• HOME
  • NEWS
  • EXPLORE
    • CAREER
      • Companies
      • Jobs
    • EVENTS
    • iGEM
      • News
      • Team
    • PHOTOS
    • VIDEO
    • WIKI
  • BLOG
  • COMMUNITY
    • FACEBOOK
    • INSTAGRAM
    • TWITTER
Friday, November 20, 2026
BIOENGINEER.ORG
No Result
View All Result
  • Login
  • HOME
  • NEWS
  • EXPLORE
    • CAREER
      • Companies
      • Jobs
        • Lecturer
        • PhD Studentship
        • Postdoc
        • Research Assistant
    • EVENTS
    • iGEM
      • News
      • Team
    • PHOTOS
    • VIDEO
    • WIKI
  • BLOG
  • COMMUNITY
    • FACEBOOK
    • INSTAGRAM
    • TWITTER
  • HOME
  • NEWS
  • EXPLORE
    • CAREER
      • Companies
      • Jobs
        • Lecturer
        • PhD Studentship
        • Postdoc
        • Research Assistant
    • EVENTS
    • iGEM
      • News
      • Team
    • PHOTOS
    • VIDEO
    • WIKI
  • BLOG
  • COMMUNITY
    • FACEBOOK
    • INSTAGRAM
    • TWITTER
No Result
View All Result
Bioengineer.org
No Result
View All Result
Home NEWS Science News Health

Machine Learning Tailoring Shows Selective Signal for Momentary Youth Resilience

by
November 16, 2026
in Health
Reading Time: 6 mins read
0
Machine Learning Tailoring Shows Selective Signal for Momentary Youth Resilience

Machine Learning Tailoring Shows Selective Signal for Momentary Youth Resilience

Share on FacebookShare on TwitterShare on LinkedinShare on RedditShare on Telegram

In an exploratory within-subject micro-randomized trial conducted in Germany, assigning mobile mental health intervention components through a person-specific machine learning algorithm generated an initial, outcome-specific proximal improvement in momentary resilience compared with random assignment, but yielded no statistically detectable benefits for momentary positive or negative affect. The investigation, published in Translational Psychiatry, sought to establish empirical evidence on whether tailoring digital micro-interventions through predictive longitudinal modeling outperforms random component allocation in daily life. Conducted between July 2022 and February 2023, the study evaluated the feasibility, preliminary safety, and momentary impact of an adaptive intervention framework known as the Artificial Intelligence for digital personalized mental health promotion in youth, or AI4U training platform.

Ecological momentary interventions deliver structured, brief psychological exercises directly into daily routines, often building on just-in-time adaptive intervention architectures designed to respond to fluctuating individual states, environmental stressors, and personal coping capacities. While traditional adaptive interventions rely on static schedules, fixed decision rules, or threshold-based clinical triggers, they frequently struggle to accommodate subtle within-person dynamics and shifting subjective states over time. To address these operational constraints, investigators designed the AI4U system to integrate intensive longitudinal data from ecological momentary assessments with recurrent neural network modeling. This architecture was configured to simulate dynamic personal trajectories and probabilistically select intervention elements predicted to offer the greatest momentary psychological utility.

The study protocol enrolled youth aged 14 to 25 recruited from both the general community and psychological counseling services, explicitly excluding individuals experiencing acute suicidality, those with current diagnosed psychiatric disorders, or youth actively engaged in specialized clinical mental health treatment. From an initial pool of 80 interested youths, 61 provided written informed consent and 58 completed baseline clinical assessments before starting the intervention protocol. After excluding nine participants who failed to supply sufficient longitudinal data for initial algorithm training, 49 individuals entered the micro-randomization phase. This participant cohort had a mean age of 20.6 years, was 78 percent female and 20 percent male, included 27 percent with lower educational attainment, and comprised 51 percent who reported a migration background.

Intervention delivery spanned a 40-day timeframe divided into an initial 10-day introductory period followed by a 30-day active micro-randomized training phase delivered via the movisensXS smartphone application alongside care as usual. During the introductory phase, participants received eight ecological momentary assessment prompts daily to gather baseline longitudinal time series, which decreased to six prompts per day during the active trial period. Within the 30-day intervention window, participants encountered up to seven decision points each day, yielding up to 210 potential randomization occasions per participant. At each eligible decision point following an assessment prompt, individuals were randomized in a 1:1 ratio to either an experimental machine learning-informed component assignment policy or an active control condition in which candidate components were selected with equal probability.

The underlying algorithmic framework implemented recurrent neural network models rooted in dynamical systems reconstruction, a computational method engineered to capture temporal dependencies within intensive time-series observations. The network employed a clipped dendritic, piecewise-linear architecture containing 30 internal units, with 15 output units mapped directly onto 15 momentary subjective assessment variables spanning mood, disappointment, fear, worry, feeling down, sadness, confidence, stress, loneliness, energy, concentration, resilience, tiredness, satisfaction, and relaxation. The model treated eight distinct psychological intervention components—derived from evidence-based techniques such as positive refocusing, mindfulness, and compassion-focused exercises—as external input vectors linearly projecting onto the internal network units, with negatively valenced items recoded so higher values consistently denoted favorable states.

Network hyperparameters were originally configured by minimizing prediction errors on an independent longitudinal assessment dataset using Generalized Teacher Forcing, setting the latent state dimension to 30, the base count to 20, teacher forcing strength to 0.4, learning rate to 0.001, and optimization duration to 1,000 epochs. Following the initial 10-day introductory phase, an individual model was fitted for each participant and retrained from scratch every night using cumulative participant data, provided the record satisfied pragmatic safeguard thresholds of at least 1.5 daily ratings on average and a minimum total of 20 ratings. At each experimental decision point, the network ingested the latest 15-item state vector alongside 15 standard normal random values, simulated subsequent trajectory paths over six steps across approximately 24 hours for each candidate component, calculated an average predicted benefit score, and converted these values into selection probabilities using a softmax distribution with temperature parameter set to one.

Primary micro-randomized trial analyses evaluated proximal time-lagged causal excursion effects on continuous standardized outcomes at the subsequent assessment point using the weighted and centered least squares method implemented via the MRTAnalysis package in R. The primary analysis sample comprised 42 participants, after excluding six individuals due to technical data-logging errors triggered by unintended server restarts and one individual who provided insufficient assessment data. Among these 42 participants, machine learning-based component assignment produced a statistically significant proximal advantage for momentary resilience relative to random assignment, yielding a marginal causal excursion effect coefficient of 0.147 with a 95 percent confidence interval spanning 0.004 to 0.290 and an associated p-value of 0.044.

In contrast to the momentary resilience findings, algorithmic selection showed no evidence of proximal benefit for momentary positive affect or negative affect. The estimated marginal causal excursion effect for positive affect was 0.012 with a 95 percent confidence interval from minus 0.11 to 0.14 and a p-value of 0.839, while the effect for negative affect was 0.093 with a 95 percent confidence interval ranging from minus 0.02 to 0.21 and a p-value of 0.103. The mean elapsed duration between intervention assignment and the subsequent assessment was 3.09 hours with a median of 2.42 hours, exhibiting no meaningful distribution discrepancy between the experimental condition median of 2.53 hours and the active control median of 2.39 hours.

Study feasibility and user acceptance were tracked using standardized instruments, including the Mobile Application Rating Scale and the Mobile App Usability Questionnaire, alongside objective application engagement logs. Of the 49 participants entering decision-point randomization, 48 completed the post-intervention assessment, reflecting a 98 percent retention rate, while participants spent an average of nearly seven hours actively engaging with the AI4U training platform. User ratings generally demonstrated high perceived usability, practical acceptability, and program satisfaction across core application features, although participant willingness to pay for the application if commercially distributed was comparatively mixed. Crucially, systematic safety monitoring throughout the active training period recorded no study-related serious adverse events, confirming the preliminary safety and operational viability of delivering automated micro-interventions to youth.

Beyond micro-randomized proximal effects, researchers tracked exploratory distal outcomes across 44 participants who completed baseline and post-intervention evaluations, measuring psychological distress via the Kessler Psychological Distress Scale, general resilience via the Connor-Davidson Resilience Scale, and emotion regulation strategies via the Cognitive Emotion Regulation Questionnaire. Uncontrolled pre-post comparisons indicated a small reduction in psychological distress with a Cohen’s d of minus 0.24, alongside moderate improvements in general resilience with an effect size of 0.55 and adaptive emotion regulation strategies with an effect size of 0.53, while maladaptive strategies showed an effect size of minus 0.20. Because distal outcome evaluations lacked an independent person-level control group, investigators explicitly noted that these improvements cannot be causally attributed to the intervention, as observed shifts could stem from regression to the mean, monitoring reactivity, or seasonal effects.

The authors outlined multiple methodological boundaries that warrant caution when interpreting these preliminary results. The final analyzed cohort of 42 youth fell short of the a priori statistical power target of 60 participants calculated to achieve 87 percent power for detecting small standardized proximal effects of 0.10, underscoring the exploratory and preliminary nature of the findings. Furthermore, the average multi-hour delay between intervention delivery and subsequent outcome assessment may have diluted transient emotional shifts, potentially explaining the absence of significant proximal impacts on affective states. The investigators also emphasized that micro-randomized trial designs evaluate the net policy value of the composite assignment mechanism rather than validating the standalone predictive accuracy or calibration of the recurrent neural network model.

Finally, the authors underscored that the study population predominantly comprised female youth presenting with mild-to-moderate psychological distress, while excluding individuals with active clinical disorders or acute psychiatric crises, thereby restricting generalizability primarily to preventative mental health promotion rather than acute psychiatric intervention. The predictive model reconstructed internal dynamics solely from observed momentary ratings and did not incorporate passive mobile sensing features, sleep tracking, physical activity, or geolocation context. Consequently, the researchers concluded that while this trial demonstrates the operational feasibility and selective proximal promise of algorithmic ecological momentary interventions, larger, fully powered micro-randomized trials and rigorous validation studies are essential to verify whether predictive forecasting performance genuinely translates into sustainable clinical utility.

Subject of Research: Micro-randomized trial of machine learning-based ecological momentary interventions for youth mental health promotion

Article Title: A machine learning-based ecological momentary intervention for mental health promotion in youth: a micro-randomized trial

Article References: Rauschenberg, C., Schirmbeck, F., Fechtelpeter, J., Wierzba, E., Hiller, S., Kahr, K., Kessler, A., Hornbacher, L., Goetzl, C., Schick, A., Durstewitz, D., Krumm, S., Koppe, G., & Reininghaus, U. (2026). A machine learning-based ecological momentary intervention for mental health promotion in youth: a micro-randomized trial. Translational Psychiatry, 16(1), Article 487. https://doi.org/10.1038/s41398-026-04475-8

Image Credits: AI Generated

DOI: 10.1038/s41398-026-04475-8

Keywords: Ecological Momentary Intervention, Micro-Randomized Trial, Machine Learning, Recurrent Neural Networks, Dynamical Systems Reconstruction, Youth Mental Health, Momentary Resilience, Positive Affect, Negative Affect, Just-in-Time Adaptive Interventions, Ecological Momentary Assessment, AI4U Training, Psychological Distress, Emotion Regulation, Mobile Health, Causal Excursion Effect, MovisensXS

News Source: Teresa Odom. (November 16, 2026). Machine Learning Tailoring Shows Selective Signal for Momentary Youth Resilience. Scienmag.

Tags: AI4U TrainingCausal Excursion EffectDynamical Systems Reconstructionecological momentary assessmentecological momentary interventionEmotion regulationjust-in-time adaptive interventionsMachine Learningmicro-randomized trialmobile healthMomentary ResilienceMovisensXSnegative affectpositive affectpsychological distressrecurrent neural networksYouth Mental Health
Share12Tweet7Share2ShareShareShare1

Related Posts

Targeted Perioperative Haemodynamic Therapy Benefits Highest-Risk Surgical Patients Most, Review Finds

Targeted Perioperative Haemodynamic Therapy Benefits Highest-Risk Surgical Patients Most, Review Finds

November 20, 2026
Medicaid Expansion Aids Justice-Involved Populations When Combined With Enrollment Assistance

Medicaid Expansion Aids Justice-Involved Populations When Combined With Enrollment Assistance

November 20, 2026

MicroRNA Molecules Offer Preclinical Clues to Reversing Breast Cancer Radiation Resistance

November 20, 2026

Adult-Onset LRBA Deficiency Identified in Patient with Rheumatoid Arthritis and Autoimmunity

November 20, 2026

POPULAR NEWS

  • Chemical and Microbial Feedback Loops Drive Soil Sickness in Medicinal Crops

    Chemical and Microbial Feedback Loops Drive Soil Sickness in Medicinal Crops

    29 shares
    Share 12 Tweet 7
  • National Survey Reveals Significant Digital Gaps Among Groups Facing Oral Health Inequities

    29 shares
    Share 12 Tweet 7
  • Review Evaluates Strategies to Prevent and Treat Mandibular Nerve Injury During Jaw Surgery

    29 shares
    Share 12 Tweet 7
  • Dental Sector Dominates Japan’s Regional Outpatient Dysphagia Rehabilitation Claims

    29 shares
    Share 12 Tweet 7

About

We bring you the latest biotechnology news from best research centers and universities around the world. Check our website.

Follow us

Recent News

Targeted Perioperative Haemodynamic Therapy Benefits Highest-Risk Surgical Patients Most, Review Finds

Mpox Research Requires Strict Containment and Standardized Biosafety Protocols Across Laboratories

Water-Atomized S7 Tool Steel Powder Shows Strong Performance in Additive Manufacturing

Subscribe to Blog via Email

Enter your email address to subscribe to this blog and receive notifications of new posts by email.

Join 89 other subscribers
  • Contact Us

Bioengineer.org © Copyright 2023 All Rights Reserved.

Welcome Back!

Login to your account below

Forgotten Password?

Retrieve your password

Please enter your username or email address to reset your password.

Log In
No Result
View All Result
  • Homepages
    • Home Page 1
    • Home Page 2
  • News
  • National
  • Business
  • Health
  • Lifestyle
  • Science

Bioengineer.org © Copyright 2023 All Rights Reserved.