Precision psychiatry through digital phenotyping - Lessons learned from three longitudinal studies in major depressive disorder

Major depressive disorder (MDD) is diagnosed on symptoms alone. Clinical endpoints rely on infrequent, subjective ratings that miss much of this variation. Objective digital markers, collected remotely and repeatedly could reduce heterogeneity and subjectivity in clinical trials.

In this lecture, Dr. Srinivasan Vairavan will share lessons from five longitudinal MDD studies in Europe and North America, with more than 2,000 participants combined: RADAR-MDD, OBSERVEMDD, CAN-BIND, CANDID and PReDicT. Together they cover recurrent, remitted, and newly treated MDD, as well as healthy volunteers. Three deep dives show how digital biomarkers map onto MDD symptom domains:

• Emotional processing. On the Facial Expression Recognition Task (FERT), more severe depressive, anxiety and functional symptoms were linked to greater negative emotional bias. Most FERT measures showed moderate 1-week test-retest reliability.
• Speech and language. In RADAR-MDD, higher symptom severity was associated with:slower speech and articulation rate
• changes in speech intensity
• fewer words, fewer positive words and fewer first-person plural pronouns

CAN-BIND replicated several of these associations: lower pitch, higher jitter, slower syllable rate and fewer total words. Many of these features showed moderate-to-good 2-week reliability.

Sociability and mobility. Lower mobility measured by phone geolocation was associated with more severe depression, and the link was stronger within individuals than between them. Homestay and location entropy predicted later changes in symptoms.

Dr. Vairavan will conclude that these findings support a multimodal digital health ecosystem for precision psychiatry. Linking “the digital patient” to “the biological patient” could support patient selection, early detection of treatment response and endpoints closer to the underlying mechanisms in drug development.