetition of self-harm is summarised in Table 16.

Table 16

Suicide risk – quality of evidence.

Narrative synthesis

Two studies with a total of approximately 10,000 participants reported associations with having a suicide plan (KAPUR2006) and carrying a suicide letter (WANG2006) as being associated with a higher risk of repetition. Both studies were adjusted for different sets of confounding factors.

However, two studies (HAW2003A, HARRISS2005B [Harriss et al., 2005b]) did not find lethality nor intent scores at the index episode as being associated with repetition of self-harm. HARRISS2005B reported that the association between repetitions and suicide intent scores was different for male and female participants.

6.2.5. Clinical evidence for risk factors for completed suicide in adults

All studies in this section included clinical populations recruited after presenting to hospital after an index episode of self-harm and followed prospectively. Therefore, the factors examined are associated with higher risk for repetition of self-harm, leading to a fatal outcome.

All risk factors reviewed below are findings from prospective studies only. Quality of evidence is presented according to the following criteria:

• Study sample – is the study representative of the population of interest with regard to key characteristics, and is it sufficient to limit potential bias to results?
• Loss to follow-up – is the loss to follow-up unrelated to key characteristics, and is it sufficient to limit potential bias?
• Putative risk factor – has this been adequately measured in study participants?
• Outcome of interest – has this been adequately measured in study participants?
• Potential confounders – have the important confounds been appropriately accounted for, limiting potential for spurious association?
• Statistical analysis – has the study used an appropriate study design that limited the potential for invalid results?

Evidence from each important outcome and overall quality of evidence are presented. The study characteristics for studies included in the meta-analysis can be found in Appendix 15c.

History of previous self-harm as a risk factor for completed suicide

Pooled adjusted data

Two studies (NORDENTOFT1993 [Nordentoft et al.,1993], SUOKAS2001 [Suokas et al., 2001]) with a total of 1,992 participants were pooled and reported an adjusted hazard ratio of 2.17 (95% CI, 1.53 to 3.09) with the general population as reference group. In comparison, COOPER2005 (Cooper et al., 2005) reported an unadjusted hazard ratio of 2.97 (95% CI, 1.6 to 5.5).

In NORDENTOFT1993 and SUOKAS2001, 10.5% and 6.7% of the participants, respectively, completed suicide during the follow-up period after their index episode.

Forty-eight per cent of participants in SUOKAS2001 had a history of self-harm prior to the index episode; the exact percentage was not reported in NORDENTOFT1993. Sixty per cent of participants had previous psychiatric treatments (SUOKAS2001), 28% had a diagnosis of ‘alcoholism’ and 15% had a personality disorder (NORDENTOFT1993). Nevertheless, in another study 40% reported no history of mental health problems (NORDENTOFT1993). Adjusted factors can be found in Table 17.

Table 17

History of self-harm – adjusted data.

Data were collected from death register and records. The follow-up period ranged from 10 to 14 years.

The quality of evidence for history of self-harm as a risk factor for completed suicide is summarised in Table 18.

Table 18

History of self-harm and completed suicide – quality of evidnce.

Narrative synthesis

Eight other studies (HAWTON1988 [Hawton & Fagg, 1988], ALLGULANDER1990, CHRISTIANSEN2007, COOPER2005, HAW2007, SKOGMAN2004 [Skogman et al., 2004], SUOKAS1991 [Suokas & Lonnqvist, 1991], ZAHL2004), with a total of approximately 39,000 participants, narratively reported a history of previous self-harm as a risk factor for completing suicide. Four studies (CHRISTIANSEN2007, COOPER2005, HAW2007, SKOGMAN2004) had adjusted this finding for other confounding variables. Two studies suggested that previous self-harm was gender specific. One study (SKOGMAN2004) reported that male repeaters were at a higher risk of completing suicide; by contrast, another study (HAW2007) reported that female repeaters were at a higher risk compared with females with fewer episodes of self-harm. Three other studies (ALLGULANDER1990, HAWTON1988, SUOKAS1991, ZAHL2004) reported this as an unadjusted factor.

Suicide intent as a risk factor for completed suicide (repetition with fatal outcome)

Pooled adjusted data

Three studies (BJORNAAS2009, COOPER2005, SUOKAS2001) with a total of approximately 10,000 participants were pooled. Suicide intent was defined differently in the three studies (see Table 19). Nevertheless, there was evidence of increased risk for those with high intent, with a pooled hazard ratio of 2.7 (95% CI, 1.91 to 3.81) being observed.

Table 19

Suicide intent – adjusted factors.

The quality of evidence for suicide intent as a risk factor for completed suicide is summarised in Table 20.

Table 20

Suicide intent as a risk factor for completed suicide – quality of evidence.

Narrative synthesis

Six studies, with a total of approximately 8,279 participants, narratively reported suicide ideation as a risk factor for subsequent completed suicide. Meta-analysis was not appropriate because none of the outcomes was comparable. Findings from three of these six studies (BJORNAAS2009, SKOGMAN2004, SUOKAS2001) had been adjusted for confounds. The reported ‘wish to die’ (SUOKAS2001) and suicidal motives (BJORNAAS2009) were reported as risk factors, and suicide ideation was found to be a risk factor for females only in SKOGMAN2004. The three other studies (HARRISS2005B, LONNQVIST1991 [Lonnqvist & Ostamo, 1991], SUOKAS1991) did not adjust for confounds. Of these, SUOKAS1991 reported that severe intention to die was predictive of subsequent suicide during follow-up. HARRISS2005B reported that suicide intent was associated with a higher risk of subsequent suicide, especially during the first year and among females.

Being male as a risk factor for completed suicide

Pooled adjusted data

Two studies (CHEN2011 [Chen et al., 2011], SUOKAS2001) with a total of approximately 2,000 participants were pooled to report an adjusted hazard ratio of 2.66 (95% CI, 1.72 to 4.11) for being male as a risk factor for completed suicide.

Suicide following an index episode of self-harm was 4.4% (CHEN2011) and 6.7% (SUOKAS2001). All participants at index episode were admitted for self-poisoning in SUOKAS2001, and 43% were admitted for overdose in CHEN2011.

In SUOKAS2001, 48% had a history of previous self-harm and 60% had received previous psychiatric treatment. No such information was provided for CHEN2011.

These two studies varied in the factors they adjusted. The adjusted factors can be found in Table 21.

Table 21

Being male – adjusted factors.

The follow-up period ranged from 7 to 14 years, with one study conducted in Taiwan and the other in Finland. There was no significant heterogeneity after pooling these studies.

The quality of evidence for being male as a risk factor for completed suicide is summarised in Table 22.

Table 22

Being male – quality of evidence.

Pooled unadjusted data

Two studies (CHEN2011, COOPER2005) with a total of 9,051 participants, reported unadjusted hazard ratios, and a pooled unadjusted hazard ratio of 2.72 (95% CI, 1.78 to 4.16) was calculated. It is important to note that unadjusted ratios do not take confounding variables into consideration, and thus findings may result from association with another unmeasured risk factor. Only one study (CHEN2011) reported both adjusted and unadjusted hazard ratios. After being adjusted for age and methods of self-harm, attenuation was observed from an unadjusted hazard ratio of 3.46 (95% CI, 1.92 to 6.26) to an adjusted hazard ratio of 2.47 (95% CI, 1.28 to 4.75).

Physical health problems as a risk factor for completed suicide

Pooled adjusted data

Two studies (COOPER2005, HOLLEY1998) with a total of approximately 8,800 participants were pooled to report an adjusted hazard ratio of 1.59 (95% CI, 0.93 to 2.72) with the general population as a reference group. HOLLEY1998 defined physical health problems as chronic and associated with high mortality and leading to significant impairment to functioning. Physical health problems were not defined in COOPER2005.

Less than 1% of participants (COOPER2005) and 6% (HOLLEY1998) completed suicide during follow-up, while 15.5% repeated self-harm (COOPER2005).

Fifty one percent of participants (COOPER2005) had a history of previous self-harm (this was not reported in HOLLEY1998). Sixty-nine per cent of participants had major depression, 24% had neuroses, and 35 to 43% reported the use of alcohol as a factor identified in the attempt (HOLLEY1998). None of the psychiatric diagnosis information was provided in COOPER2005. At least 67% participants were unemployed (HOLLEY1998). Adjusted factors can be found in Table 23.

Table 23

Physical health – adjusted factors.

Data were collected from the death register and records, while the risk factors assessed were collected from assessment forms. The follow-up period varied from approximately 4 years (COOPER2005) to 13 years (HOLLEY1998).

COOPER2005 reported an unadjusted hazard ratio of 2.68 (95% CI, 1.3 to 5.5); HOLLEY1998 did not report on this.

Another study (HAWTON1988), which could not be meta-analysed with the above studies, also reported poor physical health as a risk factor for increased risk of subsequent suicide.

The quality of evidence for physical health problems as a risk factor for self-harm is summarised in Table 24.

Table 24

Physical health – quality of evidence.

Alcohol misuse as a risk factor for completed suicide

Pooled adjusted data

Two studies (BJORNAAS2009, COOPER2005) with a total of approximately 9,000 participants were pooled to report an adjusted hazard ratio of 1.42, but the wide CI included the possibility of a small protective effect (95% CI, 0.7 to 2.8); there was also a high heterogeneity observed (I² = 65%). COOPER2005 reported an unadjusted hazard ratio of 2.11 (95% CI, 1.23 to 3.63) and BJORNAAS2009 reported an unadjusted hazard ratio of 1.6 (95% CI, 1.2 to 2.2).

Less than 1% of participants (COOPER2005) and 7% (BJORNAAS2009) completed suicide during follow-up.

Fifty-one per cent of participants (COOPER2005) had a history of previous self-harm (this was not reported in BJORNAAS2009). BJORNAAS2009 reported that 12% were addicted to opiates and 53% had no history of alcohol misuse. No information about psychiatric diagnosis was provided in COOPER2005. Both studies adjusted for participants' psychiatric history as confounding variables. BJORNAAS2009 adjusted for gender and participants' socioeconomic status. Adjusted factors can be found in Table 25.

Table 25

Alcohol misuse – adjusted factors.

Data were collected from the death register and records, while data on the assessment of alcohol misuse or ‘abuse’ were collected from psychiatric assessments. The follow-up period varied from approximately 4 years (COOPER2005) to 20 years (BJORNAAS2009), which might explain the heterogeneity.

Attenuation was examined by BJORNAAS2009 and COOPER2005, where both unadjusted and adjusted ORs were reported. The pooled unadjusted OR was 1.52 (95% CI, 0.79 to 2.94). After adjusting for previous psychiatric history and suicide intent, the adjusted OR was attenuated to 1.42 and the CI included no effect (95% CI, 0.7 to 2.88).

The quality of evidence for alcohol misuse as a risk factor for suicide completion is summarised in Table 26.

Table 26

Alcohol misuse – quality of evidence.

Narrative synthesis

One study (BECK1989 [Beck et al., 1989b]) reported an association between alcohol misuse and completed suicide, reporting both adjusted and unadjusted ORs (which could not be pooled with the above hazard ratio). However, a wide confidence interval limited the conclusions that could be drawn from this study.

There is a lack of evidence to show an association between the time of alcohol consumption and the index episode of self-harm. HOLLEY1998 reported an adjusted hazard ratio of 1.1 (95% CI, 0.6 to 2.3), providing no evidence of whether there was an effect; the time period between alcohol consumption and the episode of self-harm was not specified. Sixty-nine per cent of participants had major depression, 24% had neuroses and 35 to 43% reported the use of alcohol as a factor at the time of the attempt (HOLLEY1998). Data was based upon A&E records of whether alcohol was used as a factor in the suicide attempt.

Another study (HAW2001B) compared people who had alcohol dependence and had self-harmed with those without alcohol dependence. This was a prospective study conducted in the UK over approximately 1 to 2 years. Participants were admitted to hospital for self-harm. Forty out of 150 patients with alcohol-use disorders were selected for analysis. Of these 40 participants, 80% had previously self-harmed and 90% had a comorbid psychiatric diagnosis (mostly depression). Repetition was 45% for those with alcohol dependence and 29% for those without. More participants with alcohol dependence had consumed alcohol within 6 hours of the index episode, and those with alcohol dependence were more aggressive and impulsive with poorer problem-solving skills.

Psychiatric history as a risk factor for completed suicide (repetition with fatal outcome)

Pooled adjusted data

Two studies (COOPER2005, HOLLEY1998) with a total of approximately 9,000 participants found no evidence of an association between psychiatric history and completed suicide, with an adjusted hazard ratio of 1.22 (95% CI, 0.56 to 2.64); a high heterogeneity was observed (I² = 62%). Therefore, the meta-analysis result was inconclusive. COOPER2005 reported an unadjusted hazard ratio, which was 2.11 (95% CI, 1.22 to 3.65); HOLLEY1998 did not. Adjusted data can be seen in Table 27 and quality of evidence in Table 28.

Table 27

Psychiatric history – adjusted data.

Table 28

Psychiatric history – quality of evidence.

Narrative synthesis

Seven studies with a total of approximately 12,000 participants narratively reviewed psychiatric history as a risk factor for completed suicide. Four studies (CHRISTIANSEN2007, LONNQVIST1991, SKOGMAN2004, SUOKAS2001) adjusted their findings for confounds and therefore were more robust in their results. These studies did not report a specific psychiatric diagnosis. Past psychiatric contact (SKOGMAN2004) and being admitted to mental health hospitals (CHRISTIANSEN2007) were regarded as similar factors. The other three studies (HAWTON1988, LONNQVIST1991, SUOKAS1991) reported psychiatric history as a risk factor without adjusting their findings. HAWTON1988 reported that a diagnosis of schizophrenia was more common than other diagnoses among people who did and did not complete suicide.

6.2.6. Narrative synthesis – risk factors for repetition of self-harm in adults

Studies included in this section concern risk factors that cannot be included in meta-analysis because the outcomes reported are not suitable. These studies either did not report 95% CIs, only reported effect measures by subgroups (for example, male or female, Asian or non-Asian, and so on), only reported p-values, or comprised a mixture of people who had either self-harmed for the first time or repeatedly self-harmed (in which results were not separable). However, these factors should not be overlooked.

6.2.7. Narrative synthesis – risk factors for completed suicide in adults

Studies under this section were not included in the meta-analysis because the outcomes reported were not suitable. These studies either did not report 95% CIs, only reported effect measures by subgroup (for example, male or female, Asian or non-Asian, and so on), only reported p-values, or comprised a mixture of people who had either self-harmed for the first time or had repeatedly self-harmed (in which results were not separable). However, these variables should not be overlooked as possible risk factors.

6.2.8. Clinical evidence summary – adults

Risk factors for non-fatal repetition of self-harm

Key factors with pooled quantitative evidence

Previous self-harm and depressive symptoms are the two risk factors with most support from quantitative and narrative evidence. The majority of participants had self-harmed prior to their index episode. The pooled adjusted or unadjusted ORs of previous self-harm as a risk factor are over 2. For depressive symptoms, there is a somewhat smaller evidence base. The pooled adjusted or unadjusted OR varied, yet there was still an association. This association should also be interpreted cautiously because a number of different measures of depressive symptoms were used in the studies.

Other factors with pooled quantitative evidence

Unspecified psychiatric history has been one of the most commonly reported risk factors. Pooled quantitative synthesis showed some support for this, but the findings were not adjusted for important confounds such as age and gender. It is noteworthy that key risk factors such as previous self-harm and depression, identified above, may overlap with this factor. Moreover, these studies did not specify or define what they meant by psychiatric history; therefore uncertainties remained. Nevertheless, there was reasonable support from the studies' reported narrative findings.

Although being female is another commonly reported risk factor for non-fatal repetition, the evidence is mixed and of relatively poor quality. Two studies reported a similar pooled adjusted and unadjusted RR. However, one important limitation is that none of the studies adjusted for participants' previous self-harm, which is itself an important risk factor. Being female is often reported as being associated with self-harm. The increase in RR of repetition in females might be a consequence of its association with a first episode of self-harm rather than a repeated episode. By contrast, one study found that being male led to a higher risk of repetition. Being female might be a generic risk factor for self-harm, but it might not necessarily be associated with a higher risk of repetition.

There is evidence to suggest that being unemployed and ‘registered sick’ are associated with a higher risk of repetition. Although the studies were not pooled quantitatively, each study reported a statistically significant RR. Similarly, evidence from narrative synthesis also supports this as a risk factor.

There is mixed evidence about marital status as a risk factor for repetition of self-harm. Pooled quantitative evidence did not support this as a risk factor. However, some narrative syntheses suggest an association between not being married and repetition of self-harm.

There is mixed evidence suggesting that the existence of a suicide letter or plan might mean the individual is at higher risk for non-fatal repetition. The pooled quantitative evidence did not support this; however, some narrative syntheses do support this as a risk factor. From other narrative syntheses, there is evidence suggesting that a more violent method of index attempt is predictive of further repetition. Attempts that are regarded as violent include hanging, strangulation, suffocation and jumping from heights. A summary of risk factors can be seen in Table 29.

Table 29

Summary of risk factors for self-harm population in adults.

Factors from studies reviewed narratively

Demographics

Youth is a commonly reported risk factor. Data were not synthesised quantitatively because different studies reported different age ranges and it was unclear how some studies defined ‘youth’. This factor should not be conflated with the higher prevalence of self-harm among young people. Being at risk of self-harm may not be equivalent to being at risk for repeating self-harm.

Specific psychiatric diagnosis

There is a substantial evidence base that suggests schizophrenia and related symptoms may be a risk factor for repetition. There was some quantitative support that could not be meta-analysed and remained robust after being adjusted for separately in two studies. Although the evidence base has only been narratively reviewed, schizophrenia as a factor for repetition of self-harm should be considered. Also, alcohol misuse is an additional risk factor with a strong narrative evidence base as well as unpooled quantitative support.

6.2.9. Clinical evidence for risk factors in young people

Previous self-harm as a risk factor for repetition in young people

Pooled adjusted data

Four studies (CHITSABESAN2003 [Chitsabesan et al., 2003], MIRANDA2008 [Miranda et al., 2008], HULTEN2001 [Hulten et al., 2001], WONG2008 [Wong et al., 2008]) with a total of approximately 2,700 participants were pooled to report an adjusted OR of 3.27 (95% CI, 2.46 to 4.34). No hetereogeneity was observed. Two studies (MIRANDA2008, WONG2008) reported self-endorsed attempts and recruited a community sample. The pooled adjusted OR was 4.09 (95% CI, 1.72 to 9.74), which was higher than the pooled adjusted OR calculated from the two clinical studies (OR of 3.18 [95% CI, 2.35 to 4.29]) (CHITSABESAN2003, HULTEN2001).

Fifteen per cent (CHITSABESAN2003) and 17.2% (HULTEN2001) repeated, and 4.5% (WONG2008) and 22.5% (MIRANDA2008) self-reported repetition during follow-up.

Twenty-nine per cent (CHITSABESAN2003) and 38% (HULTEN2001) had a history of self-harm. Fifteen per cent self-reported multiple suicide attempts in MIRANDA2008. Self-reported previous self-harm was the recruitment criteria in WONG2008. Therefore, all participants self-endorsed past suicide attempts, of which 2% were attempted within the last year that the survey was conducted.

The majority of participants were diagnosed with depression and substance misuse in CHITSABESAN2003. Approximately a quarter of participants had various mood and anxiety disorders (MIRANDA2008, WONG2008). Psychiatric diagnosis was not reported in HULTEN2001. Adjusted factors can be found in Table 30 and quality of evidence in Table 31.

Table 30

Previous self-harm – adjusted factors.

Table 31

Previous self-harm – quality of evidence.

Two studies were long-term studies ranging in duration from 4 to 6 years (HULTEN2001, MIRANDA2008). The other two were short-term studies ranging from 6 to 12 months (WONG2008, CHITSABESAN2003). Three studies reported an average age of 15 years, whilst the remaining studies' participants ranged from 15 to 19 years old.

Narrative synthesis

One study (GROHOLT2006 [Groholt et al., 2006]) suggested previous self-harm history as an independent risk factor (unadjusted hazard ratio 2.8 [95% CI, 1.39 to 5.64]), which was not adjusted for.

Depressive symptoms as a risk factor for repetition in young people

Pooled adjusted data

Two studies (CHITSABESAN2003, WONG2008) with a total of approximately 1,200 participants were pooled to report a marginally significant adjusted OR of 1.05 (95% CI, 1.00 to 1.11). However, one study was conducted in a school setting, where students self-reported past suicide attempts and related outcomes in a questionnaire. Despite the difference in setting, no heterogeneity was found.

Fifteen per cent repeated (CHITSABESAN2003) and 4.5% self-reported repetition (WONG2008) during follow-up.

Twenty-nine per cent had a history of previous self-harm (CHITSABESAN2003). Self-reported previous self-harm was the recruitment criteria in WONG2008. Therefore all participants self-endorsed past suicide attempts, of which 2% were attempted within the last year the survey was conducted.

The majority of participants were diagnosed with depression and substance misuse in CHITSABESAN2003. A quarter of participants had depressive symptoms and a fifth of them had anxiety symptoms in WONG2008. Adjusted factors can be found in Table 32 and quality of evidence in Table 33.

Table 32

Depressive symptoms – adjusted factors.

Table 33

Depressive symptom – quality of evidence.

Both studies were conducted over a relatively short period ranging from 6 months (CHITSABESAN2003) to 1 year (WONG2008). In both studies, the participants' average age was 15 in the UK and Hong Kong respectively.

One study (GROHOLT2006) suggested depressive symptoms as an independent but marginal risk factor (unadjusted hazard ratio 1.05 [95% CI, 1.02 to 1.08]), which was not adjusted for.

Narrative synthesis

Three other studies (BRENT1993 [Brent et al., 1993], GROHOLT2006, NOVAKOVIC2006 [Novakovic et al., 2006]) with a total of approximately 400 participants reported depression as a risk factor. BRENT1993 reported that a diagnosis of major depression at baseline, and an affective disorder that continued through follow-up predicted repetition in young people. While NOVAKOVIC2006 reported that depressive, anxious and phobic tendencies predicted repetition, GROHOLT2006 reported hopelessness as a risk factor after adjusting for confounds; a diagnosis of depression was found to be an independent risk factor in this study.

Gender as a risk factor for repetition in young people

Pooled unadjusted data

Three studies (HAWTON1992 [Hawton et al., 1992], MIRANDA2008, WONG2008) with a total of approximately 3,600 participants reported raw data for the calculation of an unadjusted OR of 1.24 (95% CI, 0.7 to 2.17) for the age range of 10 to 19 years (moderate heterogeneity, I² = 62%). The result found no evidence of an association between gender and repetitio of self-harm in young people. MIRANDA2008 reported an adjusted OR of 2.7 (95% CI, 0.4 to 16.4). The wide confidence interval meant that no conclusion could be drawn regarding the direction or size of any association.

The repetition rates were reported as 9% (HAWTON1992) and 22.5% (MIRANDA2008). Twenty per cent reported previous self-harm and 16% had a psychiatric treatment history in HAWTON1992. Approximately one quarter of participants had various mood and anxiety disorders in MIRANDA2008.

Because the result was not significant, and confounds such as previous self-harm and psychiatric diagnosis might affect the influence of gender, evidence for gender as a risk factor for repetition in young people is inconclusive.

The quality of evidence for gender as a risk factor for repetition of self-harm in young people is summarised in Table 34.

Table 34

Gender in young people – quality of evidence.

Narrative synthesis

In a narrower age range of 12 to 14 years, HAWTON2008 [Hawton & Harriss, 2008b] reported raw data for the calculation of an unadjusted OR of 1.14 (95% CI, 0.66 to 1.98). This has not been pooled with the above studies due to the difference in age range.

Suicide intent as a risk factor for repetition in young people

Pooled adjusted data

Two studies (CHITSABESAN2003, WONG2008) with a total of approximately 1,200 participants were pooled. There was no evidence of an association between suicide intent and repetition of self-harm in young people, with an adjusted OR of 1.45 (95% CI, 0.63 to 3.37); there was considerable heterogeneity (I² = 84%). Adjusted factors can be found in Table 35 and quality of evidence in Table 36.

Table 35

Suicide intent in young people – adjusted factors.

Table 36

Suicide intent in young people – quality of evidence.

6.2.10. Narrative synthesis – young people

Studies under this section cannot be included in a meta-analysis because the outcomes reported are not suitable. These studies either did not report 95% CIs, reported effect measures by subgroup only (for example, male or female, Asians or non-Asians, and so on), reported p-values only, or comprised a mixture of people who self-harmed for the first time or repeatedly self-harmed (in which results were not separable). However, these factors should not be overlooked as risk factors.

6.2.11. Clinical evidence summary – young people

Based on the evidence review, risk factors for young people are similar to those reviewed in the section on adults.

Factors from studies reviewed narratively

Psychiatric diagnosis

A diagnosis of anxiety had some evidence supporting it as a risk factor in young people. However, a major limitation was that most findings had not been adjusted for confounding variables. Substance use had some narrative evidence reporting it as a risk factor, but it was based on self-report questionnaires conducted in school settings. There is little evidence to support a general psychiatric diagnosis (such as affective disorders) as a risk factor.

Relational problems

There are some risk factors relating to family and friendships that may be unique for young people.

A summary of risk factors for young people can be found in Table 37.

Table 37

Summary of risk factors for young people who self-harm.

6.2.12. Narrative synthesis for older adults

One study (HAWTON2006) conducted a prospective study in the UK with 20 years' follow-up, recruiting 730 older people aged 60 years or above who presented to a general hospital in Oxford following a self-harm episode. Forty-seven per cent of participants were aged between 60 and 69 years and 24% had previously self-harmed. Only 15% of the sample received psychiatric care at time of their episode. Of the 149 participants who reported suicide intent scores, nearly two thirds of the participants (65.1%) scored in the high or very high range of the Beck SIS. Repetition rate was 15.3%. It was suggested that a previous history of self-harm was the independent risk factor for suicide, with some evidence showing previous psychiatric treatment and high suicide intent being risk factors as well. This finding was confirmed by a recent, large multicentre cohort study conducted in the UK (MURPHY2011 [Murphy et al., 2011]). This study had 1,177 participants aged 60 years or above, presenting to six emergency departments in Oxford, Derby and Manchester. It was reported that 12.8% of the participants repeated within 12 months of presentation and 1.5% died by suicide within 12 months. This study concluded that previous self-harm, previous psychiatric treatment and being aged between 60 and 74 years were risk factors for repetition. There were no direct comparisons of risk factors for older adults and working-age adults. However, based on this study, it appeared that the risk factors among older adults were similar to risk factors for working-age adults.

6.2.13. Clinical evidence for risk factors in subgroups

Nine studies with psychiatric diagnosis subgroups were narratively reviewed. Participants in these studies were at risk of self-harm, but may or may not have self-harmed before. Study characteristics for each study included in the meta-analysis can be found in Appendix 15c.

6.2.14. Clinical evidence summary – subgroups

The evidence base for risk factors among psychiatric subgroups is limited. The study participants were recruited on the basis of treatment of psychiatric problem – they might or might not have self-harmed before. Therefore, the risk factors associated with psychiatric problems might not be indicative of a further repetition of self-harm (but could be a generic risk factor for self-harm). One common risk factor shared across all diagnostic subgroups was previous self-harm. People with psychiatric problems who had a previous history of self-harm might be more likely to self-harm in the future. In addition, depression might be associated with a higher risk of self-harm.

6.2.15. Prevalence of psychiatric disorders in patients who self-harm

A systematic review conducted by Hawton and colleagues (2011) was adopted in this section for narrative review. This review aimed to explore the extent to which self-harm is associated with psychiatric disorders. The authors included 46 studies, of which seven were UK-based. All participants were recruited after presenting at the hospital following an episode of self-harm. Diagnosis was made according to DSM-IV for all ages. Studies were excluded where assessment was made only for a single disorder or a retrospective diagnosis. Populations with a learning disability and those residing in psychiatric hospitals were excluded.

An overall prevalence rate of 84% (95% CI, 75 to 92%) was observed among adults or mixed samples with a very high heterogeneity (I² = 99%) in 30 studies. In the young people population (up to the age of 25 years), an overall prevalence of 81% was observed with very high heterogeneity (I² = 97%) from nine studies. Prevalences for specific disorders were reported: 61% (95% CI, 41 to 79%) mood disorders were observed, mostly in females, of which the most frequent diagnosis was depression (52%; 95% CI, 43 to 64%); anxiety had a prevalence of 37% (95% CI, 24 to 52%); substance misuse had a prevalence of 36% (95% CI, 22 to 52%) and alcohol misuse was more common than drug misuse. Prevalence was higher in males for substance misuse, personality disorders had a prevalence rate of 28% (95% CI, 18 to 39%) in all adult populations and adjustment disorders had a prevalence rate of 22% (95% CI, 6 to 45%). Lastly, psychotic disorders and eating disorders had prevalence rates of less than 10%. It was observed that the prevalence rate of multiple diagnosis (94.1%, 95% CI, 88 to 98%) was greater than single diagnosis (75%, CI, 63 to 86%). There were no major gender differences in overall prevalence rates of psychiatric disorders.

The limitations to this review lay in the heterogeneity observed in pooled prevalence rates. This could be due to the variation in diagnostic measures (research or clinical diagnosis) and different definitions of self-harm. Studies were cross-sectional, which might make them susceptible to unstable diagnosis. Nevertheless, there was a high prevalence rate (around 80 to 90%) of psychiatric disorders (most commonly depression, anxiety and alcohol misuse) among people who presented to hospital for self-harm. This underlies the importance of careful psychosocial assessment for people who self-harm, in order to treat the underlying disorders and manage their self-harm.

6.2.16. Clinical evidence for protective factors

In this section, studies that looked at protective factors that might protect against repeated self-harm or suicide were reviewed.

A meta-analysis was not conducted because the outcomes were not comparable across studies. Studies were therefore narratively reviewed.

6.2.17. Clinical evidence summary – protective factors

The evidence base for protective factors is not strong. Some narrative evidence shows that problem-solving skills are protective for further repetition. It is unclear whether the effect may diminish after adjusting for other confounding variables. In younger populations, there is some evidence regarding a healthy family environment and parental bonding as protective factors for further repetition.

6.2.18. Narrative synthesis – social care and adversity as risk factors

It is important to note the absence of some commonly reported social risk factors in the reviews above, such as childhood experience of physical abuse, sexual abuse, being a ‘looked-after child’ and other stressful childhood experiences. These studies are often conducted retrospectively, depending on participants' recollection of their childhood experiences. As a result, these did not meet the inclusion criteria set by the GDG and were not included in the above review. However, these factors cannot be overlooked.

The technical team identified a few relevant systematic or literature reviews that were deemed to cover these risk factors. In addition, some key studies were also provided by some GDG members.

It is important to take note of the limitations to these studies. Retrospective studies are subject to participants' recall bias, when recollecting childhood experiences in particular. Also, the findings from the review and studies did not specify whether these factors were associated with repetition of self-harm or incidence of self-harm behaviour.

6.3.1. Introduction

There is increasing emphasis on the assessment of risk in clinical services. Risk assessment in mental health is a broad concept that covers a judgement not only of the likelihood of an adverse outcome, such as suicide or self-harm, but also of violence, risk to children, risk of exploitation and environmental risks such as safety in the home. This guideline focuses on risk of self-harm and suicide. Risk assessment in the UK is carried out by undertaking a clinical interview and this often includes a checklist of risk factors derived from an assessment scale. In the UK, there is no consistency in the risk assessment tools used by different mental health services. Despite the widespread use of these instruments, there is no clear evidence that their use makes any difference to patient outcome. The usefulness of any particular risk assessment scale for repeated self-harm depends on the ability to distinguish those who go on to self-harm from those who do not. Whilst the risk of repeated self-harm is important, healthcare professionals will be most concerned about the risk of suicide. This is more difficult to predict given the relative rarity of suicide even in a population at high risk such as those who have self-harmed.

Risk assessment is not the same as risk management and simply assessing risk without developing a management plan contingent on the level and nature of the risk is unlikely to improve patient outcomes. Previous guidelines (NICE, 2004a) have emphasised that risk scales should not replace a full psychosocial assessment and there is evidence that the latter is associated with better outcomes (Bergen et al., 2010b; Hickey et al., 2001; Kapur et al., 2002).

A further issue to consider is the context in which the risk assessment takes place, in the emergency department after an episode of self-harm, in the community or at the point of admission to or discharge from an inpatient unit.

6.3.2. Clinical review protocol

The review protocol, including the review questions, information about the databases searched and the eligibility criteria used for this section of the guideline, can be found in Appendix 8. Further information about the search strategy can be found in Appendix 9. Also, see Table 38 for the clinical review protocol.

Table 38

Clinical review protocol.

6.3.3. Studies considered¹²

A total of 7,642 studies were identified by the electronic search. Of these, 7,573 were excluded at the screening stage on the basis of reading the title and/or abstract. The remaining 69 references were assessed for eligibility on the basis of the full text. Sixteen prospective cohort and case-control studies providing clinical evidence for risk assessment measures met the eligibility criteria for this section of the guideline. These are: BECK1985 (Beck et al., 1985), BECK1999 (Beck et al., 1999), BISCONER2007 (Bisconer & Gross, 2007), CARTER2002 (Carter et al., 2002), COOPER2006B (Cooper et al., 2006b), COOPER2007 (Cooper et al., 2007), CORCORAN1997 (Corcoran et al., 1997), GALFAVY2008 (Galfavy et al., 2008), HARRISS2005A (Harriss & Hawton, 2005a), KAPUR2005 (Kapur et al., 2005), NIMEUS1997 (Niméus et al., 1997), NIMEUS2000 (Niméus et al., 2000), NIMEUS2002 (Niméus et al., 2002), OSMAN1999 (Osman et al., 1999), OSMAN2001 (Osman et al., 2001) and WAERN2010 (Waern et al., 2010). Seven studies were identified for psychosocial assessment (BERGEN2010B, HAW2003b, HICKEY2001, KAPUR2003 [Kapur et al., 2003], KAPUR2008 [Kapur et al., 2008], OUGRIN2011 [Ougrin et al., 2011], WITTOUCK2010 [Wittouck et al., 2010]) and two studies were identified for needs assessment (CEDEREKE2007 [Cedereke & Öjehagen, 2007], KEENE2005 [Keene, 2005]).

For risk assessment, the inclusion criteria are prospective cohort or case-control studies that report sensitivity and specificity data. The populations used in the studies include people who self-harm, or have suicide ideation if the study clearly reports a history of self-harm. The studies used tools or scales (these terms are used interchangeably) to predict a repetition of self-harm or suicide.

Based on reading the full text of studies for risk assessment scales, 56 references were excluded because they were either not a self-harm population, did not look at prediction of self-harm or suicide, did not report sensitivity or specificity or did not use a risk tool or scale to predict suicide or self-harm. Studies were also excluded if it was unclear how many people in the population had self-harmed in the past. Studies that used another scale as a reference standard to measure the outcome of the study were also excluded. Studies that used a case-control design were excluded if the population of the control group was a general and not self-harm population, for example OSMAN1998. The study characteristics for studies that were included in the meta-analysis can be found in Appendix 15d, which also includes details of excluded studies.

For a full list of the scales reviewed in this chapter and the studies that have reported the predictive validity of these scales, see Table 39.

Table 39

Risk assessment scales and corresponding study ID.

6.3.4. Methods

The psychometric properties of the scales examined included sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV), using predefined cut-off scores. Sensitivity and specificity can be calculated using true positive (TP), true negative (TN), false positive (FP) and false negative (FN) values.

In this guideline, sensitivity is defined as the proportion of those who go on to repeat self-harm who have been identified as at high risk of repetition of self-harm on the basis of their scores on the risk scale or measure (sensitivity = TP / [TP + FN]). Specificity is defined as the proportion of those who do not go on to repeat self-harm and who have been identified as at low risk of self-harm repetition on the basis of their scores on the risk scale or measure (specificity = TN / [FP + TN]). Predictive values may be more useful than sensitivity and specificity in clinical practice. PPV measures the probability that a person with a positive test result really has self-harmed (TP / [TP + FP]). Finally, the negative predictive validity measures the probability that a person with a negative test result really is free of self-harm (TN / [FN + TN]).

The GDG agreed that the desired psychometric properties of scales would vary according to the outcome being predicted and the context in which the scale would be used. For example, a scale predicting repetition in routine practice should help healthcare professionals identify a person who is at high risk of repeating an episode of self-harm, without including too many who do not repeat (that is, has a low false positive rate). However, for a scale predicting suicide the consequences of missing individuals who would go on to die by suicide are so serious that the false negative rate should be very low.

Each study has been narratively reviewed and details of the impact of setting or organisational context have been included, if available, as well as the content of assessment scales.

6.3.5. Scales that predict suicide

There were six cohort studies that used scales to predict suicide (BECK1985, BECK1999, HARRISS2005A, NIMEUS1997, NIMEUS2000, NIMEUS2002).

The Beck Hopelessness Scale (BHS; Beck et al., 1974a) is a self-report instrument that consists of 20 true-false statements constructed to measure the extent of positive and negative beliefs about the future during the past week in psychiatric patients. Typical items are ‘My future seems dark to me’ and ‘I might as well give up because there is nothing I can do about making things better for myself’. Each of the 20 statements is scored 0 or 1. Responses are summed to give a score of 0 to 20. The severity of hopelessness is rated as 0 to 3 minimal, 4 to 8 mild, 9 to 14 moderate and 15 to 20 severe.

BECK1985 used the BHS in a cohort study of 207 psychiatric inpatients hospitalised for suicide ideation. Thirty-two per cent had previously attempted suicide. Participants were followed up for 5 to 10 years. The aim was to see if the BHS could predict the eventual number of suicides. Eleven out of 165 participants used in the analysis died by suicide within the study period. The results showed that, with a cut-off score of 10 or more, the BHS had a sensitivity of 90.9% and a specificity of 50.6% in identifying repetition (reported in Beck et al., 1989a). The review team for this guideline calculated a PPV of 11.6% and an NPV of 98.7%. A limitation to drawing conclusions from this study of those with suicide ideation and little history of self-harm is that it cannot necessarily be extrapolated to guiding the longer-term management of those presenting following an episode of self-harm. Another limitation in the interpretation of the results is that this study uses a lengthy follow-up period of up to 10 years, which is not useful in a clinical assessment where the concern is the risk of suicide in the shorter term.

The Scale for Suicide Ideation (SSI) was designed by Beck and colleagues (1979) to assess the severity of suicide ideation in psychiatric patients. It consists of 19 items and each item consists of three alternative statements graded in intensity using a three-point scale ranging from 0 to 2. The items assess a person's wish to die, their desire to make an active or passive attempt, the duration and frequency of suicide ideation, sense of control over making an attempt and how much preparation they have contemplated. The scores for each response are added up to give a possible total score of 0 to 38. Higher scores are associated with a greater suicide risk. The SSI – Current (SSI-C) measures a person's current intensity of specific attitudes, behaviours and plans to commit suicide (Beck et al., 1979) and the SSI at Worst Point (SSI-W) is rated when the suicide ideation is at the worst point in their lives. The SSI-C and SSI-W have high internal and good concurrent validity (Beck et al., 1997).

Beck and colleagues (1999) used the SSI-C, SSI-W and BHS in a cohort study of 3,701 outpatients evaluated at the Centre for Cognitive Therapy in Pennsylvania in the US. Of the participants, 13.3% made a prior suicide attempt. Participants were followed up for 15 years. The aim was to see if the scales could predict the eventual number of suicides. Thirty out of 3,701 participants died by suicide. The results showed that, with a cut-off score of 2 or more, the SSI-C reported a sensitivity of 53%, a specificity of 83% and a PPV of 2.4%. An NPV of 99.5% was calculated from these results. The SSI-W (cut-off score of 16 or more) had a sensitivity of 80%, a specificity of 78% and a PPV of 2.8%. An NPV of 99.7% was calculated from these results. The BHS (cut-off score of 8 or more) had a sensitivity of 90%, a specificity of 42% and a PPV of 1.3%. An NPV of 99.7% was calculated from these results. There are a number of limitations that must be addressed before coming to any firm conclusions about using these scales to predict suicide. First, the study used a sample of outpatients and only a small percentage of them had a history of self-harm. Second, this study, along with the previous studies by Beck and colleagues (such as BECK1985), also had a lengthy follow-up period, which is not useful for a clinical assessment.

The SIS (Beck et al., 1974b) is administered by interview and designed to measure the level of a person's intent to complete suicide once they have already made an attempt. It takes into consideration behaviour and attitudes before, during and after an episode. The scale comprises 15 items that are rated on a two-point Likert scale. The total score ranges from 0 to 30 and is calculated by adding up the scores of each individual item. The completion time is approximately 10 minutes and it is administered by a trained clinician. The scale can be divided into two parts. Part 1 is comprised of the first eight items, which measure the objective circumstances of self-harm. Part 2 is comprised of the remaining seven items, which measures the thoughts and feelings of a person at the time of self-harm.

HARRISS2005A examined the SIS in a cohort study on 1,049 males and 1,440 females who presented to a general hospital following self-harm. Participants were assessed by members of the psychiatric service or data were obtained by records completed in the emergency department. Participants were followed up for 3 to 7 years (5.2 mean year follow-up). The aim was to see if the SIS could predict the eventual number of suicides. Thirty male and 24 female participants used in the analysis died by suicide within the study period. The results showed that, with a cut-off score of 10, the SIS reported a sensitivity of 76.7%, a specificity of 48.8% and a PPV of 4.2% in male participants. In females, using a cut-off score of 14, the sensitivity rate was 66.7%, the specificity was 75.3% and the PPV was 4%. The NPV score calculated by the review team for this guideline was 98.6% (males) and 99.2% (females). The study also provided scores for Part 1 of the SIS in females. The sensitivity was 75% with a cut-off score of 6 and the specificity was 72.6%. The PPV remained at 4% and had an NPV of 99.4%, as calculated by the review team for this guideline. If the cut-off score was increased to 7, this yielded a higher specificity of 80.9% but a lower sensitivity of 66.7%. This study does not provide results that combine male and female subjects, which makes it difficult to generalise to a mixed male and female population of people who self-harm.

NIMEUS2002 used the SIS in a cohort study of 555 participants who were evaluated by a psychiatrist within 12 hours to 5 days following a suicide attempt, and most often as an inpatient of the medical intensive care unit. The participants were followed up from 10 months to 8 years and 10 months (with a mean time of 4 years and 6 months). Twenty-two participants died by suicide during the follow-up period. With a cut-off score of 19, the reported sensitivity for the SIS was 59%, the specificity was 77% and the PPV was 9.7%. The review team for this guideline calculated an NPV of 97.8%. This study also looked at the predictive value of the SIS with participants who were aged 55 years and above. Ten out of 88 participants in this age group died by suicide during follow-up. The reported sensitivity, specificity and PPV were 90%, 60% and 23%, respectively. The review team for this guideline calculated an NPV of 97.9%.

NIMEUS1997 used the BHS in a cohort study on 212 suicide attempters evaluated in the Medical Intensive Care Unit and during psychiatric hospitalisation. Participants were asked to participate in a suicide research programme and were followed up for a mean period of 4 years and 4 months. Thirteen out of 212 participants used in the analysis died by suicide. The results showed that with a cut-off score of 9, the BHS reported a sensitivity of 77%, a specificity of 42% and a PPV of 8%. The review team for this guideline calculated an NPV of 96.5%. When a cut-off score of 13 was used, the sensitivity was still 77% but with a specificity of 61.3% and a PPV of 13% (calculated NPV was 97.6%).

The Suicide Assessment Scale (SUAS) was developed by Stanley and colleagues (1986) and is a clinician-rated scale designed to measure changes in levels of suicidality over time. It consists of 20 items and each item is rated on a four-point Likert scale. Typical items are: ‘sadness and despondency’, ‘hostility’ and ‘anergia’. Ratings are interview based and the completion time is approximately 20 to 30 minutes.

NIMEUS2000 used the SUAS in a cohort study of 191 suicide attempters evaluated in a medical intensive care unit and asked to participate in a suicide research programme. They were followed up for 12 months and eight out of 191 died by suicide during this time. The results showed that with a SUAS cut-off score of 39 the sensitivity of the scale was 75%, the specificity was 86% and the PPV was reported as 19.4%. One important note about this study is that, due to the low prevalence of suicide, the study used a case control design to calculate the predictive validity by comparing the suicide cases with a matched control of 40 participants who did not die by suicide. Using data from the study the review team for this guideline calculated an NPV of 98.7%. A major limitation to drawing a firm conclusion about the usefulness of this scale to predict suicide, based on this study, is that a case control method was used for the analysis of predictive values and therefore any interpretations cannot be generalised to make a clinical assessment.

6.3.6. Clinical evidence summary of scales that predict suicide

There are six studies (all cohort design) that looked at predicting a fatal outcome such as suicide in people who have self-harmed. A scale that has a sensitivity of 100% means that there will be zero false negatives identified by the scale and therefore it will be unlikely to miss any people who will then go on to die by suicide. The scale that reported the highest sensitivity of 91% and the lowest false negative rate of 1 is the BHS used by BECK1985. However, there are major limitations to the interpretations of these results, such as the use of a small sample of mainly suicide ideators and a lengthy follow-up period of 5 years. Furthermore the BHS would identify 76 false positives for every true positive, severely compromising its clinical utility.

Another drawback for these scales is that they all have low PPV scores (between 1% and 13%), therefore identifying many false positives, which makes them of limited use. The low PPV scores are a result of the low prevalence of suicide. A final point to note is that the follow-up period is extremely long in some studies (between 4 and 15 years), in order to increase prevalence. The greatest concern to healthcare professionals is to be able to predict suicide in the next few weeks and months. In the shorter term, the PPV of these scales will be even lower. For these reasons, the use of scales to predict the risk of suicide cannot be recommended in clinical practice. For more information on the predictive scales, see Table 40.

Table 40

Scales that predict suicide.

6.3.7. Scales that predict a repetition of self-harm

Six cohort design studies looked at non-fatal outcomes (CARTER2002, COOPER2006B, COOPER2007, CORCORAN1997, GALFAVY2008, KAPUR2005, WAERN2010).

The Manchester Self-harm Rule (MSHR; COOPER2006B) is a clinician-rated screening tool designed for the initial assessment of self-harm in emergency departments. It contains four questions assessing history of self, past or present psychiatric treatment, and whether the service user has had a benzodiazepine overdose. The assessment of risk is divided into two categories: perceived low risk and perceived moderate or high risk.

COOPER2006B used the MSHR in a cohort study of consecutive service users who presented themselves to five hospital emergency departments following self-harm. The participants were followed up for 6 months and 373 out of 2,095 service users made a repeat attempt to self-harm, 14 of whom died by suicide. The reported sensitivity was 97% (95% CI, 95 to 98%), specificity was 26% (95% CI, 24 to 29%), PPV was 22% (95% CI, 20 to 24%) and NPV was 97% (95% CI, 96 to 99%).

COOPER2007 conducted an analysis of people who self-harmed and presented to an emergency department, 8,825 of whom completed the MSHR and 8,722 of whom had a Global Clinical Assessment (GCA) completed by emergency department clinicians or mental health specialists. The participants were followed up for 6 months, with 1,506 of those completing the MSHR and 1,481 of those completing the GCA making a repeat attempt to self-harm in that period, 59 of whom died by suicide. For the MSHR, the reported sensitivity was 94% (95% CI, 92 to 95%), specificity was 26% (95% CI, 24 to 27%), PPV was 21% (95% CI, 19 to 21%) and NPV was 96% (95% CI, 94 to 96%). For the GCA, the reported sensitivity was 85% (95% CI, 83 to 87%), specificity was 38% (95% CI, 37 to 39%), PPV was 22% (95% CI, 21 to 23%) and NPV was 92% (95% CI, 91 to 93%).

KAPUR2005 conducted a cohort study using the GCA to examine the risk of repetition in 3,828 people who presented to a hospital emergency department following self-harm. The assessment was conducted by emergency department clinicians and mental health staff. For the purpose of this guideline, only the findings of the assessment undertaken by mental health staff are presented here. Participants were followed up for 12 months and 549 patients repeated self-harm, 18 of whom died by suicide. The reported sensitivity was 17% (95% CI, 14.1 to 20.5%), specificity was 92% (95% CI, 90.7 to 92.6%), PPV was 26% (95% CI, 21.3 to 30.2%) and the review team calculated an NPV of 87%.

Kreitman and Foster (1991) developed a clinical and research scale to predict the repetition of self-harm within 12 months. The Edinburgh Risk of Repetition Scale (ERRS) has 11 variables including history of psychiatric treatment, marital status and age between 25 and 54 years old. A positive answer for each item is scored as 1 and scores are added up to give a possible total score of 0 to 11. The ERRS for research use has specific weightings for each item when it is scored.

Hawton and Fagg (1995) conducted two 1-year cohort studies in 1,180 people assessed in routine clinical practice at a UK general hospital following a suicide attempt. The study aimed to compare the performance of a clinical version of the ERRS (with non-weighted items), a research version (with weighted items) and a shorter version (comprising six items) of a scale developed in Edinburgh in 1974 (Buglass & Horton, 1974). Performance was examined in two ways: first, based on the method used by Kreitman and Foster (1991), that is, based on suicide attempts where a repeat attempt within that year indicated repetition (analysis Type A); second, based on individual persons (rather than episodes of attempts) where a repeat attempt within a 1-year period indicated repetition, measured by hospital re-admission (analysis Type B). The results showed that there was little difference in the performance of the clinical and research versions of the ERRS scales when compared with each other, using analysis Type A. These results were compared with analysis Type B, where both versions performed more poorly than in analysis Type A. The performance for the Buglass and Horton (1974) scale was similar to the clinical version of the ERRS, using analysis Type B. It is important to note that this study does not report sensitivity and specificity data, and has been reviewed to illustrate the background and development of the ERRS.

CARTER2002 used the ERRS scale by rating the items based on clinical interview, service user self-report and case notes. This cohort study used 1,317 people who self-poisoned and presented for hospital treatment. The participants were followed up for 12 months and 180 participants made a repeated presentation for self-poisoning. A cut-off score of 8 or more for male subjects and 6 or more for female subjects was used. The ERRS reported a sensitivity of 26%, a specificity of 84% and a PPV of 21%. The review team for this guideline calculated an NPV of 86.7%.

The following scales, as well the BHS and the SSI scales, which have been described above, was used in a study by GALFAVY2008.

The Hamilton Depression Rating Scale (HDRS; Hamilton, 1960) is a clinician-rated scale and consists of 17-items designed to measure the severity of depressive symptoms in people diagnosed with affective disorder of depressive type. Scores on each item are measured on a five-point Likert scale ranging from 0 to 4 (0 = absent, 1 = mild or trivial, 2 = moderate, 3 = severe, 4 = incapacitating), or they can be measured on a three-point scale (0 = absent, 1 = slight or doubtful, 2 = clearly present). The total score is the sum of the item scores and can range from 0 to 53. A score of 0 to 7 is considered clinical remission and at least 20 is low severity. It is advised to have two raters independently score a patient at the same interview and the administration time is approximately 20 to 30 minutes.

The BDI (Beck & Steer, 1987) is a self-report instrument that consists of 21 items constructed to measure the current severity of depression in psychiatric patients. Each of the items is rated on a four-point scale with scores ranging from 0 to 3). Scores for responses are added up to give a possible total score of 0 to 63. The severity of depression is rated as 0 to 9 minimal, 10 to 16 mild, 17 to 29 moderate and 30 to 63 severe.

The Reasons for Living Inventory (RFL; Linehan et al., 1983) is a 48-item self-report measure designed to assess beliefs and expectations for wanting to live as an alternative to suicide in adults and young people. As such, the scale is one of the few instruments that assesses protective factors or beliefs defending against suicidal behaviour rather than focusing on risk factors. Typical items are ‘I believe I can find a purpose in life, a reason to live’ and ‘I believe I can find other solutions to my problems.’ Each item of the inventory is rated at six levels of importance ranging from 1 (‘not at all important’) to 6 (‘extremely important’). Based on factor analyses with adults, the RFL consists of six domains of reasons for living: (1) survival and coping beliefs, (2) responsibility to family, (3) child related concerns, (4) fear of suicide, (5) fear of social disapproval and (6) moral objections. The RFL yields a total score as well as six subscale scores corresponding to each of the above domains.

GALFAVY2008 conducted a cohort study with 304 depressed psychiatric research centre participants, 54% of whom had a history of previous self-harm. Participants were administered the BHS, SSI, RFL, HDRS and BDI, and were followed up for 2 years. Fifty-two participants made a suicide attempt during follow-up, four of whom died by suicide. The BHS (cut-off score 5) had a sensitivity of 0% and a specificity of 100%. For this guideline, the NPV was calculated as 82.8%. The SSI (cut-off score 10) reported a sensitivity of 54% and a specificity of 75%. The PPV was calculated as 30.8% and the NPV as 88.7%. The RFL (0.25 probability cut off) scale reported a sensitivity of 35% and a specificity of 79%. The review team for this guideline calculated a PPV of 25.5% and an NPV of 85.4%. The HDRS (cut-off score 2) had a sensitivity of 4% and a specificity of 94%. The review team for this guideline calculated a PPV of 12% and an NPV of 82.5%. The BDI (cut-off score 16) had a sensitivity of 31% and a specificity of 83%. The review team for this guideline calculated a PPV of 27.3% and an NPV of 85.3%. When interpreting these results, it is important to note that this is not a study of a pure self-harm population but a study of depressed people, some of whom have a history of self-harm.

The SUAS has been described in Section 6.3.5. WAERN2010 used a modified version of the SUAS (Nimeus et al., 2006) in a cohort study of 162 people admitted to an emergency ward and interviewed following a suicide attempt. Participants were followed up for 6 months and 61 participants repeated a suicide attempt, including five suicides. The results showed that, with a cut-off score of 24, SUAS reported a sensitivity of 61% and a specificity of 40%. The review team for this guideline calculated a PPV of 38% and an NPV of 62.9%.

CORCORAN1997 used a statistical model created by entering 11 predictor variables into a logistic regression analysis to identify people who were at high risk of repeated self-harm. The variables included items such as any previous act of self-harm, main method of self-harm used, alcohol taken at the time of the act, and so on (for more details on the method used for identification of these variables, see CORCORAN1997). One hundred and twenty-two participants were admitted to a general or psychiatric hospital following an episode of self-harm and had their data entered into a computer. Participants were followed up for 6 months, of which 26 repeated self-harm. Results were reported for a range of cut-point probabilities, ranging from 0.2 to 0.5. With a cut-off point probability of 0.2 (which has the highest sensitivity score), the analysis gave a sensitivity of 96% and a specificity of 81%. The review team for this guideline calculated a PPV of 60% and an NPV of 99%. It is important to note that data for 100 participants (from the original sample total of 212 participants) could not be entered into the analysis because there was incomplete information for at least one of the 11 variables. The study did report, however, that there was no difference between the excluded and included participants with regards to the predictor variables.

The search identified three case control design studies (BISCONER2007, OSMAN1999, OSMAN2001) that met the inclusion criteria for this guideline. The studies looked at non-fatal outcomes and reported the sensitivity and specificity of the following scales.

The Suicide Probability Scale (SPS) by Cull and Gill (1988) is a 36-item, self-report measure designed to measure the probability of suicidal behaviour in adults and young people aged 14 years and older. Individuals rate the frequency of their subjective experience and past behaviours using a four-point Likert scale, ranging from ‘none or a little of the time’ to ‘most or all of the time’. It has a total weighted score and four subscales based on factor analysis: hopelessness, suicide ideation, negative self-evaluation and hostility. Typical items include: ‘I feel so lonely I cannot stand it’; ‘In order to punish others I think of suicide’, ‘Things seem to go well for me’; and ‘I feel I tend to be impulsive.’ The suicide probability score can be interpreted in relation to an assessed risk level: a score of 0 to 24 represents a subclinical risk level, 25 to 49 represents a mild risk level, 50 to 74 represents a moderate risk level and 75 to 100 represents a severe risk level.

The Adult Suicide Ideation Questionnaire (ASIQ; Reynolds, 1991) is a 25-item self-report measure designed to measure the frequency of suicidal thoughts in clinical and non-clinical adult populations. There are 25 descriptions of negative thoughts and behaviours that a person might experience over 1 month. Individuals rate the frequency of their experience and behaviours using a seven-point Likert scale ranging from 0 (‘I never had this thought’) to 6 (‘almost every day’), and this yields a total score with a corresponding T score and percentile score.

BISCONER2007 conducted a sensitivity and specificity analysis for the SPS and the ASIQ in a case control study on inpatients from an acute psychiatric hospital. Participants were divided into either Group 1 (n = 25), who were admitted for suicide ideation or gesture (suicide risk group), or Group 2 (n = 42), who were admitted for other reasons (comparison group) but also had a history of suicide gestures. The aim was to determine the extent to which the SPS and the ASIQ could correctly classify subjects into their groups. The results showed that with a cut-off score of 50, the SPS reported a sensitivity of 52% and a specificity of 78%. The review team for this guideline calculated a PPV of 70.8% and an NPV of 60.9%. The ASIQ, with a cut-off score of 31, reported a sensitivity of 51% and a specificity of 78%. The review team for this guideline calculated a PPV of 72% and an NPV of 59.5%.

The Suicide Behaviours Questionnaire (SBQ) was designed by Linehan (1981) to measure past suicidal thoughts and behaviour. It is a self-report measure comprising 34 items. To date, many different versions of the SBQ have been used and, furthermore, OSMAN2001 validated a revised version – the SBQ-R. This is a self-report measure comprising of four items, each touching on a different domain of suicide behaviour. These include past suicide attempt (Item 1), frequency of suicide ideation (Item 2), threat of suicidal behaviour (Item 3) and the likelihood of a future attempt (Item 4). Each item is scored using a weighted summary score and the total score ranges from 3 to 18. For Item 1, response is scored on a four-point Likert scale ranging from 1 (‘never’) to 4 (‘I have attempted suicide’), the total score, therefore, ranges from 1 to 4.

OSMAN2001 used the SBQ-R in a case control study on psychiatric inpatients. They grouped adult participants into a suicidal risk subgroup (n = 51) based on hospital admission for recent suicide attempts or serious threats, or a non-suicidal risk subgroup (n = 69) for patients who were admitted for other reasons. The young people were also divided into a suicidal risk and a non-suicidal risk subgroup based on this criteria. The analysis used a SBQ-R total score and Item 1 only from the SBQ-R, to distinguish suicidal versus non-suicidal individuals. The results showed that with a cut-off score of 8, in adults, the SBQ-R reported a sensitivity of 80%, a specificity of 91%, a PPV of 87% and an NPV of 86%. In young people, the reported sensitivity was 87%, specificity was 93%, PPV was 90% and NPV was 99%. For Item 1 of the SBQ for adults, the sensitivity was 80%, specificity was 97%, PPV was 95% and NPV was 87%. For young people, the reported scores were a sensitivity of 100%, specificity of 96%, PPV of 95% and an NPV of 100%.

OSMAN1999 used the ASIQ and the RFL scale in a case control study on psychiatric inpatients. They grouped the participants into a ‘suicide attempter group’ (n = 75) and a psychiatric control group (n =130). The suicide attempter group had made prior or current suicide attempts with an established intent to die. This was measured from assessments by intake staff using various other scale measures (Minnesota Multiphasic Personality Inventory-2, SIS, BHS and the Positive and Negative Affect Scale) as well as the ASIQ and RFL. Group assignment was further endorsed by a review of medical records. The control group did not have a history of suicide attempt. The results showed that the sensitivity (the proportion of suicide attempters that were correctly identified as suicide attempters) using a cut-off score of 14 in the ASIQ identified the maximised sensitivity of 96% and maximised specificity (the proportion of psychiatric controls who were correctly identified as non-suicide attempters) of 79%. The reported PPV was 72% and the NPV was 97%. The RFL showed that a cut-off score of 3.8 yielded greatest accuracy in giving a sensitivity of 61% and a specificity of 81.5%. The reported PPV was 65.7% and the NPV was 75.5%.

6.3.8. Clinical evidence summary of scales that predict repetition of self-harm

There were ten studies (three case control and seven cohort designed) that looked at predicting repetition of self-harm (see Table 41 and Table 42). Sensitivity, specificity, PPV and NPV scores from case control studies may be less generalisable to a real world clinical context than those obtained from cohort studies. The GDG came to the consensus that the evidence needed to potentially formulate recommendations would be derived from studies that used the stronger prospective cohort design.

Table 41

Scales that predict repetition of self-harm (prospective cohort studies).

Table 42

Scales that predict repetition of self-harm (case-control studies).

The prospective studies were examined in more detail to describe the sample and the sample size so that the utility and generalisability of the findings and precision of the estimates could be assessed. All studies with the exception of one included participants who presented to an emergency department following self-harm. The length of follow-up used by these studies varied between 6 months to 3 years. The tool that reported the highest sensitivity of 97% was the MSHR used by COOPER2006B. The limitation of this tool in terms of its clinical utility was its low specificity of only 26%. All the scales had relatively low PPVs ranging from 12 to 60%. This meant that many individuals were wrongly identified as people who would repeat self-harm, thus limiting the clinical utility of these scales and possibly resulting in unnecessary intervention in some individuals.

6.4.1. Introduction

While psychosocial assessment includes several components, the most important are the assessment of needs and the assessment of risks. The assessment of needs is designed to identify those personal (psychological) and environmental (social) factors that might explain an act of self-harm. This assessment should lead to a formulation, based upon which a management plan can be developed. The main components of an assessment of need after self-harm therefore include:

• social situation and environmental issues (including current living arrangements, work and debt, access to means of self-harm)
• family and personal relationships (including recent breakdown of a significant relationship)
• recent life events and current difficulties
• psychiatric history and mental state examination, including any history of previous self-harm and alcohol or drug use
• an assessment of physical health issues
• psychological characteristics that are known to be associated with self-harm (for example, hopelessness, problem solving ability, impulsivity)
• possible protective factors
• current episodes of self-harm including motivation for the act
• attitudes to help/care
• establishment of problem list with service user
• exposure to suicide or self-harm by others (family, environment, media, internet).

Information about the psychiatric, social and psychological factors and contexts of the act can then be brought together into a formulation that describes the antecedents of the episode of self-harm. The formulation should therefore include:

• Long-term vulnerability factors including early loss or separation from parents, difficult relationships with parents signified by rejecting or overprotective parenting styles, or abuse in early life. Although sexual abuse has been associated with self-harm, emotional or physical abuse are also important.
• Enduring psychological characteristics and psychiatric problems.
• Short-term vulnerability including current difficulties in relationships and lack of social support, work or health-related problems, drug and alcohol misuse, or exacerbation of psychological symptoms.
• Precipitating factors are likely to be stressors experienced in the few days immediately prior to self-harm. Again, relationship problems, financial worry, anniversaries, deaths or other losses can act as precipitators to the act of self-harm.

6.4.2. Narrative synthesis

The following section summarises two identified studies, which examined the assessment of need in people who self-harm.

CEDEREKE2007 aimed to look at the specific needs of people who have self-harmed by using a comprehensive needs assessment tool – the Camberwell Assessment of Need. Semi-structured interviews were conducted for participants (N = 140) 1 month and 12 months following emergency treatment for a suicide attempt. The Camberwell Assessment of Need looks at whether a need exists (as rated by a service user) and the severity of the need in 22 areas of ‘every-day living’. The most common areas of need were ‘safety to self (self-harm), psychological distress and physical health’. The aim of the study was to investigate the help that service users received from services (formal help), support from friends and family (informal help), whether help from services was considered adequate and, lastly, whether the amount of help received and its level of adequacy differed between service users who repeated suicide attempts during follow-up. There were 23 service users who repeated self-harm between 1 and 12 months, and 117 who did not. The results showed that after 1 month, a high rate of formal and informal help was received in the most common areas of need, mentioned above. But in areas of need such as intimate relationships, and needs in daytime activities, company and sexual expression, there was little (formal or informal) help received. After 12 months the need for ‘safety to self fell, but the highest rate of help received was still both in this area and psychological distress. There was no difference between repeaters and non-repeaters in that they rated the same main areas and the same severity of need. Furthermore, the help received did not differ at 1 month after the index episode of self-harm. After 12 months, however, repeaters had significantly more needs, such as ‘safety to self, psychological distress and intimate relationships’ to name a few. With regard to the help received, repeaters received more help for psychological distress, intimate relationships and company. Both groups found help from services to be adequate, with the exception of needing more information.

KEENE2005 conducted a descriptive cross-sectional study to look at the assessed needs and service use of a self-harm population. This population comprised people who had self-harmed by self-poisoning, asphyxiation, cutting, burning and other self-inflicted injuries. The first part of this study aimed to look at the assessment of need, including mental health and substance misuse needs. The second part assessed the inter-agency service use of this population, such as health and social care services, and compared it with utilisation patterns of a wider emergency department population. Results showed that 53% (n = 427) had an assessed mental health problem, 18% with drug or alcohol problems and 15% with a dual diagnosis. Only 10% had no assessed need. Results also looked at the proportion of each assessed need group and their referral to the relevant external agencies. Seventy per cent of those who had self-harmed and had mental health problems were referred to mental health services, 64% with drug problems were referred to a drug agency and 35% with alcohol problems were referred an alcohol agency or for detoxification from alcohol. Overall, 37% of those with at least one assessed need were referred to mental health services, 3% to a drug agency, 6% to an alcohol agency, 15% to social services, 16% to a GP, 15% to follow-up or outpatient services and 9% were discharged with no further service. When comparing those who self-harmed with the rest of the emergency department population over 3 years, it could be seen that the former were three times more likely to contact social services and ten times more likely to attend drug or alcohol agencies. This study described the service use of a small population of those who had self-harmed in the hope that inter-agency integrated care services could provide a better service for this population and help inform the development of integrated care initiatives.

6.5.1. Narrative synthesis

The following section summarises identified studies, which examine the psychosocial assessment of people who have self-harmed. Some studies that are relevant to answering the clinical question, investigated whether receiving an assessment had an effect on repetition of self-harm or adherence to treatment (BERGEN2010b, HICKEY2001, KAPUR2008, OUGRIN2011, WITTOUCK2010). These studies compared groups of participants who received an assessment with those who did not or those who were given treatment as usual. A further two studies (HAW2003b, KAPUR2003) looked at these outcomes but did not make a comparison between groups of participants in their study design.

HICKEY2001 conducted a study where they looked at people who presented to hospital over 2 years following self-harm, comparing those who received a psychosocial assessment with those who were discharged from hospital without an assessment, and whether these participants differed in characteristics and subsequent self-harm. A psychiatric team conducted the assessments in a general hospital in the UK. When comparing their characteristics, the non-assessed participants (n = 145) were more likely to have a history of self-harm (p < 0.02), were recorded as showing difficult behaviour (p < 0.02), were uncooperative during physical examination (p < 0.05), took early discharge from the emergency department (p < 0.0001) and were less likely to have further healthcare arrangements made (p < 0.0001), compared with the assessed participants (n = 101). The non-assessed group also had more incidence of self-poisoning (74%) as opposed self-injury, compared with the assessed group (79%).

In a follow-up study, after 12 months of the index presentation, non-assessed participants were matched on characteristics (age, sex and type of self-harm) with assessed participants (control) and there were 88 matched pairs in this comparison. The participants' GPs were contacted for information on psychosocial variables of the participants within 1 year of the index presentation. There was no significant difference between the groups in terms of psychosocial problems, although these were more common in the non-assessed group. According to the monitoring system data, more non-assessed participants repeated self-harm within 28 days of the index episode compared with assessed participants, although this was not statistically significant. After 12 months, three times as many non-assessed participants repeated self-harm compared with assessed participants. When these data were combined with GP data, the results showed that 37.5% of the non-assessed participants repeated self-harm within 12 months compared with 18.2% of the assessed participants. One limitation to this study is that it excluded participants who were in inpatient psychiatric care at the time of the index presentation; therefore, few participants had a history of psychiatric care or self-harm and so the results might not be generalisable to all people who self-harm.

KAPUR2008 carried out a large multicentre research project in the UK for people who presented to hospitals following self-harm. The study aimed to establish factors associated with receiving or not receiving a specialist psychological assessment and whether this was associated with repetition of self-harm. Taking into account various social and clinical characteristics of the 7,344 participants, and also the type of substance or method they used to self-harm, key characteristics were associated with an increased likelihood of having an assessment in hospital. These were being aged over 55 years, having current psychiatric treatment, being admitted into a medical ward and taking antidepressants. A person less likely to receive an assessment was: unemployed, used self-cutting, chose to self-discharge from hospital and attended a hospital outside working hours. Overall, a repeated subsequent attempt could not be predicted based on whether one has or has not received an assessment. However, in some hospitals having an assessment appeared to reduce the risk of repeated self-harm and these hospitals tended to have a higher proportion of assessed episodes, whereas in others having an assessment appeared to increase the risk and these hospitals had a lower proportion of assessed episodes. Only 60% of self-harm episodes resulted in an assessment, overall. The findings of this and other similar studies need to be interpreted in the context of their observational design, which means that the observed associations may not be causal and could well have been influenced by unmeasured confounders.

BERGEN2010b carried out a survival analysis to examine, first, the association between psychosocial assessment by a specialist mental health practitioner following self-harm and a subsequent repeated episode, and, second, the association between having a psychosocial assessment after an episode of self-harm and the survival time until the repeated episode. The study was conducted in three UK centres where 13,966 participants had made a hospital presentation for a first episode of self-harm in the study period (2003 to 2005) and 55.6% had received a specialist psychosocial assessment. There were 18,483 repeated episodes of self-harm within the following 2 years, which included up to the first six episodes only for each person. Participants received a specialist psychosocial assessment for 54.7% of these episodes. The following results look at ‘short term repetition’ or the time to the first repeated episode. For participants who did not have a history of psychiatric treatment, the risk of repeating an episode of self-harm was 51% (95% CI, 42 to 58%) less if they received an assessment compared with not receiving an assessment. Likewise, for participants who did have a history of psychiatric treatment, having an assessment reduced the likelihood of a repeat episode by 26% (95% CI, 8 to 34%) compared with not having an assessment, with other variants controlled. For recurrent repetition, for all six episodes, results showed that participants who did not have a history of psychiatric treatment, the risk of repetition was 57% (95% CI, 51 to 63%) less if they received an assessment at the last episode compared with not receiving an assessment, controlling for covariates. Likewise, for those who had a psychiatric treatment history, having an assessment reduced the likelihood of a repeated episode by 26% (95% CI, 11 to 41%) compared with not having an assessment.

OUGRIN2011 conducted an RCT with young people who self-harmed to examine whether therapeutic assessment compared with assessment as usual improves attendance and engagement during a 3-month follow-up period. This study was set in CAMHS settings in London. Assessment as usual involved standard psychosocial evaluation and standard disposition planning. The therapeutic assessment was carried out by trained healthcare professionals and involved the same components as assessment as usual in addition to a brief therapeutic intervention. When looking at the attendance rate of the first follow-up session, results showed that subjects in the therapeutic assessment group had better attendance rates (OR 5.19, 95% CI, 2.22 to 12.10). Subjects were also given the Strengths and Difficulties Questionnaire (SDQ) and the results showed that although there was an improvement in scores for all subjects, there was no significant differences in changes of scores between the groups (mean difference −0.37, 95% CI, −3.25 to 2.53) and this was also seen in the Children Global Assessment Scale scores (mean difference 4.49, 95% CI, −0.98 to 9.96). There was insufficient evidence to make any conclusions regarding the difference between therapeutic assessment and assessment as usual. Limitations of this study include a short follow-up period and that the effectiveness of standard clinical practice interventions for young people who self-harm is questionable.

WITTOUCK2010 conducted a longitudinal study to examine people who attempted suicide and their compliance with aftercare following standardised psychosocial assessment. The study group was assigned to assessment using the Instrument for Psychosocial Evaluation and Care of Suicide Attempters (IPEO) (n = 93) compared with people who attempted suicide and received a non-IPEO based psychosocial assessment (n = 38). Semi-structured interviews were conducted at 1 and 6 months' follow-up (FU1 and FU2, respectively) after an index episode (within the study period). Outcomes measured included contact with emergency department staff (during hospital admission), GP and mental health services. The results showed that participants who had no inpatient history (OR 2.73, 95% CI, 1.18 to 6.29) or those who had only had one inpatient admission for treatment (OR = 7.15, 95% CI, 1.43 to 35.7) were more likely to receive an IPEO-based assessment compared with participants with two or more previous inpatient admissions for treatment. There was no difference between the study groups in terms of the treatment advice received and compliance with fixed referrals; however, participants in the IPEO group were more likely to have treatment options discussed with them than the non-IPEO group (OR 3.2, 95% CI, 1.23 to 8.45). During FU1, 62% of the participants who visited their GPs did so within 1 week of their index attempt. There was no difference between the study groups in the number of participants who visited their GP. However, during the period of discharge from hospital to visiting the GP, the non-IPEO group visited their GP 3.6 times less within 1 week after the index attempt (95% CI, 1.1 to 11.9). During FU2 there was no difference in the groups regarding the regularity of visits to a GP. At FU1, there was no difference between the groups in receiving mental healthcare, outpatient mental healthcare or pharmacological treatment. However, the non-IPEO group was more likely to receive inpatient mental healthcare than the IPEO group (OR 3.1, 95% CI, 1.4 to 6.8). At FU2, no differences in treatment received remained between the two groups. There was a high dropout rate of 37% between FU1 and FU2.

HAW2003B conducted a prospective study on 135 participants who presented to a general hospital in the UK following self-harm and were given an assessment by the hospital self-harm service, a specialist service that aimed to provide a rapid psychosocial assessment and aftercare for people who have self-harmed. After the index hospital presentation, participants were followed up for between 12 and 20 months to assess repetition of self-harm, treatment adherence and satisfaction. One hundred and six patients (79%) reported how satisfied they felt with their psychiatric assessment at follow-up and a majority felt that the assessor showed understanding and that their problems were taken seriously.

Only 33 (24%) of the participants who were assessed were offered an outpatient appointment by the self-harm service because there was a criteria for this offer, such as being at high risk for further self-harm. Twenty (61%) of these participants attended their first follow-up appointment, but there was no statistical difference between the characteristics of these people and those who refused an appointment or did not turn up. Nineteen (60%) participants had a follow-up interview and the majority of these were satisfied with the care that they had received from the self-harm service. Four out of those 19 (21%) participants who reported satisfaction levels of the treatment received had also reported a further episode of self-harm, whereas 30 out of 87 (35%) participants who were not offered treatment reported a repeat episode of self-harm. Although the repetition rate of participants who received aftercare was low, there was no statistical difference between these groups.

The self-harm service provided 53 (39%) participants who were thought to be at high risk of further self-harm with an emergency telephone number if needed in a crisis. Forty one of those 53 (77%) participants were seen at follow-up and six (15%) reported a repeat episode of self-harm during this period.

When interpreting the findings of this study, it is important to note that the authors defined self-harm as including self-poisoning and self-injury but excluded self-cutting that was considered part of a repetitive pattern of self-mutilation. Another limitation to interpreting the findings of this study is that the sample size of participants offered outpatient appointments and interviewed regarding satisfaction of the psychosocial assessment that they received is small. Furthermore, experiences of the psychiatric assessment were measured after the 12- to 20-month follow-up period, where 15 participants could not then remember or report on the assessors' attitude towards their problems. This study had an original sample of 150 participants presenting to hospital for self-harm. Ninety per cent of these received an assessment; however, this study did not compare outcomes between groups of participants who received and did not receive a psychosocial assessment.

KAPUR2003 conducted a prospective study to look at the differences in six UK hospitals with regards to their management (including rate of psychosocial assessments received by participants) and direct costs associated with participants presenting for self-poisoning. Three of these hospitals had a multidisciplinary self-harm team consisting of medical or nursing staff as well as social workers who carried out the assessments. Over a 5-month study period there were 1,778 episodes of self-poisoning involving 1,306 people aged over 16 years. When looking at differences between hospitals, rates of admission following an episode of self-harm varied from 16.5 to 81.3%, rates of psychosocial assessment varied from 28.5 to 57.7%, rates of admission to a psychiatric bed varied from 1.8 to 6.2% and the rate of specific follow-up being arranged by the hospital varied from 16.3 to 58.6%. Hospital costs (including capital charges and general services) ranged from £228 to £422. The rate of repetition varied from 10.3 to 16.1%, but the difference between hospitals was not statistically significant. Furthermore, the rate was similar in the hospitals with a self-harm team (14%) and with no such team (15.2%). Six hundred and four participants who had presented in the first 8 weeks of the study were followed up to measure repetition of self-poisoning within 12 weeks of their index presentation of which 88 (14.6%) participants repeated self-poisoning. This study found that the repetition rate of participants receiving an assessment was 9.8% compared with 17.9% in those who did not receive an assessment, and this association reached statistical significance (p < 0.005) even when adjusted for differences in participants' characteristics such as age, sex, substance dependence, previous self-harm and current contact with psychiatric services (adjusted OR 0.42, 95% CI, 0.25 to 0.71, p <0.005). One limitation of this study is that it did not measure rates of repetition in a larger sample. If this had been done, there may have been more significant differences noticed between teaching and district hospitals, hospitals with or without specialist teams and in terms of aspects of management.

6.5.2. Summary

Due to the studies being very different from each other and therefore not meta-analysable, there was insufficient evidence to draw any conclusions regarding the association between psychosocial assessment and improvement in outcomes. Nevertheless, psychosocial assessment is an important part of developing care and management plans for self-harm. Reviews conducted in Section 6.2 and Section 6.3 may inform areas to explore during psychosocial assessment.

Risk assessment

Based on the evidence review conducted in Section 6.2, the following risk factors in particular should be considered when assessing risk of repeated self-harm or suicide: previous self-harm and depressive symptoms. These two factors were supported by pooled quantitative analysis. Previous self-harm before an index episode is the most robust risk factor predicting both repetition and suicide following self-harm. The size of the evidence base and the adjustment of confounding variables provide stronger support for this risk factor. Another factor, depressive symptoms, is also important but less robust.

Other risk factors, such as psychiatric illness and current and past suicidal intent, should also be taken into account. The personal and social context associated with the behaviour and any other specific antecedent factors should be noted. Individual risk and protective factors that may increase or decrease risks associated with repetition of self-harm are important as well. Interpersonal relationships with family or significant others may also lead to possible changes (positive or negative) in the level of risk. It is important to note that risk factors often overlap with each other, and measuring one may be a proxy measure for another. The association between factors does not imply any causal relations. Therefore, the evidence review in Section 6.2 is only intended to give guidance on factors to consider in psychosocial assessment, and should not be used for predicting risk.

Risk assessment tools and scales

No risk scale can be recommended for use in isolation to distinguish people at risk of repeated self-harm from those who are not. Based on the evidence reviewed in Section 6.3, there are major limitations to making a recommendation for the use of a scale alone to predict whether a person who has a history of self-harm will go on to die by suicide. The results of the risk assessment scales show that is it also almost always likely to miss people who will go on to die by suicide. The main limitation is that suicide in nature is a rare outcome, therefore, the prevalence will always be low which makes it difficult for scales, when tested, to correctly identify the probability that a person with a positive test result really has self-harmed.

There are also limitations for making a recommendation for the use of a scale alone to predict a repeated non-fatal episode of self-harm. Although some scales perform well in correctly identifying the number of people who self-harm and were classified as high risk, they perform poorly in correctly identifying those who were categorised at low risk. Furthermore, the scales will identify many individuals as high risk, who do not go on to self-harm, reducing their clinical utility. As a result, risk tools and scales should not be used to predict future repetition or suicide following self-harm; neither should risk tools or scales be used to determine treatment offers or discharge decisions. However, risk tools may be considered to prompt, add detail and help structure psychosocial assessments.

In addition, it is also good practice to identify and agree with service users the specific risks for them. Healthcare professionals should differentiate between long-term and more immediate risks, and monitor any changes in risks and associated factors.

Psychosocial assessment

Based on current literature reviewed in Sections 6.4 and 6.5, it is difficult to draw conclusions regarding the association between psychosocial assessment and improvement in outcomes. Nevertheless, an integrated psychosocial assessment should be regarded as part of the therapeutic process to engage the service user.

A comprehensive psychosocial assessment including an assessment of needs and risk should be carried out on all those who have self-harmed. This includes people from black and minority ethnic groups, children and young people, as well as people older than 65 years of age. Assessment should follow the same principles as for adults who self-harm in each subgroup.

In Chapter 4, the qualitative literature highlights the importance of exploring the meaning and functions of self-harm for each individual. Health and social care professionals should treat each episode in its own right and acknowledge each person who self-harms does so for individual reasons.

6.7.1. Recommendations

6.8.1. Research recommendation

6.8.1.1.

The effectiveness of psychosocial assessment with a valid risk scale, compared with psychosocial assessment, for the management of people who self-harm (including young people)

For people who self-harm (including young people), does the provision of psychosocial assessment with a validated risk scale, compared with psychosocial assessment alone, improve outcomes?

This question should be answered using a well-conducted RCT. The assessment should be conducted by mental health professionals in CMHTs. The main outcomes should include both hospital-reported and self-reported repetitions of self-harm. Outcomes such as service users' experience of assessment and the impact on therapeutic engagement should also be included. The duration of the study should be at least 6 months.