JMIR Publications

ABSTRACT

Background:

Anxiety and depression represent prevalent yet frequently undetected mental health concerns within the elderly population. The challenge of identifying these conditions underscores the necessity for AI-driven, remotely available tools capable of screening and monitoring mental health symptoms. A critical criterion for such tools is their cultural adaptability to ensure effectiveness across diverse populations.

Objective:

The current study aims to illustrate the preliminary transferability of two established AI models designed for detecting depression and anxiety. The models were initially trained on data from a non-verbal working memory game (1- and 2-back) in the thymia dataset, encompassing over 6,000 participants from the United Kingdom, United States, Mexico, Spain, and Indonesia. We seek to validate the models’ performance by applying it to a new dataset comprising elderly Brazilian adults, thereby exploring its generalisability across different demographics and cultures.

Methods:

The 69 Brazilian participants aged 51-92 years old were recruited with the help of Laços Saúde, a company specialising in nurse-led, holistic home care. They received a link to the thymia dashboard every Monday and Thursday for 6 months. The dashboard had a set of activities assigned to them that would take 10-15 minutes to complete, among which a game with two levels of the n-back tasks. Two Random Forest models trained on data to classify depression and anxiety based on thresholds defined by PHQ-8 ≥ 10 and GAD-7 ≥ 10, respectively, were subsequently tested on the Laços Saúde patient cohort.

Results:

The depression classification model exhibited robust performance, achieving an AUC of 0.78, specificity of 0.69, and sensitivity of 0.72. The anxiety classification model showed an initial AUC of 0.63, with a specificity of 0.58 and sensitivity of 0.64. This performance surpasses a benchmark model using only age and gender, which had AUCs of 0.47 for PHQ-8 and 0.53 for GAD-7. Re-computing the AUC scores on a cross-sectional subset of the data (the first n-back game session), we found 0.79 for PHQ-8 and 0.76 for GAD-7.

Conclusions:

This study successfully demonstrates the preliminary transferability of two AI models trained on a non-verbal working memory task, one for depression and the other for anxiety classification, to a novel demographic of elderly Brazilian adults.