Loneliness, Anxiety, and Emptiness: What Real-Time Mood Data Reveal About Teens’ Self-Injury Thoughts

When everyday feelings become early warning signals

Ask any school counselor: the moments that push a teenager toward harming themselves rarely look dramatic from the outside. They are often quiet, private, and tied to the emotions that ebb and flow throughout the day. The research paper The impact of negative emotions on adolescents’ nonsuicidal self-injury thoughts: an integrated application of machine learning and multilevel logistic models gets as close as we can to those moments by asking teens to report their feelings several times per day. In doing so, it reveals a simple but powerful insight: feelings of loneliness, anxiety, and emptiness stand out as the most consistent predictors of adolescents’ thoughts about nonsuicidal self-injury (NSSI)—even more than general sadness or anger.

This matters because early detection saves lives. While the study focuses on thoughts, not actions, those thoughts are often the first and most actionable sign that help is needed. Rather than relying on memory-based surveys that ask teens to look back over weeks or months, the study uses Ecological Momentary Assessment (EMA): brief mood check-ins delivered by smartphone three times a day for two weeks. Then the researchers pair a modern machine learning tool (a “random forest”) with a traditional multilevel logistic regression model to test which feelings predict risk at the moment it arises. The result is both precise and practical: when teens feel lonely, anxious, or empty, the odds of NSSI thoughts go up—reliably and in real time. These findings give parents, educators, and clinicians concrete targets for prevention, and they demonstrate how daily-life data can make care more responsive, humane, and effective.

Loneliness leads the pack, with anxiety and emptiness close behind

Forty-two adolescents aged 12–15 who had engaged in NSSI over the past year completed EMA prompts three times daily for 14 days, reporting feelings like depression, anxiety, loneliness, shame, self-anger, anger toward others, and emptiness, as well as whether they were experiencing NSSI thoughts. A random forest model—an algorithm that ranks which variables best forecast an outcome—found loneliness to be the most predictive (feature importance 0.40), followed by anxiety (0.18) and emptiness (0.14). In plain terms, when these feelings rose, the likelihood of NSSI thoughts rose too.

To check the robustness of those results, the team ran a multilevel logistic regression, a statistical model suited to repeated measures nested within individuals. It confirmed the pattern: for every one-unit increase in reported anxiety, the odds of NSSI thoughts increased by about 24%; for loneliness, by 19%; for emptiness, by 24%. The “how much” matters less than the “who and when”: the effects were consistent across participants, suggesting these emotional triggers are common rather than idiosyncratic. Although there was meaningful variation between individuals overall (ICC = 0.26), the study found no evidence that the strength of these emotional predictors differed substantially from one teen to another.

Put concretely: the student who eats lunch alone for the third day in a row may be at higher immediate risk than the student who is simply in a bad mood. The teen sweating through a math quiz might not just be nervous—they could also be more vulnerable to self-injury thoughts in that hour. And the kid who says they feel “numb” after an argument isn’t just checked out; that emptiness is a real-time risk signal.

Why real-time emotions outdo memory, and what that means

Much of what we know about adolescent self-injury comes from surveys that ask young people to recall how they felt over the past week or month. Memory blurs; emotions shift; context gets lost. EMA sidesteps those problems by sampling feelings in the moment, in the places teens actually live—the lunchroom, the bus ride, the bedroom before bed. That makes the signals more precise and the insights more actionable.

The study’s two-method approach strengthens confidence in the findings. The random forest is good at pattern detection in complex data—it can tell you which feelings matter most without imposing rigid assumptions. The multilevel logistic model then anchors those insights in interpretable numbers and accounts for the fact that observations are nested within people. Finding convergence between these methods is exactly what we want in applied mental health research.

Psychologically, the pattern makes sense. Decades of theory link self-injury to emotion regulation and social connection. The “interpersonal” view highlights how unmet needs for belonging can fuel distress, while family and peer stress heighten vulnerability. Here, loneliness emerged as the clearest signal—stronger than general sadness or anger. That aligns with research showing that when social bonds feel broken or brittle, teens may turn to self-injury thoughts as a way to cope, numb, or communicate pain. Anxiety fits an “overwhelm” pathway: racing thoughts, bodily tension, and fear of failure can narrow options, making harmful urges feel briefly relieving. Emptiness is equally telling. Often described as feeling hollow or disconnected from oneself, emptiness can drive a need to “feel something,” which some teens mistakenly believe self-injury might provide.

Consider a composite case: a 14-year-old who switches schools mid-year. For the first week, they sit alone on the bus (loneliness), dread first-period science (anxiety), and go home to a parent working late (emptiness). None of these moments is dramatic; together, they create a daily landscape where NSSI thoughts feel more likely. The study’s message is not that every instance of these emotions leads to harm. Rather, it’s that monitoring them in real time offers a practical map of when support is most needed—and for whom.

From insight to action: small changes that lower risk

For clinicians and school counselors: build brief, frequent check-ins that directly assess loneliness, anxiety, and emptiness. A three-item daily mood prompt during homeroom or via a confidential app can flag students who need same-day support. When a teen endorses loneliness, prioritize connection-focused interventions—peer mentoring, lunch groups, or quick “check-in walks” with staff. For anxiety spikes, teach short, on-the-spot skills: paced breathing, grounding techniques, or a two-minute “reset” before tests. For emptiness, emphasize activities that restore agency and meaning—creative tasks, values-based goals, or brief behavioral activation.

For families: ask targeted, concrete questions at predictable times. Instead of “How was your day?”, try “Did you feel alone at any point today?” or “Was there a time you felt empty or checked out?” Pair these with small, reliable rituals—shared breakfasts on test days, a 10-minute walk after school, or inviting a friend over midweek to break isolation.

For digital health and youth platforms: use these findings to drive just-in-time support. If a teen’s self-reports or passive signals (like social withdrawal patterns) suggest rising loneliness, deliver a prompt to message a trusted adult or join a moderated, supportive forum. Apps can also nudge users to schedule a specific social micro-action (text a friend, join a club meeting) rather than generic advice.

For policy and program design: expand access to low-stigma social connection. After-school drop-in spaces, peer-support programs, and structured lunch groups are not “extras”—they are protective infrastructure. Staff training should include how to spot and respond to the triad of loneliness, anxiety, and emptiness. Because the study found fairly uniform effects, schools can implement tiered supports without needing extensive individual tailoring to start.

For researchers: the integrated approach in this research paper demonstrates how pairing machine learning with multilevel models yields both prediction and interpretation. Future work can test whether reducing loneliness specifically lowers NSSI thoughts in the short term, and whether adding context (like time of day or social setting) sharpens the signal.

A clearer target for care, and a question worth asking daily

The take-home is refreshingly clear: in the daily lives of adolescents at risk, loneliness is the loudest signal, with anxiety and emptiness close behind. The study—The impact of negative emotions on adolescents’ nonsuicidal self-injury thoughts: an integrated application of machine learning and multilevel logistic models—shows that checking these feelings regularly can identify high-risk moments when support can do the most good. EMA gives us the “when,” machine learning spotlights the “what,” and multilevel logistic modeling clarifies the “how much.”

If you work with teens—or live with one—consider one simple, daily question: “Did you feel lonely today?” It’s not just small talk. It might be the earliest, most actionable clue that a young person needs connection and care right now.

Data in this article is provided by PLOS.

Tagged Adolescents, Age groups, Anxiety, Anxiety disorders, Artificial intelligence, Behavior, Children, Emotions, Families, Machine learning, Mental health and psychiatry, Neuropsychiatric disorders, Neuroses, Population groupings, Self harm, Suicide