Early warning system for depression: modeling 3 months of time-series smartphone data in 2000 participants

Early warning system for depression: modeling 3 months of time-series smartphone data in 2000 participants – MBCS Internships

The WARN-D study is aimed to building an early warning system for depression. To do so, we follow 2000 students over 2 years, including 3 months of intensive monitoring with smartwatch and smartphone data collection (website: warn-d.com, see also our recent protocol paper explaining the study in more detail: https://cpe.psychopen.eu/index.php/cpe/article/view/10075).

We have now collected most of the data and before we build prediction models, we are trying to understand our data a bit better. One of the first steps here is trying to see to what degree our ecological momentary assessment data (survey questions on smartphones, 4 times per day for 3 months) move together across time. For example, are being indicating that are in a happy mood also indicating that they are content? How does this relate to negative affect, and then all the other interesting data we have (what environment they are in, if they have in-person social interactions, stressors, and so on).

In this internship, the goal is to explore what statistical models are best suited for dynamic grouping of variables. Candidate models include time-varying principal components analysis, dynamic p-factor techniques, and perhaps dynamic time-warp models (a lesser known model I recently summarized here:https://eiko-fried.com/modeling-idiographic-and-nomothetic-dynamics-of-255-depressed-inpatients/).

We mostly work in R, have a lovely team at Leiden University, and are flexible with working hours, starting date, etc. Reach out to us if you are interested and/or have questions 🙂

Eiko Fried, Associate Professor, Leiden University

To apply for this job email your details to [staat bij de werkgever]