Introduction

Young people have repeatedly been identified as a group requiring specific attention in suicide prevention efforts. Globally, around one-third of all suicides occur among those aged 15 to 29 years [1], and rates of youth suicide are rising in many countries, including Australia [2], the United Kingdom [3], and the United States [4]. To prevent suicide, it is important to intervene in suicidal behaviors that may signal an increasing risk for suicide death, such as intentional self-harm and suicidal ideation [5]. Suicide attempt and ideation are estimated to occur in around 1% and 11% of community-based young people each year, respectively [6], while suicide mortality occurs in less than 0.02% of all 15 to 24 year olds according to national data registries (e.g., 2). Although pooled findings indicate that the odds of a future suicide attempt are only 1.88 times higher among individuals with suicidal ideation than those without [5], prior evidence shows one-third of youth with suicidal ideation will develop a suicide plan, and 60% with a plan will make a suicide attempt [7]. This risk, considered alongside the prevalence, indicates that suicidal ideation is a novel target for intervention targeting young people.

Increasing trends in youth suicide alongside the substantial and unchanging mental health burden among those aged 0 to 24 years [8] creates considerable uncertainty as to whether there is sufficient availability of treatment services and workforce capacity to meet supply and demand needs over the coming decades. Access to mental health services is already grossly inadequate in regions where suicide rates are the highest, with only 1.6% to 7.2% of those in need accessing such services in low-to middle-income countries, while access rates are 2 to 3 times greater in higher income countries [9]. Similar patterns of higher sucide rates and reduced access to mental health services are echoed in rural and remote regions of high income countries [10,11]. Developing models of treatment that are effective, translatable, and scalable is a salient priority for suicide prevention.

For young people, such services need to capitalize upon how they want to access and engage with help as they are one of the groups least likely to seek help. It is estimated that up to 70% of those aged 12 to 18 years with suicidal thoughts and/or behaviors do not actively utilize mental health services because of embarrassment and stigma, difficulty expressing suicide concerns, lack of support, and a preference for self-reliance [12]. For these reasons, technology-enabled solutions, such as smartphone interventions, may be particularly appropriate models of care for young people. Not only is smartphone ownership and use ubiquitous among those aged 13 to 29 years, but also many (40% or more) report accessing apps specifically for mental well-being [13]. This willingness to engage with technology for mental health support creates enormous potential for smartphone interventions to genuinely address service access gaps.

Adults can readily access and benefit from technology-enabled solutions that directly target suicidal ideation and/or behaviors [14,15]; however, the evidence for this is largely derived from trials of web-based programs rather than smartphone interventions. To our best knowledge, there have only been 2 published studies of brief, unsupervised smartphone interventions designed to treat suicidal ideation and/or behavior in younger adults. In both studies, these interventions were unable to demonstrate superior effects for suicidal ideation [16,17], although one study did show an intervention effect for suicidal plans and suicidal behavior [16]. This latter finding, together with meta-analytic evidence that smartphone applications have a moderate effect on depressive symptoms [18] and a small effect of anxiety symptoms [19], shows promise for these interventions. To advance the field and realize the potential of smartphone applications as effective components of suicide prevention efforts, more trials of interventions that directly target distress processes specific to suicide are needed, rather than more trials of mental health interventions that incidentally measure suicidal outcomes. This need for development and testing of purpose designed treatments is because meta-analyses have shown that interventions directly targeting suicidal behaviors can reduce ideation in the short-term, whereas broader mental health interventions are unable to [14,20].

The current knowledge gaps in the availability of, and evidence for, technology-enabled treatment models specifically for young people informed the development of a targeted self-guided smartphone application (known as LifeBuoy). The LifeBuoy application is primarily based on dialectical behavior therapy (DBT), a specific modified form of cognitive behavior therapy designed to treat persistent emotional dysregulation as a means to preventing self-harm and suicidal behaviors [21]. In DBT treatment, individuals are taught to change unhelpful ways of thinking and behaving, develop healthy distress tolerance skills, and regulate their emotions. To the best of our knowledge, the LifeBuoy application represents the first intervention of its kind to be tested in a randomized trial setting. Since many young people seek help directly online, via their phones [22], we opted to test the intervention as an app available in the app store and promoted the trial via social media. These conditions were likely to mimic the way the intervention would be delivered and used in the community if effectiveness were to be established. The primary objective of this study was to investigate the efficacy of the Lifebuoy smartphone application in reducing the severity of suicidal thoughts when compared with an attention-matched smartphone application (LifeBuoy-C). A secondary objective was to examine these effects for broader mental health outcomes of depression, anxiety, distress, and well-being.

Methods

This study is reported as per the Consolidated Standards of Reporting Trials (CONSORT) guideline (S1 CONSORT Checklist) [23]. The trial protocol (S1 Protocol) was prospectively registered on the Australian New Zealand Clinical Trials Registry (ACTRN12619001671156) and has also been published elsewhere [24]. The intervention app was registered with the Therapeutic Goods Administration Clinical Trial Notification scheme (CT-2020-CTN-00256-1-v1).

Study design and participants

This 2-arm parallel, double-blind, randomized controlled trial was conducted at the Black Dog Institute, Australia. Participants were recruited via targeted advertisements on the social media site, Facebook, between May 11, 2020 and May 22, 2020. Eligible participants were between 18 and 25 years of age, in the community (nonclinical sample), residing in Australia at the time of registration, and who responded in the positive to the question “have you experienced suicidal thoughts in the past 12 months?” Eligible individuals also had to own a smartphone (versions Android 5 and iOS 9 or higher) and be fluent in English. Participants were excluded if they had attempted suicide in the 1 month prior to trial registration and/or had a current or lifetime diagnosis of psychosis or bipolar disorder. These exclusion criteria were designed to ensure that we were able to safely manage trial participation. Ineligible participants were provided with phone numbers for appropriate 24-hour crisis support services at the conclusion of the screening and offered the opportunity to speak to the research team’s clinical psychologist. Being in current receipt of other forms of mental health treatment was not an exclusion criterion. The study was approved by the University of New South Wales Human Research Ethics Committee (HC190764) and required written opt-in consent by participants.

Results

Fig 1 shows the trial profile. Over 12 days, starting May 11, 2020, 455 participants were randomly assigned to receive LifeBuoy (N = 228) or LifeBuoy-C (N = 227). The final survey was administered on October 1, 2020. Of these, 332 (73.0%) participants completed (or partially completed) T1 survey, and 254 (55.8%) completed the T2 survey (Fig 1). There were no significant differences in attrition from assessment between the conditions at T1 (Fisher exact test p = 0.40) and T2 (Fisher exact test p = 0.09) (Fig 1). No participants formally withdrew from the study.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. CONSORT trial flow.

CONSORT, Consolidated Standards of Reporting Trials.

https://doi.org/10.1371/journal.pmed.1003978.g001

The participant sample was primarily female (n = 384, 84.4%) with a mean age of 21.5 years (SD: 2.18). Nearly one-third had completed a tertiary qualification (n = 135, 29.7%). Mental health issues were prevalent, with 88.6% (n = 403) having ever received a mental health diagnosis and 88.1% (n = 401) have ever received mental health treatment. At baseline, gender, age, and other demographic characteristics were similar between conditions (Table 1). For the total sample, the baseline mean SIDAS scores were in the range that indicates high risk of suicidal behavior (≥21; M = 22.61; SD = 8.18). The mean PHQ-9 (M = 17.15; SD = 5.64), GAD-7 (M = 12.29; SD = 5.07), DQ5 (M = 17.87; SD = 3.42), and SWEMWBS (M = 17.15, SD = 2.53) scores were also within clinically relevant ranges, indicating moderate-severe manifestations of these symptoms at baseline.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Baseline characteristics of the total sample by condition.

https://doi.org/10.1371/journal.pmed.1003978.t001

Primary outcome

There was a greater reduction in SIDAS scores from T0 to T1 for LifeBuoy (B = −8.05, 95% CI [−9.92, −6.18], t[452.90] = −8.45, p < 0.001, d = −1.02) compared to the control condition (B = −3.14, 95% CI [−5.04, −1.23], t[463.03] = −3.24, p = 0.001, d = −0.37; see Table 2, Fig 2), resulting in significantly lower SIDAS scores at T1 (B = −4.43, 95% CI [−6.70, −2.16], t[413.83] = −3.83, p < 0.001, d = 0.45) for those receiving the LifeBuoy intervention. While there was no further significant change in SIDAS scores that occurred from T1 to T2 for the LifeBuoy condition (B = −0.86, 95% CI [−2.56, 0.84], t[218.46] = −1.00, p = 0.319, d = −0.08), T1 to T2 SIDAS scores significantly decreased for the control condition (B = −1.78, 95% CI [−3.53, −0.02], t[226.05] = −2.00, p = 0.047, d = −0.20). The significantly lower SIDAS scores of the LifeBuoy condition relative to the control condition observed at T1 remained at T2 (B = −3.51, 95% CI [−6.08, −0.94], t[473.81] = −2.69, p = 0.007, d = 0.34). Over the whole T0 to T2 period, a significant time × condition interaction was also found (see Table 2), with an overall greater reduction in SIDAS scores for LifeBuoy (B = −8.91, 95% CI [−10.97, −6.85], t[530.68] = −8.50, p < 0.001, d = −1.13) versus the control condition (B = −4.91, 95% CI [−6.99, −2.83], t[531.94] = −4.64, p < 0.001, d = −0.58).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Mean scores, tests of time × condition interactions, and comparisons between time points and conditions.

https://doi.org/10.1371/journal.pmed.1003978.t002

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Comparison of SIDAS modeled mean scores at T0, T1, and T2.

SIDAS, Suicidal Ideation Attributes Scale; T0, baseline; T1, postintervention; T2, 3-month postintervention.

https://doi.org/10.1371/journal.pmed.1003978.g002

To explore the clinical significance of change in suicidal ideation, we undertook an exploratory analysis of item 5 of the SIDAS that measures life interference from suicidal ideation (“how much have thoughts about suicide interfered with the ability to carry out daily activities, such as work, household tasks or social activities”). There was a significant reduction in life interference from suicidal ideation for LifeBuoy compared to the control condition from T0 to T1, leading to lower levels of life interference from suicidal ideation for LifeBuoy at T1 but this was not maintained at T2 (S1 Text).

Harms

Participants who scored ≥21 on the SIDAS were defined as at high risk for suicidal behavior [23]. At the T1 and T2 surveys, participants in the control condition had significantly greater odds of being flagged as high risk on the SIDAS compared to the intervention condition. Correspondingly, the number needed to treat (NNT) to obtain one participant scoring <21 on the SIDAS was 5 and 8 at the T1 and T2 time points, respectively (Table 3). At the T1 survey, n = 4 participants (n = 2 in each condition) self-reported having attempted suicide in the past 30 days, while at the T2 survey, 5 participants had attempted suicide in the past 30 days, all in the control condition. Of these 9 suicide attempts, none required medical care.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Proportion of participants at high risk for suicidal behavior at T0, T1, and T2.

https://doi.org/10.1371/journal.pmed.1003978.t003

Attrition from assessment

In the attrition analysis, those who completed the T1 survey had significantly lower baseline anxiety symptoms (t[445] = 2.3; p = 0.02; S1 Table) compared to participants who did not complete it. Compared to participants who did not complete the T2 survey, those who completed it were more likely to have a university degree (χ²(1) = 6.60, p = 0.10), significantly less baseline depression (t[451] = 2.6; p = 0.01), anxiety (t[445] = 2.7; p = 0.008), distress (t[444] = 2.3; p = 0.02), and greater baseline well-being (t[438] = −2.6; p = 0.009; S1 Table). No other significant differences were found. The significant time × condition interactions and significant differences between conditions at T1 and T2 found in the primary outcome analyses of the SIDAS remained after repeating these analyses with adjustment for the variables differentially associated with attrition at T1 and T2 (see Table 4).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. Effect of LifeBuoy relative to LifeBuoy-C on the SIDAS after adjustment for baseline variables associated with attrition.

https://doi.org/10.1371/journal.pmed.1003978.t004

Discussion

This study evaluated the efficacy of a brief, unsupervised therapeutic smartphone application for young adults in the community experiencing suicidal ideation. Results show moderate intervention effects for in the immediate postintervention period (d = 0.45) and smaller benefits at the 3-month follow up (d = 0.34) in reducing the primary outcome, suicidal ideation. Although suicidal ideation reduced in both the intervention and the attention-matched control conditions over time, these reductions were superior in the LifeBuoy condition. For secondary outcomes relating to depression, generalized anxiety, distress, and well-being symptom changes in the expected direction were observed over time, with the exception that depression symptoms increased between the immediate postintervention and 3-month follow up period in the LifeBuoy condition. No significant between-condition effects were reported for any of these secondary outcomes at any time point.

Our results are consistent with pooled meta-analytic findings that digital therapeutics appear to be effective in reducing suicidal ideation [14] yet deviate from the effects reported in standalone trials. That is, prior to this study, 2 brief, unsupervised smartphone applications were tested in 4 unique RCTs involving similarly aged participants with suicidal thoughts or behaviors [16,17], neither application demonstrated an intervention effect for suicidal ideation (ps: 0.12 to 0.30). In considering why our findings may differ from previous studies, there is evidence that treatment approach may be an important moderator of efficacy [14,20]. The therapeutic models tested in these previous trials differed from that used in Lifebuoy; one application delivered acceptance and commitment therapy [17] and the other therapeutic evaluative conditioning [16]. Our findings extend prior research showing that DBT can be effectively digitized to treat suicidal ideation [14] by showing promise for this approach in young adults. It is also noteworthy that the immediate effect of Lifebuoy on suicidal ideation in our postintervention period (d = 0.45) is comparable to the pooled effects reported for the intensive face-to-face DBT treatment of suicidal ideation (e.g., d = 0.48) [32]. When delivered in its face-to-face modality, DBT is a lengthy (4 to 6 months duration) multicomponent program involving individual psychotherapy, multifamily group skills training, coaching, and weekly therapist team consultation [33]. While the intensity of face-to-face DBT is well aligned to its core purpose, i.e., to reduce intentional self-harm behaviors in those with complex mental health issues (e.g., borderline personality and bipolar disorders) [34], our findings suggest that this therapeutic approach be used for the short-term management of suicidal ideation for a less complex cohort of young adult ideators.

The absence of an intervention effect for secondary mental health and well-being outcomes was unexpected yet appears to corroborate findings from prior meta-analytic studies that even face-to-face DBT does not significantly reduce depressive symptoms among adolescents (p = 0.099; [32]) nor adults (p = 0.08; [33]). Such findings imply that effective therapeutic approaches for the treatment of suicidal ideation may not parallel those used to successfully treat common mental health conditions (e.g., depression and generalized anxiety disorder). Certainly, while cognitive behavioral therapy (CBT) has been successfully translated into digital interventions for depression and anxiety [35], it appears to have little impact on ideation or other suicidal outcomes in comparison to alternative therapies such as DBT [14]. Variation in the clinical usefulness of different treatment approaches substantiates a need to purposefully design digital interventions to a specific health outcome—particularly unsupervised digital therapeutics that are susceptible to being used as multipurpose solutions, potentially to little effect.

Our trial also offered several novel strengths. First, LifeBuoy was codesigned with young people who had a lived experience of suicide to ensure the resulting application aligned to the preferences and needs of end users. Issues of relevance, acceptability, and perceived benefits of digital therapeutics can be addressed through codesign and may help overcome issues of poor treatment engagement [36]. Pooled findings from prior research estimate that treatment dropout is 74% for unsupervised digital therapeutics [37], which is problematic as low engagement has been shown to reduce the efficacy [38]. The relatively high rates of treatment “completion” reported in the LifeBuoy condition (71.5%) does appear to support acceptability and relevance may be enhanced by codesign. Second, unlike this trial, neither of the previously referenced trials of smartphone applications [16,17] provided an application that exactly replicated the aesthetics and design of the interventional application to the control condition. Although Franklin and colleagues [16] did also use a control application, there is no evidence to suggest it mirrored the intervention program. By creating a robust digital placebo, we were able to test whether the therapeutic content contributed to improvements in key outcomes above and beyond participant expectations about LifeBuoy (i.e., the digital placebo effect) [39]. The potential for the digital placebo effect to influence efficacy is demonstrated in a prior study of a smartphone application for adolescents to self-monitor symptoms of depression; the study found that without any direct therapeutic intervention, self-monitoring significantly reduced symptoms [40]. As such, our trial design increases confidence the superior effects reported in our LifeBuoy condition are, at least in part, directly attributable to the treatment content itself. Not only does this study strengthen the evidence base for the short-term efficacy of brief unsupervised and targeted digital therapeutics in young adults [14], but it also demonstrates that LifeBuoy application has a favorable risk–benefit profile. Relative to Lifebuoy, the control condition had a 22% increase in the odds of flagging as high risk of suicidal behavior at postintervention. Also, no suicide attempts or serious adverse events were reported in the intervention condition at the final survey, signaling that LifeBuoy could be made available for young adult suicidal ideators without substantial safety concerns. To further understand the potential harms and benefits of LifeBuoy, however, this trial would have benefitted from including individuals with recent suicidal behavior. As a prior suicide attempt confers a significantly higher odds of experiencing a future suicide attempt compared to ideation [5], the inclusion of actively suicidal persons may have better positioned this study to detect serious adverse events (harms) and improve our understanding of the application’s potential to interrupt ideation-to-action processes and prevent future attempts [41]. Also, the inclusion of novel methods, such as ecological momentary assessment (EMA) and data linkage [42,43], in this trial would have improved the capture of real-time and long-term impacts of LifeBuoy on suicide ideation, attempt, and death, which would further strengthen causal inferences regarding the contribution of digital therapeutics to suicide prevention.

Close to 90% of participants had received treatment from a mental health professional prior to joining the trial. Rather than “reach the unreachable” (i.e., the nonhelp seekers [12]), this trial appeared to appeal to young adults with high rates of treatment familiarity and engagement. This finding is not new but has not received much attention; similar trials also report high rates of prior treatment engagement (60% or higher; [31,44]). This bias toward treatment engagers may be because young people are unwilling to disclose suicidal ideation mental health professionals. Prior research suggests that nondisclosure is not uncommon—and may be related to fear of stigmatization, uncertainty of repercussions, and prioritization of other mental health issues [45]. Accordingly, unsupervised digital interventions might provide a solution to these concerns, enabling young people to manage suicidal ideation in private. It will be important to explore the reasons why young people who are accessing mental health services would be motivated to engage with a targeted digital therapeutic for suicidal ideation. Such insights would help to inform decisions regarding best fit delivery pathways for enabling reach and access of digital therapeutics; if young adults are unwilling to disclose ideation to clinicians, then delivering these interventions through health systems and services may not be appropriate.

The high rates of prior treatment exposure in this study raise the question of whether treatment familiarity contributed to the effects of LifeBuoy on ideation, and whether it would work as well for those without such exposure. Certainly, the brief nature of smartphone interventions, necessitated by the physical constraints of screen sizes, creates potential for the meaning and actioning of complex therapeutic concepts to be lost in translation. Comparisons of efficacy trajectories of those with and without treatment exposure could provide novel insights into who these interventions work best for.

Several other limitations of the current study should be considered. Our sample was biased toward females, Lesbian, Gay, Bisexual, Transgender, Queer, and Intersex (LGBTQI) persons, and those who had previously received treatment. The targeted recruitment of males and nontreatment recipients using purpose designed Facebook advertisements could help address questions of generalizability. Only 1.7% of those who saw the Facebook advertisements initiated the screening for eligibility process, indicating a high rate of reach but low click conversion rates. Thoughtfully designed advertisements for specific populations may help improve conversion rates by enhancing appeal, while reconsideration of the targeted parameters of the advertisements (e.g., interests, demographics, and geography) may reduce the likelihood that they are seen by those who would be ineligible on key inclusion criteria [46].

Power analyses were calculated for our primary outcome, but not for any of our secondary outcomes or post hoc analyses. By the final survey, the attrition from assessment rate was 45%, consistent with prior clinical trials of smartphone interventions [36]. Attrition was associated with more severe baseline clinical profiles than those who remained in the study at T2. Substantial psychological effort is required to attain treatment goals—particularly for unsupervised treatment—and severe mental illness can be a major barrier to being able to sustain efforts. Individuals who dropped out may have benefitted less from our intervention, and without their actual data, efficacy may be overestimated. Strategies to improve retention should be explored in future trials. Given these limitations, the transfer of benefit of the LifeBuoy intervention to real-world settings and the full clinical meaningfulness of the findings, as well as the mechanisms underlying these effects, should be explored in further studies.

In conclusion, our findings indicate that a brief, unsupervised DBT smartphone intervention can reduce suicidal ideation in young adults, but these effects do not extend to improvements in other mental health issues. To confirm whether LifeBuoy could help to increase access to care for suicidal ideation among youth unable or unwilling to access face-to-face treatment or supplement other forms of mental health treatment, the results of this first trial need be replicated and extended to suicidal behavior in future clinical trials.

Supporting information