JMIR Publications

ABSTRACT

Background:

Early Maladaptive Schemas (EMSs) are persistent self-defeating patterns of thought, emotion, and behavior often contributing to mental health problems. These patterns are essential in Schema Therapy (ST) to understand patients’ complex mental health problems, especially their associations with EMSs and features of the manifested EMSs. However, investigation on EMSs relative to these problems remains largely unexplored among online mental health forum communities where individuals post questions regarding their mental health. This gap, hindering the translation of ST interventions into these online communities, will be explored in this study.

Objective:

We aim to (1) identify the EMSs pertinent to each forum post, (2) uncover direct associations between EMSs and mental health problems, and (3) extract features of EMSs.

Methods:

We collected posts from prominent online mental health forum communities and identified the EMSs for each post using our recently proposed AI method that uses textual entailment. These EMSs and the mental health problem linked to each post such as depression, etc. served as input for the Chi-square test of independence and Odds Ratio (OR) analysis to reveal significant associations between EMSs and mental health problems. Finally, we again applied textual entailment to identify sentences that are pertinent to each EMS from forum posts. The features common among these sentences were extracted by suitably prompting GPT-4, a state-of-the-art Large Language Model (LLM). These features helped in decoding distinct characteristics underlying each EMSs across different dimensions such as "Schema Triggers" and "Coping Responses” typically used in ST case-conceptualization.

Results:

The Chi-square test of independence and OR (95%CI) across our dataset of ~30k posts uncovered significant associations (P < .05) between EMSs and mental health problems that are also substantiated by existing subject-based studies. For Anxiety, Depression, Eating disorder (ED), Personality disorder (PD), Post-traumatic stress disorder (PTSD), and Substance use disorder (SUD), the most strongly associated EMSs were respectively "Vulnerability to Harm or Illness" (OR=5.64), "Social Isolation" (OR=3.18), "Emotional Inhibition" (OR=1.89), "Subjugation" (OR=4.22), "Mistrust" (OR=5.04), and "Failure to Achieve" (OR=1.83). Moreover, our analysis indicated that Depression, PD, and PTSD, with 12, 9, and 7 EMSs associations respectively, are multifaceted problems. Regarding the features of EMSs, our approach with GPT-4 identified an average of ~13 negative features (e.g. “Responding by self-harm or destructive behavior”) per EMS and ~2 features per dimension. These features represent their EMS, aligning with previous studies.

Conclusions:

The study uncovered multiple associations between EMSs and mental health problems and highlighted the multifaceted nature of some problems in relation to EMSs. Additionally, the features identified from our approach of group-level case-conceptualization represent their EMSs and align with observations from subject-based studies. We are hopeful that our findings will aid future adaptation of ST interventions into online mental health forum communities, ultimately benefiting their users.