Predicting suicidal behaviour

Hopelessness is often cited as a good predictor of suicidality, as touched upon earlier. For example, Beck and colleagues have found that hopelessness predicts suicide in patients who had been hospitalised because of suicide ideation (Beck et al., 1985) and in general psychiatric outpatients (Beck et al., 1990). For suicidal inpatients, a score of 10 or more (out of 20) on the Beck Hopelessness Scale successfully identified 10 of 11 patients who completed suicide in a 5-10-year follow-up (Beck et al., 1985). In a sample of almost 2000 outpatients, a score of 9 or above correctly identified 16 out of the 17 suicides that occurred in a 3-4-year follow-up. As would be expected, hopelessness outperformed depression in prediction. However, the cost of such high sensitivity (not missing those who are at risk) was poor specificity. Specificity refers to the ability to avoid labelling people as being at risk when they are not; that is, the ability to avoid false positives. In both of these studies, the rate of false positives was very high. In the inpatient study, 88% of those identified as being at risk did not commit suicide, and the rate was 98% in the outpatient study.

Other studies have used a much wider range of factors to try to predict suicidal behaviour, with similar results. There are many ways in which those who commit suicide or attempt suicide differ, on average, from those who do not. These risk factors include socio-demographic factors, such as gender, class, and employment status; psychological factors, such as poor problem solving and hopelessness; and psychiatric factors, notably depression. Sometimes completers and attempters differ from the general population in the same way, and sometimes they diverge. These factors are discussed in more detail by Maris et al. (1992) and Williams (2001). Pokorny (1983) followed almost 4800 psychiatric inpatients over a 5-year period. Sixty-seven of the group committed suicide, a rate of 1.4%. A predictive model based on a range of known risk factors correctly identified 35 of the 67

as being at high risk. This moderate success of 'hits' was offset by over 1000 false positives. Trying to improve specificity has the effect of reducing sensitivity. For example, in a similar type of study, Goldstein et al. (1991) managed to reduce the false-positive rate but at the cost of 'missing' all 46 suicides in their sample! The problem is translating group differences into prediction at the individual level. Powell et al. (2000) compared 112 people who committed suicide while in hospital with a group of randomly selected control patients from the same hospitals. There were many differences between the two groups, and a number of the variables were statistically significant risk factors (that is, clearly differentiated between the groups). However, using these factors to predict at an individual level whether someone would commit suicide was of little use: only two of the 112 patients who committed suicide had a predicted risk of suicide above 5% based on these risk factors.

Predicting parasuicide has also been shown to be difficult, though one of the problems that besets suicide prediction—low base rates—is less marked. Kreitman and Foster (1991) identified 11 factors that predicted repetition of parasuicide, including previous parasuicide, having a personality disorder, and having high alcohol consumption. Depression was not one of the identified risk factors in their study. Repetition was linked with the number of risk factors present: those who had three or fewer risk factors showed a 5% repetition rate, whereas those with eight or more had a 42% repetition rate. However, as Kreitman and Foster (1991) point out, a large majority of their repeaters were in the mid-range of risk scores, a finding which simply reflected the fact that the vast majority of the sample scored in the mid-range. If a cutoff of 8 or above was adopted for prediction, 76% of those who repeat would be missed. Adopting a lower cutoff would improve the hit rate, but it rapidly increases the number of false positives.

Predictive models fail for a number of reasons. Suicidal behaviour, especially suicide, is relatively rare, and rare behaviours are inherently difficult to predict. Many of the factors used in these predictive studies do not reflect the psychological state of individuals but instead appear to be quite distant from the actual behaviour (MacLeod et al., 1992). This can be seen most clearly in socio-demographic variables such as ethnicity, gender, and age, where, although there may be a statistical connection between these variables and suicidality, there are clearly many mediating factors that translate these variables into suicidal behaviour. More generally, predictive models, or summative checklist approaches to risk, cannot take into account the individuality of the person. Someone may have many risk factors, but these can be 'trumped' by an overriding protective factor, such as a feeling of duty or responsibility to family, or religious beliefs prohibiting suicide. Conversely, someone may have very few risk factors, but the few that they do have are very important, such as a recent major loss, and so they may be at high risk. As Shea (1999) notes, people don't kill themselves because statistics suggest that they should. The call to suicide comes from psychological pain. Each person is unique. Statistical power is at its best when applied to large populations, and at its weakest when applied to individuals. But it is the individual who clinicians must assess. (Shea, 1999, p. 11)

Letting Go, Moving On

Letting Go, Moving On

Learning About Letting Go, Moving On Can Have Amazing Benefits For Your Life And Success! Don't be held back by the past - face your guilt and fears and move on! Letting go is merely arriving at a decision, no more allowing something from the past tense to influence your life today or to cut down your inner sense of peace and welfare.

Get My Free Ebook

Post a comment