Using natural language from a smartphone pregnancy app to identify maternal depression

Area of Science:

• Reproductive Psychiatry
• Computational Linguistics
• Digital Health

Background:

• Maternal depression during pregnancy is common but often overlooked.
• Language use may reflect psychological well-being.
• Prenatal smartphone applications offer novel data streams.

Approach:

• A longitudinal observational study analyzed written language from 1,274 pregnancies via a smartphone app.
• Natural language processing (NLP) features from journaling entries were used to model depression symptoms.
• Models were evaluated for predicting depression within a 30-day window.

Key Points:

• Language features alone predicted incident depression symptoms with an Area Under the Receiver Operating Characteristic Curve (AUROC) of 0.72.
• Analysis revealed topics salient in the writing of individuals experiencing depression.
• Integrating language features with self-reported mood enhanced prediction accuracy (AUROC = 0.84).

Conclusions:

• Prenatal apps can provide insights into experiences contributing to depression.
• Sparse language data and patient mood reports can aid nuanced depression symptom identification.
• Digital tools offer a promising avenue for early detection of maternal depression.