Cognitive distortions are central to the maintenance of depression, as they bias information processing and negatively impact adaptive emotion regulation. As one manifestation of cognitive distortions, usage of absolutist words (e.g., always, never, must, completely) in written texts has been found to be indicative of underlying depression. Since absolutist word usage may allow important insights into maladaptive thinking patterns, it could be a relevant target for both monitoring and treating depressive symptoms. Therefore, we tested the relationships between absolutist word usage in spoken language with depression diagnosis, depressive symptom severity and current depressed mood. We recruited 144 age- and gender-matched participants with clinical depression (n = 48), subclinical depression (n = 48), and no history of depression (n = 48). By conducting smartphone-based ecological momentary assessments (EMA) three times daily for two weeks, participants provided ratings of depressed mood and speech samples, which included mood descriptions and mood-regulating statements. The Hamilton Rating Scale for Depression (HRSD) was administered at the end of the 2-week period to assess depressive symptom severity retrospectively. Group differences were calculated with ANOVAs, associations between depressed mood and absolutist word usage were evaluated with multilevel models, and the relationship between depressive symptom severity and absolutist word usage was calculated with regression models. Results showed more frequent absolutist word usage over the EMA phase in individuals with clinical depression compared to individuals with no history of depression. Furthermore, absolutist word usage was negatively associated with momentary depressed mood among individuals with elevated depressive symptoms. Additional analyses testing the temporal relationship showed that greater absolutist word usage was associated with higher levels of depressed mood after 12–24 h. Accordingly, absolutist word usage was positively associated with depressive symptom severity assessed after the 2-week period. Our results have several implications: First, absolutist word usage in spoken language appears to be an indicator of depression, which may be relevant for the optimization of depression assessment and monitoring approaches. Second, the finding that absolutist word usage was associated with lower depressed mood in the short-term, but higher depressed mood in the long-term provides important insights into the mechanisms of depression, and may help identify relevant treatment targets for clinicians. German Clinical Trial Registration DRKS00023670 on 19/01/2021 ( https://drks.de/search/en/trial/DRKS00023670 ).

Major depressive disorder (MDD) is a leading contributor to the global burden of disease, associated with substantial impairment in daily functioning and reduced quality of life for the individual and considerable societal costs. Farmers constitute a particularly vulnerable population, facing an elevated depression risk due to occupational, financial, and social stressors, and simultaneously encountering considerable barriers to mental health care. Personalized telephone coaching offers a promising solution for indicative prevention in this group. However, despite the evidence on its clinical effectiveness, the cost-effectiveness and cost-utility of telephone coaching for preventing MDD remain insufficiently examined. This economic evaluation aimed to evaluate the cost-effectiveness and cost-utility of a personalized telephone coaching (IG) provided by psychologists compared to treatment-as-usual plus informational material (TAU+) in farmers from a societal and social insurance perspective within a time horizon of 18 months. An economic evaluation was performed alongside a pragmatic randomized controlled trial (RCT). Health service use, patient and family expenses, and productivity costs were assessed with the adapted German version of the self-report questionnaire TiC-P (Trimbos Institute and Institute of Medical Technology Questionnaire for Costs Associated with Psychiatric Illness). Outcomes were measured in terms of depression symptom-free status (Quick Inventory of Depressive Symptomology – Self-Report (QIDS-SR16)<6) and quality-adjusted life years (QALYs; Assessment of Quality of Life (AQoL-8D)). All assessments were conducted web-based. A cost-effectiveness (CEA) and cost-utility analysis (CUA) were performed from the societal and social insurance perspective. Analyses followed the intention-to-treat principle with multiple imputation. Uncertainty around incremental cost-effectiveness ratio (ICER) estimates was characterized using bootstrapped seemingly unrelated regression equations (2,500 simulations). Sensitivity analyses assessed the robustness of the findings. In total, 314 participants were included in the RCT. From a societal perspective, the IG produced a greater number of participants achieving depression symptom-free status (ΔE=0.11, 95%CI –0.01 to 0.23) compared to TAU+ at higher costs (ΔC=€663, 95%CI 2820 to 4052; €1=US $1.12 as of 2019). The probability of the IG being cost-effective was 37% at a willingness-to-pay (WTP) threshold of €0 per depression symptom-free status and QALY gained, rising to 63% when applying a societal WTP of €20,000 per QALY gained (38% from a social insurance perspective). Analyses conducted from the social insurance perspective, as well as sensitivity analyses, indicated the dominance of TAU+ over the IG, with intervention costs as the primary cost driver. Findings suggest that MDD prevention via personalized telephone coaching yield larger effects, albeit accompanied by higher costs from both a societal and social insurance perspective within the observed time horizon. These results inform evidence-based decision-making on cost-effective MDD prevention in occupational groups, and highlight the need for comparative trials with alternative approaches. German Clinical Trial Registration: DRKS00015655. Registered on October 2, 2018. RR2-doi:10.3389/fpsyt.2020.00125

Emotionsregulation spielt eine zentrale Rolle für die psychische Gesundheit und gilt als transdiagnostischer Mechanismus in der Entstehung, Aufrechterhaltung und Behandlung psychischer Störungen. Dieser Beitrag gibt einen Überblick über zentrale Modelle und Erhebungsinstrumente und beschreibt, welche Risikofaktoren zur Entwicklung dysfunktionaler Regulationsmuster beitragen. Empirische Befunde zum Zusammenhang zwischen Emotionsregulationsdefiziten und unterschiedlichen Störungsbildern werden vergleichend dargestellt und die Bedeutung spezifischer Regulationsstrategien hervorgehoben. Darüber hinaus werden psychotherapeutische Ansätze vorgestellt, die gezielt auf die Förderung emotionaler Kompetenzen fokussieren, sowie der Forschungsstand zu ihrer Wirksamkeit dargestellt. Abschließend werden praxisrelevante Implikationen für Diagnostik, Prävention und Therapie abgeleitet und anhand eines Fallbeispiels veranschaulicht.

Abstract The effectiveness of antidepressants (ADs) in the treatment of depression is continuously discussed regarding its clinical relevance. The objective of this study was to explore thresholds of members of the public for effectiveness of ADs relative to the frequency of adverse drug reactions (ADRs) deemed acceptable. We conducted a cross-sectional online survey in Germany that included individuals with and without experiences of depression and ADs. Participants were presented a case scenario describing a moderate depressive episode, operationalized with the Montgomery Åsberg Depression Rating Scale (MADRS). Participants were asked to rate the effectiveness of different treatment options, estimate the frequency of 15 ADRs, and indicate the minimal level of treatment effectiveness they would require to accept certain ADRs. The survey was completed by 208 participants (144 female, 64 male), 44 were taking ADs. The effectiveness of ADs was significantly overestimated ( t (204) = 8.96, p < 0.001, d = 0.63). For 13 ADRs, participants selected package-insert frequency categories that were significantly lower than the corresponding reference categories. For example, to accept the potential ADR of seizures, participants required a symptom improvement of M = 23.51 ( SD = 8.82) points on the MADRS. We propose that patient-physician communication about risks and benefits of AD treatment should explicitly address individual expectations.

Internet-based psychological interventions can effectively reduce depressive symptoms in adults, but adherence remains challenging. In this preregistered individual participant data meta-analysis, we examined predictors of treatment adherence. We searched PubMed, Embase and PsycINFO on 6 February 2024 for randomized trials of internet-based interventions among adults with elevated depressive symptoms. We conducted a one-stage logit-link multilevel beta regression, with adherence defined as the proportion of completed modules post-intervention. This study included 71 trials (85 treatment arms, 8,082 participants). Lower adherence was associated with younger age (β = 0.005, standard error (SE) 0.002, P = 0.028), male gender (β = −0.163, SE 0.053, P = 0.002), lower education (β = −0.133, SE 0.05, P = 0.008) and employment (β = −0.113, SE 0.054, P = 0.037). No significant interactions were found between individual predictors and intervention format (guided versus self-guided). Higher adherence was associated with lower post-intervention depression severity, adjusting for baseline severity (β = −0.30, SE 0.04, P < 0.001). Identifying subgroups at risk of low adherence may inform targeted strategies to improve engagement and clinical effectiveness. Using data from 71 randomized controlled trials that involve 8,082 participants, the authors of this individual participant data meta-analysis examine individual- and study-level predictors of adherence to internet-based interventions for depression.

The affective sciences have predominantly investigated affective states at wakefulness. This leaves a significant research gap since affective processes during sleep remain largely unexplored. This research gap is particularly important given that evidence suggests that sleep is one critical timespan for affective processing. To date however, methodological limitations challenge the assessment of affect during sleep. To overcome these challenges, we propose a novel, contactless approach for the automated assessment of nocturnal affect using infrared recordings of facial expressions (ASANA-IF). ASANA-IF consists of three components: (1) recordings of facial expressions during sleep with infrared cameras, (2) extracting affective states from these expressions using machine learning tools for objective emotion assessment (OpenFace, a validated tool based on deep neural networks), (3) aggregating these estimates within domains of emotion (e.g., anger, fear) with an algorithm developed for this purpose. To explore the feasibility and validity of ASANA-IF, we used this approach to assess nocturnal affective states in 4 participants over 21 nights from 3 perspectives. Here, we empirically tested the hypothesis that affective states during sleep (anxious, angry, sad and happy affect) would predict self-reports of respective affective states on the following day. Findings provide evidence for the feasibility of ASANA-IF. Moreover, they confirm the hypothesis that nocturnal affective states predict levels of respective affective states on the following day. Future research should further improve this new approach and use it to address research questions associated with assessing nocturnal affect in a feasible and valid way.

OBJECTIVE:Dysfunctional beliefs play an important role in social anxiety disorder (SAD). Modifying the appraisal of such beliefs may be facilitated through emotion-based approach-avoidance modification trainings (eAAMTs). This study aimed to examine the feasibility and therapeutic potential of an eAAMT for SAD in a randomized-controlled pilot-study. METHOD:N = 30 participants with SAD were randomly allocated to an eAAMT-SAD, a non-active control condition, or a swipe-control condition (n = 28 analyzed). Participants in the eAAMT-SAD-condition received four 30-min-sessions of eAAMT-SAD within two weeks. They were instructed in using (a) swipe movements (day 1), (b) articulation of words with strong affective valence (day 2), and (c) deliberate expressions of emotion (days 3-4) to pull/push beliefs functional/dysfunctional towards/away from themself on a smartphone screen. The non-active control condition did not receive any training, participants in the swipe-control condition exclusively used swipe movements to pull/push functional/dysfunctional beliefs towards/away from themselves. Outcomes (symptom severity, agreement with (dys-)functional beliefs) were assessed immediately prior to training (T1), after the last training day (T2) and after a follow-up of 28 days (T3). RESULTS:Results demonstrated the intervention's feasibility and indicated that participants in the eAAMT-SAD condition experienced a greater reduction in anxiety symptoms compared to both control conditions, with small to moderate effect sizes (pre-post: non-active control: g = 0.42, swipe-control: g = 0.39; pre-follow-up: non-active control: g = 0.56, swipe-control: g = 0.64). CONCLUSION:Future studies should replicate these findings with larger samples and explore the feasibility and therapeutic potential of eAAMTs for other mental disorders.

Emotion regulation is a key determinant of mental health and functions as a transdiagnostic mechanism in the development, maintenance and treatment of psychological disorders. This article provides an overview of major theoretical models and assessment methods of emotion regulation and discusses risk factors contributing to the emergence of dysfunctional emotion regulation. Empirical evidence linking emotion regulation deficits to a range of mental disorders is summarized, emphasizing the role of specific regulatory strategies across disorders. Moreover, psychotherapeutic approaches explicitly targeting emotion regulation are outlined and current findings on their efficacy are presented. The article concludes with practical implications for the diagnostics, prevention and treatment, illustrated by a clinical case example.

Accurate emotion recognition is crucial for applications in mental health and affective computing. While commercial tools such as FaceReader are common for video analysis, their proprietary nature limits accessibility and integration. This study evaluates the open-source toolkit OpenDBM for predicting continuous emotion intensities, comparing a video-only approach to a multimodal model that fuses visual features with electrocardiogram (ECG) and respiration (RSP) signals. Using FaceReader-annotated video and biosignal data from 113 participants, we trained Random Forest (RF) and Gated Recurrent Unit (GRU) models. The video-only RF model achieved the best performance (MAE = 0.040), outperforming the multimodal model (MAE = 0.070). These results demonstrate the value of OpenDBM as an accessible alternative to commercial tools and suggest that visually derived labels may not fully capture emotion-related patterns present in physiological signals.

Dysfunctional beliefs are one key factor in the development and maintenance of social phobia. Modifying the appraisal of such beliefs might be achieved with the help of approach-avoidance modification trainings (AAMTs). In these trainings individuals are instructed to push dysfunctional stimuli away and pull functional ones toward themselves via joystick- or swipe-based push/pull-movements. However, the efficacy of the AAMTs could be enhanced by using high-valence words as well as facial expressions of emotions as approach/avoidance responses. The present study aimed to examine the safety and feasibility of an emotion-focused AAMT (eAAMT) targeting dysfunctional beliefs. We conducted a feasibility study with a sample of 10 participants with social anxiety. With regard to acceptability, not all predefined cutoffs where met, indicating need for further optimization of the intervention. Regarding the therapeutic potential, the social anxiety symptom severity decreased with a large effect size (g = 0.89 [0.66, 1.79]) from T1 to T2 and with a moderate effect size (g = 0.72 [0.38, 1.31]) from T1 to T3. Thus, the results of the study confirmed the clinical safety and technical feasibility of delivering the eAAMT in a laboratory setting, albeit within a Wizard-of-Oz paradigm. Furthermore, results provide preliminary evidence for its potential efficacy.Preregistration: https://osf.io/d4ye2/overview.

Perceived stress is prevalent in industrial societies, negatively impacting mental health. Smartphone-based stress management interventions provide accessible alternatives to traditional methods, but their efficacy remains modest, potentially due to limited integration of smartphone sensor technology. The primary aim of this study was to evaluate the efficacy of an 18-day smartphone-based stress management intervention, MT-StressLess with integrated heart rate (HR)-based biofeedback using built-in accelerometer sensors, compared to a waitlist control (WLC) condition. Secondary outcomes included emotion regulation skills, depressive symptoms, overall well-being, usbiality and usage data. As exploratory aims, we investigated whether the MT-StressLess version without HR-based biofeedback was also superior to the WLC condition, and whether the version with HR-based biofeedback provided additional benefits compared to the version without. In a three-arm randomized controlled trial, 166 participants were assigned to MT-StressLess with HR-based biofeedback, MT-StressLess, or the WLC condition. Linear mixed-effects models were used to analyze intervention effects over time (baseline, postintervention, and 1-month follow-up). At postintervention, MT-StressLess with HR-based biofeedback showed significantly greater reductions in perceived stress compared to the WLC condition (d = 0.41, 95% CI [0.03, 0.79]), whereas the version without biofeedback did not differ significantly (d = 0.14, 95% CI [-0.24, 0.51]). No significant differences were observed between the two active conditions (d = 0.29, 95% CI [-0.08, 0.66]). Both active conditions, however, led to significant improvements in the secondary outcomes of emotion regulation skills and well-being compared to the WLC (all ds = -0.58 to -0.27). These patterns persisted at the 1-month follow-up. Usability ratings were high, but overall adherence was moderate. The findings in the main comparison may reflect increased interoceptive awareness and self-regulation. Yet, the limited effects of the core intervention and the biofeedback component also suggest the influence of non-specific factors, such as placebo effects, outcome expectancy and user engagement, which highlights the need to better understand optimal intervention duration, motivation, reinforcement, and more individualized approaches to stress reactivity. Overall, the findings provide preliminary support for the potential of a smartphone-based intervention that includes HR-based biofeedback to reduce perceived stress compared to no intervention. As these interventions are still in their early stages, future research should explore how personalization driven by artificial intelligence and real-time physiological tracking can enhance engagement and efficacy.

INTRODUCTION:Relapse rates in individuals with alcohol use disorder (AUD) are particularly high following inpatient treatment. Innovative strategies should specifically target the transitional gap between completion of inpatient treatment and uptake of standard continuing care. This study aimed to determine whether Appstinence, a digital approach that combines a smartphone app intervention with adjunct telephone coaching, more markedly reduces the risk of relapse for 6 months after inpatient AUD treatment in comparison to a control group with access to standard continuing care. METHODS:In this multicenter clinical trial, 356 participants were randomized to the intervention (n = 175) or control group (n = 181). Eligibility criteria included diagnosis of AUD, smartphone access, no acute suicidality, and no language or neurocognitive impairments. The primary outcome was risk of relapse within 6 months after randomization, as assessed with the Timeline-Follow-Back method. Secondary outcomes included uptake of standard continuing care, hazardous alcohol consumption, craving, depression and anxiety symptom severity, and well-being. RESULTS:The intervention reduced the risk of relapse within 6 months as indicated by a log-rank test (HR: 0.72, 95% CI: 0.53-0.98, p = 0.04) and Cox regression adjusted for baseline characteristics (HR: 0.67, 95% CI: 0.48-0.92, p = 0.01). This effect increased when participants fully adhered to the intervention protocol (log-rank test: HR: 0.61, 95% CI: 0.39-0.94, p = 0.02). No significant differences were observed in secondary outcomes. CONCLUSION:Our findings provide supportive evidence for digital AUD transition treatment. Specifically, we found that, in comparison with access standard continuing care, the novel intervention more effectively reduced risk of relapse within 6 months following inpatient treatment.

Preventing mental disorders is important to avoiding clinical conditions. This study evaluated the efficacy of internet-based indicated prevention for anxiety and depressive disorders. In a three-arm randomized controlled trial, 566 adults with subthreshold anxiety (GAD-7 ≥ 5) and/or depressive symptoms (CES-D ≥ 16), but no clinical diagnosis in the past six months (MINI 6.0), were assigned to either an individually (IG-IMI, n = 186) or automatically (AG-IMI, n = 189) guided digital intervention, or waitlist control (WLC, n = 191). The digital intervention comprised 8 transdiagnostic, self-tailored, CBT-based sessions. The primary outcome was time to onset of any anxiety or depressive disorder over 12 months, assessed via blinded diagnostic interviews (MINI). AD/DD onset was 19.4% in IG-IMI, 14.8% in AG-IMI, and 30.9% in WLC. Cumulative incidence was 23.1% (IG-IMI), 20.7% (AG-IMI), and 36.0% (WLC; p < 0.001). Hazard ratios were 0.59 and 0.47; NNTs were 7.76 and 5.79. Both individually guided and automated interventions effectively reduced AD/DD incidence. Trial Registration: The study was preregistered in the German Clinical Trial Registration (DRKS00011099; https://drks.de/search/de/trial/DRKS00011099 ).

BACKGROUND:The development of automatic emotion recognition models from smartphone videos is a crucial step toward the dissemination of psychotherapeutic app interventions that encourage emotional expressions. Existing models focus mainly on the 6 basic emotions while neglecting other therapeutically relevant emotions. To support this research, we introduce the novel Stress Reduction Training Through the Recognition of Emotions Wizard-of-Oz (STREs WoZ) dataset, which contains facial videos of 16 distinct, therapeutically relevant emotions. OBJECTIVE:This study aimed to develop deep learning-based automatic facial emotion recognition (FER) models for binary (positive vs negative) and multiclass emotion classification tasks, assess the models' performance, and validate them by comparing the models with human observers. METHODS:The STREs WoZ dataset contains 14,412 facial videos of 63 individuals displaying the 16 emotions. The selfie-style videos were recorded during a stress reduction training using front-facing smartphone cameras in a nonconstrained laboratory setting. Automatic FER models using both appearance and deep-learned features for binary and multiclass emotion classification were trained on the STREs WoZ dataset. The appearance features were based on the Facial Action Coding System and extracted with OpenFace. The deep-learned features were obtained through a ResNet50 model. For our deep learning models, we used the appearance features, the deep-learned features, and their concatenation as inputs. We used 3 recurrent neural network (RNN)-based architectures: RNN-convolution, RNN-attention, and RNN-average networks. For validation, 3 human observers were also trained in binary and multiclass emotion recognition. A test set of 3018 facial emotion videos of the 16 emotions was completed by both the automatic FER model and human observers. The performance was assessed with unweighted average recall (UAR) and accuracy. RESULTS:Models using appearance features outperformed those using deep-learned features, as well as models combining both feature types in both tasks, with the attention network using appearance features emerging as the best-performing model. The attention network achieved a UAR of 92.9% in the binary classification task, and accuracy values ranged from 59.0% to 90.0% in the multiclass classification task. Human performance was comparable to that of the automatic FER model in the binary classification task, with a UAR of 91.0%, and superior in the multiclass classification task, with accuracy values ranging from 87.4% to 99.8%. CONCLUSIONS:Future studies are needed to enhance the performance of automatic FER models for practical use in psychotherapeutic apps. Nevertheless, this study represents an important first step toward advancing emotion-focused psychotherapeutic interventions via smartphone apps.

BACKGROUND:To map out the potential benefits of widely available smartphone apps for mental health, especially in contexts where face-to-face services are limited or unavailable, it is crucial to examine their efficacy compared with inactive controls. Standalone smartphone apps might offer an accessible option for individuals waiting for treatment or living in under-resourced settings. Given the currently inconclusive evidence regarding these apps, this systematic review and meta-analysis aimed to assess the efficacy and study quality of randomised controlled trials (RCTs) evaluating standalone smartphone apps for mental health. METHODS:In this systematic review and meta-analysis, based on a previously published study, we conducted an updated systematic search of PubMed, PsycINFO, Web of Science, Cochrane Clinical Trial, and Scopus for RCTs published from database inception to Nov 10, 2023. We included RCTs that examined the efficacy of standalone smartphone apps for mental health in adults (age ≥18 years) with heightened symptom severity compared with an inactive control group (eg, waitlist, informational material, and control apps). We excluded control groups that received active treatment. Two independent researchers (AV and AD) extracted summary data, which were verified by a third researcher (JKK). The effect size Hedges' g, 95% CI, and p value were calculated for each target outcome. We applied a random-effects model to all analyses due to the expected heterogeneity between RCTs. We assessed quality using the Risk of Bias 2 tool (dated Aug 22, 2019) and assessed publication bias via the Egger's test, and the Duval and Tweedie trim-and-fill analysis. The study was registered with PROSPERO, CRD42022310762. FINDINGS:We retrieved 12 705 records from electronic databases and 74 records from other sources (ie, reviews and meta-analyses on digital interventions for mental health identified through database searches and their reference lists, reference lists of other studies, trial registrations in PROSPERO, and websites of researchers in the field). Of these, we included 72 RCTs (70 reports) with 21 702 participants (of the 21 048 participants with sex or gender data, 14 208 [67%] were female, 6744 [32%] were male, and 96 [<1%] were other). At post assessment (assessment after completion of intervention), we found significant effects of apps targeting depression (33 comparisons; Hedges' g 0·45 [95% CI 0·30 to 0·60], p≤0·0001, I2=81·30%), anxiety (23 comparisons; 0·35 [0·22 to 0·48], p≤0·0001, I2=74·91%), sleep problems (14 comparisons; 0·71 [0·51 to 0·92], p≤0·0001, I2=76·17%), post-traumatic stress disorder (nine comparisons; 0·15 [0·02 to 0·28], p=0·029, I2=28·65%), eating disorders (four comparisons; 0·50 [0·29 to 0·71], p≤0·0001, I2=50·49%), and body dysmorphic disorder (three comparisons; 0·86 [0·30 to 1·41], p=0·0025, I2=74·90%) compared with inactive control groups. No significant pooled effects were found for smoking (six comparisons), self-injury (six comparisons), suicidal ideation (five comparisons), and alcohol misuse (five comparisons). Effect sizes for obsessive-compulsive disorders (two comparisons) and schizophrenia (one comparison) ranged from 0·10 to 0·96 (-0·12 to 1·51). Risk of bias was moderate to high. Publication bias was found for RCTs targeting depression and anxiety, but not for sleep problems; adjustments reduced the effect sizes for depression from 0·45 to 0·18 (0·02 to 0·34) and anxiety from 0·35 to 0·18 (0·03 to 0·32), with no change for sleep problems. INTERPRETATION:Although some outcomes showed small to medium effect sizes, these results must be interpreted cautiously given the presence of uncertainty factors, including considerable heterogeneity and moderate study quality. Heterogeneity might result from sample characteristics, assessment methods and periods, dropout rates, intervention and app components, and control conditions, limiting the generalisability of findings. Standalone smartphone apps might be offered for symptoms of depression, anxiety, and sleep problems, if no evidence-based first-line intervention is available. FUNDING:None.

Deficits in emotion regulation (ER) are associated with major depressive disorder (MDD). Therefore, enhancing ER is a promising target in the treatment of MDD. In this study, we examined ER as a mechanism of change in outpatient, group-based ER-focused psychotherapy for MDD, conducting secondary analysis of randomized controlled trial (RCT) data. Depressive symptom severity (DSS) and ER were assessed five times during an 8-week Affect Regulation Training (N = 73). Firstly, we tested cross-sectional correlations of ER and DSS at all assessment points. Then, we employed latent growth curve modeling (LGCM) to explore associations between changes in ER and DSS during treatment. Finally, we applied bivariate latent change score modeling (LCS) to test whether improvements in ER skills predicted subsequent reductions in DSS. Cross-sectional analyses revealed negative correlations between ER and DSS at all time points. Longitudinal LGCM analyses showed that an increase in ER skills during treatment was associated with a decrease in DSS. Finally, LCS analyses indicated that changes in ER predicted subsequent changes in DSS, whereas the reverse relationship was not significant. These findings support a causal relationship between the enhancement of ER skills and subsequent reduction of DSS, highlighting the importance of ER as a mechanism of change in the treatment of depression. Future research should extend these analyses to other disorders beyond depression to further clarify the role of ER as a critical transdiagnostic factor and a promising treatment target for a wide range of mental health conditions. The original RCT was registered at ClinicalTrials.gov, Nr. NCT01330485.

IntroductionThe detrimental consequences of stress highlight the need for precise stress detection, as this offers a window for timely intervention. However, both objective and subjective measurements suffer from validity limitations. Contactless sensing technologies using machine learning methods present a potential alternative and could be used to estimate stress from externally visible physiological changes, such as emotional facial expressions. Although previous studies were able to classify stress from emotional expressions with accuracies of up to 88.32%, most works employed a classification approach and relied on data from contexts where stress was induced. Therefore, the primary aim of the present study was to clarify whether stress can be detected from facial expressions of six basic emotions (anxiety, anger, disgust, sadness, joy, love) and relaxation using a prediction approach.MethodTo attain this goal, we analyzed video recordings of facial emotional expressions collected from n = 69 participants in a secondary analysis of a dataset from an interventional study. We aimed to explore associations with stress (assessed by the PSS-10 and a one-item stress measure).ResultsComparing two regression machine learning models [Random Forest (RF) and XGBoost], we found that facial emotional expressions were promising indicators of stress scores, with model fit being best when data from all six emotional facial expressions was used to train the model (one-item stress measure: MSE (XGB) = 2.31, MAE (XGB) = 1.32, MSE (RF) = 3.86, MAE (RF) = 1.69; PSS-10: MSE (XGB) = 25.65, MAE (XGB) = 4.16, MSE (RF) = 26.32, MAE (RF) = 4.14). XGBoost showed to be more reliable for prediction, with lower error for both training and test data.DiscussionThe findings provide further evidence that non-invasive video recordings can complement standard objective and subjective markers of stress.