Text Analysis of Suicide Risk in Adolescents and Young Adults

Abstract

OBJECTIVE:

Suicide is of primary public concern for adolescents and young adults (AYAs) who commonly use social media platforms to express their suicidal thoughts and behaviors. Understanding how AYAs communicate their suicide-related thoughts and behaviors in texts can support early detection of suicide risk from their social media posts. Therefore, this study sought to identify themes relevant to suicide risk in AYAs and explore words or terms used by AYAs when they described suicidal thoughts and behaviors.

METHOD:

This secondary data analysis utilized an existing data set collected from 255 AYAs between 12 and 25 years of age, who provided brief descriptions of how they and their peers expressed their experiences of self-harm, suicidal thinking, and attempts. Text analysis was conducted using KH Coder software. Three-step theory of suicide was used to guide a content analysis to explore the key themes from the narratives.

RESULTS:

A word co-occurrence network with 24 clusters of words was generated from the text analysis. These word clusters were further grouped into pain or hopelessness, connectedness, and capacity to attempt suicide in the content analysis. Six subthemes corresponding to these three themes were identified to provide detailed information: psychological or physical pain, hopelessness, relationship, help seeking, methods, and outcomes. Moreover, several slang terms and acronyms (e.g., Kermit Sewage Slide, KMS) were also identified.

CONCLUSIONS:

The findings of this study, including themes and slang terms and acronyms, are valuable to facilitate the use of terms or phrases within social media texts to identify suicide risk in AYAs.

Introduction

Suicide rates in the United States are increasing, and the Centers for Disease Control and Prevention reported an approximate 35% increase from 1999 to 2018 in suicide across all age groups, including adolescents and young adults (AYAs; Drapeau & McIntosh, 2021; Hedegaard et al., 2020; Miron et al., 2019). In the United States, suicide attempts among males and females between 10 and 24 years of age have increased at a linear trend during the recent 10 years (Curtin, 2020; Heron, 2021; Nock et al., 2013). Given this crisis, there is an urgent need to develop effective approaches to the early identification of suicide risk in AYAs.

Social media has the potential to support the early detection of suicide risk. Studies have shown that suicidal thoughts and behaviors can be detected from texts from social media (Ramirez-Cifuentes et al., 2020; Robinson et al., 2016; Tadesse et al., 2020; Wulz et al., 2021). Most AYAs, regardless of gender, socioeconomic status, or race and ethnicity, have access to a computer or own a smartphone to use the internet daily. Over 90% of AYAs use at least one social media platform, and on average, AYAs spend between 1 and 5 hours on social medial every day (Anderson & Jiang, 2018; Heffer et al., 2019; Scott et al., 2019). Studies by the Pew Research Center found that AYAs in the United States use social media regularly to interact with others such as family and close friends, share information, and express their feelings and opinions, and 45% of teenagers are online on a near-constant basis (Anderson & Jiang, 2018; Auxier & Anderson, 2021). In addition to images, videos, and audio recordings, text-based posts are common forms of sharing information or opinions on social media platforms. As such, they have become a common focus for research. Evidence has shown that suicidal ideation and behaviors can be extracted from social media texts (Arendt et al., 2019; Parrott et al., 2020; Sawhney et al., 2018). In these studies, synonyms of suicide, suicidality, suicide attempt, self-injury, or self-harm were used as a key approach to extract textual data suggestive of suicidal thoughts and behaviors. However, this approach may not be sufficiently inclusive to identify other relevant aspects or factors associated with suicide.

According to the three-step theory of suicide (3ST), suicidal action is a progression from suicidal thoughts to actual behaviors. Ideation and progression to actual plans and attempts are strongly associated (Klonsky & May, 2015). Suicidal ideation may be initiated by feeling pain (most likely is psychological and can also be physical) and hopelessness (Step 1), which becomes more intense, resulting in shifts from passive to active thoughts when connectedness, a sense of belonging and closeness with others, is absent or disrupted (Step 2), and then progresses to action when one has the capacity to make a suicide attempt (Step 3; Klonsky & May, 2015). Several characteristics from this 3ST, besides the attempt of suicide, have been identified in other studies. These traits include physical or mental pain (Mee et al., 2006; Rizvi et al., 2017; Verrocchio et al., 2016), hopelessness (Wolfe et al., 2019), and lack of meaningful support (Gunn et al., 2018). These traits are important to consider as part of the theme of suicide when using social media texts to detect people at suicide risk. The use of language can vary among persons and can be influenced by gender, culture, and age. It is known that individuals across generational groups may use varied ways to express themselves or use different terms entirely. Subramaniam and Razak (2014) found that different linguistic text patterns existed between younger and older generations regarding texts in social media posts. One of their key findings was that Generation Y (those born between 1981 and 1996) tended to use a variety of abbreviations and acronyms in their posts, such as “b4” for before and “wel” for well, compared with baby boomers (those born between 1946 and 1964). Their study implies that abbreviations and acronyms are important characteristics of linguistic patterns to be included in analyzing social media texts in younger generations. Social media data can potentially help detect suicide risk. To facilitate the use of social media texts in identifying suicide risk in AYAs, the purpose of this study was to identify themes relevant to suicide risk among AYAs and explore words or terms used by AYAs when they described suicidal thoughts and behaviors. This study is unique because we used textual data collected from AYAs and asked AYAs to use their own languages to describe suicide risk-related questions. Accordingly, we sought to answer two specific research questions:

• Research Question 1: What themes are embedded in statements describing suicide risk among AYAs?

• Research Question 2: What specific terms are used by AYAs to imply suicidal thoughts and behaviors?

Methods

This study is a secondary data analysis of an existing data set collected from community-based AYAs between the ages of 12 and 25 years living in the Intermountain West of the United States. Eligible AYAs were those who had access to complete an online survey and a self-reported reading level of fifth grade or above. All participants older than 18 years provided a written informed consent form before participating in the study. Participants who were younger than 18 years provided written assent along with written permission from a parent or guardian prior to participation. Approval for this study was obtained from the institutional review boards of the University of Utah.

Participants were recruited between March 2018 and December 2018 through advertisements and snowballing. In the primary study, the participants were asked to complete an online survey to provide a brief statement with more than two words of how they and their peers expressed their experiences of depressive symptoms. Of the sample from the primary study, 255 AYAs were included in this secondary data analysis as they provided their description of self-harm, suicidal thinking, and attempts in the online survey, which were used to study suicide risk content in this study. The majority of the sample used in this study were female (n = 148), non-Hispanic (n = 233), and Caucasian (n = 204). Participants were a mean of 19.9 years old (SD = 3.7). Around one quarter of the participants (n = 64) self-reported having depression-related diagnoses. A total of 687 statements with 9,647 words were analyzed, including 245 statements for a question of self-harm, 244 statements from a question of suicide attempt, and 198 statements for a question of suicide thoughts. Text analysis was conducted to investigate themes and terms that emerged from the suicide-related texts using KH Coder (https://khcoder.net/en/), free online software for text mining. The initial step using KH Coder was to preprocess the collected texts, which included removing whitespace and punctuations, and eliminating stop words (e.g., articles, prepositions, etc.). This process allowed the retention of words with definitive meanings. A word co-occurrence network was then generated to visualize the word clusters, that is, sets of words with similar appearance patterns or those that more frequently appeared in a phrase or sentence together. These word clusters were used to conduct a theory-guided content analysis (Hsieh & Shannon, 2005; Schreier, 2012) to identify themes based on the 3ST. As part of the content analysis, each word or term in a word cluster was further reviewed by examining how it was used (e.g., context, part of speech, etc.) in the initial statements provided by participants. Two researchers (JK and JG) worked individually to group the word clusters. They then met and resolved any discrepancy in assigning word clusters into their theme group by consensus. Subthemes within each theme were also identified to represent the subtopics within each theme. The slang terms and acronyms used by AYAs (i.e., generation terms or age-set terms) were identified by using the list of unique words.

Results

A total of 831 unique words were included in this text analysis after preprocessing of data. The majority were nouns (e.g., life, suicide; n = 324), verbs (e.g., cut, kill; n = 228), and adjectives (e.g., suicidal, sad; n = 195). Of these 747 words, the 219 words that were present at least three times in the data set were used to generate a word co-occurrence network diagram. The resulting co-occurrence network included 24 word clusters. Five clusters, “not-want,” “try-kill-yourself-you,” “cut-hurt,” “end-life,” and “suicide-attempt,” were more frequently presented in the texts compared with others (see online Supplementary 1). The 24 word clusters were grouped into three themes. Twelve were grouped into pain or hopelessness with three subthemes (psychological pain, physical pain, and hopelessness), three were grouped into connectedness with two subthemes (relationship and help seeking), and nine were grouped into capacity to attempt suicide with two subthemes (methods and outcomes; see online Supplementary 2). Four slang words related to suicide risk were found: yeet describes the act of throwing or to throw oneself out of the world (i.e., killing oneself); Kermit Sewage Slide means to commit suicide; 2meirl4meirl means “me in real life,” which is used to describe an image or video that relates to an individual’s bad situation; emo means a darker attitude. Three acronyms used in social media were identified. FML is commonly used for “fuck my life,” KMS for “kill myself,” and KYS is for “kill yourself.”

Discussion

This study used free-text statements of suicidal thoughts and behaviors generated by AYAs to analyze the themes and terms used in this generation. In response to Research Question 1, study results suggested that three themes of suicide risk were present in AYAs’ statements, which align with Klonsky and May’s (2015) 3ST. Regarding Research Question 2, this study identified several slang terms and acronyms, which are considered generational languages, relevant to AYAs’ descriptions of suicide.

Within three themes extracted in this study, pain or hopelessness was the most prominent theme, compared with others. This can be due to feeling pain, either psychological or physical, and the feeling of hopelessness was found as a common theme of suicide notes (Mee et al., 2006; Orbach et al., 2003). Feeling pain or hopelessness is strongly asocial with suicidal risk and can be used to predict future suicidal behaviors (Rizvi et al., 2017; Verrocchio et al., 2016; Wolfe et al., 2019). Romantic relationships and friendships were identified as the main social connections of AYAs. This finding was not surprising since developing positive and healthy relationships with others is important and part of AYA development. As connectedness is a key protective factor against escalating ideation of suicide (Gunn et al., 2018; Klonsky & May, 2015; Rubenstein et al., 1989), social media texts containing information related to negative social or interpersonal connections in AYAs’ lives could lead to suicidal ideation or attempts if their coping mechanisms or support was poor. “Help seeking,” the subtheme of connectedness, is an important warning of suicide risk and reflects the needs of AYAs who were at suicide risk. Seeking help is one of the reasons that individuals at risk for suicide use social media platforms (Robinson et al., 2016). The themes relevant to methods and outcomes of suicide in the capacity to attempt suicide are crucial to early identifying those who may be planning to harm themselves.

People in different generations often have different vernaculars, including specific phrases (Subramaniam & Razak, 2014). Using AYAs’ common expressions of their suicide-related thoughts and behaviors is important to detect the embedded meaning of textual data from their social media posts or self-reported notes. Several suicide-related slang terms used by AYAs hold contextual and generational meanings. These terms may not be perceived as “risky” as they are not proper English words and do not connect to other words directly related to suicidal thoughts or behaviors. Additionally, some slang terms may be used in a way that was not literal or sometimes can be used in context to be ironic or sarcastic. For example, Kermit Sewage Slide could be used as a casual or sarcastic expression; however, among AYAs, this term could also express an actual intent to commit suicide. Another example is Yeet, which means to throw something away at high velocity, may express frustration and concern in some AYAs. Acronyms are commonly used in AYAs’ social media posts. The acronyms identified from this study deliver embedded meanings of suicide risk can play a key role in analyzing social media data for early detection. Because an acronym created from the initial letter or letters of words and may contain more than one meaning or may not be recognized as an English vocabulary per se, it could lead to a missed meaning or a dismissal from alarming when the knowledge of suicide-risk acronyms is not available. For example, KMS, a common acronym, can mean as “kilometers” for most people while it can mean as “killing myself” to some AYAs. Therefore, besides using general keywords and terms describing suicide, researchers need to consider including slang terms and acronyms used in AYAs to screen those who are at high risk of suicide. Study results can serve as an initial step to future detection of AYAs’ suicide risk using their texts from social media posts or other sources, which could be a possible intervention in telemedicine research. Although the linguistic patterns identified from this study may not be directly useful for clinicians who may not necessarily have access to patients’ social media posts, a greater understanding of the themes and subthemes based on the 3ST would be useful to clinicians to assess suicide risk in their patients. The themes extracted in this study, which align with 3ST, identify key components of suicide risk that can be used to guide clinicians and nurses when they assess suicide risk for AYAs. Given that many slang terms and acronyms do not literally connect to suicidal thoughts or behaviors, they could be easily missed by clinicians and nurses when screening for suicidal risk. Using the findings of this study, we can improve the screening for suicide risk by integrating vernacular expressions used by AYAs by incorporating additional relevant word choices representing suicidal thoughts and behaviors that are relevant to AYAs during the clinical check-in. Understanding terms and slang expressions used by AYAs could lead to health care providers better understanding their patients.

This study has several limitations. First, the scope of this study was limited to the textual data collected from a convenience sample of AYAs in the primary study. Additionally, the majority of participants were White, non-Hispanic females. Since the textual data were analyzed collectively for the entire study sample, we were unable to further compare responses based on gender, race, or ethnicity. The history of suicidal thoughts and behaviors were not collected so the study results cannot be used to represent those who had an experience of suicidal thoughts and behaviors. Also, this study was limited to the quantity and quality of available text data. Some participants only provided several individual words rather than a phrase or a sentence. Without a clarified context in which these words were used, precisely interpreting the meaning of the statement was difficult. However, this limitation did not significantly affect the study analyses and subsequent results in terms of grouping the themes and identifying terms relevant to suicidal thoughts and behaviors in AYAs.

Conclusions

Early detection of suicidal thoughts and behaviors is important to implement effective, early intervention strategies to prevent suicide. AYAs actively engage in social media platforms to express themselves and socialize with others. Evidence has shown that social media data or posts can be used to identify individuals at suicide risk. The findings of this study demonstrate the feasibility of using text analyses to detect phrases and terms suggestive of suicide risk and can guide future research to develop better strategies to detect AYAs who are at suicide risk, ultimately, to prevent suicide.

Supplemental Material

sj-pdf-1-jap-10.1177_10783903221077292 – Supplemental material for Text Analysis of Suicide Risk in Adolescents and Young Adults

Supplemental material, sj-pdf-1-jap-10.1177_10783903221077292 for Text Analysis of Suicide Risk in Adolescents and Young Adults by Jia-Wen Guo, Julianne Kimmel and Lauri A. Linder in Journal of the American Psychiatric Nurses Association