A pilot predictive model based on COVID-19 data to assess suicidal ideation indirectly

Area of Science:

• Psychiatry
• Psychology
• Data Science

Background:

• The COVID-19 pandemic negatively impacted global mental health.
• Studies indicate rising rates of psychological distress and suicidal ideation (SI).
• There is a need for discreet methods to identify individuals at risk of SI.

Purpose of the Study:

• To estimate the presence of SI using indirect indicators.
• To identify demographic and psychological factors associated with SI.
• To evaluate machine learning algorithms for SI prediction.

Main Methods:

• Collected data from 1790 respondents in Slovenia via an online survey (July 2020-Jan 2021).
• Utilized machine learning algorithms (logistic regression, random forest, XGBoost, SVM) to predict SI.
• Analyzed associations between coping strategies (Brief-COPE), life satisfaction, demographics, and SI.

Main Results:

• Machine learning models (logistic regression, random forest, XGBoost) achieved an AUC of 0.83.
• Key indicators for SI included Self-Blame, increased Substance Use, low Positive Reframing, Behavioral Disengagement, relationship dissatisfaction, and younger age.
• SI was reported by 9.7% of respondents.

Conclusions:

• SI can be estimated with reasonable accuracy using indirect indicators.
• The identified factors show potential for developing a discreet screening tool for suicidality.
• Further clinical examination is recommended for individuals identified as at risk.