Based on the results of Gold et al. (21), the first aim of the current study was to investigate a suicide attentional bias in a more fine-grained sample by distinguishing between different phases of the transition from suicidal thoughts to behavior. Therefore, we compared individuals with a recent suicide attempt (suicide attempters), individuals who recently experienced SI but had never engaged in suicidal behavior (suicide ideators), and healthy controls. Based on the Cognitive Model of Suicide (12), we hypothesized that suicide ideators and attempters should display a suicide attentional bias by showing significantly greater interferences for suicide-related words compared to healthy controls, but not for positive and negative words. A suicide attentional bias in ideators and attempters should also be prevalent when comparing interferences within each subgroup. More specifically, the suicide-specific interference in ideators and attempters should be significantly greater compared to the positive and negative interferences. On the contrary, the interferences in healthy controls should not differ significantly. Despite lacking a profound theoretical foundation, prior SST research has repeatedly hypothesized a difference between ideators and attempters in suicide attentional control (13, 15, 20). However, the Cognitive Model of Suicide outlines that a suicide attentional bias represents a cognitive marker, preceding suicidal ideation and the engagement in suicidal behavior. Accordingly, ideators and attempters should not differ in their degree of a suicide attentional bias. Due to the existing empirical and theoretical inconsistencies, we refrained from formulating a directional hypothesis regarding the difference between ideators and attempters in suicide attentional control.

The second aim of the study was to add evidence to the psychometric properties of the M-SST found by Gold et al. (21). As in the study by Gold et al. (21), we analyzed the utility of stimuli used in the M-SST by including an evaluation of word material, which assessed how strongly aroused participants felt by each word category and how positively versus negatively they evaluated each word. However, compared to Gold et al. (21), who used an extended version of the M-SST with a suicide-related negative word category (e.g. suicide, destruction) and an additional suicide-related positive word category (e.g. suicide, relief), we used the M-SST with only the suicide-related negative word category. The results of Gold et al. (21) had indicated that the suicide-related negative word category (AUC = 0.72) was more selective compared to the suicide-related positive word category (AUC = 0.62) between healthy controls and patients with STBs.

Participants were recruited into one of three groups: (i) psychiatric inpatients who were hospitalized due to a current suicide attempt (suicide attempters), (ii) psychiatric inpatients who were hospitalized due to recent suicidal ideation without a lifetime history of suicide attempts (suicide ideators) and (iii) control participants without a history of psychopathology and psychotherapy (healthy controls).

Between September 2020 and May 2023, patients at a psychiatric ward of a German hospital were contacted for study participation, if they fulfilled the inclusion criteria of being hospitalized due to a) a current suicide attempt or b) recent suicidal ideation without recent or lifetime suicide attempts and c) being aged ≥ 18. Exclusion criteria included inability to speak or write German fluently, presence of cognitive impairment, color blindness, dyslexia and being currently psychotic. After checking exclusion criteria, n = 40 suicide attempters and n = 40 suicide ideators could be included in the study. After patients agreed to participate, a research assistant provided information about the purpose of the study, the voluntary nature of their participation and data storage. If patients provided informed consent, an appointment for the test session was scheduled and participants received a battery of self-report questionnaires, which they had to complete until the test session. Patients also gave informed consent to access their diagnoses in their medical records. At the scheduled appointment, the test session took place in the research assistant’s office on the psychiatric ward. During the test session, patients completed the M-SST, followed by a questionnaire assessing the evaluation of the presented word stimuli of the M-SST. Finally, a short version of the Suicidal Thoughts and Behaviors Interview (22) was administered.

The control group was recruited via flyers from local communities. Individuals were eligible, if they were a) aged ≥ 18 years and were b) without a history of psychopathology including STBs and psychotherapy. Potential participants of the control group were screened via telephone regarding the exclusion criteria, which included inability to speak or write German fluently, the presence of cognitive impairment, color blindness, dyslexia, a history of psychopathology and psychotherapy including a history of suicidal ideation or suicide attempts. If participants agreed to participate, informed consent and self-report questionnaires were sent by mail and an appointment for the laboratory session was scheduled. After obtaining informed consent, control participants attended the session in the lab of the Department of Medical Psychology and Medical Sociology at the University of Leipzig. The procedure was identical to that for patients except that a short version of a German diagnostic interview for mental disorders [Mini-DIPS, (23)] was administered instead of the SITBI to double check exclusion criteria, which also included screening questions regarding current STBs. Eventually, n = 61 healthy controls could be included in the study. Every participant received 30 € as compensation. All procedures were approved by the ethics committee of the Medical Faculty of the University of Leipzig [012/19-ek].

As described in the measures section, trials of the M-SST with incorrect responses, and single outlier response times (< 200 ms, > 4000 ms) defined as invalid trials were excluded from the analyses (21, 32, 33). We decided against the data cleansing procedure used by Cha et al. (13) and prior SST studies (e.g., 18, 19),which removed trials with reaction times (RTs) ± two standard deviations from that participants mean RT, and trials with mean RTs ± two standard deviations from the group mean RT. The procedure has been repeatedly criticized as a critical limitation of prior SST research by increasing the risk of eliminating meaningful data and decreasing the probability of detecting significant effects (18, 19).

Data on sociodemographic and clinical characteristics were missing for one suicide attempter. Sociodemographic and clinical characteristics are presented in Table 1 . The mean age of the total sample was 31.04 years (SD = 12.24) and participants were predominantly female (60.7%). Healthy controls, ideators, and attempters did not significantly differ in age and gender, but in employment, X² (8, N = 141) = 19.56, p = .012, consequently this variable was entered as a covariate in the analyses. Suicide ideators and attempters did not significantly differ in suicidal ideation, depression, and hopelessness. Ideators and attempters differed significantly in their diagnoses X² (3, N = 79) = 12.59, p = .006). Mood disorders were the most common diagnosis in both clinical samples, while personality disorders were more present in attempters, and neurotic-, stress- and somatoform disorders were more present in ideators. Furthermore, ideators indicated a shorter time period of having experienced suicidal thoughts prior to the assessment (M = 3.97 days SD = 3.87) compared to attempters (M = 6.23 days SD = 5.40, t(69) = 2.03, p = .046). Of the patients with a suicide attempt history, n = 15 (37.5%) reported a single suicide attempt and n = 24 reported two or more suicide attempts (range: 2 – 20). On average, attempters had their index suicide attempt 11.03 days (SD = 5.27 days, range 4 – 25 days) prior to the assessment.

N = 141 M-SST data sets were included, of which 272 trials with an incorrect response (1.21%) and 226 trials (1%) with an outlier response time were removed. Separately analyzed by group, we excluded 227 (2.33%) error trails from a total of 9,760 trials in the control group, 127 (1.98%) error trials from a total of 6,400 trials in the ideators group, and 144 (2.25%) error trials from a total of 6,400 trials in the attempters group. The number of excluded error trials did not significantly differ between groups, F(2, 138) = 0.33, p = .717.

The first aim of the present study was to investigate a suicidal attentional bias in healthy controls, suicide ideators and attempters using a modified version of the SST. Based on the Cognitive Model of Suicide (12), we hypothesized that suicide ideators and attempters would display an attentional bias for suicide-related words in comparison to healthy controls. When analyzing interference scores separately, our results showed a significant group effect for the suicide interference, which was confirmed by an ANOVA accounting for the repeated measures design. Due to the significant interaction effect, groups differed in dependence on interference type. More specifically, ideators and attempters showed greater interferences for suicide-related words compared to controls, which was lacking in positive and negative words. Thus, our hypothesis could be verified in line with the Cognitive Model of Suicide by demonstrating a suicide attentional bias in at-risk individuals with STBs. Our findings of the within-group analyses provide further evidence of a suicide attentional bias in ideators and attempters, showing significantly greater interferences for suicide-related stimuli compared to interferences of positive and negative words that was only found in the two patient subgroups. Contradictory to assumptions of prior SST research, ideators and attempters of the current sample did not differ in their suicide attentional bias. By using the M-SST with improved psychometric properties (21) compared to the prior SST (19, 20), insufficient psychometric properties as a reason for a lack of difference between these subgroups can be excluded. Our results therefore support the assumption of the Cognitive Model of Suicide (12) that a suicide attentional bias serves as an implicit cognitive marker of suicidality regardless of whether an individual has ever proceeded from ideation to action. Along this line, the suicide interference score demonstrated an adequate classification accuracy, differentiating between healthy controls and patients with STBs. However, the score was not able to distinguish between suicide ideators and attempters, thus confirming that this score does not differentiate between different stages of the transition from suicidal thoughts to behavior. This is in line with research on other implicit assessment tools, like the Death Implicit Association Test (D-IAT), showing no significant differences between ideators and attempters regarding their implicit associations with death (34, 35).

In terms of risk assessment, a lack of difference in SAB between ideators and attempters found in our study might indicate that the M-SST’s utility in classifying individuals at risk of attempting suicide is limited. However, in the current study the M-SST was administered to inpatients, who stayed in a clinical setting and received medication and treatment. This might have an impact on differences between ideators and attempters. To date, research is lacking on within-person processes and the temporal patterns of a SAB. Therefore, our knowledge is limited as to whether a SAB fluctuates over time and increases in high-risk situations, for example, shortly before a suicide attempt. Instead of focusing solely on between person differences, future studies should further develop behavioral tests to integrate them into an ecological momentary assessment (EMA). This could help to better understand how a SAB and other suicide-specific marker are associated with an individual’s suicide risk and whether they are modifiable through treatment.

Our second aim was to assess the psychometric properties of the M-SST in a larger sample compared to the study by Gold et al. (21). Regarding the internal consistency of the M-SST, we found good to excellent reliabilities for the total sample and the patients subgroups, which have been considerably improved compared to results of prior SST studies (19, 20), thus confirming the findings of the validation study by Gold et al. (21). There might be several reasons for this improvement: for example, the blocked stimuli presentation might reduce a possible set-switching bias (18), which represents a response bias occurring when individuals have difficulties switching between different valence categories (e.g., switching from negative to positive stimuli) and might have a potential confounding effect on interference. Moreover, having increased the number of trials might have improved internal consistency. As found by Gold et al. (21), the M-SST demonstrated lower reliabilities for the control group as compared to the patients subgroups. The lower reliabilities are due to the smaller co-variances of interferences in the control group, especially in the block with negative words (see standard deviations in Table 3 ), leading to lower correlations and thus a reduced reliability. The low standard deviations in the control group indicate that the processes of attentional control a more homogeneous in controls compared to ideators and attempters.

In contrast to prior SST studies (17, 18), the suicide interference score of the M-SST demonstrated adequate convergent validity, as it was significantly related to self-reported depression, hopelessness, and suicidal ideation, reflecting a correspondence between explicit clinical measures and an implicit suicide attentional bias. Despite this finding, convergent validity of performance-based measures should not be overestimated by considering potential self-presentation biases of explicit clinical measures and the temporal dynamic of suicide risk (7). Notably, positive, negative and suicide interference scores were significantly inter-correlated. Nevertheless, the content of the suicide-specific category was still specific enough to trigger a suicide attentional bias in patients with STBs.

Besides the assessment of implicit attentional processes, the study included the participants’ evaluation of stimuli used in the M-SST, providing important information regarding the word material’s utility. For the arousal evaluation, results mirrored the interaction effect found in interference scores, revealing that both ideators and attempters were explicitly more aroused by suicide-related words compared to controls, while ideators and attempters did not differ in their arousal ratings. This highlights that the suicide-related words of the M-SST were adequately selected to trigger an implicit suicidal schema by inducing an emotional activation in participants with STBs. Notably, suicide ideators and attempters evaluated the suicide-related words significantly more positively and less negatively compared to controls, demonstrating that individuals with STBs feel attracted to suicide-related content.

The strengths of the present study include its focus on investigating high-risk individuals having experienced recent suicidal thoughts and behavior by dividing them into ideators-only and attempters to differentiate clearly between different stages of the suicidal spectrum. More importantly, we included participants’ evaluation of the word material in order to check its utility and, thus providing valuable information for analyzing and optimizing performance-based tools. However, several limitations of this study should be taken into consideration as well. First, in our sample the time between the present suicidal crisis and M-SST administration was longer compared to prior SST studies (13, 15) whose inpatient samples had been administered the SST within 48 hours after admission. In our study, on the other hand, there were approximately nine days between admission and test administration due to several reasons. For example, all patients had to be tested for COVID-19 after admission, leading to a delay of the initial contact. Furthermore, some patients stated at the initial contact that they needed more time to decide about their study participation. Second, we did not include psychiatric controls without STBs. Thus, we were unable to determine if a suicide attentional bias is uniquely linked to STBs or represents a general marker of psychopathology, which should be addressed in future research. Third, we tested the M-SST in a German sample and participants were rather young in age; the results of the present study may thus not be generalizable to diverse populations and participants of older age. Fourth, our study had a cross-sectional design and no prospective suicide attempt data was available, precluding our ability to determine the predictive validity of suicide-related interference of the M-SST in relation to future suicide attempts. Furthermore, we were not able to examine the temporal dynamics of a suicide-attentional bias and to determine whether a SAB is state-related or behaves trait-like. For example, recent hospitalization for suicide risk as well as treatment of STBs might influence a SAB. Therefore, investigating temporal fluctuations of such an implicit marker is of high clinical relevance and future studies should examine a SAB during hospitalization as well as post-discharge by adapting and integrating the M-SST into EMA to examine critical change processes of this implicit suicidogenic marker. Finally, we used a microphone instead of keys for measuring reaction times with the aim of reducing potential cognitive interference due to possible key searching behavior. This method might be less standardized with regard to the measurement of reaction times and incorrect answers. However, the psychometric properties found by Gold et al. (21) were replicated by using a different sample, which suggests a high test standardization. Nevertheless, future studies should compare the psychometric properties of the M-SST using keys for the potential integration into studies with an online format or an EMA design.

The study aimed to assess a suicide attentional bias in a suicidal high-risk sample in comparison to healthy individuals who had never experienced STBs by using a modified SST. Our results revealed that the M-SST is able to detect a suicide attentional bias in individuals with STBs and addresses doubts about the existence of a suicide-specific attentional control deficit in individuals with STBs. Our findings support the assumption that a suicide attentional bias can be viewed as an implicit cognitive marker of suicide vulnerability, independently of having exceeded the threshold to suicidal behavior in the past. This finding has noteworthy clinical implications with regard to suicide-risk assessment and commonly recommended strategies for coping with suicidal thoughts and urges.

Given that the incidence of suicidal ideation in the general population is high (36), and many suicide attempts occur among individuals not currently in clinical settings, the M-SST may be a helpful tool for detecting people at risk outside acute clinical settings. Furthermore, the M-SST demonstrated good psychometric properties and, as a next step, future studies with a prospective design should address questions about the predictive validity of the M-SST with regard to future suicide attempts and how the M-SST could be used in combination with other assessments, including self-report measures and clinical expertise.

Besides clinical risk assessment, our results also have clinical implications for the treatment of STBs. As a suicide attentional bias (SAB) indicates, individuals with an activated suicide schema have difficulties to redirect their attention away from suicidal stimuli that may include distressing suicide-related negative affect, suicidal thoughts, and urges. In these situations, different coping strategies tend to be recommended, for example, individuals should distract themselves from their aversive feelings and thoughts through activities or mindfulness, or should focus on something positive in order to cope with suicidal thoughts and urges. However, the question arises, whether, and in which situations, individuals are able to distract themselves from suicidal thoughts and urges, when they exhibit dysfunctional implicit processes, such as an implicit SAB. Individuals with a (high) SAB may not be able to successfully cope with suicidal thoughts and urges due to a biased suicide-specific attention regulation. Therefore, frequently recommended intervention strategies like “self-distraction” or “positive refocusing” should be reconsidered in light of this implicit suicidal mode, as prior studies have shown that distracting from suicidal thoughts and urges was of limited help (37, 38).

To date, our knowledge regarding the influence of implicit processes on suicidal crises in general (e.g. their interaction with emotion-regulation, decision-making and coping with stress during an acute suicidal state), and implicit suicide-specific within-person processes in particular, is limited. Therefore, further research on implicit between- and within processes is urgently needed to gain a more integrative and holistic understanding of a complex problem such as suicidality. By improving the psychometric properties of the Suicide Stroop Task, an important step has been taken in this direction.