News

Feb 20, 2025

Randomized controlled study of a digital data driven intervention for depressive and generalized anxiety symptoms | npj Digital Medicine

npj Digital Medicine volume 8, Article number: 113 (2025) Cite this article Metrics details As mental health disorders like Major Depressive Disorder and Generalized Anxiety Disorder rise globally,

npj Digital Medicine volume 8, Article number: 113 (2025) Cite this article

Metrics details

As mental health disorders like Major Depressive Disorder and Generalized Anxiety Disorder rise globally, effective, scalable, and personalized treatments are urgently needed. This 16-week prospective, decentralized, randomized, waitlist-controlled study investigated the effectiveness of a digital data-driven therapeutic integrating wearable sensor data with a mobile app to deliver personalized CBT-based interventions for individuals with depressive and generalized anxiety symptoms. 200 adults were randomized to intervention or control groups, with 164 completing the study. The intervention group demonstrated significant reductions in depressive (mean change = −5.61, CI = −7.14, −4.08) and anxiety symptoms (mean change = −5.21, CI = −6.66, −3.76), compared to the control group, with medium-to-large effect sizes (r = 0.64 and r = 0.62, P < 0.001). Notably, these improvements were also observed in participants with clinically significant depression and anxiety, further reinforcing the potential of digital therapeutics in targeting more severe cases. These findings, combined with high engagement levels, suggest that data-driven digital health interventions could complement traditional treatments, though further research is needed to assess their long-term impact.

As we approach the end of the first quarter of the 21st century, we face an escalating challenge from the previous century in treating mental health empirically, equitably, and at scale1,2. Currently, mental health issues, primarily anxiety and depression, continue to be a considerable cause of global disability across all age groups, with a peak in prevalence among individuals aged 45–493. In 2017, depression moved to the 3rd overall leading cause of disability, up from 4th in 1990, indicating a persistent and pervasive impairment3. Globally an estimated 5% of the adult population experiences depression, and more than 700,000 people die by suicide every year, rendering it the fourth leading cause of death in 15–29-year-olds4. Through the Global Burden of Disease5 lens, the impact of depression and anxiety has been amplified from 1990 to 2019, with depression moving from 19th to 13th and anxiety from 34th to 24th in terms of their global burden rankings. This trend has significant consequences for individuals, communities, and health systems trying to manage the global impact6. Those experiencing symptoms of either or both depression and anxiety often struggle in all life domains, including social, vocational, and activities of daily living and they report poor quality of life4,7,8. Therefore, it is imperative to provide effective treatment options to reduce symptoms. Yet, the etiology of depression and anxiety is heterogeneous, complicating the provision of treatments that effectively alleviate the individual’s symptoms9,10,11. This variability also makes the identification of which treatments work and for whom to be an open question in clinical research12,13,14. Despite these challenges, digital innovations in 21st century healthcare offer a beacon of hope. These technologies enable personalized, evidence-based treatments with the potential to improve treatment outcomes15.

Digital innovation began to emerge in healthcare with the advent of mobile health (mHealth). Over the past two decades, this movement was accelerated by the proliferation of smartphones and was further integrated into health systems worldwide during the COVID-19 pandemic16. Focusing on digital health interventions (DHIs) for the management and treatment of Major Depressive Disorder (MDD) and Generalized Anxiety Disorders (GAD), we observe a plethora of studies17,18,19,20 incorporating web-based, app-based, self-guided and human-guided modalities, that innovate on the tenets of Cognitive Behavioral Therapy (CBT). CBT remains the dominant therapeutic approach due to several key features: the Thoughts, Feelings, and Behaviors (TFB) Cycle, Cognitive Distortions, Cognitive Restructuring, and Behavioral Activation21. These elements have proved to be effective in reducing the burden of MDD and GAD21.

Several meta-analyses22,23 have occurred to evaluate the efficacy of DHIs for depression, generally yielding favorable results. Specifically, a meta-analysis examining the non-inferiority of digital CBT compared to traditional face-to-face CBT found no significant difference in effectiveness24. Similarly, a recent meta-analysis of DHI studies for anxiety also showed no difference in treatment effectiveness compared to in-person therapy 22. These findings highlight the potential for DHIs to revolutionize the way MDD and GAD are treated. DHIs broaden the scope of treatment modalities to better align with user needs, reduce treatment duration compared to face-to-face therapy, triage clinical resources based on symptom severity, expand access to treatment, and address the heterogeneity of symptoms24,25.

Layering digital innovation along with the core principles of CBT has proven to be effective26,27, however, engagement and retention with DHIs can be challenging28,29. At the patient level, it is important to understand the user’s perceptions of DHIs along with their preferences, when deciding which digital modalities to provide30. At the intervention level, various engagement strategies have been utilized in studies29,31, with one potential solution for enhancing the patient experience being the utilization of data to drive personalized mental health treatment32,33.

Data-driven interventions bring together various data modalities captured via smartphones, web-based programs, and wearables to personalize clinical interventions. These modalities encompass a wide range of data sources, including physiological information like heart rate, skin conductivity, and temperature, as well as physical activity measures such as sleep patterns, stress and activity levels, and social interaction metrics34,35,36. This rich array of data helps shape a detailed digital twin37, offering a comprehensive view of the daily patterns and physiological state of an individual, which leads to a better understanding of the patient’s mental health status38,39.

Given the heterogeneous nature of MDD and GAD, data-driven approaches have the potential to provide more precise treatments by delivering interventions at the time of need, tailored to the unique symptomatology of the individual40,41. As a result, these interventions may be more effective, leading to more favorable mental health outcomes40,41. Further exploration is needed to determine whether data-driven interventions enhance engagement and retention with DHIs, if such engagement increases exposure to evidence-based interventions like CBT, and ultimately, if this translates into a reduction in depression and anxiety symptoms.

In this work, we introduce data-driven DHIs as part of a digital therapeutic for depressive and generalized anxiety symptoms, expanding on a previous pilot study that showed favorable results in terms of feasibility, acceptability, engagement, and potential impact on depressive and anxiety symptoms and quality of life outcomes42. This randomized controlled study evaluates the efficacy of a 16-week, standardized digital therapeutic that leverages CBT as the primary treatment modality. The intervention utilizes data collected from a wearable sensor, and a mobile app to support the delivery of DHIs. A provider is also incorporated to serve as a digital navigator, while also supporting goal setting, participant engagement, and monitoring. The primary objective of this study is to determine whether data-driven DHIs can achieve greater reductions in depressive and anxiety symptoms, as measured by PHQ-9 and GAD-7 scores, compared to a control group. We hypothesize that integrating real-time physiological data with CBT-based interventions will deliver a more precise treatment approach, driving positive outcomes for individuals with depressive and generalized anxiety symptoms, and advancing the “Digital” wave of CBT15,43.

Overall, 1772 individuals expressed interest in participating in the study and were assessed for eligibility, with 1105 of them being ineligible and 667 eligible. As illustrated in Fig. 1, out of the 1105 ineligible study candidates, 17 did not meet age criteria, 20 had minimal depressive or anxiety symptoms, 382 were taking psychotropic medication, 72 had been diagnosed with bipolar disorder, 358 with an eating disorder, 128 with a psychotic disorder, 131 with a personality disorder, 237 with PTSD, while 174 suffered from substance abuse and 313 had suicidal or self-harm thoughts. Therefore, 667 candidates were invited to join the study and 200 accepted the invitation and provided consent. These 200 study participants were randomly allocated between the experimental and control groups (i.e., 100 participants per group). For the experimental group, dropout is defined as discontinuing weekly check-ins and engagement with weekly DHIs, while for the control group, as not completing the final (16-week) assessment. Thus, the overall dropout rate for the study was 18% (n = 29 for the experimental and n = 7 for the control group), corresponding to 29% and 7% for the experimental and control groups, respectively.

Flow diagram illustrating the progression of participants through the study. The number of participants at each step of the process is indicated in brackets.

Baseline demographics and disorder severity characteristics were generally similar between the experimental and control groups (Table 1). The average age was 33.2 and 35.53 respectively with female/male distributions of 78/22 and 76/24. Regarding depressive symptom severity, 28.5% (57/200) of participants exhibited mild symptoms, 35.5% (71/200) moderate symptoms, 23.5% (47/200) moderately severe symptoms, and 10% (20/200) severe symptoms. For anxiety symptom severity, 39.5% (79/200) of participants reported mild symptoms, 34% (68/200) moderate symptoms, and 18% (36/200) severe symptoms.

In Table 2, we present the obtained statistics for the primary outcomes. It can be observed that no significant differences occur between the two groups at baseline. Notably, a medium-to-large effect size is showcased for the experimental group in both within-group comparisons across and between-groups comparisons at mid and post-intervention. Specifically, for the experimental group at mid-intervention time point, effect sizes of r = 0.61 (P < 0.001) for depressive symptoms (PHQ-9) and r = 0.75 (P < 0.001) for anxiety symptoms (GAD-7) were obtained in the within-group comparison. Between groups, at the same time point, effect sizes of r = 0.42 (P < 0.001) for both PHQ-9 and GAD-7 were observed. Similarly, at post-intervention, the within-group effect sizes for the experimental group were r = 0.83 (P < 0.001) for both PHQ-9 and GAD-7, while between-group comparisons revealed an effect size of r = 0.64 (P < 0.001) for PHQ-9 and r = 0.62 (P < 0.001) for GAD-7. Overall, for the depressive symptoms (PHQ-9), we observed a 45% relative reduction (baseline to post) in the experimental group with 32% of the participants (n = 32) surpassing the MCID threshold. At the same time, for anxiety symptoms (GAD-7), the overall relative reduction (baseline to post) was 50%, with 39% of the participants (n = 39) surpassing the MCID threshold. In total, 42% of the participants (n=42) showed MCID in either depressive or anxiety symptoms, and 29% (n = 29) showed MCID in both. In general, 60% of the participants (n = 60) showed a score reduction in either of the scales and 52% (n=52) showed a reduction in both. The average PHQ-9 and GAD-7 scores at baseline, mid and post-intervention evaluations for the experimental and control groups are presented in Fig. 2.

A Average PHQ-9 scores at baseline, mid-intervention, and post-intervention. Vertical error bars represent the standard error of the average PHQ-9 scores. The solid line refers to the experimental group and the dashed line to the control group. B Average GAD-7 scores at baseline, mid-intervention, and post-intervention. Vertical error bars represent the standard error of the average GAD-7 scores. The solid line refers to the experimental group and the dashed line to the control group.

Next, we investigate in more detail the effect of the different factors by incorporating also the variance to the results from each individual in an LMM. The mixed model for PHQ-9 suggested a significant variability at the level of individual scores, (SD = 3.56, 95% CI = 3.08, 4.03), with a substantial explanatory power (conditional R2 = 0.61). The fixed effect (main and interaction) coefficients estimates are shown in Table 3 along with the corresponding CI and P values. A statistically significant interaction of group at both time points is observed (b = −1.80, CI = −2.51,−1.09, P < 0.001 for mid and b = −2.79, CI = −3.50, −2.08, P < 0.001 for post). This affects the slope of the line PHQ-9 vs time, suggesting that when the group variable is positive (i.e. for the experimental group) the score decreases. The main effect of the group was also found to be statistically significant (b = −1.49, CI = −2.08,−0.91, P < 0.001). The Bonferroni adjusted estimated mean difference between the two groups at post is 5.52 (SE = 0.75, P < 0.001, 95% CI 3.30, 7.74).

Similarly, for the GAD-7 scores, the fitted LMM showed a conditional R2 = 0.61 with a significant variability at the level of individual scores, (SD = 3.33, 95% CI = 2.88, 3.76). The corresponding fixed effect (main and interaction) coefficients, along with the corresponding CIs and P-values, are shown in Table 4. A statistically significant interaction of group at both time points is observed (b = −1.81, CI = −2.47, −1.16, P < 0.001 for mid and b = −2.85, CI = −3.51, −2.2, P < 0.001 for post). This affects the slope of the line GAD-7 vs time, suggesting that when the group variable is positive (i.e. for the experimental group) the score decreases. In addition, the effect of the group is statistically significant (b = −1.35, CI = −1.89, −0.80, P < 0.001). This adds a negative offset to the score (thus acting similar to a constant) when the group variable is positive (i.e. experimental group). The Bonferroni adjusted estimated mean difference between the two groups at post is 5.3 (SE = 0.69, P < 0.001, 95% CI = 3.24, 7.36).

In Figs. 3 and 4, the results of the sensitivity analysis concerning the interaction of the group with the time variable for the PHQ-9 (Fig. 3) and GAD-7 (Fig. 4) outcomes, respectively, are presented. More specifically, in each of these figures we show the estimate and the corresponding confidence interval for the group × t2,1 (light-colored bars) and group × t3,1 (dark-colored bars) interactions corresponding to mid- and post-intervention, for different k values. Statistically significant results are shown with solid lines, while the cases with no statistical significance are shown with dashed lines. Both for the PHQ-9 and the GAD-7, the missing values of the dropouts that were imputed have been multiplied by k. Regarding the comparison between the MCAR and MAR scenarios, a minimal impact on the interaction coefficients is observed. Concerning the different MNAR scenarios, for PHQ-9, the inference at mid, remains the same up to k = 1.5, while for GAD-7 it is statistically significant up to k = 1.2. Similarly for post, the inference for both PHQ-9 and GAD-7 is robust for all tested scenarios. Based on these results, for the primary outcomes, our models’ inferences at post are robust, considering various worse-case MNAR scenarios for the dropouts with increasing values of up to 2.5 times for both outcomes.

Sensitivity analysis of the group × time interaction coefficient estimates for PHQ-9 scores across different missing values scenarios. In the MCAR scenario, only data from participants who completed the questionnaires in all evaluation periods are used. In the MAR scenario, missing data is imputed via multiple imputations. For the MNAR scenarios, imputed values are adjusted by a parameter k. Vertical bars represent 95% confidence intervals, and the horizontal dashed line indicates the threshold for which the inference reverses. Light-colored lines show mid-baseline changes in PHQ-9 scores, while dark-colored lines reflect post-baseline changes. Solid vertical lines denote statistically significant outcomes, and dashed vertical lines indicate non-significant results.

Sensitivity analysis of the group × time interaction coefficient estimates for GAD-7 scores across different missing values scenarios. In the MCAR scenario, only data from participants who completed the questionnaires in all evaluation periods are used. In the MAR scenario, missing data is imputed via multiple imputations. For the MNAR scenarios, imputed values are adjusted by a parameter k. Vertical bars represent 95% confidence intervals, and the horizontal dashed line indicates the threshold for which the inference reverses. Light-colored lines show mid-baseline changes in GAD-7 scores, while dark-colored lines reflect post-baseline changes. Solid vertical lines denote statistically significant outcomes, and dashed vertical lines indicate non-significant results.

In Table 5, we present the statistics for the secondary outcomes. At baseline no significant differences were observed between the two groups. The intervention demonstrated medium-to-large effect sizes for the experimental group both for a within-group comparison across time and between the two groups at mid and post. Specifically, for the experimental group at the mid-intervention time point, effect sizes of r = 0.57 (P < 0.001) for SWLS and r = 0.4 (P<0.001) for LISAT-11 were obtained in the within-group comparison. Between groups, at the same time point, effect sizes of r = 0.19 (P = 0.035) and r = 0.32 (P<0.001) were obtained for SWLS and LISAT-11 respectively. Similarly, at post-intervention, the within-group effect sizes for the experimental group were r = 0.81 (P<0.001) for SWLS and r = 0.74 (P<0.001) for LISAT-11, while between-group comparisons revealed r = 0.47 (P<0.001) for SWLS and r = 0.59 (P<0.001) for LISAT-11. Overall, for the SWLS scale, an average increase of 27.19% was observed with 53% of the participants (n=53) exhibiting an improvement. Accordingly, for the LISAT-11 scale, the average increase was 18.89% with 45% of the participants (n=45) demonstrating an improvement. In general, 45% of the participants (n=45) showed an improvement in both scales while 58% of them (n=58) showed an improvement in either of the scales. The average SWLS and LISAT-11 scores at baseline, mid and post-intervention evaluations for the experimental and control groups are presented in Fig. 5.

A Average SWLS scores at baseline, mid-intervention, and post-intervention. Vertical error bars represent the standard error of the average SWLS score. The solid line refers to the experimental group and the dashed line to the control group. B Average LISAT-11 scores at baseline, mid-intervention, and post-intervention. Vertical error bars represent the standard error of the average LISAT-11 score. The solid line refers to the experimental group and the dashed line to the control group.

As for the case of the primary outcomes, we investigate in more detail the effect of the different factors by incorporating also the variance to the results from each individual in an LMM. The LMM for SWLS suggested significant variability at the level of individual (SD = 4.57, CI = 4.03, 5.09) and had substantial explanatory power with R2 = 0.72. The fixed effect (main and interaction) coefficients, CIs, and P-values are shown in Table 6. A statistically significant interaction of group at both time points is observed (b = 0.92, CI = 0.26, 1.59, P = 0.006 for mid and b = 2.30, CI = 1.63, 2.96, P < 0.001 for post). This affects the slope of the line SWLS vs time, suggesting that when the group variable is positive (i.e., for the experimental group) the score increases. A significant main effect was also found for group (b = 1.17, CI = 0.46,1.87, P = 0.001) and at post (b = 4.46, CI = 1.87, 7.05, P < 0.001). The Bonferroni adjusted estimated mean difference between the two groups at post was found to be 4.79 (SE = 0.83, CI = 2.32, 7.26, P < 0.001).

Regarding the LISAT-11 scores, the LMM had substantial explanatory power with R2=0.715 and suggested a significant variability at the individual level (SD = 0.62, CI = 0.55 0.70). The fixed effect (main and interaction) coefficients along with the corresponding CIs and P-values are shown in Table 7. A statistically significant interaction of group at both time points is observed (b = 0.11, CI = 0.02, 0.21, P = 0.017 for mid and b = 0.32, CI = 0.23, 0.42, P < 0.001 for post). This affects the slope of the line LISAT vs time, suggesting that when the group variable is positive (i.e., for the experimental group) the score increases. A significant main effect of time at post (b = 0.43, CI = 0.06, 0.80, P = 0.02) and for group (b = 0.21, CI = 0.12, 0.31, P < 0.001) was also observed. The Bonferroni adjusted estimated mean difference between the two groups at post was found to be 0.79 (SE = 0.11, CI = 0.45, 1.13, P < 0.001).

In Figs. 6 and 7, the results of the sensitivity analysis concerning the interaction of the group with the time variable for the SWLS (Fig. 6) and LISAT-11 (Fig. 7) responses, respectively, are illustrated. More specifically, in each of these figures, we present the estimate and the corresponding confidence interval for the group × t2,1 (light-colored bars) and group × t3,1 (dark-colored bars) interactions corresponding to mid- and post-intervention, for different k values. Results with statistical significance are shown with solid lines, while non-significant results are represented by dashed lines. For SWLS and LISAT-11, the missing values of the dropouts that were imputed are divided by k. For the MCAR and MAR scenarios, the estimated values are similar, but with larger standard errors, especially for the SWLS. For this outcome also, the interaction coefficient at mid is non-significant under the MAR assumption. With respect to the different MNAR scenarios, for SWLS, the inference at mid is non-significant, while for post it is robust up to k = 1.5. Similarly, for LISAT-11, the inference at mid is the same only for MAR (k=1) and it reverses for k = 2.5. The inference at post remains for values up to k = 1.5. Based on these results, for the life satisfaction questionnaires, the impact of missing values in MNAR scenarios do not affect the sign of the inference at the end of the program, even if the dropouts would have had 34% worse scores.

Sensitivity analysis of the group × time interaction coefficient estimates for SWLS scores across different missing values scenarios. In the MCAR scenario, only data from participants who completed the questionnaires in all evaluation periods are used. In the MAR scenario, missing data is imputed via multiple imputations. For the MNAR scenarios, imputed values are adjusted by a parameter k. Vertical bars represent 95% confidence intervals, and the horizontal dashed line indicates the threshold for which the inference reverses. Light-colored lines show mid-baseline changes in SWLS scores, while dark-colored lines reflect post-baseline changes. Solid vertical lines denote statistically significant outcomes, and dashed vertical lines indicate non-significant results.

Sensitivity analysis of the group × time interaction coefficient estimates for LISAT-11 scores across different missing values scenarios. In the MCAR scenario, only data from participants who completed the questionnaires in all evaluation periods are used. In the MAR scenario, missing data is imputed via multiple imputations. For the MNAR scenarios, imputed values are adjusted by a parameter k. Vertical bars represent 95% confidence intervals, and the horizontal dashed line indicates the threshold for which the inference reverses. Light-colored lines show mid-baseline changes in LISAT-11 scores, while dark-colored lines reflect post-baseline changes. Solid vertical lines denote statistically significant outcomes, and dashed vertical lines indicate non-significant results.

Overall, participants showed very high engagement during the 16-week study with an average of 77.1% (98.36% - per protocol, meaning only participants who completed the intervention were included in the analysis) being active on a weekly basis with 2.61 (3.4 - per protocol) days of activity. During every week, 60.66% (77.16% - per protocol) of participants were engaging with the sensor, spending 57.65 (74.4 - per protocol) minutes on average in the mobile app with 19.42 (24.98 - per protocol) interactions. Finally, participants attended 74.31% (95.44% - per protocol) of the weekly check-ins and completed on average 7.83 (10.02 - per protocol) DHIs per week. Specifically, participants completed 3.64 (4.61 - per protocol) exercises, 2.99 (3.92 - per protocol) emotion journals, 46% of which were prompted by the FES, and 1.2 (1.49 per protocol) reviews of assigned educational modules. In Fig. 8, we present boxplots illustrating the distributions of the engagement metrics, offering a visual summary of central tendencies and variability. No adverse events were reported in the intervention group.

Boxplots illustrating engagement metrics for the intention-to-treat group (dark-colored bars) and the per-protocol group (light-colored bars). Whiskers represent the range, extending from the minimum to the maximum observed values. Boxes indicate the interquartile range (IQR), with the median represented by a horizontal line within each box.

The current decentralized study aimed to evaluate the efficacy of a digital therapeutic intervention for depressive and generalized anxiety symptoms. Specifically, we hypothesized that data-driven DHIs, delivered through a structured 16-week digital therapeutic program, would significantly reduce symptoms of depression and anxiety, enhance participants’ quality of life, and serve as an engagement tool, thereby positively affecting retention and adherence rates. The targeted population included individuals with a spectrum of depressive and anxiety symptoms and demonstrated the requisite technical literacy to effectively engage with the digital therapeutic platform within the study framework. Participants were randomly assigned to the intervention and waitlist-control groups to ensure the validity of the outcomes.

The digital therapeutic was delivered via a mobile app. The content and tools within the app have been designed to promote ease of use and support consistent engagement throughout the 16-week intervention period. The primary mechanism of action was CBT in addition to mindfulness, biopsychology and positive psychology. The use of CBT and other psychological frameworks aligns with existing digital health interventions that have shown positive results in reducing symptoms of anxiety and depression44,45,46,47. Additionally, participants used a wrist-worn sensor to continuously capture their physiological data, enabling data-driven DHIs along with key behavioral and emotional insights. This integration of wearable technology mirrors the growing trend in digital health for improving patient outcomes through more tailored and responsive approaches48,49,50,51. Each participant was also supported by a dedicated digital navigator, who helped establish and monitor goals, and directed them towards DHIs that aligned with their objectives.

Based on participant outcomes, we believe the study objectives were successfully achieved. Specifically, we observed significant reductions in both depression and anxiety symptoms, accompanied by an improvement in quality of life. Additionally, the majority of participants demonstrated consistent engagement with the DHIs. These results suggest that the provided data-driven DHIs, called Feel DTx for depressive and generalized anxiety symptoms, is an effective treatment solution.

Focusing on the study’s primary endpoints, the intervention demonstrated notable improvements, with significant between-group differences and consistent progress within the experimental group. The between-group effect sizes observed at mid (r = 0.42 for both PHQ-9 and GAD-7, P < 0.001) were further increased at post, with the within-group comparison showing medium to large effect sizes (PHQ-9: r = 0.83, GAD-7: r = 0.83, P < 0.001, at post). Furthermore, a mixed model analysis indicated significant group × time interactions for both PHQ-9 and GAD-7, which further demonstrates the efficacy of the intervention at the individual level. These results indicate robust reductions in symptoms across both depression and anxiety, aligning with findings from previous studies of digital health interventions for mental health26,52,53,54. Similar improvements were observed for quality of life outcomes, including SWLS and LISAT-11, further reinforcing the intervention’s positive impact. A sensitivity analysis of the LMMs, conducted to examine the impact of missing values, showed that the overall conclusions were robust under various missing value mechanisms (MAR and MNAR). Although the absolute value for the group × time interaction decreased in some models, the direction of the effect remained consistent, underscoring the intervention’s efficacy.

For the secondary endpoints, the quality of life measures showed significant improvements corroborating the clinical outcomes. For example, the SWLS showed moderate effect sizes at mid (r = 0.19, P = 0.03) and large effect sizes at post (r = 0.47, P < 0.001). Similarly, the LISAT-11 demonstrated significant improvements with effect sizes of r = 0.32 (P < 0.001) at mid and r = 0.59 (P < 0.001) at post. Along with the significant group × time interactions obtained by the mixed model analysis of the SWLS and LISAT-11 scores, these results underscore the positive impact of the intervention on participants’ quality of life over time. These findings align with various studies on digital health interventions, which have reported enhanced quality of life and well-being as key outcomes of such programs55,56,57,58,59. However, most studies highlight the need for larger studies to better understand the long-term effectiveness, scalability, and optimal deployment of digital technologies across diverse patient populations and settings.

To further contextualize these findings, we examined the intervention’s effects within symptom ranges typically indicative of diagnostic caseness for MDD and GAD (PHQ-9 ≥ 10 and GAD-7 ≥ 10)60,61. For participants with MDD in the experimental group, a within-group comparison revealed substantial improvements, with effect sizes of r = 0.85 (P< 0.001) at mid-intervention and r = 0.91 (P < 0.001) at post-intervention. Between-group comparisons for MDD showed moderate to large effect sizes, with r = 0.47 (P < 0.001) at mid and r = 0.62 (P < 0.001) at post-intervention. Mixed model analyses for MDD outcomes demonstrated significant group × time interactions both at mid (b = −2.66, CI = −3.51, −1.80, P < 0.001) and post-intervention (b = −3.52, CI = −4.38, −2.65, P < 0.001), with the model explaining 57.4% of the variability in outcomes (conditional R2 = 0.574). Similarly, for participants with GAD in the experimental group, within-group effect sizes were r = 0.91 (P < 0.001) at mid and r = 0.92 (P < 0.001) at post-intervention. Between-group effect sizes for GAD were moderate at mid (r = 0.36, P = 0.004) and large at post-intervention (r = 0.65, P < 0.001). Mixed model analyses for GAD revealed significant group × time interactions at mid (b = −1.89, CI = −2.80, −0.98, P < 0.001) and post-intervention (b = −3.42, CI = −4.34, −2.50, P < 0.001), with the model accounting for 58.5% of variability (conditional R2 = 0.585). These results underscore the robust efficacy of the intervention in reducing symptoms for participants meeting clinical thresholds for MDD and GAD, highlighting its potential for targeted treatment of clinically significant cases.

The clinical effectiveness demonstrated by this study significantly contributes to the growing body of evidence suggesting that DHIs, within the framework of a digital therapeutic, are a viable alternative to treatment as usual for depression and anxiety 24,25. A common theme among additional analyses is the challenge in drawing comparisons across studies due to the extensive variability in study design62,63. Therefore, we focus on studies that have CBT as a treatment model in common with this study. These studies have indicated that a primary mechanism of action within CBT is the “homework” assignments, such as emotion journaling, cognitive restructuring exercises, and behavior activation, supplemented by psychoeducation from a therapist and adjunct educational materials64. As many studies have shown, compliance with homework in face-to-face therapy has been challenging65,66,67. In this study, DHIs served as the homework component for CBT, along with other evidence-based interventions. Remarkably, patients demonstrated significant engagement with the provided homework, with the weekly average of completed DHIs among treatment graduates being 10. This total included an average of 4.61 exercises, 3.92 emotion journals, and 1.49 reviews of assigned educational modules. Notably, the intervention aimed for an average of 2 exercises and 1 review of educational material per week, and participants consistently met or exceeded these targets. Additionally, the feature for labeling cognitive distortions, which was introduced in week 4, was utilized in 38% of emotion journals. This indicates strong engagement with the intervention’s core components.

Additionally, approximately 46% of emotion journals were completed in response to notifications from the wearable device, which monitors participants’ emotional states. This suggests that personalized, data-driven DHIs can effectively support adherence to a primary mechanism of action in CBT66. The objective measurement of compliance with DHIs in this study offers future opportunities to explore the connection between homework engagement and treatment outcomes, in line with previous theoretical frameworks67 and study findings68,69.

Expanding our analysis of participant engagement, we observed that 98% of treatment graduates were active on a weekly basis, averaging 3.4 days of activity, 74.4 minutes in the mobile app, and 95.44% attendance at weekly check-ins. Additionally, the study showed a 71% completion rate. Over the 16-week study period, 72.4% of dropouts occurred within the first 4 weeks, with only 0.07% dropping out after the midpoint of the treatment. The initial dropout rates are consistent with typical patterns observed in adults seeking mental health treatment. However, the very low dropout rate beyond week 8 is atypical70. This high level of retention suggests that digital therapeutics using personalized DHIs may be superior to face-to-face therapy in terms of adherence for the treatment of MDD and GAD.

As we move beyond evaluating clinical effectiveness and treatment engagement, it is crucial to explore what is required in the future to establish clinical meaningfulness of DHIs. The current landscape for consumers, providers, and stakeholders is both exciting and confusing, as more research is needed to differentiate between clinically meaningful advancements and consumer marketing16. First, we must consider the significance of clinical meaningfulness, which can be defined as the relevance of a statistically effective treatment to a real-world application71. However, determining its true impact on an individual has been debated for the past three decades72,73,74. In our view, clinical meaningfulness should encompass several key factors, based on previously established analyses of digital solutions in mental health23,75,76,77,78,79: objective and subjective reduction in symptoms, improvement in daily functioning, ease of access to treatment (including availability and affordability), fidelity and adherence to the intervention’s mechanism of action, usability, and interoperability with healthcare systems and providers. To ensure that digital therapeutics for MDD and GAD are truly clinically significant future research should examine these factors in greater depth. A particularly promising area of exploration involves the use of sensor-based digital health technologies in precision medicine, other than supporting increased engagement and symptom reduction as observed in this study. For instance, wearables could be employed to capture the pathophysiology of individual anxiety and depression, similar to studies in other conditions80,81,82,83,84,85. This approach has the potential to address symptom heterogeneity and provide objective data that can be combined with subjective patient responses to diagnostic questionnaires, thereby offering a more comprehensive understanding of mental health conditions. Additionally, future studies should investigate how digital therapeutics can complement psychotropic medications, further enhancing their effectiveness and enabling more personalized treatment plans. By considering these broader implications, we can better understand how digital health interventions fit into the larger landscape of mental health care and refine their role in the treatment of MDD and GAD.

While the findings of this study are very promising, several limitations should be considered when interpreting the results. First, although the sample size was adequate for our statistical analyses, it may not fully represent the diversity of the broader population. This limitation affects the generalizability of the results, as different demographic and clinical characteristics could influence the effectiveness of the digital therapeutic across more diverse populations. Second, the study’s focus on a 16-week intervention period does not account for the long-term sustainability of the observed symptom improvements and quality of life. Without extended follow-up, it remains unclear whether the benefits of the digital therapeutic persist over time or if participants might experience relapse after the intervention ends. Future studies with longer follow-up periods will be necessary to determine the durability of these effects. Third, the reliance on self-reported data to assess depressive and anxiety symptoms, as well as quality of life introduces potential bias. Participants may have reported their symptoms in a manner that reflects social desirability or inaccurate recall, which could affect the reliability of the results. Fourth, the study may be subject to selection bias, as participants were required to have a certain level of technical literacy to engage with the digital therapeutic. This exclusion of individuals with limited access to or familiarity with digital technologies may have skewed the sample towards those more comfortable using such tools, which could affect the broader applicability of the findings to populations with lower technical literacy or access. Fifth, while the study compares the digital therapeutic to a waitlist-control group, it does not include comparisons with other active treatment modalities, such as face-to-face therapy or pharmacotherapy. This limits the ability to assess the relative effectiveness of the digital therapeutic in comparison to established treatments, which would be valuable for understanding its place in the broader treatment landscape for MDD and GAD. Finally, the comorbidity of depression and anxiety was not specifically examined in this study. As the intervention may affect individuals with both conditions differently, future studies should consider including separate groups for participants with depression, anxiety, and comorbid diagnoses to better understand how the intervention may work for each group.

In conclusion, this study provides compelling evidence for the efficacy of a data-driven digital therapeutic called the Feel DTx, in reducing symptoms of depression and anxiety and improving quality of life among individuals with depressive and generalized anxiety symptoms. The high engagement and adherence underscore the potential of digital health interventions to complement or even surpass traditional treatment modalities. Moreover, the intervention demonstrated robust efficacy, with significant improvements observed for participants meeting clinical thresholds for MDD and GAD, further supporting its potential as a targeted treatment for clinically significant cases. However, while these findings are promising, further research is essential to assess the long-term sustainability of these benefits, investigate the effectiveness of the intervention across more diverse populations, and address limitations related to self-reported data, selection bias, and technological factors. As digital therapeutics continue to evolve, they offer a promising avenue for personalized, accessible, and scalable mental health care, but their integration into clinical practice will require careful consideration of these factors.

This was a prospective, decentralized, randomized waitlist-controlled study assessing the efficacy of a data-driven digital therapeutic, the Feel DTx, for depressive and/or generalized anxiety symptoms. The study design featured two parallel groups: an experimental group receiving the Feel DTx intervention and a waitlist control group. The study was conducted in accordance with the Declaration of Helsinki and approved by the institutional review board of the Neurosciences and Precision Medicine Research Institute “Costas Stefanis” (approval 72/26072021). Written informed consent was obtained from all patients before joining the study.

Participant recruitment and enrollment took place from September 2021 to March 2023. Participants for both groups were recruited via the following channels: (i) candidate referrals from the National and Kapodistrian University of Athens; (ii) social media ads and (iii) word of mouth. The study’s inclusion and exclusion criteria were evaluated based on self-reported candidate responses to an eligibility questionnaire. The validity of these responses was also confirmed during the first introductory check-in. Eligible participants were adults who had mild to severe depressive (PHQ-9≥ 5) and/or generalized anxiety (GAD-7≥ 5) symptoms, and were also required to own an Apple or Android-based smartphone (bring your own device) and be able to provide informed consent. Patients were excluded if they fulfilled any of the following criteria: (i) personality disorders; (ii) psychotic disorders; (iii) bipolar disorder; (iv) eating disorders; (v) suicidal or self-harm thoughts; (vi) psychotropic medication; (vii) substance abuse. These exclusion criteria were assessed via self-report on the eligibility questionnaire, as well as through follow-up clinical interviews conducted by the provider.

All study activities took place remotely. Interested study candidates could express their intent to participate through a custom webpage which provided general information about the study scope and intervention. Candidates were then redirected to an online eligibility assessment form, which included a series of questionnaires assessing the study’s inclusion and exclusion criteria. Eligible individuals were invited to participate in the study, while ineligible candidates were disqualified and received appropriate communication. After obtaining informed consent, the candidate was enrolled in the study and randomly assigned to the experimental or waitlist control group.

Participants of the intervention group were mailed a wearable sensor (see Intervention section below) and underwent a remote enrollment process. During this process, a team member conducted a video call with each participant to introduce the study’s objectives and procedures, provide a demonstration of the mobile app and wearable sensor functionalities and guide them through the app’s installation and registration. Participants of this group then downloaded, installed, and registered to the mobile app. Participants who were allocated to the waitlist control group did not receive any intervention. They were given the opportunity to join the intervention following the end of their involvement in the study. Participants of both groups were followed-up until study completion or withdrawal. It should be noted that no monetary incentive was provided to the participants of both groups.

Patient safety was closely monitored throughout the study by providers and a psychiatrist on the study team using a comprehensive approach to detect any potential harms, adverse events (AEs), or serious adverse events (SAEs). The monitoring process involved systematic evaluations at baseline, mid, and post-intervention to ensure participant safety and well-being. All study participants were regularly assessed for the occurrence of any AEs through both self-reporting and provider-led assessments conducted during weekly check-ins. Participants were encouraged to promptly report any symptoms or health changes to the study team. A structured questionnaire was utilized to identify potential AEs/SAEs, including unanticipated events. Although no AEs or SAEs were observed during the study, all assessments were carefully documented and regularly reviewed by the study team. Periodic reviews of the AE/SAE data were conducted, with reporting to the last author (CP), to ensure compliance with ethical safety standards.

Following baseline assessments and obtaining informed consent, eligible candidates were randomized 1:1 to either the experimental or the control group. Randomization was performed remotely using a computer-generated random number sequence implemented with the random Python package86. For every pair of participants, two random numbers were generated in an independent and identically distributed (i.i.d.) manner using the random method from the associated software. The first random number was used to select one of the participants, and the second indicated the group to which they should be allocated. The remaining participant was then allocated to the opposite group.

To maintain allocation concealment and minimize bias, group allocation was done by C.T., who was not otherwise involved in the intervention delivery or data analysis. Additionally, researchers involved in the analysis of the primary and secondary outcomes were blinded to group assignment to reduce the risk of outcome assessment bias. However, due to the nature of the study, it was not possible to blind the intervention providers, as they were aware of the participant’s group allocation during the intervention.

Participants allocated to the experimental group received the Feel DTx, a data-driven digital therapeutic developed by Feel Therapeutics for the management and treatment of depressive and/or generalized anxiety symptoms in adults. The primary mechanism of action of the intervention is CBT in combination with biopsychology, mindfulness and positive psychology DHIs. These DHIs include clinical interventions, exercises, data-driven journaling and educational content, organized into a structured 16-week program.

For example, in Week 1 (You Are in Charge of Change), participants engage in the SMART Goals exercise (Fig. 9A), where they review the SMART (Specific, Measurable, Achievable, Relevant, Time-bound) framework to set personalized goals for the program. Participants record their goals in the “Notes” section, helping them establish clear objectives and track progress. This exercise is designed to support structured goal-setting and encourage sustained engagement throughout the intervention. Later, in Week 10 (Being in the Present), participants practice Grounding Techniques: Six Ways to Practice Grounding (Fig. 9B). This exercise introduces various grounding strategies that help individuals regain awareness of the present moment and manage distressing thoughts or emotions. Participants are encouraged to try at least one technique and reflect on their experience in the “Notes” section. This exercise aligns with the weekly theme on mindfulness and grounding, reinforcing the program’s emphasis on present-moment awareness. The number of exercises assigned to each participant depends on their individual needs and the stage of the intervention. On average, participants receive two exercises per week, with the flexibility to repeat them as often as needed. Additionally, each participant is provided with one educational module (bundle of education materials) per week, which they can also review as many times as they need. The exercises and educational material were available during the entire week and participants could complete them at their convenience.

Screenshots of two example exercises from the digital intervention. A The SMART Goals exercise (Week 1) helps participants set structured goals using the SMART framework, supporting goal-oriented behavior. B The Grounding Techniques exercise (Week 10) introduces six strategies for enhancing present-moment awareness and emotional regulation.

The first three weeks are foundational, building upon the participants’ knowledge. The following five weeks focus on CBT, biopsychology and positive psychology and the final weeks expand on the TFB cycle, developing skills and resilience. The DHIs promote the development of mental resilience through mind-body awareness, cognitive restructuring, behavior modification, and emotion regulation techniques. The goal is to help participants with self-management, reduce depression and anxiety symptoms, and further improve their quality of life according to their personal objectives in the program.

The DHIs are further enhanced with passively collected from the Feel Emotion Sensor (FES) data and the participant input. The FES is a proprietary multi-sensor wearable device that continuously and unobtrusively collects the participants’ physiological data including Electrodermal Activity (EDA), Heart Rate Variability (HRV), Skin temperature and other relevant metrics. These physiological features are transmitted to the Mental Health Biomarkers platform via the mobile app, where they are processed in real-time to inform personalized interventions42,80,87. Leveraging the data collected from the FES, significant emotion moments are identified, when a participant’s physiological responses indicate a heightened emotional state88,89. These moments are marked by changes in the physiological features such as increased heart rate, decreased HRV or fluctuations in skin conductance. A proprietary machine-learning algorithm detects these patterns, triggering a notification that prompts participants to complete an emotion journaling flow (TFB Cycle)42. This process helps participants reflect on their emotional experiences and track their responses to specific moments of emotional intensity. In addition to the automatic notifications triggered by physiological data, participants also have the option to manually journal any experienced emotions, following the same process. As part of the emotion journaling flow, participants are provided with the opportunity to complete a DHI designed to further support the management of the emotions they are experiencing.

Each week of the program includes a 15-minute personalized check-in with a provider. There are a total of 16 check-ins, administered remotely via video conference embedded in the mobile app. During the introductory check-in, which lasts 45 minutes, the participant establishes motivation to engage in the program and sets program goals. Each check-in has an agenda designed to leverage the participant data to personalize the program, reinforce the weekly content, and review progress toward goals. The provider’s role during check-in is that of a digital facilitator who utilizes the data collected by the mobile app to personalize the program and provide targeted feedback. The provider is also responsible for monitoring progress on the participant’s SMART90 goals and motivating the participant to take action. All providers’ skills and experience align with the scope of work defined by the National Board of Health and Wellness coaches91. Providers are trained on the program guidelines and adhere to the weekly program agenda.

For example, in Week 7, the check-in focuses on reinforcing the participant’s progress, reflecting on their emotional experiences, and setting new goals. The provider begins by reviewing the participant’s emotion journals, noting significant patterns or changes in their emotional states, triggers, thoughts, and behaviours. If the participant has been journaling feelings of frustration or anxiety, the provider may highlight these emotions and discuss any physiological data from the Feel Emotion Sensor (e.g., increased Electrodermal Activity or Heart Rate Variability) to deepen the understanding of how their body responds to these emotions. For instance, if the participant reports heightened anxiety before a specific event, the provider may review the Catching Hot Thoughts exercise, where the participant identifies automatic negative thoughts and reflects on their emotional triggers.

The provider also encourages the participant to reflect on exercises from the previous week that focus on relaxation techniques and cognitive restructuring. For example, the provider may explore how well the participant has practiced focusing on their breath or engaging in self-reflection, asking questions such as: “How did you feel after completing the exercises? Did you notice any changes in your emotional state or physical sensations?”. Then, the provider introduces new content, building on the participant’s progress. For example, they might discuss how to identify deeper cognitive beliefs underlying emotional responses or how to recognize moments when the participant’s thoughts diverge from reality. By making these connections, the provider helps the participant build a stronger awareness of their emotional and thought patterns.

The above components (i.e. DHIs, physiological data, and weekly check-ins) are directly integrated into the Feel DTx mobile app which has been designed to engage participants on their treatment journey and promote behavior activation. The app serves as the Feel DTx digital front door, guiding the participant through program onboarding and program steps. Moreover, the clear and coherent user experience within the app ensures that the participant can complete the DHIs and effortlessly track their progress in the program anytime.

The primary outcomes of this study were the progression of depressive and anxiety symptoms of the experimental and control group participants. These were evaluated using the PHQ-960 and GAD-792 scales at the baseline (week 0), mid (week 8), and post-intervention (week 16). Secondary outcomes included the progression of the participants’ global and domain-specific life satisfaction and were assessed using the Satisfaction with Life Scale (SWLS)93 and the Life Satisfaction Questionnaire (LISAT-11)94, respectively. They were also assessed at the baseline, mid and post-intervention period (16th week). All questionnaires were administered remotely via the mobile app. At the mid and post-intervention evaluation periods (i.e. weeks 8 and 16), the study staff made three attempts to collect the data from participants in both groups.

Assuming a medium to large effect size (f = 0.3), a significance level of a = 0.05 and a power of 0.8 for two groups and three assessment times, a repeated measures ANOVA test yielded a minimum sample size of 55 participants per group (i.e., experimental and control). To account for participants who might drop out or not respond to either the middle or end-of-intervention assessments, 100 participants per group were finally enrolled, ensuring that the study is sufficiently powered.

We used descriptive statistics (mean values, proportions) in order to capture the main aspects of participant recruitment and demographic characteristics. Furthermore, we utilized a two-sample, two-tailed Kolmogorov-Smirnov test95 to compare the age distribution and chi-square tests96 to assess differences in the gender, depressive and generalized anxiety symptom severity distributions in the two groups. The threshold for statistical significance was set at a P-value of .05.

For the primary (i.e. PHQ-9 and GAD-7 scores) and secondary (i.e. LISAT-11 and SWLS scores) study outcomes, we utilized the mean values and the corresponding standard errors to describe the distributions of the scores for each group at each evaluation period (i.e., baseline, mid, and post). In addition, we reported the mean change in scores between mid and baseline and between post and baseline separately for each group, along with their 95% confidence intervals (CI). To analyze differences within and between groups, we performed two-tailed statistical tests. For within-group comparisons relative to baseline at mid and post, paired samples tests were utilized. For between-group comparisons at each evaluation period, independent samples tests were employed. Before selecting the specific statistical tests, we assessed whether the distributions of scores and their differences met the assumption of normality using the Kolmogorov-Smirnov test97. The results indicated that the distributions did not meet normality assumptions. As a result, we opted for non-parametric tests. Specifically, for within-group comparisons, we used the two-tailed Wilcoxon signed-rank test98, which evaluates whether the distribution of differences is symmetrical around zero. For between-group comparisons, we employed the two-tailed Mann-Whitney U test99, which assesses whether there is a difference in central tendency between groups. These tests are non-parametric and do not assume any specific form for the underlying data distributions. For all statistical tests, the cutoff P-value for statistical significance was set at 0.05. To enhance interpretability, for each statistical test we reported effect sizes using the rank biserial correlation (r), which resembles the difference between the proportion of the data that, according to the statistical test, shows favorable and unfavorable results100. For the PHQ-9 and GAD-7 questionnaires, a favorable result corresponded to a decrease in the respective score, whereas for the LISAT-11 and SWLS, it was associated with an increase in the score. Finally, minimal clinically important differences (MCID) were defined as changes of at least 5 points for the PHQ-9 scores101 and at least 4 points for the GAD-7 scores102.

Additionally, we investigated the impact of various factors (e.g., time, group, gender, age, participant) and accounted for correlations between the data by treating the different evaluation periods (baseline, mid and post) as repeated measurements. We employed linear mixed models (LMM) following an Intention-to-Treat (ITT) approach as the main analysis framework for both primary and secondary outcomes. A separate model has been constructed for each primary and secondary outcome. We have considered the gender, age, group and time variables as fixed effects, along with the interaction of the first three with time, while random intercept for each participant has been included. The time variable has been encoded with control encoding and we, thus, create two auxiliary variables, one regarding the comparison between mid and baseline (variable t2,1) and one regarding the comparison between post and baseline (variable t3,1). The group and gender variables have been encoded using deviation encoding (group=1 for the experimental group and gender=1 for females) and thus the corresponding coefficients regard main effects. The age variable was treated as continuous. The LMMs were fitted using all available data and with restricted maximum likelihood. For each model, we report the estimates, 95% CIs and P-value for the coefficients of the fixed effects. Moreover, as a measure of the variability among the grouping levels (participants), we report the standard deviation of the random effects as captured by the model, along with the 95% profile likelihood CI. For the explanatory power of the model, we report the conditional R2 which captures the amount of explained variance for the entire model including both fixed and random effects.

Our dataset contains missing data in the clinical and life satisfaction questionnaires responses. More specifically, the missing values occur in the 8th or 16th week assessment, and come from dropouts or participants who completed the program but did not fill in the respective questionnaires. In total, the missing values account for 13.6% of our data. Depending on the mechanism of missingness103, the missing data can either be missing completely at random (MCAR), missing at random (MAR) or missing not at random (MNAR). In this work, we explore the impact of missing values under different scenarios103 by conducting the following sensitivity analyses104: i) complete case analysis where we utilize data from participants that completed the questionnaires in all evaluation periods (i.e. participants with no missing data) and ii) analysis utilizing multiple imputations (MI) for the missing data. The complete case analysis assumes MCAR data and in case this assumption is not met, significant bias is introduced in the results. On the other hand, MI treats data as MAR (imputation is done based on other observations in the dataset) and reduces the underestimation of the variance of the effect. To also account for the MNAR case, we employ a sensitivity parameter k to multiply the imputed values that need to be treated as MNAR105. Therefore, the imputed values can either increase (k > 1) or decrease (k < 1), while for k = 1 they do not change compared to the MAR case. The majority of the missing values (approximately 10% of the total data) regarded the dropouts from the experimental group and were, thus, expected to have the most significant impact on our results. Consequently, we treated this subset of the missing data under MAR and the different MNAR scenarios (k ∈ {1, 1.2, 1.5, 2, 2.5}), while the rest of the missing data were treated as MAR (k = 1). For the sensitivity analyses, we focus on the group x time interaction at both mid and post time points, which corresponds to the effect of the intervention on the two groups.

The exploratory analysis and statistical tests were carried out in Python programming language using the open-source scipy package106. The LMMs were fitted using the lme4 package107 for the R programming language. Multiple imputation was performed using the mice108 package in R with predictive mean matching and m = 50 imputations. To maintain a family-wise Type I error rate of 5% for the primary outcomes, we applied a Bonferroni correction to account for multiple testing across the two primary outcomes (PHQ-9 and GAD-7) and two endpoints (mid and post). This adjustment set the significance level to 0.0125 (0.05/4) for the LMMs and corresponding sensitivity analyses. For the secondary outcomes, the significance level was set at .05.

Regarding participant engagement, we report the average weekly active participants, followed by the averages of the weekly mean values for the following metrics: number of days per week during which the participants engaged with any of the components of the intervention, number of attended weekly provider check-ins, time spent in the app, number of completed DHIs, number of completed exercises, emotion journals and reviews of assigned educational content, as well as for the percentage of participants that wore the sensor. We report the averages following a per protocol (i.e. including only participants that completed the study109) and an ITT approach. For the latter, a value of 0 was assumed for all the metrics of interest and for all dates after the last day of the intervention for the participants that dropped out.

Ethical approval for this trial was received from the Neurosciences and Precision Medicine Research Institute “Costas Stefanis” (approval 72/26072021). The described procedures were conducted according to the ethical principles of the Declaration of Helsinki. All study participants provided written consent to participate in the study upon being informed of the aims, the voluntary nature of their participation, their right to withdraw at any time without negative consequences as well as the anonymization and publication of their data. After obtaining written consent, the patient was enrolled in the study.

All participant data collected during the study was stored and processed in a secure cloud-based infrastructure in Europe. For privacy reasons and to adhere to the General Data Protection Regulation, all data was pseudonymized before any processing or insight extraction.

The data and materials will be made available by the authors upon request, without undue reservation.

The underlying code for this study will be available from the corresponding authors on reasonable request.

Davenport, S., Darby, B., Gray, T. & Spear, C. Access across America: State-by-state insights into the accessibility of care for mental health and substance use disorders. Tech. Rep., Milliman (2023).

Kazdin, A. E. Addressing the treatment gap: A key challenge for extending evidence-based psychosocial interventions. Behav. Res. Ther. 88, 7–18 (2017).

Article Google Scholar

James, S. L. et al. Global, regional, and national incidence, prevalence, and years lived with disability for 354 diseases and injuries for 195 countries and territories, 1990-2017: a systematic analysis for the global burden of disease study 2017. Lancet 392, 1789–1858 (2018).

Article Google Scholar

World Health Organization (WHO). Depressive disorder (depression) (2023).

Institute for Health Metrics and Evaluation (IHME). Global Burden of Disease (GBD) (2024).

Vos, T. et al. Global burden of 369 diseases and injuries in 204 countries and territories, 1990-2019: a systematic analysis for the global burden of disease study 2019. Lancet 396, 1204–1222 (2020).

Article Google Scholar

World Health Organization (WHO). Anxiety disorders (2023).

Rapaport, M. H., Clary, C., Fayyad, R. & Endicott, J. Quality-of-life impairment in depressive and anxiety disorders. Am. J. Psychiatry 162, 1171–1178 (2005).

Article Google Scholar

Herrman, H. et al. Reducing the global burden of depression: a lancet-world psychiatric association commission. Lancet 393, e42–e43 (2019).

Article Google Scholar

Athira, V. K. et al. An overview of the heterogeneity of major depressive disorder: Current knowledge and future prospective. Curr. Neuropharmacol. 18, 168–187 (2020).

Article CAS PubMed Central Google Scholar

Drzewiecki, C. & Fox, A. Understanding the heterogeneity of anxiety using a translational neuroscience approach. Cogn. Affect Behav. Neurosci. 24, 228–245 (2024).

Article PubMed Central Google Scholar

Huibers, M. J. H., Lorenzo-Luaces, L., Cuijpers, P. & Kazantzis, N. On the road to personalized psychotherapy: A research agenda based on cognitive behavior therapy for depression. Front. Psychiatry 11 (2021).

Cohen, Z. D. & DeRubeis, R. J. Treatment selection in depression. Annu. Rev. Clin. Psychol. 14, 209–236 (2018).

Article Google Scholar

Szuhany, K. L. & Simon, N. M. Anxiety Disorders: A Review. JAMA 328, 2431–2445 (2022).

Article CAS Google Scholar

Marsch, L. Digital health data-driven approaches to understand human behavior. Neuropsychopharmacol. 46, 191–196 (2021).

Article Google Scholar

Istepanian, R. S. H. Mobile health (m-health) in retrospect: The known unknowns. Int. J. Environ. Res. Public Health 19 (2022).

Harty, S. et al. Implementing digital mental health interventions at scale: one-year evaluation of a national digital CBT service in Ireland. Int. J. Ment. Health Syst. 17 (2023).

Carl, J. R. et al. Efficacy of digital cognitive behavioral therapy for moderate-to-severe symptoms of generalized anxiety disorder: A randomized controlled trial. Depression Anxiety 37, 1168–1178 (2020).

Article Google Scholar

Venkatesan, A., Rahimi, L., Kaur, M. & Mosunic, C. Digital cognitive behavior therapy intervention for depression and anxiety: Retrospective study. JMIR Ment. Health 7, e21304 (2020).

Article PubMed Central Google Scholar

Szigethy, E. M. et al. A digital cognitive-behavioral intervention for depression and anxiety among adolescents and young adults. Psychiatr. Serv. 74, 437–440 (2023).

Article Google Scholar

Kristina, F. & Majella, B. The key principles of cognitive behavioural therapy. InnovAiT 6, 579–585 (2013).

Article Google Scholar

Firth, J. et al. The efficacy of smartphone-based mental health interventions for depressive symptoms: a meta-analysis of randomized controlled trials. World Psychiatry 16, 287–298 (2017).

Article PubMed Central Google Scholar

Himle, J. A., Weaver, A., Zhang, A. & Xiang, X. Digital mental health interventions for depression. Cogn. Behav. Pract. 29, 50–59 (2022).

Article Google Scholar

Moshe, I. et al. Digital interventions for the treatment of depression: A meta-analytic review. Psychol. Bull. 147, 749–786 (2021).

Article Google Scholar

Pauley, D., Cuijpers, P., Papola, D., Miguel, C. & Karyotaki, E. Two decades of digital interventions for anxiety disorders: a systematic review and meta-analysis of treatment effectiveness. Psychological Med. 53, 567–579 (2023).

Google Scholar

Tokgöz, P., Hrynyschyn, R., Hafner, J., Schönfeld, S. & Dockweiler, C. Digital health interventions in prevention, relapse, and therapy of mild and moderate depression: Scoping review. JMIR Ment. Health 8, e26268 (2021).

Article PubMed Central Google Scholar

van Orden, M. L. et al. Preliminary effects of a digital mental health intervention for depression and anxiety. Clin. eHealth 5, 44–51 (2022).

Article Google Scholar

Ng, M. M., Firth, J., Minen, M. & Torous, J. User engagement in mental health apps: A review of measurement, reporting, and validity. Psychiatr. Serv. 70, 538–544 (2019).

Article PubMed Central Google Scholar

Lipschitz, J., Pike, C., Hogan, T., Murphy, S. & Burdick, K. The engagement problem: A review of engagement with digital mental health interventions and recommendations for a path forward.Curr. Treat Options Psychiatry 10, 119–135 (2023).

Article PubMed Central Google Scholar

Patel, S. et al. The acceptability and usability of digital health interventions for adults with depression, anxiety, and somatoform disorders: Qualitative systematic review and meta-synthesis. J. Med. Internet Res. 22, e16228 (2020).

Article PubMed Central Google Scholar

Saleem, M. et al. Understanding engagement strategies in digital interventions for mental health promotion: Scoping review. JMIR Ment. Health 8, e30000 (2021).

Article PubMed Central Google Scholar

Kooiman, B. E. A. M. et al. Congruency of multimodal data-driven personalization with shared decision-making for stayfine: individualized app-based relapse prevention for anxiety and depression in young people. Front. Psychiatry 14 (2023).

Liang, Y., Zheng, X. & Zeng, D. D. A survey on big data-driven digital phenotyping of mental health. Inf. Fusion 52, 290–307 (2019).

Article Google Scholar

Prakash, J., Chaudhury, S. & Chatterjee, K. Digital phenotyping in psychiatry: When mental health goes binary. Ind. Psychiatry J. 30, 191–192 (2021).

Article PubMed Central Google Scholar

Torous, J. et al. The growing field of digital psychiatry: current evidence and the future of apps, social media, chatbots, and virtual reality. World Psychiatry 20, 318–335 (2021).

Article PubMed Central Google Scholar

Ahmed, A. et al. Wearable devices for anxiety & depression: A scoping review. Computer Methods Prog. Biomedicine Update 3, 100095 (2023).

Article Google Scholar

Erol, T., Mendi, A. F. & Doğan, D. The digital twin revolution in healthcare. 2020 4th International Symposium on Multidisciplinary Studies and Innovative Technologies (ISMSIT) 1–7 (2020).

Spitzer, M., Dattner, I. & Zilcha-Mano, S. Digital twins and the future of precision mental health. Front. Psychiatry 14, 1082598 (2023).

Article PubMed Central Google Scholar

Fagherazzi, G. Deep digital phenotyping and digital twins for precision health: Time to dig deeper. J. Med. Internet Res. 22, e16770 (2020).

Article PubMed Central Google Scholar

Scala, J. J., Ganz, A. B. & Snyder, M. P. Precision medicine approaches to mental health care. Physiology 38, 82–98 (2023).

Article CAS Google Scholar

Hsin, H. et al. Transforming psychiatry into data-driven medicine with digital measurement tools. npj Digital Med. 1 (2018).

Tsirmpas, C. et al. Feasibility, engagement, and preliminary clinical outcomes of a digital biodata-driven intervention for anxiety and depression. Front. Digital Health 4 (2022).

Wilhelm, S. et al. Cognitive-behavioral therapy in the digital age: Presidential address. Behav. Ther. 51, 1–14 (2020).

Article Google Scholar

Lepping, P. et al. Clinical relevance of findings in trials of CBT for depression. Eur. Psychiatry 45, 207–211 (2017).

Article CAS Google Scholar

DiMauro, J., Domingues, J., Fernandez, G. & Tolin, D. F. Long-term effectiveness of cbt for anxiety disorders in an adult outpatient clinic sample: A follow-up study. Behav. Res. Ther. 51, 82–86 (2013).

Article Google Scholar

Oud, M. et al. Effectiveness of cbt for children and adolescents with depression: A systematic review and meta-regression analysis. Eur. Psychiatry 57, 33–45 (2019).

Article Google Scholar

Springer, K. S., Levy, H. C. & Tolin, D. F. Remission in cbt for adult anxiety disorders: A meta-analysis. Clin. Psychol. Rev. 61, 1–8 (2018).

Article Google Scholar

Sheikh, M., Qassem, M. & Kyriacou, P. A. Wearable, environmental, and smartphone-based passive sensing for mental health monitoring. Front. Digital Health 3 (2021).

Sano, A. et al. Identifying objective physiological markers and modifiable behaviors for self-reported stress and mental health status using wearable sensors and mobile phones: Observational study. J. Med Internet Res 20, e210 (2018).

Article PubMed Central Google Scholar

Hickey, B. A. et al. Smart devices and wearable technologies to detect and monitor mental health conditions and stress: A systematic review. Sensors 21 (2021).

Patel, S. & Saunders, K. E. Apps and wearables in the monitoring of mental health disorders. Br. J. Hospital Med. 79, 672–675 (2018).

Article Google Scholar

Lewkowitz, A. K. et al. The effect of digital health interventions on postpartum depression or anxiety: a systematic review and meta-analysis of randomized controlled trials. Am. J. Obstet. Gynecol. 230, 12–43 (2024).

Article Google Scholar

Moshe, I. et al. Digital interventions for the treatment of depression: A meta-analytic review. Psychological Bull. 147, 749–786 (2021).

Article Google Scholar

Kaihara, T. et al. Efficacy of digital health interventions on depression and anxiety in patients with cardiac disease: a systematic review and meta-analysis. Eur. Heart J. - Digital Health 3, 445–454 (2022).

Article Google Scholar

Castro, R. et al. What are we measuring when we evaluate digital interventions for improving lifestyle? A scoping meta-review. Front. Public Health 9 (2022).

Obrero-Gaitán, E. et al. Digital and interactive health interventions minimize the physical and psychological impact of breast cancer, increasing women’s quality of life: A systematic review and meta-analysis. Cancers 14 (2022).

Victoria-Castro, A. M. et al. Impact of Digital Health Technology on Quality of Life in Patients With Heart Failure. JACC: Heart Fail. 12, 336–348 (2024).

CAS Google Scholar

Pimentel-Parra, G. A., Soto-Ruiz, M. N., San Martín-Rodríguez, L., Escalada-Hernández, P. & García-Vivar, C. Effectiveness of digital health on the quality of life of long-term breast cancer survivors: A systematic review. Semin. Oncol. Nurs. 39, 151418 (2023).

Article Google Scholar

Gentili, A. et al. The cost-effectiveness of digital health interventions: A systematic review of the literature. Front. Public Health 10 (2022).

Kroenke, K., Spitzer, R. & Williams, J. The PHQ-9. J. Gen. Intern. Med. 16, 606–613 (2001).

Article CAS PubMed Central Google Scholar

Löwe, B. et al. Validation and standardization of the Generalized Anxiety Disorder Screener (GAD-7) in the general population. Med. care 46, 266–274 (2008).

Article Google Scholar

Kambeitz-Ilankovic, L. et al. A systematic review of digital and face-to-face cognitive behavioral therapy for depression. NPJ Digital Med. 5 (2022).

Luo, C. et al. A comparison of electronically-delivered and face to face cognitive behavioural therapies in depressive disorders: A systematic review and meta-analysis. eClin. Med. 24 (2020).

Kazantzis, N., Brownfield, N. R., Mosely, L., Usatoff, A. S. & Flighty, A. J. Homework in cognitive behavioral therapy: A systematic review of adherence assessment in anxiety and depression (2011-2016). Psychiatr. Clin. North Am. 40, 625–639 (2017).

Article Google Scholar

Helbig, S. & Fehm, L. Problems With Homework in CBT: Rare Exception or Rather Frequent? Behavioural Cogn. Psychother. 32, 291–301 (2004).

Article Google Scholar

Gaudiano, B. A. Cognitive-behavioural therapies: achievements and challenges. Evid.-Based Ment. Health 11, 5–7 (2008).

Article Google Scholar

Kazantzis, N. & Miller, A. R. A comprehensive model of homework in cognitive behavior therapy. Cogn. Ther. Res. 46, 247–257 (2022).

Article Google Scholar

Haller, E. & Watzke, B. The role of homework engagement, homework-related therapist behaviors, and their association with depressive symptoms in telephone-based cbt for depression. Cogn. Ther. Res. 45, 224–235 (2021).

Article Google Scholar

Mausbach, B. T. et al. The relationship between homework compliance and therapy outcomes: An updated meta-analysis. Cogn. Ther. Res. 34, 429–438 (2010).

Article Google Scholar

Swift, J. K. & Greenberg, R. P. Premature discontinuation in adult psychotherapy: a meta-analysis. J. Consulting Clin. Psychol. 80, 547–559 (2012).

Article Google Scholar

Sharma, H. Statistical significance or clinical significance? a researcher’s dilemma for appropriate interpretation of research results. Saudi J. Anaesth. 15, 431–434 (2021).

Article PubMed Central Google Scholar

Jacobson, N. S. & Truax, P. Clinical significance: a statistical approach to defining meaningful change in psychotherapy research. J. Consulting Clin. Psychol. 59, 12–19 (1991).

Article CAS Google Scholar

Eisen, S. V., Ranganathan, G., Seal, P. & Spiro, r,A. Measuring clinically meaningful change following mental health treatment. J. Behav. Health Serv. Res. 34, 272–289 (2007).

Article Google Scholar

Vaganian, L. et al. Critical consideration of assessment methods for clinically significant changes of mental distress after psycho-oncological interventions. Int. J. Methods Psychiatr. Res. 29, e1821 (2020).

Article PubMed Central Google Scholar

Gan, D. Z. Q., McGillivray, L., Han, J., Christensen, H. & Torok, M. Effect of engagement with digital interventions on mental health outcomes: A systematic review and meta-analysis. Front. Digital Health 3 (2021).

Morrison, C. Digital mental health: Findings of a desktop horizon scan for global leaders & digital innovation opportunities. (University of Strathclyde, Glasgow, 2022).

Murray, E. et al. Evaluating digital health interventions: Key questions and approaches. Am. J. Preventive Med. 51, 843–851 (2016).

Article Google Scholar

Yeager, C. M. & Benight, C. C. If we build it, will they come? issues of engagement with digital health interventions for trauma recovery. mHealth 4, 37 (2018).

Article PubMed Central Google Scholar

ICER-PHTI. Assessment Framework for Digital Health Technologies. https://phti.com/wp-content/uploads/sites/3/2023/09/ICER-PHTI-Assessment-Framework-for-Digital-Health-Technologies.pdf (Accessed: 2024-08-20).

Andrikopoulos, D. et al. Machine learning-enabled detection of attention-deficit/hyperactivity disorder with multimodal physiological data: a case-control study. BMC Psychiatry 24 (2024).

Greco, A., Valenza, G., Lanata, A., Rota, G. & Scilingo, E. P. Electrodermal activity in bipolar patients during affective elicitation. IEEE J. Biomed. Health Inform. 18, 1865–1873 (2014).

Article Google Scholar

Maxim, O., Yashar, B., John, M. K., Georgios, P. & Gari, D. C. Objective identification and analysis of physiological and behavioral signs of schizophrenia. J. Ment. Health 24, 276–282 (2015).

Article Google Scholar

Mourtakos, S. et al. Resilience of the hellenic navy seals assessed by heart rate variability during cognitive tasks. Physiol. Behav. 239, 113437 (2021).

Article CAS Google Scholar

Mourtakos, S. et al. Assessment of resilience of the hellenic navy seals by electrodermal activity during cognitive tasks. Int. J. Environ. Res. Public Health 18, 4384 (2021).

Article PubMed Central Google Scholar

Mourtakos, S. et al. Endocannabinoids and heart rate variability alterations after exposure to prolonged intensive physical exercise of the hellenic navy seals. Int. J. Environ. Res. Public Health 19, 28 (2021).

Article PubMed Central Google Scholar

Van Rossum, G. & Drake Jr, F. L. Python tutorial (Centrum voor Wiskunde en Informatica Amsterdam, The Netherlands, 1995).

Tsirmpas, C. et al. A digital mental health support program for depression and anxiety in populations with attention-deficit/hyperactivity disorder: Feasibility and usability study. JMIR Form. Res 7, e48362 (2023).

Article PubMed Central Google Scholar

Egger, M., Ley, M. & Hanke, S. Emotion Recognition from Physiological Signal Analysis: A Review. Electron. Notes Theor. Computer Sci. 343, 35–55 (2019).

Article Google Scholar

Bota, P. J., Wang, C., Fred, A. L. N. & Plácido Da Silva, H. A review, current challenges, and future possibilities on emotion recognition using machine learning and physiological signals. IEEE Access 7, 140990–141020 (2019).

Article Google Scholar

Bailey, R. R. Goal setting and action planning for health behavior change. Am. J. Lifestyle Med. 13, 615–618 (2019).

Article Google Scholar

National Board for Health & Wellness Coaching. Scope of practice (2024).

Spitzer, R. L., Kroenke, K., Williams, J. B. W. & Löwe, B. A Brief Measure for Assessing Generalized Anxiety Disorder: The GAD-7. Arch. Int. Med. 166, 1092–1097 (2006).

Article Google Scholar

Diener, E., Emmons, R., Larsen, R. & Griffin, S. The Satisfaction With Life Scale. J. Personality Assess 49, 71–75 (1985).

Article CAS Google Scholar

Fugl-Meyer, A. R., Bränholm, I.-B. & Fugl-Meyer, K. S. Happiness and domain-specific life satisfaction in adult northern Swedes. Clin. Rehab. 5, 25–33 (1991).

Article Google Scholar

Yadolah, D.Kolmogorov–Smirnov Test, 283–287 (Springer New York, New York, NY, 2008).

Wuensch, K. L.Chi-Square Tests, 252–253 (Springer Berlin Heidelberg, Berlin, Heidelberg, 2011).

Jr, F. J. M. The Kolmogorov-Smirnov Test for Goodness of Fit. J. Am. Statistical Assoc. 46, 68–78 (1951).

Rey, D. & Neuhäuser, M. Wilcoxon-signed-rank test. In Lovric, M. (ed.) International Encyclopedia of Statistical Science, 1658–1659 (Springer, Berlin, Germany, 2014).

McKnight, P. E. & Najab, J.Mann-Whitney U Test, 1–1 (John Wiley & Sons, Ltd, 2010).

Kerby, D. S. The simple difference formula: an approach to teaching nonparametric correlation. Comprehensive Psychol. 3, 11.IT.3.1 (2014).

Kroenke, K. Enhancing the clinical utility of depression screening. CMAJ 184, 281–282 (2012).

Article PubMed Central Google Scholar

Toussaint, A. et al. Sensitivity to change and minimal clinically important difference of the 7-item generalized anxiety disorder questionnaire (gad-7). J. Affect. Disord. 265, 395–401 (2020).

Article Google Scholar

Dziura, J., Post, L., Zhao, Q., Fu, Z. & Peduzzi, P. Strategies for dealing with missing data in clinical trials: from design to analysis. Yale J. Biol. Med. 86, 343–358 (2013).

PubMed Central Google Scholar

Blozis, S. Bayesian pattern-mixture models for dropout and intermittently missing data in longitudinal data analysis. Behav. Res 56, 1953–1967 (2024).

Article Google Scholar

Fiero, M., Hsu, C. & Bell, M. A pattern-mixture model approach for handling missing continuous outcome data in longitudinal cluster randomized trials. Stat. Med. 36, 4094–4105 (2017).

Article PubMed Central Google Scholar

Virtanen, P. et al. SciPy 1.0: Fundamental Algorithms for Scientific Computing in Python. Nat. Methods 17, 261–272 (2020).

Article CAS PubMed Central Google Scholar

Douglas, B., Martin, M., Ben, B. & Steve, W. Fitting linear mixed-effects models using lme4. J. Stat. Softw. 67, 1–48 (2015).

Google Scholar

van Buuren, S. & Groothuis-Oudshoorn, K. mice: Multivariate imputation by chained equations in r. J. Stat. Softw. 45, 1–67 (2011).

Article Google Scholar

Shah, P. B. Intention-to-treat and per-protocol analysis. CMAJ 183, 696–696 (2011).

Article PubMed Central Google Scholar

Download references

The authors thank Maria Nikolakopoulou for her significant contribution to the study and for providing insightful input and support during the manuscript preparation and revision process.

Feel Therapeutics Inc., San Francisco, CA, USA

Panagiotis Fatouros, Charalampos Tsirmpas, Dimitrios Andrikopoulos & Sharon Kaplow

First Department of Psychiatry, Aiginition Hospital Medical School National and Kapodistrian University of Athens, Athens, Greece

Konstantinos Kontoangelos

Neurosciences and Precision Medicine Research Institute “Costas Stefanis”, University Mental Health, Athens, Greece

Konstantinos Kontoangelos & Charalabos Papageorgiou

You can also search for this author in PubMed Google Scholar

You can also search for this author in PubMed Google Scholar

You can also search for this author in PubMed Google Scholar

You can also search for this author in PubMed Google Scholar

You can also search for this author in PubMed Google Scholar

You can also search for this author in PubMed Google Scholar

P.F., C.T., K.K., and C.P. contributed to the conceptualization and methodology of the study, whereas S.K., K.K., and C.P. contributed to the selection of measures. P.F. and D.A. were responsible for the data analyses. P.F., D.A., and S.K. wrote the first draft of the paper and all authors edited multiple drafts and reviewed the final manuscript.

Correspondence to Panagiotis Fatouros or Dimitrios Andrikopoulos.

C.T. is employed by Feel Therapeutics Inc., receives a salary, and owns a large share of the company stocks. D.A., P.F., and S.K. are employed by Feel Therapeutics Inc., receive a salary, and own options in the company. The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.

Reprints and permissions

Fatouros, P., Tsirmpas, C., Andrikopoulos, D. et al. Randomized controlled study of a digital data driven intervention for depressive and generalized anxiety symptoms. npj Digit. Med. 8, 113 (2025). https://doi.org/10.1038/s41746-025-01511-7

Download citation

Received: 18 September 2024

Accepted: 12 February 2025

Published: 19 February 2025

DOI: https://doi.org/10.1038/s41746-025-01511-7

Anyone you share the following link with will be able to read this content:

Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative