Brief Digital Interventions for Psychological Distress: An AI-Enhanced Response-Adaptive Randomized Clinical Trial

Jill Newby; Sunil Gupta; Leonard Hoon; WuYi Zheng; Alexis E Whitton; Kit Huckvale; Eileen Stech; Andrew Mackinnon; Manisha Senadeera; Artur Shvetcov; Julian Berk; Aimy Slade; Michael J Spoelma; Jin Han; Joanne R Beames; Rena Logothetis; Omar Dabash; Stefanus Kurniawan; Rajesh Vasa; Kon Mouzakis; Stuart Cameron; Akash Agarwal; Joshua Asbury; Joost Funke Kupper; Aliza Werner-Seidler; Simon Rosenbaum; Henry Cutler; Svetha Venkatesh; Helen Christensen

doi:10.1001/jamanetworkopen.2025.40502

. 2025 Oct 31;8(10):e2540502. doi: 10.1001/jamanetworkopen.2025.40502

Brief Digital Interventions for Psychological Distress

An AI-Enhanced Response-Adaptive Randomized Clinical Trial

Jill Newby ^1,^2,^✉, Sunil Gupta ³, Leonard Hoon ³, WuYi Zheng ¹, Alexis E Whitton ¹, Kit Huckvale ^1,⁴, Eileen Stech ^1,², Andrew Mackinnon ¹, Manisha Senadeera ³, Artur Shvetcov ^1,^5,⁶, Julian Berk ³, Aimy Slade ¹, Michael J Spoelma ¹, Jin Han ^1,⁷, Joanne R Beames ^1,⁸, Rena Logothetis ³, Omar Dabash ^1,⁴, Stefanus Kurniawan ³, Rajesh Vasa ³, Kon Mouzakis ³, Stuart Cameron ³, Akash Agarwal ³, Joshua Asbury ³, Joost Funke Kupper ³, Aliza Werner-Seidler ^1,⁵, Simon Rosenbaum ⁵, Henry Cutler ⁹, Svetha Venkatesh ³, Helen Christensen ^1,⁵

¹Black Dog Institute, University of New South Wales Sydney, New South Wales, Sydney, Australia

²School of Psychology, University of New South Wales Sydney, New South Wales, Sydney, Australia

³Applied Artificial Intelligence Initiative, Deakin University, Victoria, Melbourne, Australia

⁴Centre for Digital Transformation of Health, The University of Melbourne, Victoria, Melbourne, Australia

⁵Discipline of Psychiatry and Mental Health, University of New South Wales Sydney, New South Wales, Sydney, Australia

⁶Department of Psychological Medicine, Sydney Children’s Hospitals Network, Sydney, New South Wales, Australia

⁷Division of Arts and Sciences and Centre for Global Health Equity, New York University Shanghai, Shanghai, People’s Republic of China

⁸Centre for Contextual Psychiatry, Department of Neuroscience, KU Leuven, Leuven, Belgium

⁹Centre for the Health Economy, Macquarie University, New South Wales, Sydney, Australia

Accepted for Publication: August 28, 2025.

Published: October 31, 2025. doi:10.1001/jamanetworkopen.2025.40502

^✉

Corresponding Author: Jill Newby, PhD, Black Dog Institute, University of New South Wales Sydney, Hospital Road Randwick, NSW 2031, Australia (j.newby@unsw.edu.au).

Author Contributions: Drs Newby and Gupta had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. Drs Newby and Gupta were co–first authors. Dr Christensen was senior author.

Concept and design: Newby, Gupta, Huckvale, Stech, Mackinnon, Han, Beames, Logothetis, Mouzakis, Rosenbaum, Cutler, Venkatesh, Christensen.

Acquisition, analysis, or interpretation of data: Newby, Gupta, Hoon, Zheng, Whitton, Stech, Mackinnon, Senadeera, Shvetcov, Berk, Slade, Spoelma, Dabash, Kurniawan, Vasa, Cameron, Agarwal, Asbury, Funke Kupper, Werner-Seidler, Rosenbaum.

Drafting of the manuscript: Newby, Gupta, Whitton, Senadeera, Berk, Spoelma, Beames, Dabash, Asbury, Venkatesh, Christensen.

Critical review of the manuscript for important intellectual content: Newby, Gupta, Hoon, Zheng, Whitton, Huckvale, Stech, Mackinnon, Shvetcov, Berk, Slade, Spoelma, Han, Beames, Logothetis, Dabash, Kurniawan, Vasa, Mouzakis, Cameron, Agarwal, Funke Kupper, Werner-Seidler, Rosenbaum, Cutler, Christensen.

Statistical analysis: Newby, Gupta, Whitton, Mackinnon, Senadeera, Shvetcov, Berk, Spoelma, Cameron, Rosenbaum.

Obtained funding: Newby, Gupta, Hoon, Han, Beames, Werner-Seidler, Cutler, Christensen.

Administrative, technical, or material support: Newby, Hoon, Zheng, Whitton, Stech, Slade, Han, Beames, Logothetis, Dabash, Kurniawan, Vasa, Cameron, Agarwal, Funke Kupper, Rosenbaum.

Supervision: Newby, Gupta, Mouzakis, Venkatesh.

Conflict of Interest Disclosures: Dr Newby reported receiving grants from Australian National Health and Medical Research Council, Australian Medical Research Future Fund, Knowledge Partnership Platform Australia-Indonesia (KONESKI), Wellcome Trust, National Institute of Health Research, and HCF Research Foundation outside the submitted work. Dr Gupta reported receiving grants from Deakin University during the conduct of the study. Dr Whitton reported receiving grants from Wellcome Trust outside the submitted work. Dr Berk reported receiving grants from Deakin University during the conduct of the study. Dr Han reported receiving grants from Foundation for Australia-China Relations, the Universitas 21 Health Sciences Group, and Shanghai Municipal Education Commission; and personal fees from Asian Development Bank, the Rici Foundation, Shanghai Mental Health Center, and the Australian Embassy Beijing outside the submitted work. No other disclosures were reported.

Funding/Support: This work was supported by the Commonwealth of Australia Medical Research Future Fund grant MRFAI000028 Optimising Treatments in Mental Health Using AI. Dr Cutler is funded by a National Health and Medical Research Council Senior Principal Research Fellowship (1155614). Drs Newby, Whitton, Werner-Seidler, and Rosenbaum are funded by National Health and Medical Research Council Investigator Grants (Dr Newby: grant 2008839; Dr Whitton: grant 2017521; Dr Werner-Seidler: grant 1197074; Dr Rosenbaum: grant 2017506). Dr Beames is funded by a Marie Skłodowska-Curie fellowship (101063326). The team is funded by an Australian National Health and Medical Research Council Centre of Research Excellence in Depression Treatment Precision (2024796).

Role of the Funder/Sponsor: The funders had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

Data Sharing Statement: See Supplement 3.

Additional Contributions: We would like to thank Adryon Joubert-de Villiers, Bachelor of Graphic Design, UX/UI, Black Dog Institute, Jenny Vuu, Bachelor of International and Global Studies, International Business and Government, and International Relations, UX/UI, Black Dog Institute, Caroline Fitzgerald, Bachelor of Clinical Exercise Physiology, Exercise Physiology, University of New South Wales, Matthew Slarke, Master of Clinical Psychology, Clinical Psychology, Black Dog Institute, Marya Bautista, Bachelor of Design, UX/UI, Black Dog Institute, Vera Kravchuk, Bachelor of Visual Communication, UX/UI, Black Dog Institute, Matt Lee, Bachelor of Arts, Media, Graphic Design, Contractor, Dean Winder, Bachelor of Media (Screen and Sound), Marketing and Communications, Black Dog Institute, and Sandrine Chevassu, Videography, Contractor, for their significant contributions to the design and development of the study app and digital intervention content. We would like to thank Smrithi Ravindra, Bachelor of Psychology (Honours), Black Dog Institute, and Jodie Rosenberg, Bachelor of Science (Computing), Research Officer, Black Dog Institute, for their assistance with the project administration. We would also like to thank Alasdair Grant, Bachelor of Science (Computing), Software Development, Deakin University, and Scott Barnett, Bachelor of Science (Computer Science and Software Engineering)/Bachelor of Engineering (Robotics and Mechatronics), Software Development, Deakin University, for their technical insight and feedback. We also thank Truyen Tran, PhD, AI Algorithm for Health Care, Deakin University, and Santu Rana, PhD, AI Algorithm for Optimisation, Deakin University, for their technical discussions on the adaptive trial methodology. We are also very grateful to Debbie Agnew, Master of Marketing, Marketing and Communications, Black Dog Institute, Zoe Jenkins, Bachelor of Design, Marketing and Communications, Black Dog Institute, and Bianca Cruz, Marketing and Communications, Black Dog Institute, for their advice on social media recruitment strategy and content; and to David Jung, PhD, Research Technology, University of New South Wales, for helping secure data governance approvals. We would also like to thank the university and college student organizations and counselling services and Lived/Living Experience Advisors for their feedback on design and intervention content. All persons named were compensated for their work on the study.

^✉

Corresponding author.

PMCID: PMC12579342 PMID: 41171275

Key Points

Question

Which brief digital intervention most effectively reduces mild, moderate, and severe levels of psychological distress among college students?

Findings

This artificial intelligence–enhanced response-adaptive randomized clinical trial involving 1282 distressed college students found that smartphone-based physical activity and mindfulness were most effective for severe distress and that physical activity and sleep hygiene were most effective for mild distress; no significant differences were observed for moderate distress. The trial’s novel methods improved efficiency by reducing control group allocation.

Meaning

These findings support using brief smartphone-based physical activity for mild and severe distress, mindfulness for severe distress, and sleep hygiene for mild distress.

Abstract

Importance

A large proportion of college students report experiencing psychological distress. Smartphone app–based interventions may alleviate distress, but their effectiveness across severity levels is unclear. Artificial intelligence (AI)–enhanced response-adaptive randomized clinical trials may offer an efficient method to evaluate competing interventions.

Objective

To compare the effectiveness of 3 brief, 2-week self-guided smartphone application–based interventions (physical activity, mindfulness, sleep hygiene) or an active control (ecological momentary assessment [EMA]) for reducing psychological distress among college students with mild, moderate, or severe distress.

Design, Setting, and Participants

This population-based AI-enhanced response-adaptive randomized clinical trial included 1282 participants with distress scores of 20 or more on the 10-item Kessler Psychological Distress Scale in 12 minitrials from November 9, 2021, with final follow-up on February 17, 2023. Participants’ distress was categorized as mild, moderate, or severe based on their normalized 21-item Depression Anxiety Stress Scale (DASS-21) scores at screening.

Interventions

After a 2-week onboarding period of daily EMA, participants were assigned by a contextual multi-armed bandit algorithm to 1 of 4 two-week self-guided app interventions: physical activity, mindfulness, sleep hygiene, or a control that continued EMA.

Main Outcomes and Measures

The primary outcome was change in psychological distress (DASS-21 total score) from week 2 (before intervention) to week 4 (after intervention). The primary end point was after the intervention (4 weeks). Secondary outcomes included DASS-21 subscale scores, self-reported physical activity, mindfulness, sleep quality, and app engagement, usability, and satisfaction. Analysis was performed on an intention-to-treat basis.

Results

A total of 1282 individuals (mean [SD] age, 23.5 [5.2] years; 950 women [74.1%]) participated: physical activity (n = 305), mindfulness (n = 453), sleep hygiene (n = 431), or a control that continued EMA (n = 93). Among 349 participants with severe distress, physical activity (n = 79) and mindfulness (n = 180) were significantly more effective than the EMA control (n = 29) in reducing DASS-21 total scores (physical activity vs control: standardized mean difference [SMD], 0.62 [95% CI, 0.23-1.02]; mindfulness vs control: SMD, 0.53 [95% CI, 0.19-0.87]) and sleep hygiene (n = 61) (physical activity vs sleep hygiene: SMD, 0.50 [95% CI, 0.16-0.84]; mindfulness vs sleep hygiene: SMD, 0.41 [95% CI, 0.13-0.69]). Among 494 participants with mild distress, physical activity (n = 161) and sleep hygiene (n = 224) were significantly more effective than control (n = 37) in reducing DASS-21 total scores (physical activity vs control: SMD, 0.58 [95% CI, 0.30-0.86]; sleep hygiene vs control: SMD, 0.47 [95% CI, 0.20-0.73]). No significant group differences were observed among participants with moderate distress (n = 439).

Conclusions and Relevance

In this AI-enhanced response-adaptive randomized clinical trial among college students, physical activity and mindfulness were most effective for severe distress, while physical activity and sleep hygiene were most effective for mild distress. These findings can guide personalized mental health interventions for college students. This trial improved efficiency by minimizing control group allocation but had reduced power to detect significant group differences.

Trial Registration

http://anzctr.org.au Identifier: ACTRN12621001223820

This randomized clinical trial compares the effectiveness of 3 brief smartphone app–based interventions (physical activity, mindfulness, and sleep hygiene) or an active control (ecological momentary assessment) for reducing mild, moderate, and severe psychological distress among college students.

Introduction

More than 1 in 5 college students experience psychological distress,¹ which affects academic performance, employment, relationships, and quality of life.^1,2 Brief smartphone application–based interventions targeting cognitive, emotional, and lifestyle factors offer a scalable means to improve college students’ mental health.³

Mindfulness, physical activity, and sleep hygiene are promising interventions for distress. Mindfulness reduces distress via improved attention, concentration, self-compassion, and present-focused awareness.^4,5 Physical activity reduces distress by improving self-confidence and increasing cardiorespiratory fitness levels.⁶ Sleep hygiene reduces distress by promoting positive sleep habits that improve sleep quality.⁷ Despite the growing number of interventions for distress, the limited direct comparisons make it difficult to determine their relative effectiveness. There is limited understanding of how intervention effectiveness varies by distress severity. This lack of evidence hinders the development of tailored or stratified treatment approaches and personalized treatment recommendations for college students with distress.

Artificial intelligence (AI)–enhanced bandit-based response-adaptive trials may offer a more efficient approach than traditional randomized clinical trials (RCTs) for identifying the most effective interventions and evaluating effects in different participant subgroups. Unlike traditional RCTs that allocate participants equally across trial arms, in bandit-based response-adaptive designs, data from previous trial participants are processed by an AI algorithm that dynamically adjusts the probability that new participants will be allocated to different trial arms, in favor of better-performing interventions. This method may optimize resource allocation by directing more participants toward interventions that show promise, thereby enhancing statistical power to discern differences in treatment effects between conditions and within participant subgroups,⁸ and improving overall participant outcomes.

Although AI-enhanced response-adaptive trials have been used in other fields of medicine,^9,10,11 they have never been used to compare the effectiveness of mental health interventions, to our knowledge. The primary goal of this contextual bandit-based AI-enhanced response-adaptive trial was to determine the relative effectiveness of brief smartphone app–based mindfulness, physical activity, and sleep hygiene and an active control for reducing psychological distress in subgroups of college students with mild, moderate, or severe levels of distress. As a secondary goal, we report on the feasibility and outcomes of this new method for detecting differences in intervention effects across distress subgroups.

Methods

Study Design

This 4-arm AI-enhanced response-adaptive randomized clinical trial compared mindfulness, physical activity, sleep hygiene, and an active control (ecological momentary assessment; EMA) in reducing psychological distress among college students (trial protocol¹² in Supplement 1). This study is reported in line with the Consolidated Standards of Reporting Trials (CONSORT) reporting guideline.^13,14,15 The trial ran across 12 minitrials from 2021 to 2023 (recruitment from October 29, 2021; first participant enrolled November 9, 2021; final follow-up on February 17, 2023; total duration, 16 months) and was approved by the University of New South Wales Sydney Human Research Ethics Committee. Participants were recruited via social and traditional media and college organizations and provided electronic informed consent to participate.

Each minitrial lasted 4 weeks (eFigure 1 in Supplement 2). Participants completed baseline assessments on their smartphone (week 0) followed by a 2-week onboarding period of daily EMA to facilitate smartphone-based passive sensing data collection (for a secondary study) and to familiarize participants with the app. Two weeks after baseline (week 2), they completed preintervention assessments, were assigned to an intervention or control for 2 weeks, and then completed a postintervention assessment (week 4). All interventions were then made available to all participants, who completed a week 12 follow-up assessment. The trial continued until a significant difference was found between the most effective and second-most effective interventions within each distress severity group (mild, moderate, and severe), or up to 12 minitrials.

Eligible Participants and Enrollment

Inclusion criteria were age 18 years or older; Australian resident; enrolled at a higher education institution; advanced, fluent, or native English speaker; owned an eligible smartphone (iPhone 6S or Android 5 or later) with an active mobile number and internet access; and a self-rated psychological distress score of 20 or more on the 10-item Kessler Psychological Distress Scale.^16,17 Exclusion criteria were high self-rated suicidal ideation on the Suicide Ideation Attributes Scale (scores ≥21)¹⁸; self-reported psychosis or bipolar disorder; previous participation in the trial; circumstances that would prevent adequate trial participation in the next 2 months (eg, travel); and/or inability to safely undertake a physical activity intervention. Participants could undertake other concurrent treatments but were discouraged from starting new treatments while participating.

Randomization and Allocation

Each participant took part in 1 minitrial only. In the first minitrial, participants were equally assigned to 1 of 4 interventions, independent of distress severity, in a round-robin sequence (sleep hygiene, mindfulness, EMA, and physical activity) based on the order in which the participants completed their first baseline assessment task. In subsequent minitrials, allocation was determined by a contextual multi-armed bandit AI algorithm,¹⁹ which updated an underlying model of the effectiveness of each condition within mild, moderate, and severe distress groups (severity based on the participant’s normalized baseline scores on the 21-item Depression, Anxiety, Stress Scale [DASS-21]²⁰) (eMethods 1 in Supplement 2).

The contextual multi-armed bandit AI algorithm¹⁹ (eMethods 2, eFigure 2, and eFigure 3 in Supplement 2) aimed to identify the most effective intervention for reducing distress as quickly as possible for the 3 distress subgroups. The algorithm explored intervention outcomes (preintervention to postintervention changes in distress) to ensure that 1 or more trial arms were not discarded from the trial until the best-performing intervention emerged, and to ensure the minitrials maximized statistical power for key outcome comparisons while controlling for false detection rates. There was no minimum or maximum allocation to trial arms.

Participants were allocated after they completed their baseline assessment. Their allocation was revealed to them within the app, after they completed the preintervention assessment (ie, after the initial 2-week EMA period). Participants and operational staff were unblinded. All other investigators, trial staff, and statisticians were blinded until the final analysis was completed.

Interventions

All interventions were brief, self directed, standardized, accessible anytime, and delivered via a smartphone app. See the trial protocol¹² in Supplement 1 (to access and view intervention content, contact the corresponding author) and eTable 1 in Supplement 2 for details. Each intervention intentionally involved a similar time commitment and effort over the 2-week period. A daily reminder encouraged engagement. Participants could not delay their intervention.

Mindfulness

After a brief introductory video, guided audio meditations of less than 5 minutes were released sequentially, every second day. Content included instructions to enhance mindfulness during everyday activities and adopt nonjudgmental, accepting, and self-compassionate responses and present-focused awareness.

Physical Activity

After a brief introductory infographic about the guidelines and benefits of physical activity, participants set a physical activity goal (eg, increasing steps). The intervention also included an evidence-based, 7-minute, high-intensity interval training guided video.²¹

Sleep Hygiene

This intervention included infographics on topics designed to shift behaviors and daily habits to improve sleep quality and daytime alertness (eg, sleep environments and stimulus control).²²

EMA Control

This condition included twice-daily surveys of mood (state emotions, adapted from the International Positive and Negative Affect Short-Form²³) and associated behavioral responses. EMA was an active comparator designed to control for various factors such as app engagement and repeated assessment.

Primary Outcome

The primary outcome was self-rated psychological distress, assessed using the DASS-21 total score²⁰ at the postintervention assessment (week 4). Scores range from 0 to 126, with higher scores indicating higher levels of distress; DASS-21 scores are multiplied by 2 to derive the full DASS-42 total scores.

Secondary Outcomes

Secondary outcomes were scores on the DASS-21 depression, anxiety, and stress subscales²⁰; the Physical Activity Vital Sign (PAVS), which measured the number of minutes engaged in moderate-to-strenuous physical activity²⁴; item 6 of the Modified Pittsburgh Sleep Quality Index, which measured sleep quality (range, 0-3 [higher scores indicated better quality sleep])²⁵; and a bespoke single item that measured mindfulness (range, 0-4 [higher scores indicate more mindfulness]).

Two items from the Credibility Expectancy Questionnaire²⁶ measured participants’ perceived credibility and expected intervention benefit. App engagement metrics included the self-reported daily log of time spent engaging in the intervention, module access, completion, and time spent on pages (collected within the app).²⁷ App usability and satisfaction were assessed with items from the System Usability Scale²⁸ and mHealth App Usability Questionnaire.²⁹ Additional self-report questionnaires and smartphone sensor data for digital phenotyping were collected and will be reported separately.¹²

Sample Size Calculation

Sample size calculation was based on fixed trial estimates. We aimed to recruit at least 120 participants for each of the 12 minitrials, with 80 expected to start (20 per arm in minitrial 1). Allowing for 20% attrition, 64 participants were expected to complete postintervention assessments. Traditional power estimation tools are not well suited to adaptive allocation schemes. Assuming recruitment of a minimum of 192 participants in each of the 2 best-performing arms, it would have 80% power (2-tailed α = .05) to detect a 0.29 (small) between-groups effect size.

Statistical Analysis

Refer to eMethods 3 and eFigure 4 in Supplement 2 and Huckvale et al¹² for the statistical analysis plan. Analyses were conducted in R, version 2023.09.0+463 (R Project for Statistical Computing). Because the adaptive trial design results in unequal probabilities of allocation across interventions, we used the Horvitz-Thompson estimator with inverse propensity weighting. This method reweights each observation by the inverse of its intervention allocation probability, allowing for estimation of marginal intervention effects as if all arms had been equally sampled. This shifts the estimand from a potentially biased sample mean to an unbiased population-level mean effect.

Mixed-model repeated-measures analyses of variance with an unstructured covariance structure were used with all available data from the intention-to-treat sample to assess intervention (and control) effectiveness in reducing distress in each severity group (mild, moderate, and severe). Planned contrasts between each trial arm compared the differences in preintervention with postintervention changes in distress (DASS-21 total score) and secondary outcomes (eg, sleep quality). These contrasts included fixed effects for intervention, time, and their interactions as covariates. The unstructured covariance matrix allowed for the modeling of within-individual correlations across time points to allow for estimation in the change in outcome scores before and after the intervention. All tests were 2-sided. To control for type I error across multiple comparisons and repeated interim analyses, we combined the Benjamini-Hochberg³⁰ procedure with an α-spending approach (eTable 2 in Supplement 2). Interim analyses were planned after minitrials 4 and 8, with the final analysis after minitrial 12. At each testing point, a portion of the overall α (.05) was allocated using an O’Brien-Fleming–type α-spending function. Within each time point, the Benjamini-Hochberg procedure was applied to adjust significance thresholds across intervention comparisons, allowing greater sensitivity while controlling the false discovery rate.

Planned interim analyses were conducted after minitrials 4 and 8 (eTable 3 and eTable 4 in Supplement 2). If any condition was significantly more effective in reducing distress than all other conditions within a specific distress severity group, it could be removed from subsequent minitrials so that the effect of other interventions could be estimated with greater precision. One-sided Welch t tests with Satterthwaite-adjusted degrees of freedom were used. All 4 conditions were retained in all minitrials because, at both interim analyses, no condition was more effective than all other conditions in any distress group (eTable 5 in Supplement 2). Secondary outcomes were analyzed at the study’s conclusion.

Results

Enrollment and Participant Characteristics

The sample of 1282 individuals (mean [SD] age, 23.5 [5.2] years; 950 women [74.1%], 280 men [21.8%], and 48 nonbinary or gender-neutral individuals [3.7%]) was predominantly undergraduate students (1004 [78.3%]), with 467 (36.4%) identifying as sexually diverse (Table 1). eFigure 5 in Supplement 2 shows DASS-21 subscale severity ranges by distress group. The Figure shows the participant flow diagram. A total of 1394 individuals were allocated to a condition, 1282 (92.0%) completed the postintervention assessment, and 701 (50.3%) completed the follow-up assessment.

Table 1. Sample Demographic Characteristics.

Charactaristic	Total sample (N = 1282)	Mindfulness (n = 453)	Physical activity (n = 305)	Sleep hygiene (n = 431)	Ecological momentary assessment (n = 93)
Age, mean (SD), y	23.5 (5.2)	23.3 (5.0)	23.5 (5.2)	24.0 (5.6)	22.9 (4.1)
Student type, No. (%)
Undergraduate or TAFE student	1004 (78.3)	361 (79.7)	227 (74.4)	340 (78.9)	76 (81.7)
Postgraduate coursework	218 (17.0)	75 (16.6)	56 (18.4)	71 (16.5)	16 (17.2)
Postgraduate research	60 (4.7)	17 (3.8)	22 (7.2)	20 (4.6)	1 (1.1)
International student status, No. (%)
Domestic (Australian)	1208 (94.2)	431 (95.1)	285 (93.4)	407 (94.4)	85 (91.4)
International	74 (5.8)	22 (4.9)	20 (6.6)	24 (5.6)	8 (8.6)
Gender, No. (%)
Woman or female	950 (74.1)	347 (76.6)	217 (71.1)	312 (72.4)	74 (79.6)
Man or male	280 (21.8)	91 (20.1)	72 (23.6)	105 (24.4)	12 (12.9)
Nonbinary or gender neutral	48 (3.7)	14 (3.1)	15 (4.9)	12 (2.8)	7 (7.5)
Did not disclose	4 (0.3)	1 (0.2)	1 (0.3)	2 (0.5)	0
Sex, No. (%)
Female	1005 (78.4)	366 (80.8)	234 (76.7)	325 (75.4)	80 (86.0)
Male	275 (21.5)	87 (19.2)	70 (23.0)	105 (24.4)	13 (14.0)
Did not disclose	2 (0.2)	0	1 (0.3)	1 (0.2)	0
Sexual orientation, No. (%)
Heterosexual	740 (57.7)	245 (54.1)	185 (60.7)	269 (62.4)	41 (44.1)
Homosexual	79 (6.2)	28 (6.2)	21 (6.9)	21 (4.9)	9 (9.7)
Bisexual	315 (24.6)	120 (26.5)	71 (23.3)	100 (23.2)	24 (25.8)
Other nonheterosexual^a	73 (5.7)	29 (6.4)	17 (5.6)	19 (4.4)	8 (8.6)
Not known	64 (5.0)	29 (6.4)	8 (2.6)	18 (4.2)	9 (9.7)
Did not disclose	11 (0.9)	2 (0.4)	3 (1.0)	4 (0.9)	2 (2.2)
Indigenous status, No. (%)
Not Aboriginal or Torres Strait Islander	1245 (97.1)	438 (96.7)	301 (98.7)	417 (96.8)	89 (95.7)
Aboriginal	22 (1.7)	9 (2.0)	4 (1.3)	9 (2.1)	0
Torres Strait Islander	1 (0.1)	1 (0.2)	0	0	0
Unknown	9 (0.7)	3 (0.7)	0	3 (0.7)	3 (3.2)
Did not disclose	5 (0.4)	2 (0.4)	0	2 (0.5)	1 (1.1)
Primary language, No. (%)
Arabic	1 (0.1)	1 (0.2)	0	0	0
Cantonese	16 (1.2)	7 (1.5)	0	7 (1.6)	2 (2.2)
English	1176 (91.7)	418 (92.3)	281 (92.1)	396 (91.9)	81 (87.0)
Hindi	7 (0.5)	2 (0.4)	2 (0.7)	2 (0.5)	1 (1.1)
Italian	1 (0.1)	1 (0.2)	0	0	0
Mandarin	13 (1.0)	2 (0.4)	6 (2.0)	3 (0.7)	2 (2.2)
Punjabi	2 (0.2)	0	1 (0.3)	1 (0.2)	0
Spanish	4 (0.3)	3 (0.7)	0	1 (0.2)	0
Vietnamese	10 (0.8)	4 (0.9)	2 (0.7)	3 (0.7)	1 (1.1)
Other^b	50 (3.9)	15 (3.3)	12 (3.9)	17 (3.9)	6 (6.5)
Did not disclose	2 (0.2)	0	1 (0.3)	1 (0.2)	0
Ancestry, No. (%)
Australian	851 (66.4)	310 (68.4)	199 (65.2)	287 (66.6)	55 (59.1)
Chinese or Cantonese	107 (8.3)	33 (7.3)	28 (9.2)	36 (8.4)	10 (10.8)
Dutch	35 (2.7)	9 (2.0)	8 (2.6)	13 (3.0)	5 (5.4)
English	278 (21.7)	101 (22.3)	68 (22.3)	93 (21.6)	16 (17.2)
German	47 (3.7)	24 (5.3)	7 (2.3)	14 (3.2)	2 (2.2)
Greek	25 (2.0)	10 (2.2)	5 (1.6)	9 (2.1)	1 (1.1)
Indian	50 (3.9)	14 (3.1)	7 (2.3)	25 (5.8)	4 (4.3)
Irish	91 (7.1)	41 (9.1)	17 (5.6)	26 (6.0)	7 (7.5)
Italian	39 (3.0)	14 (3.1)	10 (3.3)	11 (2.6)	4 (4.3)
Scottish	76 (5.9)	24 (5.3)	14 (4.6)	27 (6.3)	11 (11.8)
Vietnamese	20 (1.6)	5 (1.1)	5 (1.6)	8 (1.9)	2 (2.2)
Other^b	210 (16.4)	70 (15.5)	60 (19.7)	63 (14.6)	17 (18.3)
Did not disclose	3 (0.2)	1 (0.2)	1 (0.3)	1 (0.2)	0 (0.0)

Open in a new tab

Abbreviation: TAFE, technical and further education (ie, vocational education).

^{^a}

Encompasses labels referring to abrosexuality, asexuality, bicuriousity, demisexuality, omnisexuality, pansexuality, and queerness.

^{^b}

Encompasses all languages or ancestries that are not listed; participants were not required to specify what these were.

Primary Outcome

Table 2 presents the observed mean change scores on the primary outcome, DASS-21 total score for each trial arm and severity group, pooled across minitrials. Table 3 presents the standardized mean differences (SMDs), 95% CIs, and significance tests for bias-corrected and log-transformed DASS-21 total scores. eFigure 6 and eTables 6, 7, and 8 in Supplement 2 show observed DASS-21 total scores and the participant allocation per minitrial for each trial arm and severity group.

Table 2. Unweighted Observed Mean Values for DASS-21 Total Scores, Before and After Intervention, and Change by Severity and Intervention Group.

Severity group	Observed DASS-21 score, mean (SD)		Pre-post change
Severity group	Preintervention	Postintervention	Pre-post change
Mild
Physical activity	33.6 (12.6)	24.5 (13.4)	−8.2 (13.5)
Mindfulness	30.1 (13.0)	23.5 (12.1)	−6.1 (12.2)
Sleep hygiene	31.7 (13.7)	24.9 (14.3)	−7.3 (14.7)
EMA	27.4 (14.0)	26.6 (14.5)	−0.6 (7.6)
Moderate
Physical activity	46.4 (11.8)	34.8 (15.4)	−10.8 (16.1)
Mindfulness	50.1 (13.2)	38.8 (16.8)	−12.1 (17.9)
Sleep hygiene	45.4 (12.9)	36.3 (15.3)	−9.8 (14.8)
EMA	47.3 (11.9)	44.3 (18.0)	−3.0 (17.8)
Severe
Physical activity	66.4 (18.7)	50.4 (24.8)	−16.6 (23.2)
Mindfulness	63.2 (17.7)	48.9 (21.9)	−14.5 (20.6)
Sleep hygiene	65.7 (20.0)	56.7 (21.4)	−7.6 (19.8)
EMA	69.4 (19.4)	65.4 (22.4)	−4.4 (23.3)

Open in a new tab

Abbreviations: DASS-21; 21-item Depression Anxiety Stress Scale (scores range from 0 to 126, with higher scores indicating higher distress); EMA, ecological momentary assessment.

Table 3. Group Comparisons of the Before and After Changes in Bias-Corrected and Log-Transformed DASS-21 Total Scores.

Comparison	Mild distress			Moderate distress			Severe distress
Comparison	SMD (95% CI)	P value	BH-adjusted P value	SMD (95% CI)	P value	BH-adjusted P value	SMD (95% CI)	P value	BH-adjusted P value
Mindfulness vs control	0.34 (0.01 to 0.67)	.06	.03	0.47 (0.09 to 0.84)	.02	.008	0.53 (0.19 to 0.87)^a	.005	.03
Physical activity vs control	0.58 (0.30 to 0.86)^a	.001	.007	0.45 (0.02 to 0.88)	.03	.02	0.62 (0.23 to 1.02)^a	.003	.01
Sleep hygiene vs control	0.47 (0.20 to 0.73)^a	.006	.01	0.39 (0.01 to 0.76)	.04	.02	0.12 (−0.26 to 0.50)	.30	.04
Physical activity vs mindfulness	0.24 (−0.03 to 0.51)	.06	.02	−0.01 (−0.34 to −0.31)	.46	.05	0.09 (−0.22 to 0.40)	.26	.04
Sleep hygiene vs mindfulness	0.13 (−0.12 to 0.39)	.18	.04	−0.08 (−0.31 to 0.15)	.24	.03	−0.41 (−0.69 to −0.13)^a	.005	.02
Physical activity vs sleep hygiene	0.11 (−0.10 to 0.31)	.16	.04	0.07 (−0.25 to 0.38)	.34	.04	0.50 (0.16 to 0.84)^a	.003	.007

Open in a new tab

Abbreviations: BH, Benjamini-Hochberg; DASS-21; 21-item Depression Anxiety Stress Scale (scores range from 0 to 126, with higher scores indicating higher distress); SMD, standardized mean difference.

^{^a}

P value lower than BH-adjusted critical P value.

For participants with mild distress, physical activity (n = 161) and sleep hygiene (n = 224) were significantly more effective in reducing DASS-21 total scores than the control (n = 37) (physical activity vs control: SMD, 0.58 [95%CI, 0.30-0.86]; sleep hygiene vs control: SMD, 0.47 [95% CI, 0.20-0.73]) (Table 3). For participants with moderate distress, group differences were not statistically significant. For participants with severe distress, physical activity (n = 79) and mindfulness (n = 180) were significantly more effective than control (n = 29) in reducing DASS-21 total scores (physical activity vs control: SMD, 0.62 [95% CI, 0.23-1.02]; mindfulness vs control: SMD, 0.53 [95% CI, 0.19-0.87]). Both physical activity (n = 79) and mindfulness (n = 180) were also significantly more effective than sleep hygiene (n = 61) in reducing DASS-21 total scores for participants with severe distress (physical activity vs sleep hygiene: SMD, 0.50 [95% CI, 0.16-0.84]; mindfulness vs sleep hygiene: SMD, 0.41 [95% CI, 0.13-0.69]). When we repeated the analyses controlling for preintervention DASS-21 total scores, the same pattern of results was observed.

Secondary Outcomes

Secondary outcomes for DASS-21 depression, anxiety, and stress subscales are available in eTables 9, 10, and 11 in Supplement 2 with significance tests in eTables 12, 13, and 14 in Supplement 2. For participants with severe distress, both physical activity (n = 79) and mindfulness (n = 180) were superior in reducing DASS-21 depression scores compared with the control (n = 29) (physical activity vs control: SMD, 0.61 [95% CI, 0.18-1.04]; mindfulness vs control: SMD, 0.61 [95% CI, 0.23-1.00]) and compared with sleep hygiene (n = 61) (physical activity vs sleep hygiene: SMD, 0.50 [95% CI, 0.15-0.85]; mindfulness vs sleep hygiene: SMD, 0.50 [95% CI, 0.22-0.78]) (eTable 12 in Supplement 2).

For participants with mild distress, physical activity (n = 161) was superior to mindfulness (n = 72) in reducing DASS-21 anxiety scores (physical activity vs mindfulness: SMD, 0.35 [95% CI, 0.04-0.66]) (eTable 13 in Supplement 2). For participants with mild distress, all active interventions were better than the control in reducing DASS-21 stress scores (physical activity [n = 161] vs control [n = 37]: SMD, 0.65 [95% CI, 0.36-0.93]; mindfulness [n = 72] vs control [n = 37]: SMD, 0.57 [95% CI, 0.23-0.90]; sleep hygiene [n = 224] vs control [n = 37]: SMD, 0.47 [95% CI, 0.20-0.74]) (eTable 14 in Supplement 2). For the group with severe distress, physical activity (n = 79) was superior to both control (n = 29) (physical activity vs control: SMD, 0.61 [95% CI, 0.23-1.00]) and sleep hygiene (n = 61) (physical activity vs sleep hygiene: SMD, 0.55 [95% CI, 0.23-0.87]) in reducing DASS-21 stress scores.

Outcomes for sleep quality, mindfulness, and PAVS (physical activity levels) are shown in eTables 15, 16, and 17 in Supplement 2, and significance testing results are presented in eTables 18, 19, and 20 in Supplement 2. In all severity subgroups, there were no differences between trial arms in changes in sleep quality. At all levels of distress severity, the mindfulness intervention was associated with greater reported mindfulness relative to all other arms (SMDs ranged from 0.49 [95% CI, 0.18-0.80] to 0.87 [95% CI, 0.44-1.29]) (eTable 19 in Supplement 2).

For participants with mild distress, changes in self-reported physical activity (PAVS scores) were greater in the physical activity condition (n = 161) compared with the control (n = 37) (SMD, 0.53 [95% CI, 0.06-1.00]) (eTable 20 in Supplement 2). For participants with severe distress, changes in physical activity (n = 79) were greater in the physical activity arm compared with both mindfulness (n = 180) (SMD, 0.32 [95% CI, 0.07-0.57]) and sleep hygiene (n = 61) (SMD, 0.56 [95% CI, 0.20-0.92]).

Intervention Engagement

eTable 21 in Supplement 2 displays intervention expectancy, engagement, and usability ratings. Of the participants allocated to an intervention, 83.2% (1066 of 1282) accessed it, 29.3% (375 of 1282) accessed all content, and 61.7% (791 of 1282) accessed half of all content. Nearly two-thirds of participants (68.8% [818 of 1189]) attained our a priori criterion of acceptable engagement. The interventions were acceptable to participants.

Comparison of Sample Size With Fixed Allocation RCT

We calculated the sample size needed for a fixed allocation RCT to detect the smallest significant effect (SMD, 0.41) observed in this trial. Allowing for within-group correction for multiple testing (.05/6), a sample of 146 participants per intervention, per severity subgroup would be required, totaling 1752 participants.

Discussion

In this bandit-based AI-enhanced response-adaptive RCT, we compared the effectiveness of brief smartphone app–based mindfulness, physical activity, and sleep hygiene interventions as well as an EMA active control for reducing psychological distress among college students and showed that the effectiveness of these interventions varied across distress severity subgroups. For students with severe distress, physical activity and mindfulness were more effective than sleep hygiene and EMA. For students with moderate distress, no differences were found. For those with mild distress, physical activity and sleep hygiene were more effective than EMA. Changes in the target behaviors were observed in the desired directions.

Despite being brief, 2-week interventions, the between-group effect sizes relative to EMA control were comparable with meta-analyses of longer app interventions with active controls for depression (Hedges g = 0.19)³¹ and anxiety (Hedges g = 0.22).^32,33 User-centered design features (eg, notifications and evolving character)¹² may have contributed to usability, satisfaction, and engagement relative to past research.^34,35

These results support the use of brief smartphone app–based interventions to provide on-demand, anonymous, low-intensity mental health support to college students with distress. These results may optimize outcomes through improved treatment precision, by selecting the most effective treatment for an individual based on their distress levels. Physical activity, a nonstigmatizing intervention with mental and physical health benefits,³⁶ was the most effective and recommended intervention for mild and severe distress. Results also support using mindfulness for severe distress (in line with meta-analyses of samples of individuals with depression and anxiety^37,38), but not mild distress. In contrast, sleep hygiene was effective only for mild but not severe distress, suggesting more intensive insomnia treatments may be needed for students with severe distress.⁷ Given that no clear intervention recommendations can be made for participants with moderate distress, future trials should focus on this subgroup and explore the reasons for the lack of group differences (eg, power and symptom heterogeneity).

Implications for AI-Enhanced Adaptive Trials

Adaptive trials have advantages over standard RCTs including efficiency (less time and resources) as well as ethical and health benefits.^8,39 Our study investigated multiple subgroups, demonstrating the feasibility and capability of this method to investigate personalized intervention recommendations. Although future comparisons with standard RCTs are needed (eg, in stroke research),⁴⁰ the algorithm varied intervention allocation across minitrials, and more participants were allocated to better-performing interventions (n = 305-453) than the control (n = 93). Overall, participants experienced more health benefits compared with a standard RCT because fewer participants were allocated to less-effective conditions. However, the smaller control group, coupled with strict significance thresholds, reduced the power to detect group differences.

Future research should identify the ideal use case for the AI-enhanced adaptive-response trial design, including the ideal interventions, type and timing of outcomes, and participant features used in the contextual multi-armed bandit algorithm. We used baseline self-reported psychological distress because it is strongly associated with digital intervention outcomes,⁴¹ is easily assessed, and is expected to change with brief interventions. However, psychological distress may reflect varied factors (eg, acute stress related to college or mental disorders). Future research should explore using other participant characteristics in the algorithm (eg, treatment preferences, history, and sociodemographic characteristics) to help inform more personalized treatment recommendations.

Limitations

This study has some limitations. First, participants were not blinded to their intervention. Second, self-reported outcomes were measured immediately after the intervention, with no long-term follow-up data. Third, the interventions were brief. Longer interventions may be needed, especially for severe symptoms. Fourth, the sample, primarily English-speaking, female, domestic students recruited via social media, may limit generalizability. Fifth, the findings from smartphone app–based interventions may not apply to in-person interventions.

Conclusions

This AI-enhanced response-adaptive RCT, conducted with college students experiencing elevated psychological distress, revealed differences in the effectiveness of brief smartphone app–based interventions for mild and severe distress, but not moderate distress. Physical activity was the most effective intervention for mild and severe distress, mindfulness was effective for severe distress, and sleep hygiene was effective for mild distress.

This AI-enhanced response-adaptive trial was feasible and offered some advantages over traditional RCTs, such as reduced allocation to the underperforming control and detecting varying treatment effects based on distress severity. This finding shows the potential for treatment personalization at scale. Future research should compare the efficiency and utility of AI-enhanced response-adaptive trials with conventional RCTs in mental health.

Supplement 1.

Trial Protocol and Statistical Analysis Plan

jamanetwopen-e2540502-s001.pdf^{(64.7MB, pdf)}

Supplement 2.

eFigure 1. Trial Schematic

eMethods 1. Method of Calculating Severity Scores

eMethods 2. Contextual Multi-Arm Bandit Algorithme

Figure 2. Overview of the Contextual Multi-Arm Bandit (MAB) Algorithm Framework

eFigure 3. Illustration of How the Contextual MAB Algorithm Allocates Participants to an Interventione

Table 1. Intervention Content

eMethods 3. Statistical Analysis Plan

eFigure 4. Hypothesis Testing Procedure and Processing of DASS-21 Scores Including Bias Adjustment and Log Transformation

eTable 2. Maximum P Value for Each Distress Severity Group at Each Interim Analysis and Final Analysis

eTable 3. Detailed Breakdown Used to Calculate Final Maximum P Values for Each Distress Severity Group

eTable 4. Number of Participants Per Distress Severity Group Included in the Interim Analyses and Final Analysis

eTable 5. Interim Hypothesis Tests After Mini-Trials 4 and 8 for DASS-21 Total Scores

eResults.

eFigure 5. Number of Participants in the Mild, Moderate, and Severely Distressed Subgroups With Scores in the Normal/Mild, Moderate, and Severe/Extremely Severe Ranges on the DASS-21 Depression, Anxiety and Stress Subscales at Baseline

eFigure 6. Mean Change in DASS-21 Total Score From Pre- to Post-Intervention (Top), and the Number of Participants (Bottom) Allocated Across Each of the 12 Mini-Trials, for Mild (a), Moderate (b), and Severe (c) Distress Groups

eTable 6. Observed DASS-21 Total Scores for the Mild Distress Group by Intervention Group Across Each of the Twelve Mini-Trials

eTable 7. Observed DASS-21 Total Scores for the Moderate Distress Group by Intervention Group Across Each of the Twelve Mini-Trials

eTable 8. Observed DASS-21 Total Scores for the Severe Distress Group by Intervention Group Across Each of the Twelve Mini-Trials

eTable 9. Unweighted Observed Means and Standard Deviations for DASS-21 Depression Subscale Scores Pre- and Post-Intervention, and Change by Severity and Intervention Group

eTable 10. Unweighted Observed Means and Standard Deviations for DASS-21 Anxiety Subscale Scores Pre- and Post-Intervention, and Change by Severity and Intervention Group

eTable 11. Unweighted Observed Means and Standard Deviations for DASS-21 Stress Subscale Scores Pre- and Post-Intervention, and Change by Severity and Intervention Group

eTable 12. Group Comparisons of the Pre- to Post-Changes in Bias Corrected and Log Transformed DASS-21 Depression Subscale Scores

eTable 13. Group Comparisons of the Pre- to Post-Changes in Bias Corrected and Log Transformed DASS-21 Anxiety Subscale Scores

eTable 14. Group Comparisons of the Pre- to Post-Changes in Bias Corrected and Log Transformed for DASS-21 Stress Subscale Scores

eTable 15. Unweighted Observed Means and Standard Deviations for Sleep Quality Scores Pre- and Post-Intervention, and Change by Severity and Intervention Group

eTable 16. Unweighted Observed Means and Standard Deviations for Mindfulness Scores Pre- and Post-Intervention, and Change by Severity and Intervention Group

eTable 17. Unweighted Observed Means and Standard Deviations for Physical Activity Vital Sign Scores Pre- and Post-Intervention, and Change by Severity and Intervention Group

eTable 18. Group Comparisons of the Pre- to Post-Changes in Bias Corrected and Log Transformed Sleep Quality Scores

eTable 19. Group Comparisons of the Pre- to Post-Changes in Bias Corrected and Log Transformed Mindfulness Scores

eTable 20. Group Comparisons of the Pre- to Post-Changes in Bias Corrected and Log Transformed Physical Activity Vital Sign Scores

eTable 21. Descriptive Statistics for the App User Engagement Metrics Across Groups

eReferences.

jamanetwopen-e2540502-s002.pdf^{(1.5MB, pdf)}

Supplement 3.

Data Sharing Statement

jamanetwopen-e2540502-s003.pdf^{(16.9KB, pdf)}

References

1.Auerbach RP, Alonso J, Axinn WG, et al. Mental disorders among college students in the World Health Organization World Mental Health Surveys. Psychol Med. 2016;46(14):2955-2970. doi: 10.1017/S0033291716001665 [DOI] [PMC free article] [PubMed] [Google Scholar]
2.Stallman HM. Psychological distress in university students: a comparison with general population data. Aust Psychol. 2010;45(4):249-257. doi: 10.1080/00050067.2010.482109 [DOI] [Google Scholar]
3.Oliveira C, Pereira A, Vagos P, Nóbrega C, Gonçalves J, Afonso B. Effectiveness of mobile app-based psychological interventions for college students: a systematic review of the literature. Front Psychol. 2021:12:647606. doi: 10.3389/fpsyg.2021.647606 [DOI] [PMC free article] [PubMed] [Google Scholar]
4.Dawson AF, Brown WW, Anderson J, et al. Mindfulness-based interventions for university students: a systematic review and meta-analysis of randomised controlled trials. Appl Psychol Health Well Being. 2020;12(2):384-410. doi: 10.1111/aphw.12188 [DOI] [PubMed] [Google Scholar]
5.Chen B, Yang T, Xiao L, Xu C, Zhu C. Effects of mobile mindfulness meditation on the mental health of university students: systematic review and meta-analysis. J Med Internet Res. 2023;25:e39128. doi: 10.2196/39128 [DOI] [PMC free article] [PubMed] [Google Scholar]
6.Singh B, Olds T, Curtis R, et al. Effectiveness of physical activity interventions for improving depression, anxiety and distress: an overview of systematic reviews. Br J Sports Med. 2023;57(18):1203-1209. doi: 10.1136/bjsports-2022-106195 [DOI] [PMC free article] [PubMed] [Google Scholar]
7.Friedrich A, Schlarb AA. Let’s talk about sleep: a systematic review of psychological interventions to improve sleep in college students. J Sleep Res. 2018;27(1):4-22. doi: 10.1111/jsr.12568 [DOI] [PubMed] [Google Scholar]
8.Pallmann P, Bedding AW, Choodari-Oskooei B, et al. Adaptive designs in clinical trials: why use them, and how to run and report them. BMC Med. 2018;16(1):29. doi: 10.1186/s12916-018-1017-7 [DOI] [PMC free article] [PubMed] [Google Scholar]
9.Burnett T, Mozgunov P, Pallmann P, Villar SS, Wheeler GM, Jaki T. Adding flexibility to clinical trial designs: an example-based guide to the practical use of adaptive designs. BMC Med. 2020;18(1):352. doi: 10.1186/s12916-020-01808-2 [DOI] [PMC free article] [PubMed] [Google Scholar]
10.Bothwell LE, Avorn J, Khan NF, Kesselheim AS. Adaptive design clinical trials: a review of the literature and ClinicalTrials.gov. BMJ Open. 2018;8(2):e018320. doi: 10.1136/bmjopen-2017-018320 [DOI] [PMC free article] [PubMed] [Google Scholar]
11.Allender S, Hayward J, Gupta S, et al. Bayesian strategy selection identifies optimal solutions to complex problems using an example from GP prescribing. NPJ Digit Med. 2020;3(1):7. doi: 10.1038/s41746-019-0205-y [DOI] [PMC free article] [PubMed] [Google Scholar]
12.Huckvale K, Hoon L, Stech E, et al. Protocol for a bandit-based response adaptive trial to evaluate the effectiveness of brief self-guided digital interventions for reducing psychological distress in university students: the Vibe Up study. BMJ Open. 2023;13(4):e066249. doi: 10.1136/bmjopen-2022-066249 [DOI] [PMC free article] [PubMed] [Google Scholar]
13.Hopewell S, Chan AW, Collins GS, et al. CONSORT 2025 statement: updated guideline for reporting randomised trials. BMJ. 2025;389:e081123. doi: 10.1136/bmj-2024-081123 [DOI] [PMC free article] [PubMed] [Google Scholar]
14.Dimairo M, Pallmann P, Wason J, et al. ; ACE Consensus Group . The Adaptive Designs CONSORT Extension (ACE) statement: a checklist with explanation and elaboration guideline for reporting randomised trials that use an adaptive design. BMJ. 2020;369:m115. doi: 10.1136/bmj.m115 [DOI] [PMC free article] [PubMed] [Google Scholar]
15.Liu X, Rivera SC, Moher D, Calvert MJ, Denniston AK; SPIRIT-AI and CONSORT-AI Working Group . Reporting guidelines for clinical trial reports for interventions involving artificial intelligence: the CONSORT-AI Extension. BMJ. 2020;370:m3164. doi: 10.1136/bmj.m3164 [DOI] [PMC free article] [PubMed] [Google Scholar]
16.Kessler RC, Andrews G, Colpe LJ, et al. Short screening scales to monitor population prevalences and trends in non-specific psychological distress. Psychol Med. 2002;32(6):959-976. doi: 10.1017/S0033291702006074 [DOI] [PubMed] [Google Scholar]
17.Andrews G, Slade T. Interpreting scores on the Kessler Psychological Distress Scale (K10). Aust N Z J Public Health. 2001;25(6):494-497. doi: 10.1111/j.1467-842X.2001.tb00310.x [DOI] [PubMed] [Google Scholar]
18.van Spijker BAJ, Batterham PJ, Calear AL, et al. The Suicidal Ideation Attributes Scale (SIDAS): community-based validation study of a new scale for the measurement of suicidal ideation. Suicide Life Threat Behav. 2014;44(4):408-419. doi: 10.1111/sltb.12084 [DOI] [PubMed] [Google Scholar]
19.Auer P, Cesa-Bianchi N, Fischer PJML. Finite-time analysis of the multiarmed bandit problem. Machine Learning. 2002;47:235-256. doi: 10.1023/A:1013689704352 [DOI] [Google Scholar]
20.Lovibond SH, Lovibond PF. Manual for the Depression Anxiety Stress Scales. 2nd ed. Psychology Foundation of Australia; 1995. [Google Scholar]
21.Klika B, Jordan C. High-intensity circuit training using body weight: maximum results with minimal investment. ACSM’s Health & Fitnest Journal. 2013;17(3):8-13. doi: 10.1249/FIT.0b013e31828cb1e8 [DOI] [Google Scholar]
22.Chung KF, Lee CT, Yeung WF, Chan MS, Chung EW, Lin WL. Sleep hygiene education as a treatment of insomnia: a systematic review and meta-analysis. Fam Pract. 2018;35(4):365-375. doi: 10.1093/fampra/cmx122 [DOI] [PubMed] [Google Scholar]
23.Thompson ER. Development and validation of an internationally reliable short-form of the Positive and Negative Affect Schedule (PANAS). J Cross Cult Psychol. 2007;38(2):227-242. doi: 10.1177/0022022106297301 [DOI] [Google Scholar]
24.Greenwood JLJ, Joy EA, Stanford JB. The Physical Activity Vital Sign: a primary care tool to guide counseling for obesity. J Phys ActHealth. 2010;7(5):571-576. doi: 10.1123/jpah.7.5.571 [DOI] [PubMed] [Google Scholar]
25.Buysse DJ, Reynolds CF III, Monk TH, Berman SR, Kupfer DJ. The Pittsburgh Sleep Quality Index: a new instrument for psychiatric practice and research. Psychiatry Res. 1989;28(2):193-213. doi: 10.1016/0165-1781(89)90047-4 [DOI] [PubMed] [Google Scholar]
26.Devilly GJ, Borkovec TD. Psychometric properties of the credibility/expectancy questionnaire. J Behav Ther Exp Psychiatry. 2000;31(2):73-86. doi: 10.1016/S0005-7916(00)00012-4 [DOI] [PubMed] [Google Scholar]
27.Barnett S, Huckvale K, Christensen H, Venkatesh S, Mouzakis K, Vasa R. Intelligent Sensing to Inform and Learn (InSTIL): a scalable and governance-aware platform for universal, smartphone-based digital phenotyping for research and clinical applications. J Med Internet Res. 2019;21(11):e16399. doi: 10.2196/16399 [DOI] [PMC free article] [PubMed] [Google Scholar]
28.Brooke J. SUS: a quick and dirty usability scale. In: Jordan PW, Thomas B, Weerdmeester BA, McClelland IL, eds. Usability Evaluation in Industry. CRC Press; 1995:189. [Google Scholar]
29.Zhou L, Bao J, Setiawan IMA, Saptono A, Parmanto B. The mHealth App Usability Questionnaire (MAUQ): development and validation study. JMIR Mhealth Uhealth. 2019;7(4):e11500. doi: 10.2196/11500 [DOI] [PMC free article] [PubMed] [Google Scholar]
30.Chen SY, Feng Z, Yi X. A general introduction to adjustment for multiple comparisons. J Thorac Dis. 2017;9(6):1725-1729. doi: 10.21037/jtd.2017.05.34 [DOI] [PMC free article] [PubMed] [Google Scholar]
31.Firth J, Torous J, Nicholas J, et al. The efficacy of smartphone-based mental health interventions for depressive symptoms: a meta-analysis of randomized controlled trials. World Psychiatry. 2017;16(3):287-298. doi: 10.1002/wps.20472 [DOI] [PMC free article] [PubMed] [Google Scholar]
32.Firth J, Torous J, Nicholas J, Carney R, Rosenbaum S, Sarris J. Can smartphone mental health interventions reduce symptoms of anxiety? a meta-analysis of randomized controlled trials. J Affect Disord. 2017;218:15-22. doi: 10.1016/j.jad.2017.04.046 [DOI] [PubMed] [Google Scholar]
33.Linardon J, Cuijpers P, Carlbring P, Messer M, Fuller-Tyszkiewicz M. The efficacy of app-supported smartphone interventions for mental health problems: a meta-analysis of randomized controlled trials. World Psychiatry. 2019;18(3):325-336. doi: 10.1002/wps.20673 [DOI] [PMC free article] [PubMed] [Google Scholar]
34.Baumel A, Muench F, Edan S, Kane JM. Objective user engagement with mental health apps: systematic search and panel-based usage analysis. J Med Internet Res. 2019;21(9):e14567. doi: 10.2196/14567 [DOI] [PMC free article] [PubMed] [Google Scholar]
35.Torous J, Nicholas J, Larsen ME, Firth J, Christensen H. Clinical review of user engagement with mental health smartphone apps: evidence, theory and improvements. Evid Based Ment Health. 2018;21(3):116-119. doi: 10.1136/eb-2018-102891 [DOI] [PMC free article] [PubMed] [Google Scholar]
36.Hutchesson MJ, Whatnall MC, Yazin N, et al. Health behavior interventions for university students measuring mental health outcomes: a scoping review. Front Public Health. 2022;10:1063429. doi: 10.3389/fpubh.2022.1063429 [DOI] [PMC free article] [PubMed] [Google Scholar]
37.Spijkerman MP, Pots WT, Bohlmeijer ET. Effectiveness of online mindfulness-based interventions in improving mental health: a review and meta-analysis of randomised controlled trials. Clin Psychol Rev. 2016;45:102-114. doi: 10.1016/j.cpr.2016.03.009 [DOI] [PubMed] [Google Scholar]
38.Baer R, Crane C, Miller E, Kuyken W. Doing no harm in mindfulness-based programs: conceptual issues and empirical findings. Clin Psychol Rev. 2019;71:101-114. doi: 10.1016/j.cpr.2019.01.001 [DOI] [PMC free article] [PubMed] [Google Scholar]
39.Harrer S, Shah P, Antony B, Hu J. Artificial intelligence for clinical trial design. Trends Pharmacol Sci. 2019;40(8):577-591. doi: 10.1016/j.tips.2019.05.005 [DOI] [PubMed] [Google Scholar]
40.Broglio K, Meurer WJ, Durkalski V, et al. Comparison of bayesian vs frequentist adaptive trial design in the Stroke Hyperglycemia Insulin Network Effort trial. JAMA Netw Open. 2022;5(5):e2211616. doi: 10.1001/jamanetworkopen.2022.11616 [DOI] [PMC free article] [PubMed] [Google Scholar]
41.Haller K, Becker P, Niemeyer H, Boettcher J. Who benefits from guided internet-based interventions? a systematic review of predictors and moderators of treatment outcome. Internet Interv. 2023;33:100635. doi: 10.1016/j.invent.2023.100635 [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials