In this case-control study, 13-to-19-year-old adolescents and their parents or primary caregivers were recruited sequentially as they presented to National University Hospital’s Emergency Medicine Department or associated clinical or community services during the study period. Cases were adolescents (N = 61) who had attempted suicide, while controls (N = 63) were adolescents who presented with acute medical conditions without severe morbidity, matched to cases by age, ethnicity and gender at group level. 68 and 71 parents or main caregivers were recruited for cases (27 fathers, 40 mothers) and controls (19 fathers, 48 mothers) respectively. One case was excluded as they did not meet eligibility criteria. Five controls were excluded due to inability to contact (N = 1) or documented past suicide attempt or self-harm (N = 4). The final sample consisted of 252 participants (60 cases, 67 parents; 58 controls, 67 parents).

Adolescents’ average age was 16.2-years-old (sd = 1.85; range = 13–19). 73.7% were female. Majority were of Chinese ethnicity (50.8%), followed by Indian (24.6%), Malay (20.3%) and others (4.3%). Participants identified their religious affiliation as Muslim (26.3%), Hindu/Sikh (21.1%), Christian (18.6%), no religion (18.6%) and Buddhist/Taoist (15.2%). 71.2% of the sample attended secondary school, 28.0% attended post-secondary school, and one adolescent was in primary school. There were no significant case-control differences in any demographic measure. An outline of demographic characteristics broken down according to participant groups (cases and controls) is given in Table 1.

Ethical approval for this project had been obtained from the Domain-Specific Review Board of the National Healthcare Group.

To differentiate suicidal intent of the adolescents in the case and control groups, the Columbia-Suicide Severity Rating Scale (C-SSRS) (21) was used, which measures suicidal ideation, its severity, and suicidal behavior. The Adolescent Stress Questionnaire (ASQ) (22) was used to profile domains of adolescent stress resulting from various life stressors (α = 0.96). The present study utilized the Dimensions of Temperament Survey-Revised (DOTS-R) (23) to profile dimensions of adolescent temperament (α = 0.62 to 0.89). Parenting style was assessed via the Parental Control Scale (PCS) (24) and Parental Acceptance-Rejection Questionnaires (PARQ) (25) that were administered to both adolescents and their parents. The PCS (α = 0.73) assesses the perceptions of parental behavioral control as experienced by the adolescent or administered by the adult themselves, while the PARQ (α = 0.89) examines the parent-adolescent relationship in terms of dimensions of acceptance and rejection. Axis I mental disorders were examined using the Youth’s Inventory-4 Self Report (YI-4) (26), which evaluates emotional and behavioral disorders based on the Diagnostic and Statistical Manual-IV (DSM-IV) diagnostic criteria (α = 0.97). Axis II disorders were chiefly assessed via the Personality Diagnostic Questionnaire–4 (PDQ-4) (27), which examines 12 personality disorder traits (PDTs) from the DSM-IV (α = 0.94). Lastly, a self-constructed questionnaire comprising three questions was administered to evaluate adolescents’ help-seeking behavior. Further details of all measurement instruments are displayed in Table 2.

Data analysis was conducted on the IBM SPSS Statistics version 25. Pearson Chi-square test or Fisher’s Exact test, whichever was more appropriate, was used to compare the categorical variables between case and control group, and p-value and odds ratio (OR) were reported. For some variables in which the cross table contained a cell with observed value zero, a modified Haldane-Anscombe (28) correction was applied, whereby 0.5 was added to each cell for calculation. Student’s t-test was conducted to compare the continuous variables between case and control groups, and p-value, mean difference with 95% CI were reported.

Across 13-to-19-year-old adolescents, cases had higher total stress levels compared to controls. Scores were significantly higher for suicide attempters than controls in all domains of the ASQ apart from emerging adult responsibility, as seen in Table 3 and illustrated by Figure 1. All factors were found to be statistically significant among females (p < 0.01, M_A–B = 2.28 – 7.93, 95% CI [0.56, 11.67]), except for emerging adult responsibility (p = 0.81). However, for male adolescents, home life (p = 0.005, M_A–B = 10.20, 95% CI [3.28, 17.12]), peer pressure (p = 0.005, M_A–B = 6.50, 95% CI [2.09, 10.91]), romantic relationships (p = 0.004, M_A–B = 6.35, 95% CI [2.16, 10.54]), financial pressure (p < 0.001, M_A–B = 5.50, 95% CI [2.66, 8.34]) and emerging adult responsibility (p = 0.004, M_A–B = 2.91, 95% CI [0.85, 4.97]) were significant risk factors.

Stratification of the study population into two age groups of 13-to-16-year-olds (secondary school students) and 17-to-19-year-olds (post-secondary school students) provided insight into the differential impact of stressors, given their differences in school experiences and phase of social development. For 13-to-16-year-olds, significant differences across all factors (p < 0.01, M_A–B = 2.69 – 9.83, 95% CI [1.34, 14.04]) were established between suicide attempters and controls, except for that of emerging adult responsibility (p = 0.32). For 17-to-19-year-olds, only home life (p = 0.018, M_A–B = 6.55, 95% CI [1.15, 11.96]), romantic relationships (p = 0.012, M_A–B = 4.38, 95% CI [1.02, 7.73]) and peer pressure (p = 0.021, M_A–B = 4.43, 95% CI [0.70, 8.17]) were statistically significant at p < 0.05.

Overall, statistically significant case-control differences were greatest for interpersonal relationships, which is consistent with previous research reporting that chronic interpersonal stress results in increased suicidal ideation (29). High relational stress could additionally reflect a lack of socio-emotional support, which may render an individual more vulnerable, since social support has been found to be a protective factor against suicide (30). Peer pressure having the largest effect size (d = 0.98) in the overall sample, as well as in the female (d = 0.91) and 13-to-16-year-old age group (d = 1.24) is unsurprising, since social acceptance is important for adolescents, forming a large part of one’s identity development (31). Interestingly, emerging adult responsibility was the only stressor that did not significantly differ between cases and controls, even in the 17-to-19-year-old age group. A potential explanation might be that adolescents face the same stress vis a vis transitioning to adulthood, and the difference between suicide attempters and controls lies in their ability to cope. Findings for adolescent temperament, as discussed below, provides additional insight.

In the older 17-to-19-year-old group, romantic relationships (d = 0.74) as a stressor had the greatest effect size, indicating that romantic relationships may crucially influence the risk of suicide attempt in older adolescents. School performance and attendance as risk factors were significant only for younger adolescents and females. As such, when screening students at risk for suicide attempt, using school performance and attendance as tools for early identification might not be accurate for older adolescents and males. Furthermore, in male adolescents, financial pressure had the largest effect size (d = 1.48). Financial literacy training may be considered to provide adolescents with resources and instill hope through knowledge of the availability of options. Finally, in younger compared to older adolescents, significant case-control differences were found in more avenues of stress, indicating that risk factors for suicide attempt are more salient and thus might be easier to identify when adolescents are younger. This elevates the importance of early identification in suicide prevention efforts.

As seen in Table 4 and illustrated by Figure 2, adolescents who attempted suicide tended to (a) avoid new persons/situations (approach/withdrawal); (b) be less adaptable to changes in routine/environment (flexibility/rigidity); (c) have a negative outlook (mood); and (d) have an irregular sleep-wake cycle (rhythmicity-sleep) significantly more than controls. Male adolescent cases had a significantly greater negative outlook (p = 0.003, M_A–B = –5.38, 95% CI [–8.82, 1.95]), irregular sleep-wake cycle (p = 0.009, M_A–B = –3.78, 95% CI [–6.53, 1.03]) and higher general activity levels (p = 0.04, M_A–B = 3.14, 95% CI [0.22, 6.06] than controls, while female cases tended to have a negative outlook (p < 0.001, M_A–B = –4.24, 95% CI [–6.46, 2.03]), avoid new persons/situations (p = 0.005, M_A–B = –2.27, 95% CI [–3.84, 0.71]) and be less adaptable to changes (p = 0.001, M_A–B = –2.07, 95% CI [–3.31, 0.83]).

Results from the present study corroborate with existing literature and extend their applicability to a multi-ethnic Asian population. For instance, previous studies found that adolescents who were more withdrawn in new social situations and less adaptable to changes were prone to developing depressive symptoms (32) and the tendency to express negative thoughts had been found associated with adolescent suicide attempt (33). Irregular sleep-wake patterns in adolescent suicide attempters have additionally been documented (34). Previously, Hurtig et al. (35) had reported that hyperactive and inattentive adolescents were more likely to have suicidal ideation compared to those who were not. In this sample, however, higher general activity levels were found for male but not female suicide attempters, compared to controls, although this was only significant at p < 0.05.

As a result of these tendencies toward withdrawal, inflexibility, low mood and irregular sleep patterns, adolescent suicide attempters may have less adaptive coping strategies when faced with similar life stressors as other adolescents. When compounded by greater intensity of life stressors, as previously discussed, their risk for suicidal behavior is amplified. Although, it is possible that these temperamental characteristics do not independently influence adolescents’ ability to cope but may reflect underlying mental disorders. Future research may investigate the combined or independent effects of these risk factors through regression analyses. Regardless, with these trait tendencies in mind, suicide prevention programs should be devised to support the development of effective coping strategies. The internalizing nature of many of these temperament traits may additionally cause these adolescents to fall under the radar. Implementing a regular check-in or buddy system may prove beneficial.

From the adolescents’ perspective, findings indicate that mothers of adolescents who attempted suicide were significantly more aggressive, neglecting, rejecting, and cold, while fathers were perceived to be significantly more aggressive, rejecting and cold (Table 5 and Figure 3). Neither perceived maternal nor paternal control differed significantly between cases and controls. From the parents’ perspective, there were no significant case-control differences except for neglect (p = 0.044, M_A–B = 2.08, 95% CI [0.058, 4.09]) and maternal rejection (p = 0.042, M_A–B = 2.16, 95% CI [0.076, 4.25]).

These findings are in line with results from the ASQ, which indicated large case-control differences in the dimension of home life. Previous studies have similarly found low levels of both maternal and paternal affection as perceived by adolescents who had attempted suicide (6). Additionally, the present study’s results for perceived parental hostility are consistent with Wagner and Cohen’s (36) finding that harsh maternal discipline was associated with adolescent suicide attempt. The present findings expand existing literature by adding that perceived paternal aggression is likewise higher among adolescent suicide attempters. Parental neglect and rejection had previously been found to increase the likelihood of suicide attempt (37). However, the present results indicate that only mothers’ but not fathers’ neglect was associated with attempted suicide.

With respect to parents’ perception of their own parenting style, surprisingly, no differences were found apart from that of neglect and maternal rejection. This could potentially be explained by (a) a lack of awareness regarding how their own behavior was perceived by their child, (b) modification of parenting practices following their child’s suicide attempt, or (c) social desirability bias resulting in inaccurate questionnaire responses. As such, when evaluating parenting styles, more weight should be given to adolescent- rather than parent-reports. Interventions must be developed to help parents modify their parenting styles.

For Axis I disorders, cases were more likely than controls to show symptoms of schizophrenia, eating disorders, disruptive behavior disorders, experience of disturbing events, mood disorders, and anxiety disorders, as shown in Table 6 (Figure 4). Cases were also more likely to show symptoms for more than two disorders; 90.0% of cases versus 41.4% of controls reported symptoms of at least two Axis I disorders (p < 0.001, OR = 12.75, 95%CI [4.73, 34.39]). The difference between cases and controls increased for possessing at least 3 Axis I disorder traits (85.0% cases, 27.6% controls; p < 0.001, OR = 14.88, 95% CI [5.97, 37.07]) and at least 4 Axis I disorder traits (80.0% cases, 20.7% controls; p < 0.001, OR = 15.33, 95% CI [6.26, 37.58]). Greater co-morbidity or co-occurrence of symptoms could thus be an independent risk factor for adolescent suicide attempt.

Significant differences were also found between cases and controls for 10 out of 12 PDTs (Figure 5 and Table 6), particularly antisocial, borderline, depressive, and avoidant PDTs. Cases were also more likely to have a combination of PDTs compared to controls; 95.0% of cases versus 56.9% of controls possessed at least 2 PDTs (p < 0.001, OR = 14.39, 95% CI [4.03, 51.36]). A significant case-control difference was noted on the “Suspect” scale (Table 6), which suggests more cases were either lying, responding randomly, or not taking the questionnaire seriously compared to controls. No difference was found on the Too Good Scale, indicating that both cases and controls under-reported to the same extent.

Significant case-control differences for Axis I Disorders were consistent with findings from previous studies (38). Biological predispositions, combined with life stressors, place one at higher risk of developing psychiatric disorders (39). The higher intensity of stressful life events experienced by adolescent suicide attempters, as evidenced by their ASQ scores, could lead to increased vulnerability of developing mental disorders, which in turn leads to greater risk of suicidal outcomes (40). Furthermore, the higher proportion of comorbid Axis I disorder traits among cases compared to controls suggests that comorbidity, not just the presence of Axis I disorders, amplifies the risk of suicidality among adolescents and should be considered when identifying high-risk adolescents and developing suicide prevention strategies.

Results for PDTs among adolescent suicide attempters were similar to what had been found in a study by Brent et al. (41), where suicide attempters showed a greater number of symptoms associated with avoidant and borderline personality disorders compared to controls. Most existing studies primarily examined effects of borderline personality disorder on suicide attempt. The present findings highlight the importance of additionally considering other Axis II disorders when investigating risk factors. Importantly, histrionic PDTs differed insignificantly between cases and controls, highlighting the fact that adolescents who attempt suicide may not be simply attention-seeking but should be taken seriously.

There were no statistically significant case-control differences with respect to whether any form of help was sought at all. Regarding specific help-seeking avenues, suicide attempters sought help from counselors/psychologists and telephone hotlines significantly more than controls, as seen in Table 7 (Figure 6). However, the proportion of attempters who sought help from telephone lines was small compared to other help-seeking avenues. For male adolescents, there were no statistically significant case-control differences in any help-seeking avenue, whereas for females, seeking help from counselors/psychologists (p = 0.001, OR = 10.14, 95% CI [2.14, 48.02]) and general practitioners (p = 0.026, OR = 12.71, 95% CI [0.68, 237.4]) was significantly higher among suicide attempters. There was no significant difference between males and females in terms of overall help-seeking behavior (p = 0.21).

Help-seeking behavior was expected to be more prevalent in cases than controls, especially from friends and family compared to healthcare professionals (42, 43). However, this was not observed in the present sample. While help-seeking from professionals may have resulted from higher school referrals for psychiatric help among adolescent suicide attempters, students have the choice to follow through or refuse these appointments. Future studies should examine the ways and rates of referral, and the factors that influence following through with these appointments.

Low levels of help sought from family members may indicate low perceived familial support, which aligns with adolescents’ perceptions of their parents’ parenting styles, as discussed above. Moreover, while help-seeking behavior may reflect social norms and stigma around mental health, it may also reflect the accessibility and structure of primary care and mental health services (42). Introduction of more community mental health services, in addition to campaigns to increase mental health literacy, could increase help-seeking behavior.

A main limitation of this study is that retrospective self-report, which was this study’s primary avenue of data collection, may be subjected to inaccuracies of under- or over-reporting due to social desirability or recall bias, for caregivers and/or adolescents. To supplement the measure of adolescent temperament, collection of biomarkers may provide further insight and objective validation to the questionnaire data collected. This could include utilizing a biomarker-based temperament measure (44). Another key limitation of this study is its smaller sample size of 118 adolescents, with a higher representation of female participants. Gender bias should be accounted for. In addition, researchers should consider conducting a similar population profile study of adolescents who completed suicide through post-mortem retrospective assessments, to compare against those who survived a suicide attempt. This may inform a more holistic understanding of suicidal behavior. Future research may also consider a longitudinal approach to examining the relationship between these risk factors and adolescent suicidality. Finally, future studies should examine the ways in which the risk factors examined here interact with one another.