Reasons for living and dying in suicide attempters: a two-year prospective study

The sample consists of participants who were admitted at the emergency unit of the Bern University General Hospital following an attempted suicide. The present study, which was part of a larger randomized controlled trial [16], included 60 patients who gave written informed consent and received routine psychiatric treatment (inpatient, day patient, and individual outpatient care). Exclusion criteria were insufficient mastery of the German language, serious cognitive impairment, psychotic disorder, and residency outside the hospital catchment area. Table 2 presents baseline characteristics and clinical diagnoses of the patients. Participants were Caucasians, had an average age of 39 years, and half of them were female. Twenty-five percent were married and 32% had children. Sixty percent were diagnosed with an affective disorder, 47% with a neurotic and stress-related disorder, and 33% a substance abuse disorder. The mean number of diagnoses was 2 (M = 2.1, SD = .93, range 1–4). Twenty-two percent reported one prior suicide attempt; 35% had a history of multiple (two or more) prior suicide attempts.

All patients took part in a single clinical interview at baseline, which included a structured suicide risk assessment using the “Suicide Status Form” [12]. The median delay time between the suicide attempt and the interview was 17.5 (interquartile range 9.0–34.5) days. They filled out the DEMO) [17], the BDI [18], the BSS [20], and the SSF-III [13] at baseline (T1) and after 6 (T2), 12 (T3), and 24 (T4) months. Six months (T2) after baseline 71.7% of the participants completed the questionnaires; 12 months (T3) after baseline the response rate was 60% of the baseline sample and at the two-year follow-up (T4), 63.3% of the baseline sample took part. Complete data over all included measurement points was obtained by 46.7% of the baseline sample. Data on repeated suicide attempts were primarily assessed with the DEMO [17]. Additionally, we complemented self-reported data by searching medical records and contacting involved health professionals in order to complete data regarding suicidal behavior. Within 12 months after baseline, one patient died by suicide and three died of natural causes.

Overall attrition was significantly associated with a higher number of RFL at baseline (r = .26, p = .040), stronger psychological pain at the time of the suicide attempt (r = .29, p = .025), and not being in outpatient treatment between T2 and T3 (r = −.33, p = .049). Attrition at T2 was associated with fewer suicide attempts (r = −.34, p = .007), stronger psychological pain at the time of the suicide attempt (r = .27, p = .033), and fewer outpatient and inpatient treatments before baseline (r = − .35, p = .008, r = −.34, p = .007, resp.). Attrition at T3 was associated with stronger psychological pain at the time of the suicide attempt (r = .26, p = .041) and more psychotherapy sessions during the previous 6 months before the index suicide attempt (r = .27, p = .040). Attrition at T4 was associated with a higher number of RFL at T1 (r = .26, p = .048), a lower number of psychotherapy sessions in the 6 months prior to T3 (r = .40, p = .016). Under the assumption of missing at random, we used multiple imputations for missing data (see data analysis section).

Under the assumption of missing at random, we computed multiple imputations for all variables included in the analyses based on Markov Chain Monte Carlo simulations with the Bayesian estimator in Mplus version 7.4 [22]. A total of 40 imputed datasets were generated and results are reported on pooled analyses of these datasets. Sensitivity analyses yielded similar results using the imputed and the non-imputed datasets. Pearson correlation were used for two continuous variables, tetrachoric correlations for two binary, biserial correlations for one continuous and one binary or ordered polytomous variable, and polychoric correlations for a binary and an ordered polytomous variable. To analyze predictors of the BSS score and repeated suicide attempts, we used robust linear and logistic regression analyses. In the multiple regression analyses, we entered all correlations with a p-value <.09 simultaneously.

We conducted cross-sectional path analyses at baseline to investigate whether the number of RFD mediated the relationship between the BDI score and BSS score at baseline. Furthermore, we conducted longitudinal analyses with BDI at baseline, number of RFD at T2, and BSS at T3. This temporal sequence of the variables suggests a directional link from the predictor to the outcome variables via the mediator. It accounts for the implication of the temporal relation with depression occurring before the RFD and the RFD occurring before suicide ideation and thus the notion that the depression at baseline affects the number of RFD 6 months later and the RFD affect the level of suicide ideation at the 12-month follow-up [23].

After the index suicide attempt at baseline participants reported significantly more RFL (M = 3.45, SE = 1.35) than RFD (M = 1.90, SE = 0.55, t(59) = 5.71, p < .001), which corresponds to a strong effect (d = 1.00). Therefore, our hypothesis that patients immediately after a suicide attempt had more RFD than RFL was rejected. Even though significance level was just not reached, participants with at least one prior suicide attempt produced more RFD (M = 2.24, SE = 0.28) than participants with no prior suicide attempt (M = 1.46, SE = 0.25, t(58) = − 1.96, p = .054, d = 0.53). There was no difference regarding RFL between those with and without prior suicide attempt (M = 3.38, SE = 0.22 vs. M = 3.54, SE = 0.27, t(58) = .44, p = .660, d = 0.12).

Table 3 presents the zero-order correlations of socio-demographic variables, RFL and RFD at baseline and 6 month (T2) later, BDI at baseline, BSS scores at T1 - T4, and repeated suicide attempts. RFL and RFD responses at baseline did not correlate (r = −.05, 95% BCa CI [−.31, .24], p = .697). The number of RFL at baseline and at T2 were not significantly related to socio-demographic and clinical sample characteristics or any other variables (all p > .05). However, RFD at baseline correlated with BDI at baseline (r = .56, 95% CI [.37, .75], p < .001) and BSS at baseline (r = .70, 95% CI [.54, .86], p < .001), BSS at T2 (r = .53, 95% CI [.31, .76], p < .001), and BSS at T4 (r = .29, 95% CI [.04, .53], p = .021).

Table 4 presents the results of the simple and multiple regression analyses for predicting the BSS score at baseline, the six-month (T2), and two-year follow-up (T4). Simple regression analyses showed that being in a relationship was significantly associated with a higher BSS score at baseline and BSS at T2. Having children at baseline was significantly related to a lower BSS score at baseline and at T2. RFD at baseline showed the strongest association with BSS score at baseline with an explained variance of 50%. More suicide attempts prior to the baseline, a higher BDI at baseline, and more RFD at baseline consistently predicted a higher BSS score at baseline, at T2, and at T4 with declining effect sizes across time. RFD at baseline still explained 9% of the variance of the BSS score at T4. In contrast, RFL were not associated with suicide ideation.

As a next step, we computed multiple regression analyses to investigate the unique contribution of all significant predictors in the simple regression analyses controlling for the other predictors. Concurrent associations at baseline confirmed the unique contributions of being in a relationship, having children, a higher number of RFD, and a higher level of depression for predicting suicide ideation. However, previous suicide attempts were no longer significant. Again, RFD at baseline were the strongest predictor; all variables together explained 64% of the variance of suicide ideation at baseline (BSS T1).

We examined longitudinal associations of the predictor variables at baseline on BSS score at T2 controlling for BSS at baseline. Multiple regression analyses showed that only BSS score at baseline remained a significant predictor and had a unique contribution to the BSS at T2. The explained variance was again 65%. Regarding the long-term associations after 2 years (T4), no baseline predictor had a unique contribution to the BSS at T4. The amount of explained variance of the BSS score at T4 declined to 26%.

Table 5 presents the results of the path models testing whether RFD mediated the association between depression (BDI) and suicide ideation (BSS). We found a significant indirect effect (STD b = .29, p < .0001) of the level of depression on the level of suicide ideation via the number of RFD. The direct relationship between depression and suicide ideation remained significant, but the standardized regression coefficient declined from .63 to .34. Thus, a partial mediation was confirmed. The model explained 57% of the variance of suicide ideation at baseline.

As a next step, we repeated these analyses using longitudinal data to test whether RFD at T2 mediated the relationship between the BDI at baseline and BSS at T3. Results confirmed a significant indirect effect (STD b = .19, p = .024) even though smaller than the cross-sectional indirect effect. The direct effect of depression at baseline on suicide ideation 12 month later disappeared, confirming a full mediation. The model explained 22% of the variance of suicide ideation at 12-month follow-up.

Table 4 reports the results of the logistic regression analyses to predict suicide reattempts. Within the 12-month period after baseline, 23.3% (n = 10) of the participants reported another suicide attempt. In line with the previous analyses, BSS and BDI at baseline predicted suicide reattempts after 12 months (T1-T3) in the simple regression analyses. The number of RFD at baseline just missed the level of significance (p = .051). In contrast to the prediction of suicide ideation, having a partner, having children or previous suicide attempts before baseline did not predict suicide reattempts. Multiple regression analyses explained 32% of the variance in suicide reattempts within 1 year but no single predictor remained significant. Within the two-year period from baseline, 36.7% (n = 16) of the participants reported another suicide attempt. BDI and BSS at baseline, but not RFD at baseline predicted significantly suicide reattempts after 2 years (T1 – T4) in the simple regression analyses. The multiple regression model explained 16% of the variance and again no single predictor remained statistically significant.

We investigated the number of RFL and RFD responses and their association with baseline and future suicide ideation and suicide reattempts. Against our hypothesis and a previous study where highly suicidal individuals reported more RFD than RFL [14], we found that study participants reported significantly more RFL than RFD after the index suicide attempt. The numbers of RFD and RFL in our sample was comparable to non-suicidal individuals in Harris et al.’s study [14]. The concept of the suicidal mode may explain this finding. The suicidal mode is an out-of-the-ordinary state of mind which has a time-limited nature of activation [24]. When the mode is activated, the person is cognitively and affectively restricted to suicidal thoughts and feelings of hopelessness and helplessness. In most cases the suicide attempt leads to a reduction of inner tension and the deactivation of cognitive restriction—a cathartic effect. Afterwards, individuals often feel relief and again have access to life-oriented goals [25]. It might be that at time of data collection important RFL were again accessible for study participants.

We further found that the number of RFL and RFD responses at baseline were unrelated. This suggests that RFL and RFD represent two different psychological dimensions rather than a continuum. Interestingly, the number of RFL did not correlate with any socio-demographic and clinical sample characteristics. Against our hypothesis, the number of RFL was not associated with the degree of depression at baseline nor suicide ideation and suicide reattempts during the two-year follow-up. Thus, the number of RFL was not a marker for suicide risk. This is not in line with the finding of Cwik et al. [10], where RFL moderated the association between depression and suicide ideation, and contradicts previous research, where fewer RFL were associated with an increased risk of suicidal thoughts and reattempts [8, 9]. A possible explanation for this finding might be the different samples. Some of the previous studies have investigated only suicide ideators, not attempters [9‐11]. These might be two different populations. RFL may have a protective effect in ideators, but this effect might disappear in attempters who experienced a strong impulse to act on suicide ideation.

In accordance with our hypothesis, a higher number of RFD was associated with higher levels of depression at baseline as well as a higher degree of suicide ideation at baseline and during the two-year follow-up. Similarly, participants with a history of suicide attempts showed more depression and suicide ideation compared to individuals with no prior attempt and also produced significantly more RFD then the latter. This is in line with results by Gutierrez et al. [11], where internal risk factors were more critical for identifying suicidal individuals than protective factors.

Moreover, multiple regression analyses showed that being in a relationship and having children were predictive for suicidal thinking at baseline. While having children was associated with decreased suicide ideation, having a partner was positively associated with suicide ideation and might be a risk factor for having suicidal thoughts. Possibly, study participants with a partner tended to perceive themselves as a burden to their partner, which could be expressed by a higher level of suicidal thoughts. This would be in line with findings of the interpersonal theory of suicide [26], where a person’s perception of being a burden to others was a predictor for suicide ideation. Furthermore, previous suicide attempts were significantly associated with suicide ideation at baseline and follow-up in the simple regressions, but did not become a significant predictor in the multiple regressions. Therefore, these characteristics may indicate a higher risk for suicide ideation when assessed in an interview or questionnaire, which is clinically relevant in terms of identifying individuals at risk for suicidal behaviors. However, they do not seem to uniquely contribute to the development of suicide ideation when including the other predictors and thus are probably not targets for interventions.

In our study, the number of RFD was the strongest predictor for suicide ideation at baseline. This implies that the number of RFD is a strong risk factor for recent suicidal thoughts. Also, depression was a strong predictor for suicide ideation at baseline which is in line with previous results [5]. However, mediation analyses showed that the influence of depression on suicide ideation at baseline was partially explained by the number of RFD. The mediation effect was replicated longitudinally. These results again emphasize the substantial role of RFD on suicidal thinking in the short and long run.

At follow-up, only suicide ideation at baseline was a significant predictor for suicide ideation after 6 months and no clinical baseline predictor became significant in order to predict suicidal thoughts after 2 years. These findings indicate that the number of RFD and other clinical predictors at baseline are primarily related to the current suicidal crisis and that their predictive power in regard to future suicidal states can be seen as only limited.

Regarding suicide reattempts, simple regressions showed that the number of RFD almost reached the significance level and depression, as well as suicide ideation at baseline were significant predictors for reattempting suicide within the one-year follow-up period. In contrast, the number of RFL had no influence on suicidal behavior during 12 months after the index suicide attempt. This again underpins the substantial role of RFD as a motivational risk factor of suicidal behavior. In the multiple regressions, RFD, BDI, and BSS explained 32% of the variance in suicide reattempts, but no single predictor remained significant. This indicates that in our study no specific clinical risk factor was predictive for reattempting suicide within 1 year. It was rather the interplay of all three factors increasing the risk of suicidal behavior and, therefore, depression, suicide ideation, and the number of RFD should be assessed in order to evaluate the risk of suicidal behavior after a recent suicide attempt. Within 2 years after the index suicide attempt, clinical baseline predictors were only marginally predictive in regard to a new suicidal crisis indicating that their predictive power is time limited. The results suggest that a reattempt at least 12 month after the index attempt is more influenced by predictors, which are temporally closer to that event. Moreover, research has shown that previous suicide attempts are a reliable and robust risk factor for future suicidal behavior [26, 27]. Surprisingly and against these findings, previous suicide attempts were not predictive for a repeated suicide attempt in our study. In our sample, 43% of the participants had not experienced a past suicide attempt, thus, it can be argued that there was not enough variability to assess the prediction.

There are a several limitations that should be taken into account. Data on repeated suicide attempts were primarily assessed with self-report questionnaires. This might be related to issues with under- and over-reporting of suicidal behavior, and a divergence between self-reports and hospital-based data [28]. We addressed this issue by complementing self-reported data with searching medical records and contacting involved health professionals. Moreover, median delay time between the index suicide attempt and assessment at baseline was 17.5 days. This was for example due to surgery or medical interventions after the suicide attempt or longer reflection of patients regarding study participations. This time interval might have led to a recall bias, which might have resulted in an overrepresentation of RFL and an underrepresentation of RFD. Furthermore, participants were instructed to write down a maximum of five RFD and RFL, which might have result to a ceiling effect. The recent study was not able to answer the question whether specific reasons were more or less related to an increased suicide risk. Further analyses will explore this notion and will be reported elsewhere. Finally, the relatively small sample size decreases the statistical power and increases the likelihood of Type II errors. For that reason we decided not to carry out a Bonferroni correction resulting in a further loss of power. Nevertheless, multiple hypothesis testing without a statistical correction can increase the risk of Type I error. Therefore, the results of this study can only be considered as preliminary and replications in larger samples are needed.