Patients generate daily a broad range of data, e.g., biosensor data from smartphones and wearables, smartphone usage data, or self-ratings. This symposium will focus on using these data in individuals with affective disorders and address the questions: how can such data be used by patients and at-risk groups to improve their self-management and by health professionals as a basis for treatment decisions?
The BipoLife A3 project aims to recognize early warning symptoms for affective episodes to facilitate initiation of targeted therapeutic steps. For this purpose, activity data is continuously recorded using mobile sensing via smartphone and compared with an individual threshold value (currently N=132 patients with Bipolar Disorder included). Using data (365 days, N=30) from the BipoSense study, we will report analyses on digital phenotypes prospectively predicting upcoming episodes of Bipolar Disorders. The ultimate goal is to prevent new illness episodes by real-time analyses and classification of prodromal symptoms. The WARN-D project (360 measurement points, N~1000) uses both smartwatch data and ecological momentary assessment data collected via smartphones to build a personalized early warning system for depression. We report on the utility and promises as well as challenges and pitfalls of smartwatch data in depression research. The STEADY project used a smartphone and wearable sensor-based system for the intensive, long-term monitoring (approx. 1 year, N=23) of patients with Major Depressive Disorder. Objective data for sleep and voice parameters were associated with self-reported core depressive symptoms that were assessed daily via Ambulatory Assessment.
Affective disorders are serious, chronically relapsing conditions with an increased risk of suicide. As up to now no clinically relevant tools for stratifying subgroups or predicting outcomes in affective disorders have been established, digital phenotyping may fill a gap for patients and individuals at-risk.
08:30 Uhr
Mobile sensing for the detection of early warning signs in bipolar disorder
S. Matura (Frankfurt am Main, DE)
Details anzeigen
Autor:innen:
S. Matura (Frankfurt am Main, DE)
E. Mühlbauer (DE)
M. Bauer (DE)
U. Ebner-Priemer (DE)
P. Ritter (DE)
H. Hill (DE)
F. Beier (DE)
A. Reif (Frankfurt am Main, DE)
E. Severus (DE)
Bipolar disorder is a severe, chronic relapsing disorder with a greatly increased risk of suicide. Timely identification of characteristic early warning signs before the onset of an episode makes it possible to avert it, or to mitigate its course. This is the scope of the BipoLife A3 project.
The BipoLife A3 project is a randomized, multi-center, observer-blind, active-control, parallel group trial within a nationwide research project (with 9 university centers from all over Germany, project coordinator: TU Dresden). The aim of the project is to examine the effectiveness of a smartphone based automated feedback about ambulatory assessed early warning signs in prolonging states of euthymia and therefore preventing hospitalization. For this purpose, activity data (e.g., number of steps per day, frequency and duration of conversations, frequency of switching on the smartphone, etc.) are continuously recorded via smartphone using mobile sensing and compared with an individual threshold value. If this threshold is undershot or overshot, an automated notification is sent to the therapist. In the control arm, the participants' activity data is also recorded via mobile sensing. However, there is no threshold-based intervention here. Additionally to mobile sensing, the psychopathologic state of all participants is assessed every four weeks over 18 months during the intervention phase. The primary outcome measure is the time from randomization to the occurrence of an affective episode. It will be investigated to what extent the automated detection of early warning signs and a threshold-dependent intervention can extend the time interval until the occurrence of a manifest affective episode.
Currently the data is analysed. A total of 132 patients with bipolar disorder were randomized. 104 patients have successfully completed the study. During the presentation, the study design with focus on mobile sensing will be presented.
08:52 Uhr
Prospectively predicting upcoming episodes in bipolar disorders using digital phenotypes
U. Ebner-Priemer (Karlsruhe, DE)
Details anzeigen
Autor:in:
U. Ebner-Priemer (Karlsruhe, DE)
Digital phenotyping promises to unobtrusively obtain a continuous and objective stream of symptomatology from patients’ daily lives. The prime example are bipolar disorders, as smartphone parameters directly reflect bipolar symptomatology. The ultimate goal is to prevent new illness episodes by real-time analyses and classification of prodromal symptoms. However, such an early warning system would require that digital phenotype-based prodromal symptoms can be detected significantly before a full-blown episode. But, how long is our window of opportunity? Days or just a few hours?
We conducted the BipoSense study, i.e. frequent (biweekly) dimensional and categorical expert ratings and daily self-ratings over a 12 months period in 29 patients with bipolar disorder. Digital phenotypes were monitored continuously. Combining gold-standard categorical expert ratings with dimensional self and expert ratings using structural equation modelling, we obtained two latent outcomes (mania and depression) with statistically meaningful factor loadings that dynamically varied over 299 days.
Latent digital phenotypes of sleep and activity were associated with same-day latent manic psychopathology, suggesting that psychopathological alterations in bipolar disorders relate to domains (latent variables of sleep and activity) and not only to specific behaviours (such as the number of declined incoming calls). In addition, further analyses showed that digital phenotypes like calls and steps show significant elevations several days before new upcoming episodes.
In summary, digital phenotyping hold its promises for an early-warning tool, but randomized trials are warranted.
09:14 Uhr
Using smartwatch and smartphone data to build an early warning system for depression
C. Rieble (Leiden, NL)
Details anzeigen
Autor:in:
C. Rieble (Leiden, NL)
Depression is common and debilitating, but treatment effectiveness remains disappointing. Experts agree that focusing on prevention is crucial. However, it is not yet possible predict who is at risk of depression onset in the near future – but this information is needed to know who to intervene on and when. By developing the personalized early warning system WARN-D, we aim to tackle this barrier to implementing successful, tailored prevention programs. The WARN-D study follows 2,000 students in the Netherlands over two years. For a three-month period, we collect both self-report (ecological momentary assessment) and activity tracking data (smartwatches). To reliably predict depression onset, we draw from complexity science and the network approach to psychopathology. We conceptualize depression as a complex, dynamical, biopsychosocial within-person system of problems influencing each other over time. That means that depression symptoms are not seen as a mere consequence of depression as an underlying cause – instead, they are seen as problems that cause and uphold each other. When people get stuck in vicious cycles, this can be seen as a transition from a healthy to a clinical state (i.e., depression). Crucially, the relations between the symptoms (i.e., the network of symptoms), may hold important information for upcoming transitions into depression. Across different disciplines studying complex systems, a variety of early warning signals (EWS), like critical slowing down have been proposed. These EWS can indicate that a system will transition from one stable state to another (e.g. from healthy to depressed). In our talk, we will discuss the promises and challenges of using easily collectable smartwatch data to enhance subjective self-reports in networks of depressive symptoms. We will then review whether applying the EWS known from other disciplines to these mixed-data symptom networks could help reliably forecast transitions into depression.
09:36 Uhr
Ideographic time-series analyses of objective sleep and voice data in Major Depressive Disorder
H. Reich de Paredes (Frankfurt am Main, DE)
Details anzeigen
Autor:innen:
H. Reich de Paredes (Frankfurt am Main, DE)
C. Sander ( Leipzig, DE)
B. Siepe (Frankfurt, DE)
H. Oppenheimer (Frankfurt, DE)
E. Wetzel (DE)
S. Ludwig (Leipzig, DE)
A. Kliem (Berlin, DE)
U. Hegerl (Frankfurt, DE)
Background: Patients generate daily a broad range of data, e.g., biosensor data from smartphones and wearables, smartphone usage data, self-ratings, environmental, and GPS data. The current contribution is going to explore the use of these data for the monitoring and digital phenotyping of patients with depressive disorders (DD).
Methods: A sensor-based system (STEADY), consisting of a smartphone, a smartphone-application, two wristbands, and a bed sensor, was developed to assess daily self-reported symptoms and behaviour (Ambulatory Assessment, AA) and continuous objective behavioural and physiological parameters (e.g., heart rate, sleep, speech). An observational n-of-1 trial was conducted with N=23 participants with DD, providing data for 357±160 days (AA, 2017-2019). Participants answered self-report questionnaires in regular intervals during the study period, received regular calls from the study centre, and a financial compensation.
Results: Idiographic symptom networks based on AA data could be estimated and visualized by time-varying vector autoregressive models. First results show that sensor-based sleep data explained 8.20% (total sleep time) and 9.77% (time in bed) of forecast error variance in self-rated depression severity (PHQ-2, AA) over a ten day period. In 5/ 5 participants with continuously recorded audio files (144±67 days), meaningful time-lagged cross-correlations were observed between voice parameters and PHQ-2 ratings (time lag: +/- 0-3 days).
Conclusions: Intensive longitudinal assessment of self-report and sensor-based data over time spans of one year and more was viable in patients with DD. Analyses showed heterogenous symptom patterns over time for each participant. This encourages the idea of using individual smartphone and wearable data for the personalized monitoring of DD. Possible applications, e.g. for the (self-)management of DD or new therapeutic interventions (just-in-time adaptive interventions, JITAIs) will be discussed.