How to improve testing when trying to predict inmate suicidal behavior

Abstract

Objective

To measure the predictive power of the Suicide Probability Scale (SPS) in a male inmate population (federal penitentiary) with the added contribution of actuarial data.

Method

SPS scores and data from the files of 518 inmates were analyzed in relation to their suicidal behaviors over the following 10 years.

Results

During this period, 12 inmates committed suicide (2.32%), 43 engaged in non-lethal self-harm (8.3%) and 15 expressed serious suicidal intention (2.9%), for a total of 70 (13.51%) who manifested at least one form of suicidal behavior. The records of the 518 inmates allowed identifying seven actuarial variables (out of 24 documented) that distinguished the group that acted out. These variables were tested in combination with the SPS score to determine the best predictive models of suicidal behavior. Depending on type of suicidal behavior and on observation period, the following seven variables could prove useful in improving the predictive capacity of the SPS: age, prior suicidal behavior, borderline personality disorder, length of sentence, number of sentences, prior incarceration in a provincial prison, and juvenile priors. However, analyses did not allow developing a better predictive model for the specific subgroup of suicide completers.

Conclusions

SPS is improved when adding actuarial data.

Introduction

Suicide is a much more frequent phenomenon in prison settings than in the general population (Fazel et al., 2010, Konrad et al., 2007). Research has allowed identifying factors that explain this higher prevalence (Haines and Williams, 2003, Harvey and Liebling, 2001, Wichmann et al., 2000). However, the conclusions reached in various studies have been contradictory at times (Lohner & Konrad, 2007) owing to differences in how the variables examined were grouped and defined. What's more, the absence of control groups comparable to the population under investigation has made it harder to generalize results.

Despite these difficulties inherent in the research process, Lohner and Konrad (2007) identified various suicide risk factors to emerge from control-group studies of suicidal inmates. Those investigated in a prison population of adult males sentenced to terms of at least two years include the following: being single; having been homeless; having committed a violent offense; prior imprisonment; multiple disciplinary reports; having been placed in solitary confinement; victim of intimidation; presence of mental disorder; having received psychological or psychiatric treatment; history of self-harm; prior suicide attempts; history of self-harm in the homes of significant others; and psychotropic drug abuse or dependence. All in all, we know that the rate of mental disorders is higher in prisons and that mental disorders are strongly associated with suicide (Harris & Barraclough, 1997). This is especially true with opiate-dependent persons, who represent a specific group to be considered when planning suicide prevention in prisons (Bird, 2008, Gore, 1999).

All these offenders seem to carry these risks with them to prison and continue to carry it with them after they are released (Daigle & Naud, 2012). Under this essentially psychological take on the problem referred to as importation theory, inmate suicidal behaviors are primarily the product of previous experiences and personal characteristics (Camilleri, McArthur, & Webb, 1999). On the other hand, being imprisoned constitutes a new stressful event on top of any previous others experienced even for healthy inmates, especially as it cuts the individual off from a host of key resources (Konrad et al., 2007). According to this more sociologically oriented explanation known as deprivation theory, situational and environmental factors are what primarily account for the high suicide rates among inmates. Between these two theories, there is evidence to suggest that the former weighs more heavily in the balance.

This balance between importation and deprivation has been studied as regard three perspectives: (1) long-term risk of dying (Fergusson and Lynskey, 1995, Lattimore et al., 1997, Laub and Vaillant, 2000, McDonald, 1998, Pritchard et al., 1997, Sattar, 2003, Thompson et al., 2006, Tremblay and Paré, 2002); (2) risk of dying before imprisonment or after release (Binswanger et al., 2007, Björk and Linqvist, 2005, Forsman and Holmberg, 1998, Pratt et al., 2010); (3) risk of dying inside vs. outside prison (Brophy, 2003, Fleming et al., 1992, Kariminia et al., 2007, Sattar, 2003, Shaw et al., 2003). The results of most of these studies lend further support to importation theory. Accordingly, offenders could be suicidal throughout their life and even more so when outside prison (Daigle & Naud, 2012).

Regardless of these two opposite or complementary perspectives, however, it is important to keep in mind that the correctional system remains responsible for treating those in its charge. As Liebling (1998, p. 62) stated, “whilst a number of risk factors are, to a large degree, set on arrival within the institution, the effects of the additional stress presented by the prison environment can be manipulated by staff and managers to decrease the risk of suicide”.

Various tools for predicting suicidal behavior have been developed around the world to predict suicidal behavior (Arboleda-Florez and Holley, 1988, Arboleda-Florez and Holley, 1989, Blaauw and Kerkhof, 2006, Blaauw et al., 2001, Dahle et al., 2005, Eyland et al., 1997, Frottier et al., 2009, Mills and Kroner, 2004, Mills and Kroner, 2005, Perry and Olason, 2008, Polvi, 1997, Power and Moodie, 1997, Sherman and Morschauser, 1989, Wichmann et al., 2000). Many of these are very specific to correctional settings in that often they were constructed on the basis of risk factors identified in prison populations. However, psychometric instruments employed with the general population can prove useful with inmates as well. The objective of this paper is to explain how such an instrument, the Suicide Probability Scale (SPS; Cull & Gill, 1988) can be useful, especially with the added contribution of actuarial data.

The SPS is a self-report measure for predicting suicidal behavior. It was developed with a sample of suicide attempters and conceived for use with adults and adolescents 14 years of age or over. The instrument requires respondents to be able to read at a fourth-grade level. It consists of 36 items rated on a 4-point Likert-type scale and takes 20 min to complete.

Its designers reported internal consistency of .93, test–retest reliability of .92, split–half reliability of .93, and four factors. Eisenberg, Hubbard, and Epstein (1990) reported 100% sensitivity and 50% specificity in a study with 1397 Veterans Administration patients. Gutierrez, Osman, Barrios, and Kopper (2001) found a Cronbach's alpha coefficient of .92. Tatman, Green, and Karr (1993) obtained comparable reliability in a study with a group of 217 adolescents (alpha coefficient of .90 for the overall test). In her review of the test, Range (1989) concluded that the SPS is a short, clear screening instrument with very good psychometric properties given its brevity.

In validating a French-language version of the SPS, Labelle, Daigle, Pronovost, and Marcotte (1998) repeated the reliability studies with three Canadian populations (150 university students, 1092 adolescents, 600 male inmates). Internal consistency for the total score was relatively high (.87, .93, .89, respectively) and comparable to that reported by the instrument's creators.

Cull and Gill (1988) performed a factor analysis of the various subscales, the different groups and the SPS total score (no eigenvalues are reported). The correlation matrix yielded significant results between all the subscales (r ranging from .42 to .92). The study by Labelle et al. (1998) obtained similar results. All correlation coefficients were found to be significant and the strongest correlation was noted between the Hopelessness and Suicide Ideation subscales. Correlations varied from .57 to .93, attesting to the instrument's internal validity.

Though Cull and Gill never researched the SPS's predictive validity themselves, they did suggest that it be done with a population at high risk, and over a precise period of time. Yet, despite widespread use of the SPS in various settings and in research on suicide prevention, very few studies have bothered to verify the actual validity of this scale in predicting suicidal behavior (Brown, 2001).

Larzelere, Smith, Batenhorst, and Kelly (1996) showed, with adolescents in group-home treatment (552 boys, 277 girls), that the SPS total score had a certain capacity to predict suicide attempts in the three years ahead: in the group of respondents who obtained a weighted total score of 70 or more, 8.1% engaged in suicidal behavior during their stay at the home. However, the likelihood of false negatives was high: 52% of the youths who attempted suicide had obtained a weighted total score of 69 or less. For many, then, low risk at admission did not mean absence of suicidal behavior thereafter.

Though instruments like the SPS are less specific to correctional settings, they tend to be more elaborate. They generally take into account cognitions, emotions and behaviors specific to suicidal persons in the general population. In this regard, Naud and Daigle (2010) demonstrated that the SPS, used as a sole tool, had by itself a certain predictive capacity in a prison population of adult males at the beginning of their sentence. This prospective study documented over a period of 11.5 years the suicidal behaviors of inmates previously evaluated with the SPS. The SPS probability score and the scores on the instrument's four subscales were found to be significantly higher in the group that engaged in at least one of three types of suicidal behavior: suicide completion, non-lethal self-harm and serious suicidal intention. However, the SPS did not differentiate the inmates who went on to complete suicide. The instrument's capacity to predict suicidal behavior proved limited in terms of sensitivity. In this regard, the SPS's low and high risk levels correctly identified 36% of inmates who later engaged in suicidal behavior. The instrument obtained a specificity rating of 85%. Despite these limitations, the SPS remains a useful and reliable instrument for screening for suicidal behavior in correctional settings.

For the purpose of reaching a clinical judgment, it has been proposed that the SPS's predictive capacity might be enhanced by merely taking into account also a series of precise actuarial data. In this regard, Larzelere et al. (1996) sought to boost the instrument's predictive capacity in a sample of incarcerated young offenders (552 males, 277 females) by considering two variables only: age and history of suicidal behavior. Their attempt, however, proved unsuccessful.

In addition, existing screening instruments have been shown to have two shortcomings. First, the various studies that have used them have recognized that they present a margin of error of 25%. In other words, a certain number of false positives (persons found to be at risk but who actually are not) and false negatives (persons found to be non-suicidal but who actually will be) were to be expected. Second, Stolberg, Clark, and Bongar (2002) expressed an interesting caveat with regard to “suicide risk prediction formulas” (sic), pointing out that most research has focused on suicide attempters and not suicide completers. Consequently, the factors covered by these instruments were often no more than indicators of risk for suicide attempt and/or repeated self-injury. If suicide completers differed from these two groups, then the various “formulas” used to construct instruments would be less specific to suicide completion notwithstanding the many points in common between the different forms of suicidal behavior. In this context, our main hypothesis was that the SPS's predictive capacity might be enhanced by merely taking into account also a series of precise actuarial data.