Optimizing Existing Mental Health Screening Methods in a Dementia Screening and Risk Factor App: Observational...

Area of Science:

• Digital health
• Machine learning in healthcare
• Mental health screening

Background:

• The Mindstep app aims to enhance dementia screening by evaluating cognition and risk factors.
• It incorporates clinical risk factors like prodromal symptoms and mental health disorders.
• Validated scales like the Patient Health Questionnaire (PHQ-9) and Generalized Anxiety Disorder Scale (GAD-7) are used for depression and anxiety screening, with shorter versions (PHQ-2/GAD-2) available.

Purpose of the Study:

• To create a brief, app-based screening method for affective disorders.
• To maintain the brevity of short questionnaires while improving precision.
• To leverage machine learning with existing app data for enhanced screening.

Main Methods:

• Developed single questions to capture symptoms from PHQ-9 and GAD-7.
• Combined these questions with PHQ-2/GAD-2 and anonymized risk factors from 2235 Mindstep users.
• Trained machine learning models using app data (age, joke response, functional impairment) to predict PHQ-9/GAD-7 outcomes.
• Utilized 10-fold cross-validation and a holdout testing set for model development and comparison.

Main Results:

• The model demonstrated superior performance in predicting PHQ-9 screening cutoffs compared to PHQ-2 (AUC difference 0.04, P=.02).
• Performance for GAD-7 prediction was equivalent to GAD-2 (AUC difference 0.00, P=.42).
• Regression models accurately predicted total scores for PHQ-9 (R²=0.655) and GAD-7 (R²=0.837).

Conclusions:

• An app-adapted PHQ-4 was created using brief summary questions and machine learning.
• This tool effectively utilizes additional app-collected data for screening.
• The developed app-based tool shows promising superior or equivalent performance to established screening methods for affective disorders.