1 - (SYM 111) Detecting Peak Daily Shame in Body Dysmorphic Disorder Using Passive Smartphone Data

Introduction: Body dysmorphic disorder (BDD) is a chronic, debilitating, and common psychiatric illness associated with high rates of suicidal thoughts and completed suicide (Phillips et al., 2005; Phillips, 2007; Phillips & Menard, 2006). BDD involves persistent concern about imagined or slight flaw(s) in one’s appearance and unhelpful compulsive behaviors performed to fix or hide the perceived appearance flaw(s). Negative emotions – particularly shame – are elevated in people with BDD (Weingarden et al., 2016; Weingarden et al., 2018) and correlate with risk for suicide in BDD (Weingarden et al., 2017; Weingarden et al., 2018). Therefore, it is critical that we have reliable and valid tools to assess negative emotion states in this high-risk condition. Retrospective self-report measures – the field’s go-to assessment approach – are subject to recall biases, average one’s experiences over broad time frames, and place burden on patients. Alternatively, sensor-based digital phenotyping to detect negative emotion states has the potential to yield accurate, real-time, and low-burden measurement of emotions.

Method: This study aimed to use passively collected smartphone sensor data to detect same-day peak shame states in 87 adults with moderate to severe BDD. Models of shame states were built using daily summary statistics (e.g., time spent at home, distance traveled from home, steps) derived from passively collected raw GPS and accelerometer smartphone data over 3 months; outcome variables included peak daily ecological momentary assessment (EMA) ratings of shame across 28 days. We used both a linear statistical model (cumulative link mixed model [CLMM]) and a non-linear machine learning model (random forest) to build digital phenotypes of daily peak shame, and we evaluated model success using prediction accuracy and feature importance (Ren et al., 2023; Jacobson et al., 2020). 


Results: Data collection is completed and analyses will be completed in spring 2024. Initial results from random forest show that daily time at home, steps, distance traveled, and entropy are important features for predicting peak daily shame.

Conclusion: Risk for suicide is notably high in people with BDD, and shame is a well-documented risk correlate in this population. Developing unobtrusive real-time methods for detecting shame by using smartphone sensor data can enable just-in-time intervention opportunities, as a next step toward reducing risk for suicide in BDD.