The emerging possibilities of machine learning, artificial intelligence, and mobile technology may substantially advance how we can measure and identify important mechanisms, predict mental health outcomes and personalize interventions in psychiatry. Machine learning methods can be trained to learn dynamic models that may uncover and predict risk trajectories that can be targeted in the early stages of mental disorders. The rapid rise and now widespread distribution of wearable and handheld devices has opened a broad range of possibilities for collecting vast amounts of temporally highly resolved and ecologically valid data. This, in turn, provides the basis for mHealth interventions such as ecological momentary interventions that allow for experience-, time- and context-specific targeting of underlying momentary mechanisms and for tailoring treatment to individual needs. As such, this reflects one of the best opportunities for personalized medicine in the digital space. However, these rapid developments also pose many challenges such as questions around data ownership and governance, engagement with digital interventions, digital exclusion, staff concerns, and ethical issues.
This symposium brings together researchers from various strands of digital mental health to showcase some of its enormous opportunities but also critically reflect on and discuss these rapidly evolving developments. Tim Hahn will discuss promises, problems and perspectives of automatized machine learning analyses in mental health research. Georgia Koppe will follow on from this to showcase the potential of deep learning methods in mobile sampling and intervention. Matthias Schwannauer will critically discuss the digital revolution and its impact on individuals in mental healthcare. Finally, Mary Rose Postma will report data on the role of self-esteem as a putative momentary mechanism and present a novel ecological momentary intervention for targeting this mechanism in early psychosis.
Deep learning methods in mobile sampling and intervention
Georgia Koppe, Mannheim (Germany)
Details anzeigen
Autor:in:
Georgia Koppe, Mannheim (Germany)
Portable and wearable devices such as smartphones and sensors bear unprecedented opportunities to improve health care for individuals suffering from psychiatric disorders. The devices allow us to collect (and potentially feed back) vast amounts of data predictive of mental health and well-being in real-time and real-life, while being widely accessible and affordable. The hereby collected data could in principle serve to predict upcoming symptoms, identify predecessors thereof (and associated tipping points), and ultimately provide the basis for personalized ecological momentary interventions and treatment. However, the opportunities for health care that emerge with such data also come with great yet unsolved challenges. For instance, the data we collect may be obtained from multiple different modalities (e.g. categorical ratings vs. continuous GPS-based signals), may be sampled at different rates, and may harbour pertinent information at different temporal scales (e.g., millisecond typing dynamics vs. daily fluctuations in sleep-wake rhythm). Predictive features most likely rely on a combination of all of these data streams while we often lack clear hypotheses on how to combine and model them. Here, I will talk about one approach to tackle these problems by employing statistical models based on recurrent neural networks (RNNs). RNNs are AI based tools particularly suited for the analysis of sequential time series data. I will demonstrate how these models may be harnessed in the context of mobile health data in order to predict upcoming mental states, to simulate the effect of various environmental inputs such as medication or social encounters, as well as to gain a deeper understanding on the dynamics and contingencies underlying behaviour.
Changing the time and form of intervention in psychosis: early intervention by targeting self-esteem in daily life
Mary Rose Postma, Maastricht (Netherlands)
Details anzeigen
Autor:in:
Mary Rose Postma, Maastricht (Netherlands)
Title: Along the continuum: fluctuations in momentary self-esteem and psychotic experiences.
Background: Self-esteem has been recognized as a putative mechanism in the development and maintenance of psychosis. Uncertainty still exists about how fluctuations in self-esteem relate to psychotic experiences. The present talk therefore will elaborate on the associations between (fluctuations in) momentary self-esteem and psychotic experiences in daily life.
Methods: Experience sampling data were collected for 46 individuals presenting with an at-risk mental state (ARMS), 51 individuals with first episode psychosis (FEP), and 53 controls, to investigate associations between (fluctuations in) self-esteem and psychotic experiences within and across FEP, ARMS, and controls, using linear mixed models.
Results: ARMS individuals and FEP individuals reported lower momentary self-esteem (both p < 0.001) and greater variability and instability in momentary self-esteem than controls. In all three groups we found that lower momentary self-esteem as well as greater variability in self-esteem were associated with a greater intensity of psychotic experiences (all p < 0.001). Instability in momentary self-esteem was associated with psychotic experiences only in controls.
Conclusion: The presented findings in this talk suggest momentary self-esteem to be fitted within a framework of phenomenological continuity where it can be targeted using ecological momentary interventions aimed at reducing the intensity of psychotic experiences and preventing illness progression at an early stage.
This talk will further introduce the SELFIE study that aims to investigate such an ecological momentary intervention for targeting self-esteem in youth with early childhood trauma.