Identifying ICU survivors and relatives with post-traumatic stress disorder using text mining: An explorative study

Area of Science:

• Medical Informatics
• Psychiatry
• Critical Care Medicine

Background:

• Post-traumatic stress disorder (PTSD) affects a significant portion of intensive care unit (ICU) patients post-discharge.
• Early detection of PTSD risk is crucial for timely intervention and improved patient outcomes.

Purpose of the Study:

• To evaluate the accuracy of text mining on self-narratives for identifying ICU patients and their relatives at risk of PTSD.
• To explore the potential of natural language processing in PTSD screening.

Main Methods:

• A prospective cohort study involving discharged ICU patients and surviving relatives.
• Utilized an online screening tool incorporating self-administered questionnaires and self-narratives.
• Text mining analysis was performed on narratives collected 3 to 6 months after ICU discharge, with the Trauma Screening Questionnaire (TSQ) as a reference standard.

Main Results:

• 15% of participants showed indications of PTSD based on TSQ scores.
• Text mining of self-narratives demonstrated reasonable predictive performance for PTSD risk (AUROC of 0.67) compared to TSQ.
• Key predictive terms identified included 'happen,' 'again,' and 'done,' though context is needed for interpretation.

Conclusions:

• Text mining of self-narratives is a viable method for predicting PTSD risk in ICU survivors and relatives.
• This approach aids in identifying individuals requiring mental health support after ICU discharge.
• The prevalence of PTSD indications underscores the importance of encouraging help-seeking behavior.