Participants and procedure
The PATH Through Life Project is a population-based study examining the health and well- being of people who were initially 20–24, 40–44, and 60–64 years of age (Anstey et al. 2012). Each cohort is being followed up every four years over a total period of 20 years. Participants were randomly sampled from the electoral rolls for the city of Canberra, Australia, and in the neighbouring town of Queanbeyan. Results presented here concern only the 20s cohort, with data from the fourth wave of interviews conducted in 2011–2012, when participants were aged 32–38 (the majority being 33–37). The rationale for including only this cohort cross-sectionally is because the INQ items were only included for that cohort and only at the most recent assessment. At the first wave, interviews were completed with 2,404 in the 20–24 year age-group, of which, 1242 (51.7%) were female and 1162 (48.3%) were male. The participation rate of those who were found to be in the appropriate age range was 58.6%. Follow-up interviews were completed by 1,191 (49.5%) participants (42.5% male, 57.5% female) at wave 4, 12 years after the initial interview, with a further 95 partial completions (53.5% total). The response rate was markedly lower than for previous interview waves (88.6% and 79.7% at Waves 2 and 3), due to reduced project funding that required the interviews be largely conducted online. At Wave 3, there were no significant differences in the rates of suicidal ideation ( = 1.65, p = 0.199), suicidal behaviours ( = 0.02, p = 0.881), presence of anxiety ( = 3.41, p = 0.065) or presence of depression ( = 1.56, p = 0.212) between those who did and did not complete Wave 4. Those who completed Wave 3 but not Wave 4 had significantly less education (14.0 vs 14.4 y; F 1, 1964 = 35.1, p < 0.001) and females had significantly higher rates of assessment completion (68% of F vs 58% of M; = 22.2, p < 0.001). After exclusion on the basis of missing outcome data was made, the sample size for the regression analyses was n=1,177. Items used in the present analyses were based on a self-completed online survey. Approval for the research was obtained from The Australian National University’s Human Research Ethics Committee (protocol #2010/542). All participants provided written informed consent to participate in the study.
Measures
The suicidal ideation outcome was based on endorsement on one yes/no item from the Psychiatric Symptom Frequency scale (Lindelow et al. 1997): “In the last year have you ever thought about taking your own life?”. The IPT constructs of perceived burdensomeness and thwarted belongingness were assessed using seven and five items respectively from the Interpersonal Needs Questionnaire (INQ) (Van Orden et al. 2008). The INQ, derived from the Interpersonal Theory of Suicide, was developed to measure thwarted belongingness and perceived burdensomeness. A validation study by Van Orden et al. (Van Orden et al. 2012) supported the two constructs being distinct but related and reliable. An example item of the INQ is “These days the people in my life would be better off if I were gone”. These items are rated on a seven-point scale from 1 “Not at all true for me”, through 4 “Somewhat true for me”, to 7 “Very true for me”, with scores based on the mean item response ranging from 1–7. Acquired capability for suicide was assessed using five items of the Acquired Capability for Suicide Scale (ACSS) (Van Orden et al. 2008). An example item of the ACSS is: “Things that scare most people don't scare me”. Responses for these items are rated from 0 “Not at all like me” to 4 “Very much like me”, with the acquired capability score assessed as the mean of items, ranging from 0–4. The scale showed good reliability, discriminant and convergent validity (Van Orden et al. 2008). The risk factors examined were gender, age, years of education completed, marital status, number of recent stressful events, number of lifetime traumas, positive and negative interactions from friends and family (Schuster Social Support Scale; (Schuster et al. 1990)), ruminative style (Butler & Nolen-Hoeksema 1994), mastery (Pearlin & Schooler 1978), personality traits, SF-12 physical and mental health scores (Ware et al. 1996), self-reported presence of generalized anxiety and presence of major depression. All questionnaires showed acceptable to good psychometric properties (Schuster et al. 1990; Butler & Nolen-Hoeksema 1994; Pearlin & Schooler 1978; Ware et al. 1996; Rosenman 2002). A count of stressful events in the past six months was identified from a list of 16 events: suffered illness/injury/assault, relative suffered illness/injury/assault, parent/child/partner died, close family friend/relative died, broke off a relationship, serious problem with friend/neighbour/relative, career crisis, thought would soon lose job, partner thought they would soon lose job, partner had career crisis, marriage separation, unemployment, being fired, financial crisis, legal problems, or having something valuable lost or stolen. Lifetime traumas were assessed as a count of adverse experiences from 10 items, including combat experience, life- threatening accident, natural disaster, witnessing injury or death, rape, sexual molestation, physical attack or assault, being threatened with a weapon/held captive/kidnapped, being tortured or a victim of terrorism, or other extremely stressful/upsetting event (Rosenman 2002). Social support was assessed using summed measures of both negative and positive support from family and friends (Schuster et al. 1990). The items were “How often do friends make you feel cared for?”, “How often do friends express interest in how you are doing?”, “How often do family make you feel cared of?” and “How often do family express interest in how you are doing?” Responses were given on a four-point scale ranging from “often” to “never”. Three personality traits of neuroticism, extroversion and psychoticism were measured at the initial interview, twelve years before the outcome data were assessed, using the Eysenck Personality Questionnaire-Revised (Eysenck et al. 1985). These traits tend to be highly stable (four-year reliability correlations of 0.56-0.74). Presence of major depressive episode was assessed using the nine-item Patient Health Questionnaire (PHQ-9(Spitzer et al. 1999)) based on the algorithm identified by the authors of the scales, specifically, presence of anhedonia or feelings of depression (first two items of PHQ-9) and five or more of the nine PHQ-9 items being rated as “more than half the days” or higher (or “several days” or higher for the suicidal ideation item). An example item is “How often have you been bothered by little interest of pleasure in doing things?”. In a review of Wittkampf et al. (Wittkampf et al. 2007), a sensitivity of 0.77 (0.71–0.84) and a specificity of 0.94 (0.90–0.97) was found for the PHQ-9. Presence of Generalized Anxiety Disorder was assessed using the GAD-7 scale (Spitzer et al. 2006), which was also scored using the authors’ diagnostic algorithm (see (Spitzer et al. 1999)) based on ratings of “more than half the days” or “nearly every day” on the first item and at least three subsequent items. An example item is “how often have you been bothered by feeling nervous, anxious or on edge?”. Reliability and validity are excellent (Cronbach's α=0.92, AUC: 0.91). With a cut-off point of ≥10, sensitivity is 0.89 and specificity is 0.82 among primary care participants (Eysenck et al. 1985). Both the PHQ-9 and the GAD-7 are based on past two weeks.
Analysis
Two exploratory factor analyses were conducted, the first with items from the shortened version of the INQ, and the second with the shortened items of the ACSS. Descriptive statistics for the sample were tabulated by gender. Differences in potential risk factors between female and male participants with and without suicidal behaviour were assessed using F values from one- way ANOVAs and chi-square statistics for continuous and categorical variables respectively. Next, logistic regression analyses were used to test the predictions of the IPT model for suicidal ideation in males and females. The variables were centred to reduce multicollinearity. The independent variables for the model were the risk factors hypothesized by the IPT to predict ideation (perceived burdensomeness, thwarted belongingness and the interaction of these two constructs). The rates of suicidal ideation based on levels of perceived burdensomeness and thwarted belongingness were plotted using tertile splits of scores on these constructs. The logistic regression model for plans and attempts were not tested because prevalence was too low (n= 18, 1.4%). The hypothesized interaction between acquired capability and ideation could not be tested in the model for plan/attempts, as all participants who reported a plan or attempt also experienced ideation. As thwarted belongingness and perceived burdensomeness are distinct, but related constructs (Martin et al. 2004), we included the constructs as covariates in each of the models to assess risk factors for each of the constructs independent of the other constructs. Linear regression models were used to assess the association between the potential risk factors and the three IPT constructs: perceived burdensomeness, thwarted belongingness and acquired capability for suicide in separate models for males and females. We tested whether the models for males explained more variance than the models for females using Chow tests (Chow 1960). To test the accuracy of the model for males compared to females, area under the ROC curve analysis (Hanley & McNeil 1982) were conducted. Finally, where possible differences were observed in the effects for males and females (i.e., one significant and the other non-significant, or both significant but of different magnitude), we examined a linear regression model that included the effects of gender, the risk factor and the interaction between gender and the risk factor. We then interpreted discrepant effects as indicating gender differences only when this interaction term was significant.
SPSS version 20 was used for all analyses. Because of the exploratory nature of the analyses, alpha was set at P<.01.