Alcohol use disorder (AUD) is a major public health burden. While cross-sectional studies link AUD to reduced cognitive control and heightened risky decision-making, the temporal dynamics of these associations in everyday life remain unclear. Therefore, we deployed a battery of gamified smartphone tasks measuring cognitive control and decision-making in a one-year longitudinal ecological momentary assessment (EMA) study with N=603 participants diagnosed with mostly mild-to-moderate AUD, who completed tasks monthly and reported alcohol consumption and stress every two days.

In this presentation, we first show that within-person fluctuations in decision-making predict alcohol consumption in the following month: as predicted, participants consumed more alcohol following higher risk-taking in a mixed reward context, less alcohol following higher risk-taking in a loss context, and less alcohol following increased information sampling. We further show that both between-person differences and within-person changes in working memory capacity (WMC) buffer stress-associated drinking.

Next, diving deeper into the distribution of drinking data, we show that while decreased response inhibition is not associated with increased median monthly drinking, it is associated with increased binge drinking in the following month, suggesting that cognitive control may be specifically linked to the most harmful drinking patterns.

Finally, we report psychometrics of a newly developed mobile Simon task and give a first peek into findings from a study using the task to predict daily drinking.

Together, these findings demonstrate how smartphone tasks can refine our understanding of how different cognitive processes relate to distinct drinking outcomes in AUD and point toward real-time, mechanism-based interventions.

Low mood has been linked to negative cognitive biases, but how mood and cognition interact over time and in the real world is unclear. Here, we used ecological momentary assessment (EMA) and remote brain stimulation to capture and modulate real-time dynamics of cognition and mood in natural environments.

Across three EMA studies, participants (total N=1,121), tracked their mood 2–3 times per day for 6–8 weeks (median assessments=88). In a fourth study, participants additionally applied at-home transcranial direct current stimulation (tDCS) for two weeks to alter cortical excitability in prefrontal cortical regions implicated in cognition (N=54 tDCS, N=47 sham, median assessments=102). Across all studies, we passively assessed cognitive processing speed using digital questionnaire response time (DQRT). Data were analyzed using hierarchical linear models.

 Contemporaneously, DQRT became slower when negative mood increased (-0.136>partial correlation<0.129, p<.05). This pattern was consistent across 34/37 mood variables assessed. In lagged analyses, higher negative mood predicted slower DQRT at the next time point (-0.027>β<0.067, p<.05). Strongest predictors of future DQRT were feelings of worry and anxiety, with no evidence for reverse associations. There were no differences between tDCS and sham across the study duration for mood (η_p²=0.02, p=.45) or DQRT (η_p²=0.005, p=.79).

Using data on real-time dynamics of mood and cognition, we find that low mood predicts slower cognitive processing speed without evidence for the reverse direction. This finding may inform mechanistic accounts of cognitive deficits in mental health disorders. Contrary to our expectation, there was no evidence supporting stimulation-induced effects on mood or cognition.

Computational phenotyping of mental health involves identifying behavioural and computational markers of psychopathology that could aid timely and tailored prediction, prevention, and treatment. Such phenotyping is typically based on between-person differences in performance on cognitive tasks, but existing studies taking this approach have shown only limited ability of such between-person markers to predict mental health. In line with these findings, we present results from a recent preregistered study (N = 1,259) in which we used gamified, smartphone-based learning and decision-making assessments and machine learning to create a profile of behavioural, computational, and demographic features to predict depressive symptoms. Whilst yielding significant effects, these were primarily driven by demographics, with task-based predictors playing only a minor role. It remains possible, however, that within individuals, fluctuations in such task-based, cognitive markers predict fluctuations in mental health. Leveraging technological developments that facilitate remote and high-frequency assessments of cognitive markers and mental health in large samples, we also present a smartphone-based Ecological Momentary Assessment (EMA) study examining how within-person fluctuations in learning and decision-making relate to fluctuations in depressive symptoms, and how a short metacognitive intervention causally alters these fluctuations. Such within-person phenotyping approaches promise to provide more insight into the temporal dynamics of mental health conditions, tailor prediction and treatment to individual clinical trajectories, and gain insight into the causal direction of associations between markers and mental health. Ultimately, this could aid the development of digital early warning signals and just-in-time interventions based on fluctuations in cognitive markers.

Objective. We present a feasible routine outcome monitoring (ROM) system embedded in inpatient routine care that provides clinicians and clients with personalized questionnaires and data-informed feedback on the therapeutic process. Grounded in a complex systems perspective, the approach enables the identification of general change processes from individual psychotherapy trajectories.

Method. Data were collected from 404 clients in an inpatient treatment setting between 2013 and 2020. Clients completed the ISR and DASS-21 before and after treatment. In addition, they developed an individualized case conceptualization using idiographic system modelling (ISM; Schiepek, 2016), which informed the daily assessment of personalized items. These data were integrated into biweekly feedback sessions. ISM content was examined using thematic analysis. To analyse change dynamics, a principal component analysis was applied to each client’s multivariate time series, resulting in one person-specific time series per client. These time series were then classified into change profiles and related to outcome measures.

Results. Case conceptualizations were highly individualized, with 87% of concepts being unique. At the same time, thematic analysis identified seven overarching themes extending beyond a purely symptom-focused diagnostic perspective. Based on the individualized questionnaires, five general change profiles were identified, including sudden change (>60%), gradual change, and no change. The average time-series length was 36 days (SD = 15). Clients who showed change in their individualized process measures also showed better post-treatment outcomes.

Discussion. ROM should move beyond symptom tracking towards individualized, process-oriented feedback in order to be therapeutically meaningful and to better capture (non-linear) & clinically relevant change.