This study investigates the longitudinal effects of brain stimulation treatments on decision-making and reward learning. Here, we use daily assessments of reinforcement learning behavior in patients with depression who receive neuromodulation treatment to study potential associations between changes in decision-making and reward learning and changes in clinical symptoms.
Brain stimulation techniques are increasingly applied to treat mental disorders, especially when psychotherapy and psychopharmacotherapy fail (Conroy \& Holtzheimer, 2021). The efficacy of treating treatment-resistant depression has been demonstrated using electroconvulsive therapy (ECT; Espinoza \& Kellern, 2022; Subramanian et al., 2022), transcranial magnetic stimulation (TMS, Trapp et al., 2025), and vagus nerve stimulation (VNS, Kamel et al., 2022). With regard to transcutaneous auricular vagus nerve stimulation (taVNS), a motivation-enhancing effect has been demonstrated in individuals with depression (Ferstl et al., 2024). It is unclear to what extent these stimulation techniques also affect cognitive processes, such as learning and decision-making behavior, and whether such changes can explain treatment success. Temporary cognitive side effects, such as memory impairments, are known to occur with ECT treatments (Landry et al., 2021). Initial evidence regarding learning and decision-making behavior is available for taVNS, in which the vagal auricular branch is noninvasively stimulated at the ear. Acute taVNS stimulation at the left ear showed a reduced learning rate particularly during punishment (Kühnel et al., 2020), as well as increased reward sensitivity (Weber et al., 2021). Additionally, TMS over the left dorsolateral prefrontal cortex was shown to alter the ratio of reward-to-punishment learning rates (Biernacki et al., 2023). However, these initial findings on learning and decision-making behavior and the underlying computational parameters relate to acute stimulation. Long-term effects of brain stimulation techniques on theses processes, as used in routine clinical practice (e.g., eight TMS sessions or daily stimulation of the vagus nerve) remain unclear to date. Likewise, the influence of fluctuations in mental state, such as mood, on learning and decision-making behavior over the course of a treatment, as well as potential changes in this influence after successful treatment, remain mostly unexplored. However, a recent study found, that fluctuations in metabolic state affected reward sensitivity and punishment learning rates in obesity (Kühnel et al., 2025). In the context of effort-based decision-making, it has been shown that fluctuations in motivation influence reward sensitivity (Hewitt et al., 2025). It is also not yet possible to say to what extent stimulation-induced changes in clinical symptoms are reflected in possible changes in decision-making and reward learning. However, increasing evidence of altered computational mechanisms in patients points towards this potential connection. For example, higher punishment learning rates were observed in individuals with depression (Pike \& Robinson, 2022) and anhedonia was linked to reduced reward sensitivity (Huys \& Browning, 2025). Also, greater temporal discounting of future rewards was observed in individuals with depression compared to those without a diagnosis (Amlung et al., 2019). An open question remains, that is whether, as symptoms improve, changes in cognitive processes also diminish. Based on the above, the following objectives have been established: First, the acute and mid-term effects of the mentioned neurostimulation therapies on computational processes of decision-making and reward learning will be investigated over the course of a clinical application. We expect, that brain stimulation lowers punishment learning rates (Hypothesis 1). Subsequently, the effects of the various stimulation methods on punishment learning rates will be compared (acutely after stimulation as well as over the course of the treatment). In addition, the influence of state fluctuations, such as mood, on computational parameters of decision-making and reward learning, such as learning rates, will be modeled. We expect, that fluctuations in mood, motivation, and metabolic state are associated with changes in punishment learning rates (Hypothesis 2). Next, we investigate the effect of brain stimulation on working memory. We expect, that ECT impairs working memory compared to the other treatments (Hypothesis 3). Finally, we expect that changes in punishment learning rates predict changes in symptoms of depression, general well-being, and somatic symptoms (Hypothesis 4), as well as cognitive side effects on working memory (Hypothesis 5). Here, we use daily assessments of reinforcement learning with a gamified online task (Neuser et al., 2023) over the course of 8 weeks to track longitudinal effects of brain stimulation on decision-making and reward learning in patients with depression who receive neuromodulation treatment (N = 100). Daily assessments are accompanied by an ecological momentary assessment (EMA) of mood and metabolic states. At least five runs of the reinforcement learning task should be played before the first stimulation as a baseline. Additionally, working memory (backward digit span task), depressive symptoms (BDI-II), well-being (WHO-5), and somatic symptoms (PHQ-15) are measured online at the start of the study before the first treatment, after 4 weeks, and after 8 weeks at the end of the study. Patients will be recruited directly from the clinic after brain stimulation treatment is indicated for them..
Study Type
OBSERVATIONAL
Enrollment
100
Section of Medical Psychology, Department of Psychiatry & Psychotherapy, Faculty of Medicine, University of Bonn
Bonn, Germany
Punishment learning rates
The primary outcome are punishment learning rate estimates from a computational reinforcement learning model. Learning rates will be compared within and between stimulation conditions. Reinforcement learning is repeatedly measured with a bandit task with fluctuating reward probabilities (reward learning task). Reward learning behavior will be collected online over up to 60 runs, each including 150 trials.
Time frame: Assessed online up to 5 times before the first stimulation treatment and daily throughout the study (8 weeks).
Stimulation-induced mid-term changes in the backward digit span
The primary outcome is the change in the backward digit span. Changes over time will be compared within conditions and between ECT and the other conditions. The backward digit span is the maximum sequence length that could be repeated correctly backwards in the digit span task (at least one out of two correct). The task runs two sequences per sequence length, starting at two digits and increasing by one digit up to eight digits. The task is terminated if both sequences of a sequence length could not be repeated correctly backwards. Scores range from 0 to 16 and are the number of correctly repeated sequences.
Time frame: Assessed before the first stimulation treatment and after 4 and 8 weeks (10 minutes).
Correct choices in the reward learning task
The outcome describes the percentage of correct choices (wins) collected during the reward learning task. Choices are collected in each of the 150 trials per run.
Time frame: Assessed online at least 5 times before the first stimulation treatment and daily throughout the study (8 weeks).
Reward sensitivity
The primary outcome is reward sensitivity estimates from a computational reinforcement learning model reflected by the inverse temperature parameters. Changes over time will be compared within and between stimulation conditions. Reinforcement learning is repeatedly measured with an bandit task with fluctuating reward probabilities (reward learning task). Reward learning behavior will be collected online over up to 60 runs, each including 150 trials.
This platform is for informational purposes only and does not constitute medical advice. Always consult a qualified healthcare professional.
Time frame: Assessed online at least 5 times before the first stimulation treatment and daily throughout the study (8 weeks).
Weighting of learned values and rewards at stake (lambda)
The primary outcome is the weighting of win probability to reward magnitude weighting (lambda) from a computational reinforcement learning model. Changes over time will be compared within and between stimulation conditions. Reinforcement learning is repeatedly measured with an bandit task with fluctuating reward probabilities (reward learning task). Reward learning behavior will be collected online over up to 60 runs, each including 150 trials.
Time frame: Assessed online at least 5 times before the first stimulation treatment and daily throughout the study (8 weeks).
Stimulation-induced mid-term changes in reward learning rates
The primary outcome is the change over time in reward learning rate estimates from a computational reinforcement learning model. Changes over time will be compared within and between stimulation conditions. Reinforcement learning is repeatedly measured with an bandit task with fluctuating reward probabilities (reward learning task). Reward learning behavior will be collected online over up to 60 runs, each including 150 trials.
Time frame: Assessed online at least 5 times before the first stimulation treatment and daily throughout the study (8 weeks).
Association between mood state and punishment learning rates
The primary outcome is the association (regression coefficient) quantifying the relationship between mood state and punishment learning rates. Punishment learning rates are measured in the reward learning task. Mood state is calculated from two 0-100 visual analog scales (state happiness minus state sadness), which are asked before every run of the reward learning task.
Time frame: Assessed online at least 5 times before the first stimulation treatment and daily throughout the study (8 weeks).
Association between metabolic state and punishment learning rates
The primary outcome is the association (regression coefficient) quantifying the relationship between metabolic state and punishment learning rates. Punishment learning rates are measured in the reward learning task and metabolic state is asked before every run of the reward learning task on a 0-100 visual analog scale (sated - hungry).
Time frame: Assessed online at least 5 times before the first stimulation treatment and daily throughout the study (8 weeks).
Association between changes in punishment learning rates and changes in depressive symptoms
The primary outcome is the association (regression coefficient) quantifying the relationship between changes in punishment learning rates and changes in depressive symptoms. The regression model predicts changes in depressive symptoms from changes in punishment learning rates. Depressive symptoms are measured with the BDI-II (total scores, ranging from 0 to 63). Punishment learning rates are measured in the reward learning task.
Time frame: Depressive symptoms are assessed before the first stimulation treatment and after 4 and 8 weeks. Reward learning is assessed online at least 5 times before the first stimulation treatment and daily throughout the study (8 weeks).
Association between motivational state and punishment learning rates
The primary outcome is the association (regression coefficient) quantifying the relationship between motivational state and punishment learning rates. Punishment learning rates are measured in the reward learning task and motivational state is asked before every run of the reward learning task on a 0-100 visual analog scale (not motivated - motivated).
Time frame: Assessed online at least 5 times before the first stimulation treatment and daily throughout the study (8 weeks).
Association between changes in punishment learning rates and changes in mental well-being
The primary outcome is the association (regression coefficient) quantifying the relationship between changes in punishment learning rates and changes in mental well-being. The regression model predicts changes in mental well-being from changes in punishment learning rates. Mental well being is measured with the WHO-5 (total scores, range from 0 to 30). Punishment learning rates are measured in the reward learning task.
Time frame: Mental well being is assessed before the first stimulation treatment and after 4 and 8 weeks. Reward learning is assessed online at least 5 times before the first stimulation treatment and daily throughout the study (8 weeks).
Association between changes in punishment learning rates and changes in somatic symptoms
The primary outcome is the association (regression coefficient) quantifying the relationship between changes in punishment learning rates and changes in somatic symptoms. The regression model predicts changes in somatic symptoms from changes in punishment learning rates. Somatic symptoms are measured with the PHQ-15 (total scores, range from 0 to 30). Punishment learning rates are measured in the reward learning task.
Time frame: Somatic symptoms are assessed before the first stimulation treatment and after 4 and 8 weeks. Reward learning is assessed online at least 5 times before the first stimulation treatment and daily throughout the study (8 weeks).
Association between changes in the backward digit span and changes in punishment learning rates
The primary outcome is the association (regression coefficient) quantifying the relationship between changes in the backward digit span and changes in punishment learning rates. The backward digit span is measured in the backward digit span task. Punishment learning rates are measured in the reward learning task.
Time frame: Assessed online at least 5 times before the first stimulation treatment and daily throughout the study (8 weeks).
Response times in the reward learning task
The outcome describes response times collected during the reward learning task. Response times (milliseconds) are collected in each of the 150 trials per run.
Time frame: Assessed online at least 5 times before the first stimulation treatment and daily throughout the study (8 weeks).
Backward digit span
The primary outcome is the maximum sequence length that could be repeated correctly backwards (at least one out of two correct). The task runs two sequences per sequence length, starting at two digits and increasing by one digit up to eight digits. The task is terminated if both sequences of a sequence length could not be repeated correctly backwards. Scores range from 2 to 8.
Time frame: Assessed before the first stimulation treatment and after 4 and 8 weeks (10 minutes).
BDI-II (Beck Depression Inventory-II)
Questionnaire assessing depressive symptoms. Will be associated with behavioral outcomes. Total scores range from 0 to 63, with higher scores indicating greater depressive symptom severity.
Time frame: Assessed before the first stimulation treatment and after 4 and 8 weeks (10 minutes).
WHO-5 (World Health Organization-Five Well-Being Index)
Questionnaire assessing mental well being. Will be associated with behavioral outcomes. Total scores range from 0 to 30, with higher scores indicating better mental well being.
Time frame: Assessed before the first stimulation treatment and after 4 and 8 weeks (3 minutes).
PHQ-15 (Patient Health Questionnaire-15)
Questionnaire assessing the severity of somatic symptoms. Will be associated with behavioral outcomes. Total scores range from 0 to 30, with higher scores indicating higher severity of somatic symptoms.
Time frame: Assessed before the first stimulation treatment and after 4 and 8 weeks (5 minutes).