Staff Characteristics and Baseline Job Role Differences
Rates of attrition were equivalent in the two groups of randomly assigned staff. Follow-up participation rates were 73.5% and 72.2%, respectively, for staff in training and control schools. Attrition was not associated with any baseline measure nor differentially predicted by staff condition as a function of baseline (training condition by baseline measure interaction).
Staff in trained and untrained schools were comparable on baseline measures (, column 3). These null differences paralleled the minimal differences between trained and non-trained groups on demographic and job role characteristics described earlier. The scaled variables all had approximate normal distributions as judged by relatively low levels of skewness and kurtosis. As we expected, the single item of ‘ask about suicide’ had a small mean and high skewness; this variable was further analyzed by dichotomizing and using logistic regression. There was adequate distribution of scores on baseline measures, with the exception of Gatekeeper Reluctance scores; few staff indicated reluctance to engage with suicidal students.
Baseline gatekeeper training variables and communication with students by training status (Intention to Treat) and job role
Staff in different job roles had substantially different baseline levels of knowledge, appraisals, and behaviors (, columns 4 and 5). Health/Social Service staff had the highest knowledge, the most positive gatekeeper appraisals, and highest rates of gatekeeper behaviors and communication. Support staff reported lowest levels. Staff reported large differences in asking students about suicide by job role (ANOVA F=146.69, p <0.001): 72% of Health/Social Service staff asked one or more students about suicide in the prior 6 months, compared to 22% of Administrators, 7% of Teachers and 1% of Support Staff. In our analyses below examining impact of training, we take into account the important differences by job role.
We found very consistent findings of a positive training impact on staff knowledge and appraisals, whereas training impact on behaviors was concentrated within staff in certain job classes and in those communicating with distressed youth about suicide at baseline. ITT analyses are summarized in and AT analyses in . Means and standard deviations are shown in columns 1 and 2, and training impact ES’s are shown in column 3.
Intention to Treat Analyses: Gatekeeper Training Variables and Communication at 1-year Follow-up by Training Status
As Treated Analyses: Gatekeeper Training Variables and Communication with Students at 1-year Follow-up by Training Status
Both ITT and AT analyses yielded highly comparable findings regarding main effects of training ( and , column 3). Regression analyses indicated that time 2 scores were strongly predicted by baseline scores in both groups (not shown). An average of one year post-training, we found a moderate, positive training condition effect on knowledge (ITT ES 0.41); very large positive effects on perceived preparedness (ES 1.21), self-evaluated knowledge (ES 1.32), efficacy (ES 1.22), and access to services (ES 1.07). On reluctance, the overall effect was non-significant for ITT but positive for AT analyses (AT ES 0.37). No overall training effect for suicide identification behaviors was found, although a significant condition by baseline interaction for Ask Students about Suicide showed differential impact of training as a function of staff baseline level (described in detail below). As expected, training did not impact staff communication with students, either Natural Gatekeeper or Ask Students about Distress. We found no evidence that receiving refresher training had an effect on time 2 outcomes. Furthermore, duration since training was unrelated to any time 2 outcome measure as well. None of the results were different when we adjusted for the single baseline variable that differed between the groups at baseline (years employed), by repeating analyses with staff years employed added as a covariate.
The analyses showed significant school-level effects on knowledge, appraisals and behaviors, from both ITT and AT analyses that included random effects for staff members’ current school placement (Column 5 in and ). Those intraclass correlations for school-level effects were larger for As-Treated analyses, i.e., 0.06 for appraisals of preparation, 0.097 for efficacy to perform gatekeeper role, and 0.088 for asking students about suicide.
We also found that training affected staff differently depending upon their baseline levels of attitudes and behaviors. In column 4 of and we summarize how the training effects differed by participants’ baseline scores, with the relative change being a ratio of change in these slopes for the trained versus untrained groups relative to that in the untrained group. Significance levels were based on testing of homogeneity of slopes. A negative ratio (in combination with positive main effects discussed earlier) indicates a greater gain from training for participants who had low scores at baseline compared to the gain for those with higher scores at baseline. Participants with the lowest appraisals at baseline had the largest gains from training on Gatekeeper Preparedness, Self-Evaluation of Knowledge, and Access to Services. shows scores on Self-evaluation of Knowledge at time 2 as a function of baseline scores for trained and non-trained staff to illustrate the largest training impact for staff with lowest baseline scores.
Training Impact on Self-Evaluation of Suicide Prevention Knowledge at Time 2 as a Function of Baseline Self-Evaluation.
In contrast, training had a greater impact on staff suicide identification behaviors with students for those with higher baseline scores, indicated by positive ratio of slopes comparing training impact for trained versus untrained staff on Ask Students About Suicide for ITT (Relative Change=0.27; 95% CI: 0.03, 0.51) and for AT (Relative Change=0.31; 95% CI: 0.07, 0.54). Because of the low frequency of staff asking students about suicide, we also examined differences by dichotomizing this variable. For staff not asking students about suicide at baseline (i.e., 86% of non-trained and 87% of trained staff), there was no training effect: 14% in the trained group asked students about suicide at time 2 compared to 16% in the un-trained group. Among those staff asking about suicide at baseline (i.e., 14% in the non-trained group and 13% in the trained group), 100% in the trained group asked one or more students about suicide at time 2 (mean number of students M=3.14 SD=1.2) compared to only 58% of controls (M=1.79 SD=1.8). Training therefore had a positive effect on asking about suicide for the 13% of staff already asking students about suicide before training and had little benefit for the 87% of remaining staff.
Training Impact by Job Classification
We next report AT findings separately by job role, focusing first on the two largest groups, teachers and support staff (AT analysis). Whereas QPR training had positive impact for teachers and support staff on knowledge and appraisals, only teachers showed positive training impact on suicide identification behaviors. Training had large positive effects on increasing perceived preparedness for both groups (ES for teachers=1.53; 95% CI: 1.17, 1.90; ES for support staff=1.04; 95% CI: 0.77, 1.31); on self-evaluated knowledge (ES for teachers=1.76; 95% CI: 1.39, 2.14; ES for support staff=1.77; 95% CI: 1.42, 2.13); on gatekeeper efficacy (ES for teachers=1.89; 95% CI: 1.50, 2.28; ES for support staff=1.07; 95% CI: 0.79, 1.36); and on access to services (ES for teachers=1.52; 95% CI: 1.18, 1.86; ES for support staff=1.04; 95% CI: 0.75, 1.32). Teachers with lower baseline appraisals gained most from training (Relative Change −0.24 to −0.62), whereas no significant differences in training impact as a function of baseline appraisals was found for support staff. Training had a positive effect for teachers on knowledge (ES=0.58; 95% CI: 0.20, 0.94) but no significant effect for support staff.
For teachers, although the overall training effect on asking students about suicide was non-significant (ES=0.24; 95% CI: −0.13, 0.61), examination of a significant condition by baseline level interaction (Relative Change=2.00; 95% CI: 0.15, 3.84) showed that training increased self-reported asking behavior primarily for teachers already asking students about suicide at baseline. Training also increased teachers’ appropriate referral behaviors with suicidal students (ES = 0.42; 95% CI: 0.08, 0.75). In contrast, no support staff in the trained group reported asking students about suicide at time 2 (ES = 0.00), and training for support staff also had a directionally negative effect on referral behaviors that was non-significant (ES = −0.65; 95% CI: −1.34, 0.03; p = 0.08).
For staff in health/social service roles, training had very large positive effects on knowledge (ES = 1.11; 95% CI: 0.17, 2.04), perceived preparedness (ES = 1.78; 95% CI: 0.90, 2.65), self-evaluated knowledge (ES = 2.31; 95% CI: 1.12, 3.51), efficacy (ES = 1.98; 95% CI: 0.82, 3.09), and on lower gatekeeper reluctance (ES = 1.14; 95% CI: 0.23, 2.06). For health/social service staff, training also had a non-significant, directionally positive relationship with Ask Students about Suicide (ES = 0.66; 95% CI: −0.23, 1.56; p = 0.17).
For administrators, QPR training had directionally positive, but comparatively smaller, effects. We found large positive training effects for administrators after one year on knowledge (ES = 1.03; 95% CI: 0.25, 1.81) and efficacy (ES = 1.02; 95% CI: 0.24, 1.81).
Prediction of Changes in Asking About Suicide
To assess whether the main behavioral variable changed during training, we compared the distributions of Ask Students about Suicide from Time 1 to Time 2 by training condition. For the trained group, there was a significant increase in the numbers of times each staff member reported asking about suicide, with one more student asked for every four staff members (by paired t-test mean difference of 0.26, CI: 0.14, 0.38; p < 0.0001). In contrast, the gain in the non-trained group was a more modest one additional student reported asked for every ten staff members (by paired t-test mean difference of 0.10, CI: −0.01, 0.22; p = 0.08). A direct comparison of these two differences by training condition indicated this difference did not quite reach significance (CI: −0.02, 0.32; p = 0.07). Overall, we found that seven staff members need to be trained (NNT) in order to result in a staff member asking one more student about suicide (NNT = 1/(1/4 – 1/10)).
The Role of Staff-Student Communication in Predicting Identification Behaviors by Staff
In this section, we extend our analysis of training impact on Ask Students about Suicide, our primary outcome behavior, by adding predictors related to staff-student communication. Because the variable Ask Students about Suicide had such a skewed distribution, we used Poisson modeling with random school effects to examine the impact. We found significant impact on baseline levels of Asking Students about Suicide (β = 3.47 se = 0.79, p < 0.0001), Natural Gatekeeper relationship (β = 0.35 se=0.10, p < 0.001) and three-way interactions involving these two variables and training (β = 0.60 se=0.23 p = 0.01). Plots of these interactions indicated that the highest predicted number of students asked about suicide were from staff who were highest on this variable at baseline and simultaneously high on Natural Gatekeeper Relationship.
Contextual Predictors in Asking About Suicide
To examine differences in training across schools, we added school contextual variables to individual level baseline variables in predicting Ask Students about Suicide at Time 2. School level averages of knowledge and appraisals were computed based on Time 1 scores for those in our sample. Only Appropriate Referral Behaviors showed any indication of predictive ability in Poisson multilevel modeling, and this effect was marginal (β = 0.44 se = 0.22, p = 0.06).
Testing Mediational Pathways through Knowledge and Appraisals to Identification Behaviors by Staff
Earlier we found that training impacted appraisals through interactions involving these respective variables at baseline. Furthermore, we found an interactive effect of training and baseline asking about suicide on that variable at Time 2. In this section, we investigate whether knowledge and appraisals mediate this latter relationship. To examine these mediation-moderation effects, we predicted Time 2 Ask Students about Suicide using all previously found main effects and interactions, and then added both Time 1 and Time 2 knowledge and appraisals as predictors. If mediation pathways do go through knowledge or appraisals, the coefficients involving training should be diminished. The interactions changed very little with these potential mediators. For example, the three-way interaction involving asking about suicide, natural gatekeeper relationship, and training coefficient changed from β = 0.58 (se = 0.30, p = 0.05) without these mediators to β = 0.50 (se = 0.29, p = 0.09). Also, second order interactions involving training changed little as well, and the main effect of training remained non-significant after mediators were added. Thus we did not find evidence that gains in staff knowledge and appraisals mediated the relationship between QPR training and asking students about suicide.
Help-Seeking Attitudes: Students with and without suicide attempt
Students who reported a suicide attempt were significantly less likely to report positive help-seeking attitudes regarding asking an adult for help at school. Of 2,059 students surveyed, 7.3% (150/2059) indicated that they had tried to kill themselves in the last year. That rate was comparable to the proportion of students who indicated a suicide attempt in the prior year survey (6.8%) in the same District and also similar to YRBS rates (YRBSS, 2005
). Also, the odds for girls reporting a suicide attempt was 1.68 times that for boys (95% CI: 1.19, 2.35; p = .01). Attempt rates were comparable for 8th
Help seeking attitudes differed between students with and without prior suicidal behaviors (). For students who reported a suicide attempt, their odds of reporting that they would talk to a counselor or other adult at school if they needed help was less than half as likely as those with no attempt (17.8% vs. 37.8%), or believed a counselor or other adult at school could help them (21.9% vs. 46.5%). Students with a suicide attempt were also less likely to report that friends would want them to talk to an adult at school (34.9% vs. 44.6%) or their family would want them to talk to an adult at school (35.4% vs. 53.4%).
Eighth and Tenth Graders’ Positive Help-Seeking Attitudes as a Function of Self-reported Suicide Attempt vs. No Attempt