Digital phenotyping of depression during pregnancy using self-report data.

Area of Science:

• Perinatal mental health
• Digital health interventions
• Computational linguistics

Background:

• Depression is a frequent complication of pregnancy, often undetected and undertreated.
• Mobile health (mHealth) tools offer potential for early identification of depression symptoms in pregnant individuals.

Purpose of the Study:

• To investigate the utility of patient-reported data from a prenatal mobile health app for predicting depression risk during pregnancy.
• To assess the individual and combined predictive value of various data types, including mood, symptoms, and written language.

Main Methods:

• An observational cohort study involving 2062 pregnancies.
• Utilized a prenatal support app to collect self-reported data on patient history, mood, pregnancy symptoms, and written language.
• Employed selective inference and lasso modeling to analyze data and predict depression risk over 30- and 60-day periods.

Main Results:

• Depression prediction models demonstrated varying effectiveness (AUC 0.64-0.83).
• Models incorporating personal history, daily mood, and acute pregnancy symptoms were most predictive.
• Daily mood emerged as the strongest predictor of subsequent depression symptoms; natural language inputs enhanced accuracy and contextual understanding.

Conclusions:

• Simple, patient-reported data, including natural language, via digital tools can aid in earlier depression symptom identification.
• Provides a more nuanced understanding of the lived context of maternal depression.
• Findings are limited to digitally literate, self-motivated pregnant individuals using the app.