Characterizing Online Activities Contributing to Suicide Mortality Among Youth

Background and Purpose.

Suicide is the third leading cause of death among youth ages 10 - 24 in the U.S.A. In his 2021 advisory, the U.S. Surgeon General identified social media as a key factor in recent declines in youth mental health, as research links higher use of social media, longer screen times, and harmful online activities (e.g., cyberbullying) with youth suicide ideation. However, it remains unclear which online behaviors are linked with higher risk of youth suicide deaths. Since the population who die by suicide has different characteristics than those expressing suicidal ideation (e.g., gender, means of death), it is important to study factors related to suicide death as an epidemiologically distinct phenomenon. However, it is difficult to link information about social media usage to death records.

Methods.

Our mixed-methods approach responds to this challenge by developing a set of themes focused on risk factors for suicide mortality in online digital spaces and a framework to model these themes at scale. To analyze online activities among youth who have died by suicide, we use the NVDRS registry of suicide deaths in the U.S.A. NVDRS data include a narrative summary of the law enforcement and medicolegal investigations into each death. We analyze online activities described in 29,124 narratives among youth ages 10 to 24 in a ten year period when social media use among youth was high, from 2013 through 2022. First, our team of eight authors conducted a thematic analysis on these NVDRS narratives. We started with an inductive approach to describe online activities in the narratives, and then grouped these descriptions into 13 themes aligned with two suicide theories. Next, we prompted a large language model, Llama3.1-8B-Instruct, to detect each of these themes in NVDRS narratives. We developed a novel prompting approach to classify decedents' online engagement in the narratives and evaluated the performance of the LLM against 595 cases labeled by the authors. Using this data, we examined the demographic, social, and contextual variables associated with a given social media experience to better understand trends related to social media and adolescent suicide.

Results.

The 13 themes were related to harm to self, harm to others, interpersonal interactions, and activity levels online. The most common themes were Disclosure (44% of online activities), Online Conflict (13%), and Suicidal Content Consumption (11%). We find that many themes are likely driven by the COVID-19 stay-at-home lockdowns, including a rise in problems with online school, intensifying of online activity, and online relationships during the pandemic period. We also find a connection between online activities and greater lethality of means, as using a firearm for means of death was positively associated with the presence of online activities in the decedent’s narrative.

Conclusion.

The broad range of behaviors captured by our themes suggests that there are several key points of intervention that are often overlooked in analyses of suicide risk online. Our work has implications for suicide prevention research in computational and policy domains.