Suicidal behaviour not only includes suicides, but also those suicide attempts which do not result in the patient's death. Mental phenomena, such as suicidal impulses or unconsummated ideations can also be included, as well as a wide range of behaviours which are pernicious to the patient's health without a previous explicit declaration of a suicidal intention [6,7].

A variety of factors are associated with an increased risk of suicide and attempted suicide [8], including psychiatric disorders [9-14], feelings of hopelessness and impulsivity [15,16], history of previous suicide attempts [9,17,18], age, sex and race [19-24], marital status [25,26], occupation [27], comorbidity [28,29], adverse childhood experiences [30], family history [31,32], and accessibility to weapons, especially firearms [33].

A recent paper [34] highlights the similarities and differences among heterogeneous groups of patients, some of whom committed suicide, and others who survived the attempt. The authors conclude that depressive symptoms are present in both groups as a shared variable.

Some of the individuals who attempt to commit suicide (by physically inflicting self-harm or ingesting a product at toxic dose levels) are taken to the Emergency Departments of General Hospitals, while others remain in their family or social environment after the attempt, with no specific health care; no official statistics on attempted suicides are available.

The association of psychiatric illness as a predictor of suicide attempts has been reported in the literature [9,10]. More than 90% of patients who attempt suicide have a major psychiatric disorder [11,12], and 95% of patients who committed a suicide attempt had a psychiatric diagnosis [13]. A very recent project [40] from the World Health Organisation (WHO) produced results on the association between the diagnosis of Mental Disorders and Suicidal Behaviours in several countries (including Spain), supported by worldwide data obtained from surveys in general populations (108,664 people from 21 countries on five continents answered the surveys). One of its conclusions is that the presence of a psychiatric diagnosis with DMS-IV Mental Disorder criteria is a strong predictor for the appearance of suicide ideations and behaviours, as well as for consummated suicide. This applies generally to both economically developed countries and to developing countries, except for one important difference: among the Mental Disorders associated with higher suicide risk, Affective Disorders stand out in developed countries, while the association is higher with diagnoses other than Affective Disorders, such as Impulse-Control Disorders, Substance Use Disorders and Post-Traumatic Stress Disorders, in developing countries.

In a large sample of patients in Sweden [10], the authors concluded that the category of psychiatric diagnosis coexisting with a suicide attempt is a variable which influences the future risk of consummated suicide. Planned reduction of the risk of future suicide among those patients requires implementing care plans, mainly in the first two years following the current suicide attempt, in particular among those patients diagnosed with Unipolar and Bipolar Affective Disorder, and Schizophrenia.

Although the risk of suicide increases with age, it has been demonstrated that in young adults, suicide attempts are more frequent than in older adults [19,20]. These findings, in relation with age, are consistent with those of the present study. In a study of 10,892 suicides and 57,439 attempted suicides among hospital-admitted individuals in 8 states, groups with high attempted suicide rates were teenagers, young adults, women, and people of Caucasian and Afro-American origin aged 25 to 44 years [19]. Another study in Korea with 344 attempted suicides found that a significantly higher number of women than men were admitted to emergency rooms due to attempted suicide during the study period [21]. In a random sample of 3021 adolescents aged 14-24 years, the females who attempted suicide showed suicidal thoughts and suicide attempts significantly more often, and suicide attempts at a far younger age than the males [23].

A systematic review of studies on the epidemiology of suicide published from 1997 to 2007 showed that non-fatal suicidal behaviours are more prevalent among women and young, unemployed, unmarried individuals, with low levels of education and suffering from a psychiatric disorder [1]. Our results are consistent in relation with age, gender and psychiatric disorders, but they do not coincide with the aforementioned study with regard to the variables of educational level and unemployment. In the univariate analysis we found a higher rate of suicide attempts among working populations, and with higher educational levels. The utilisation of health care services is probably related to socioeconomic status and educational level. Nonetheless, owing to the fact that we only have information on those patients who finally went to the hospital, we cannot make inferences regarding this point. In the multivariate generalised estimating equation logistic regression analysis, the association with educational level and employment disappears.

The variable most strongly associated with suicide attempts in the study is that of a prior history of attempted suicide. Ecological studies showed a positive correlation between rates of attempted suicide and suicide rates for both sexes among young people in Europe [41]. Prospective monitoring studies of patients hospitalised as a result of a suicide attempt show that the risk of suicide and all causes of mortality were very high immediately after discharge [9]. In a systematic review of data published on the following-up of patients subsequent to failed suicide attempts [42], the authors found that, in a one-year follow-up period subsequent to the failed suicide attempt, the repeat attempt rate was15%. There is a strong correlation between suicide attempt and subsequent death by suicide. In a study of a group of patients who attended the General Hospital after a self-harm episode, the presence of previous self-harm episodes is a risk factor for the onset of future suicidal behaviours [6]. This implies that outpatient care after attempted suicide is of paramount importance. Despite this fact, the identification of specific individuals who will go on to attempt and consummate suicide is currently unfeasible, owing to the scant sensitivity and specificity of the identification procedures available, and the low base rate of this behaviour [17,18].

Similar findings with respect to age, sex and previous suicide attempts have been reported in other publications. Thus, a study on a European level shows how the person-based suicide attempt rates were higher among women than among men and the highest person-based rates were found in the younger age groups [37]. Moreover, more than 50% of the individuals attempting suicide made more than one attempt [37]. Another study shows connection between suicide attempts and the individuals in the youngest age group, who reported suicidal ideation at baseline and psychiatric disorders, especially depression and drug abuse [38]. A systematic sample of 114 patients from consecutive cases of attempted suicide referred to a general hospital in Helsinki shows that a high proportion of individuals attempting suicide (82%) suffered from comorbid mental disorders [39].

In turn, in a population-based study conducted in Spain [2] among young women, mental disorders and psychiatric comorbidity and recent ideation of suicide were identified as high risk factors for an attempted suicide. Episodes of severe depression are the diagnosis representing the greatest risk in relation to the presence of the ideation of suicide and attempted suicide.

Another selection bias is associated with the use of hospital patients as a comparison group for identifying risk factors associated with suicide attempts.

The study is clearly conducted on admitted patients, some of whom are admitted owing to attempted suicide and others who are not (comparison group). Thus, the inference is not intended to be the general population, rather the sub-set of patients who are admitted to hospital. Nonetheless, given the magnitude of the problem in this sub-set of patients, we feel that it is relevant to be able to define the variables associated with the attempted suicide, while clearly accepting that these are patients who have been admitted for medical-surgical problems.

Given that the controls used in the hospital study are patients who also suffer from psychiatric disorders, we are aware of the presence of Berkson's bias [45,46], which occurs when hospital controls are used as a reference group. If the controls are hospitalised due to an exposure that is also related to the disease under study, then the measure of effect may be weakened; i.e. biased towards the null hypothesis of no association. This produces a systematically higher exposure rate among the comparison group, an therefore distorts the odds ratio with an underestimation of the suicide attempt risk. In this sense, we can confirm that the odds ratios associated with the mental disorder in the present study are lower than those that could be obtained in a population-base study. This has been confirmed by comparing the results of the present study with the population-based study by Gabilondo et al. (2007) [2], which shows how the presence of mental disorders is associated with a significant increase in attempted suicides, with higher odds ratio values than found in the present study, pointing towards an underestimation of the risk of attempted suicide depending on the type of controls used, as could have been expected.

Despite the underestimation of this effect, our findings are consistent with those from other studies and meta-analyses which show how psychiatric illness is a strong predictor of suicide [9,10,47]. A psychiatric disorder has also been identified as the strongest risk factor for attempted suicide [48-51]. More than 90 percent of patients who attempt suicide have a major psychiatric disorder [11,12]. Among patients with depression, a history of suicide attempts correlated most strongly with feelings of worthlessness [43]. Concurrent personality disorder was also strongly correlated with suicide attempts in depressed patients [43].

In relation to previous suicide attempts, we consider that one explanation for the extremely high association of previous suicide attempts with a current suicide attempt in this study could be that after the initial attempt, individuals may tend to stay with their family, being referred to the hospital more often after a subsequent suicide attempt. There may be an association between attending hospital after a suicide attempt and repeated suicide attempts. This effect, which is described as surveillance bias or detection bias, occurs when one group is monitored more closely than the other group, and it may lead to an outcome being diagnosed more often in the more closely monitored group, but not because it actually occurred more frequently in that group. This association with previous suicide attempts was also identified in a nationwide study in Finland [9], and in a prospective study [17] in which the authors found that the strongest single factor for suicide is prior history of attempt suicide.

There may also be an information bias due to the fact that the diagnoses of mental disorders were not based on standardised diagnostic interview schedules, but instead were clinical diagnoses made by the psychiatrists treating the patients.

As a number of patients were admitted on multiple occasions, the results may be affected by the lack of independence among admissions. The analysis was conducted with hospitalisation as the unit of analysis, including 5,234 admissions from 4,509 patients. As previously mentioned, this means that 725 (16.1%) were readmissions, some concerning the same patients. To overcome this problem, the statistical analysis was performed using generalised estimating equations, in order to take into account the correlated nature of the data. More specifically, multivariate generalised estimating equation logistic regression models were used to identify those variables associated with suicide attempts [35]. This analysis takes into account the fact that the same patient could be admitted more than once during the study period. Generalised estimating equations are an extension of the generalised linear model, which takes into account this within-group correlation. They assume that the dependent variable of analyses can be expressed as a linear function of the independent variables through a monotonic differentiable link function, and that the variance of the dependent variable is a known function of its expectation. The parameters of the model are then estimated subsequent to the specification of the intra-cluster dependency correlation matrix.

The risk of suicide attempts decreases with age. Psychiatric illness is a strong predictor of suicide attempt. The strongest single factor predictive of suicide attempts is a prior history of attempted suicide.

This study in a hospital setting shows similar findings to population based studies.