Smartphone language and resting‐state <scp>EEG</scp> indicators of self‐focused attention prospectively predict major depressive disorder risk in adolescents

Lilian Y. Li, Nayoung Kim, Esha Trivedi, Sarah E. Sarkas, Madeline M. McGregor, Aishwarya Sritharan, Katherine Durham, Ivan Alekseichuk, Allison M. Letkiewicz, Vijay A. Mittal, David Pagliaccio, Nicholas B. Allen, Randy P. Auerbach, Stewart A. ShankmanJournal of Child Psychology and Psychiatry 2025 · DOI

TL;DR

This study reveals that both how adolescents use language on their smartphones (specifically, first-person singular pronouns) and their brain activity (resting-state EEG alpha power) can predict future risk of major depressive disorder (MDD). Higher alpha power is protective, while increased "I" language signals higher risk, suggesting these are distinct but related markers of self-focused attention that could be used for early detection and intervention.

What Problem Does This Paper Address?

Major depressive disorder (MDD) is a significant public health concern, particularly among adolescents, and its recurrent nature poses a substantial burden. Identifying reliable risk markers that persist even after remission is crucial for prevention and early intervention. Previous research has pointed to maladaptive self-focused attention as a core feature of depression, which can be measured through two seemingly parallel avenues: linguistic analysis of language (specifically, the use of first-person singular pronouns like "I") and neurophysiological measures (resting-state EEG alpha oscillations). However, the precise relationship between these linguistic and neural indicators, and their combined predictive power for future MDD onset, has remained unclear. This paper aims to bridge this gap by investigating how these two markers of self-focused attention, captured through both smartphone language and EEG, prospectively predict MDD risk in adolescents.

Methods

The researchers employed a longitudinal design, recruiting adolescents aged 13-18 who either had remitted MDD or no history of psychiatric disorders. At baseline, participants underwent resting-state EEG recordings and had their smartphone keyboard inputs passively collected for 12 months.

  • Linguistic Analysis: Smartphone data was processed to extract the proportion of first-person singular pronouns ("I," "me," "mine") relative to all pronouns used. This "linguistic self-focused attention score" was calculated from approximately 2.3 million messages.
  • EEG Analysis: Resting-state EEG data was collected during eyes-closed conditions, a state known to maximize alpha power. The researchers focused on alpha oscillations (8-13 Hz) in the posterior regions, a frequency band associated with intrinsic, self-focused attention.
  • MDD Assessment: Participants were followed for 12 months, during which their occurrence of major depressive episodes (MDEs) was assessed through retrospective interviews.
  • Statistical Modeling: Cox proportional hazards models were used to assess the predictive utility of linguistic and neural indicators for MDD risk. Linear regression was employed to examine the association between alpha power and pronoun usage. A mediation model was constructed to explore whether pronoun usage mediated the relationship between alpha power and MDD risk, and to investigate potential suppression effects.

Datasets

The study utilized data from:

  • Smartphone Keyboard Input Data: Passively collected over 12 months from participants' smartphones, comprising approximately 2.3 million messages. This data was used to derive the proportion of first-person singular pronouns.
  • Resting-State EEG Data: Collected at baseline from 126 adolescents.
  • Adolescent Longitudinal Interval Follow-Up Evaluation (A-LIFE): A retrospective interview used to assess weekly MDD severity over the 12-month follow-up period.
  • DSM-5 version of the Kiddie Schedule for Affective Disorders and Schizophrenia (K-SADS-PL): Used at baseline to assess lifetime psychiatric diagnoses.
  • Wechsler Abbreviated Scale of Intelligence-II (WASI-II): Used to assess verbal intelligence.
  • Children's Depression Rating Scale-Revised (CDRS-R): Used at baseline to assess depression symptom severity.

No new benchmarks were introduced; the study leverages existing methods for data collection and analysis.

Key Insights and Ideas

  • Dual Nature of Self-Focused Attention: The study elegantly demonstrates that self-focused attention, a known risk factor for depression, is not monolithic. It can be indexed by both linguistic markers (pronoun usage) and neural markers (alpha power), and these markers appear to have distinct, even opposing, roles in predicting depression risk.
  • Protective Role of Alpha Power: Counterintuitively, higher resting-state alpha power was associated with a decreased risk of developing MDD. This suggests that alpha oscillations might be involved in cognitive control mechanisms that buffer against depressive rumination.
  • "I" Language as a Red Flag: Increased use of first-person singular pronouns on smartphones was a significant predictor of increased risk for MDD. This aligns with the idea that excessive self-focus, particularly negative self-focus, is a hallmark of depression.
  • Smartphone Data's Potential: The study validates the use of passively collected smartphone language data as a viable and scalable method for assessing a key psychological construct (self-focused attention) in real-world settings.
  • Suppression Effect Hint: The mediation analysis hinted at a suppression effect, suggesting that the protective role of alpha power might be amplified when accounting for the risk associated with "I" language. This implies a complex interplay between neural and linguistic mechanisms.

Key Results

  • Pronoun Usage Predicts MDD: Greater usage of first-person singular pronouns prior to an MDE significantly increased the risk for developing an MDE during the 12-month follow-up (Hazard Ratio [HR] = 2.02, p < .001).
  • Alpha Power Protects Against MDD: Greater baseline resting-state alpha power significantly decreased the risk for MDE during follow-up (HR = 0.78, p = .001).
  • Alpha Power Predicts Pronoun Usage: Higher alpha power predicted greater subsequent usage of first-person singular pronouns (β = 0.17, p = .004).
  • Mediation Analysis: A marginal suppression effect was observed, where accounting for first-person singular pronoun usage amplified the protective association between alpha power and MDD risk (indirect effect p < .10).
  • Sample Characteristics: Out of 126 participants, 29 (23.0%) developed an MDE during the 12 months.

Limitations and Future Work

The authors acknowledge several limitations:

  • Noisy Language Data: The inclusion of messages from all smartphone apps, not just social communication, likely introduced noise into the language data, potentially diluting effect sizes and making it harder to discern context.
  • Broad Alpha Power Measurement: Using whole-brain alpha power might have masked region-specific or frequency-specific effects that could be more strongly related to cognitive control or self-referential processing.
  • Exclusion of Non-White Participants: Due to poor EEG data quality (e.g., braided hair), a disproportionate number of non-White participants were excluded, limiting the generalizability of findings to this population. Future research should focus on best practices for EEG preparation to improve inclusivity.

Future work suggested includes:

  • Refined Linguistic Analysis: Investigating the context of pronoun usage (e.g., positive vs. negative messages) and exploring other linguistic markers.
  • More Granular EEG Analysis: Examining alpha sub-bands, alpha power and phase, and source localization to disentangle distinct alpha mechanisms.
  • Clinical Translation: Leveraging passive smartphone monitoring for real-time identification of excessive self-focus and developing targeted interventions, such as third-person self-talk, to prevent MDD recurrence.

Summary updated

Abstract

Findings highlight functionally distinct alpha mechanisms and provide support for smartphone-based first-person singular pronoun usage as a neurobehavioral risk factor and a potentially promising intervention target for adolescent MDD.
BibTeX
@article{li2025smartphone,
  author = {Lilian Y. Li and Nayoung Kim and Esha Trivedi and Sarah E. Sarkas and Madeline M. McGregor and Aishwarya Sritharan and Katherine Durham and Ivan Alekseichuk and Allison M. Letkiewicz and Vijay A. Mittal and David Pagliaccio and Nicholas B. Allen and Randy P. Auerbach and Stewart A. Shankman},
  title = {{Smartphone language and resting‐state <scp>EEG</scp> indicators of self‐focused attention prospectively predict major depressive disorder risk in adolescents}},
  journal = {Journal of Child Psychology and Psychiatry},
  year = {2025},
  doi = {10.1111/jcpp.70096},
}
Figure