The review, led by Xiling Li, Cunyue Gong, Hegen Li, and Heng Yang of the Department of Oncology at Longhua Hospital, Shanghai University of Traditional Chinese Medicine, is a scoping review of the emerging field known as digital phenotyping. The technique involves harvesting behavioral and physiological signals from smartphones, wearable sensors, and other consumer devices without requiring patients to do anything beyond live their ordinary lives. Accelerometers log movement, GPS receivers trace patterns of mobility and isolation, and screen-use records reveal changes in how people interact with their devices. Together, these data streams form a kind of behavioral fingerprint that can shift measurably when mental health deteriorates.

To map the state of the field, the researchers searched five major biomedical databases, PubMed, Web of Science, CINAHL, the Cochrane Library, and Embase, from their earliest records through July 2026, supplementing the systematic search with a targeted sensitivity search. Two reviewers independently screened records and extracted data on study design, participant characteristics, data collection methods, mental health outcomes, and the assessment tools used to validate them. The team adhered to the PRISMA extension for scoping reviews, a reporting standard designed to make such evidence syntheses transparent and reproducible, and the full protocol was made publicly available for scrutiny.

The harvest from that search was modest but telling: 17 studies encompassing 829 participants with cancer. The numbers themselves are a finding. For a disease that affects millions of people worldwide each year, and for which psychological distress is among the most common and disabling burdens, the evidence base for passive digital monitoring remains strikingly thin and fragmented. Most of the included studies were small, used heterogeneous methods, and focused on depression and anxiety, the outcomes most frequently assessed with validated instruments such as the Hospital Anxiety and Depression Scale and the Patient Health Questionnaire-9.

Technically, the studies converged on a familiar toolkit. Wearable devices and smartphones dominated passive data collection, with accelerometers and GPS sensors being the most commonly deployed hardware. From raw sensor streams, researchers derived features across five broad domains: physical activity, location and mobility, sleep, cardiac physiology, and device-use behavior. Step counts and measures of moderate-to-vigorous physical activity quantified how much patients moved; GPS trajectories yielded metrics of mobility range and time spent at home versus places of interest; actigraphy captured sleep duration and the regularity of rest-activity rhythms; and heart rate and heart rate variability offered windows into autonomic nervous system function, which is known to shift with stress and mood.

The consistent signal emerging from these studies is that the body’s movement tells a story about the mind. Higher levels of physical activity were generally associated with lower depression, anxiety, sadness, and psychological distress across the reviewed studies. Conversely, greater sedentary behavior and disrupted sleep or flattened rest-activity rhythms were associated with higher psychological symptom levels. These associations echo a much broader literature in psychiatric digital phenotyping, where reduced mobility, social withdrawal reflected in GPS data, and disturbed circadian patterns have repeatedly been shown to correlate with depressive states. What the new review adds is evidence that the same relationships hold in a population whose activity and sleep are already perturbed by tumors and toxic therapies.

That complication matters. Cancer treatment itself alters sleep, saps energy, and confines patients to their homes, so a naive interpretation of digital markers borrowed from healthy populations risks misreading treatment side effects as psychological decline. The review’s authors are careful on this point, noting that the current evidence is insufficient to support routine clinical implementation of digital phenotyping in oncology. Passive data can describe what is happening, but translating those descriptions into clinical decisions, determining when a drop in step count or a fragmentation of sleep warrants an intervention, requires validation studies that these small, heterogeneous investigations cannot yet provide.

The methodological challenges are considerable. Sensor data are noisy, affected by device type, wear time, and individual habits. Feature definitions vary from study to study, making results difficult to pool or compare. Privacy and consent raise their own questions: continuous location tracking and behavioral surveillance, even for benevolent purposes, demand robust safeguards and clear communication with patients. And because depression and anxiety were overwhelmingly the chosen outcomes, domains such as fear of cancer recurrence and psychological distress during rehabilitation remain comparatively unexplored, despite being central to the lived experience of survivorship.

Still, the potential of the approach is difficult to overstate. Conventional assessment places the burden on patients to complete questionnaires and on clinicians to administer them at intervals, an arrangement that both burdens patients and misses the dynamic, day-to-day fluctuations in psychological state that predict clinical deterioration. Digital phenotyping inverts that model. Data accumulate silently in the background, and patterns that unfold over days rather than months become visible. Combined with brief ecological momentary assessments, in which patients answer short prompts on their phones, passive sensing could eventually give oncology teams a continuous, ecologically valid picture of mental health that no clinic-based scale can match.

The review’s authors call for larger-scale, standardized, multicenter prospective studies to move the field forward, and that prescription is likely to shape the next phase of research. If such trials confirm that sensor-derived features can anticipate psychological crises before patients report them, digital phenotyping could become a routine part of supportive cancer care, alerting clinicians to intervene earlier and more precisely. For now, the message of this careful synthesis is one of measured optimism: the technology works, the early signals are coherent, and the infrastructure to make it clinical medicine is the task that remains.

Subject of Research: Use of digital phenotyping with smartphones and wearables to monitor mental health outcomes in patients with cancer

Article Title: Digital phenotyping for monitoring mental health outcomes in patients with cancer: a scoping review

Article References: Li, X., Gong, C., Li, H., & Yang, H. (2026). Digital phenotyping for monitoring mental health outcomes in patients with cancer: a scoping review. BMC Cancer. https://doi.org/10.1186/s12885-026-17160-2

Image Credits: AI Generated

DOI: 10.1186/s12885-026-17160-2

Keywords: digital phenotyping, cancer, mental health, depression, anxiety, wearable technology, smartphones, psychological distress, scoping review, GPS tracking, sleep monitoring, oncology

The motivation is a widening crisis. Epidemiological research cited in the study indicates that roughly 30 to 35 percent of college students worldwide report symptoms of depression, anxiety, or psychological distress during their college years, and longitudinal data suggest a 40 percent rise in reported mental health issues among students over the past decade. Students with untreated conditions show a 20 to 25 percent lower grade point average and are twice as likely to leave their degree programs than their peers. Traditional assessment tools, however, struggle to keep pace with this burden.

Self-report instruments such as the Patient Health Questionnaire-9 for depression and the Perceived Stress Scale introduce recall bias and social desirability effects, impose a high burden on respondents, and offer only cross-sectional snapshots rather than longitudinal views of wellbeing. Clinical interviews, while diagnostically useful, cannot be scaled to screen large populations and are often unavailable because of stigma, cost, and insufficient campus counseling services. The episodic character of traditional assessment also cannot reflect the dynamic contextual variation of mental state that typifies the college experience, leaving clinicians and educators with an incomplete picture precisely during high-stress periods such as examinations or project deadlines.

Digital phenotyping through smartphones offers an alternative. Phones are rich sensor platforms that can record behavioral and environmental predictors of mental health without any user interaction. The study builds on the StudentLife dataset, which contains ten weeks of multimodal measurements from 48 Dartmouth undergraduates, including physical activity through accelerometry, sleep cycles inferred from phone monitoring, social engagement through conversation detection, geographical pathways, and self-reported feelings. The sensing infrastructure ran in the background on Android devices, recording roughly 672,000 hours of passive data alongside 1,920 completed self-report surveys.

The computational challenge lies in combining heterogeneous data streams with disparate sampling rates, patterns of missingness, and predictive relevance. Simple concatenation or late averaging of modalities fails to capture the complex nonlinear interactions between behavioral domains that define human psychology. The researchers therefore designed a cross-modal attention network with modality-specific encoders—temporal convolutional networks with dilated convolutions for continuous sensor streams, dense feedforward networks with embedding layers for sparse survey data—followed by multi-head scaled dot-product attention that learns dependencies between modalities bidirectionally, and a hierarchical fusion strategy combining weighted concatenation with a transformer layer.

Crucially, the architecture was not hand-designed. The team introduced the Addax Optimization Algorithm, a metaheuristic inspired by two behaviors of the addax antelope native to the Sahara: foraging, in which the animals track rainfall to locate areas with abundant vegetation, and excavation, in which they dig depressions in the sand to rest and shield themselves from sandstorms. In the algorithm, each addax represents a candidate neural architecture, the foraging phase drives exploration of the search space, and the digging phase refines solutions locally. The novel version adds a chaotic circular mapping mechanism to prevent premature convergence and a Lévy flight mechanism that introduces large jumps to escape local optima.

The search space encompassed about 47 parameters—28 discrete choices such as encoder type, attention head count, and fusion strategy, plus 19 continuous hyperparameters including learning rate and dropout rate. The optimization objective combined regression loss, classification loss, model complexity regularization, and an attention sparsity term that encourages interpretable cross-modal interactions, with at least 70 percent of attention weights required to be near zero. The Addax algorithm reached an optimal fitness of 0.894 in 42 iterations, outperforming particle swarm optimization, genetic algorithms, ant colony optimization, and Bayesian optimization, with statistically significant differences confirmed by repeated-measures ANOVA. Full optimization took 8.7 hours, a 12.1 percent improvement over Bayesian optimization.

The optimized network initially achieved 87.4 percent accuracy for depression detection on the StudentLife test set, with sensitivity of 85.6 percent and an area under the ROC curve of 0.923. Stress classification reached 83.7 percent accuracy, and the predicted mental health index correlated strongly with GPA. Modality weight analysis revealed that survey data dominated depression prediction, followed by sleep patterns and activity data, while activity data was the strongest predictor of stress. Attention weights shifted systematically with the academic calendar: location and activity were strongest during exam weeks, while sleep dominated depression prediction during regular coursework. Ablation studies confirmed that removing the attention mechanism reduced depression accuracy by 13.6 percentage points.

But the test set contained only seven students, a limitation the authors confronted directly. When the model was evaluated on the College Experience Study dataset—a longitudinal study of over 200 students collected over five years, including passively collected sensor, survey, and brain-imaging data—zero-shot transfer achieved 72.3 percent depression classification accuracy, and fine-tuning on 140 local participants raised this to 83.9 percent. The authors conclude that the initial 87.4 percent result can be interpreted as overoptimistic. Even with fine-tuning, a false positive rate of roughly 16 percent implies that for every 100 students flagged, about 15 or 16 are healthy—too high for autonomous screening or policy-level deployment.

The authors propose a phased, human-in-the-loop implementation instead. The model could serve as an early notification tool for counseling staff during predictable peak stressors such as examinations, project deadlines, and term transitions, with human judgment ultimately used for decision-making. Confidence thresholds above 0.95 would reduce flags by roughly 60 percent while preserving about 70 percent of true positives. Before any real-world deployment, the researchers recommend prospective validation across 3 to 5 institutions with at least 500 total participants, continuous institutional calibration, regular bias audits, and privacy-preserving methods such as on-device processing or federated learning. Only once false negatives for severe depression fall below 3 percent could the technology approach actual decision-making, and even then it would supplement rather than replace existing clinically focused systems.

Subject of Research: Deep learning-based multimodal mental health assessment in college students using passive smartphone sensing

Article Title: Metaheuristic-optimized cross-modal attention networks for multimodal mental health assessment in college students

Article References: Liu, W., & Qian, C. (2026). Metaheuristic-optimized cross-modal attention networks for multimodal mental health assessment in college students. Frontiers in Systems Neuroscience, 20, Article 1847214. https://doi.org/10.3389/fnsys.2026.1847214

Image Credits: AI Generated

DOI: 10.3389/fnsys.2026.1847214

Keywords: digital phenotyping, cross-modal attention network, Addax Optimization Algorithm, neural architecture search, passive smartphone sensing, college student mental health, depression detection, StudentLife dataset, College Experience Study, metaheuristic optimization, multimodal deep learning, stress prediction

The review, which followed PRISMA 2020 guidelines and screened more than 3,300 records from six databases before including 330 studies, makes clear why the black-box problem matters so much in this field. Machine learning models can already predict diagnosis, treatment response, and relapse risk from structural and functional MRI, diffusion tensor imaging, EEG, speech, electronic health records, and genetic data. But deep neural networks that achieve high accuracy often cannot explain which features drove their decisions. In medicine, where an incorrect prediction can carry serious consequences and where regulatory frameworks such as the GDPR emphasize accountability and explainability, that opacity is a barrier to clinical adoption. XAI aims to bridge the gap between complex algorithms and human clinical reasoning by revealing influential features and decision pathways.

The technical toolkit is diverse. Model-specific techniques such as Grad-CAM generate class-specific heatmaps that highlight diagnostically important regions in medical images, with Grad-CAM++ improving localization when multiple pathological areas overlap. Model-agnostic approaches such as SHAP and LIME can explain nearly any machine learning model by quantifying the contribution of individual features, though at substantial computational cost. Layer-wise Relevance Propagation and Deep Taylor Decomposition trace predictions through layers of computation to produce detailed relevance maps, while counterfactual explanations show how small changes in input would flip a prediction, supporting a form of causal reasoning. Explanations can be global, describing model behavior across an entire dataset, or local, explaining why one particular patient was flagged as high risk. Newer directions combine XAI with self-supervised and federated learning to protect privacy, and with multimodal fusion frameworks that dynamically weight contributions from MRI, EEG, genomics, and clinical data.

In neuroimaging, the results are striking but nuanced. One deep learning model trained on grey matter volume features from 925 participants across nine international centers achieved an average AUC of 82 percent under leave-one-center-out validation, rising to 83 percent when white matter and cerebrospinal fluid features were added, with peak performance of 96 percent at the best-performing site after ComBat harmonization corrected for scanner differences. Crucially, SHAP analysis revealed that the model relied on biologically plausible regions, including the thalamus, pallidum, amygdala, hippocampus, and postcentral gyrus, findings consistent with the existing literature. Grey matter volume contributed most to classification, while the other structural measures added only marginal gains. Earlier studies reported similarly impressive numbers: a VGG16-based convolutional network with transfer learning reached 84.3 percent accuracy on fMRI data, and an ensemble framework called EMPaSchiz achieved 87 percent accuracy using resting-state fMRI from antipsychotic-naïve patients.

EEG offers a cheaper and more scalable alternative, capturing abnormal neural oscillations with millisecond temporal resolution. Traditional machine learning pipelines using wavelet decomposition, entropy measures, and connectivity features have achieved accuracies between roughly 80 and 99 percent, while deep architectures including CNNs, BiLSTM models, and graph convolutional networks have reported results above 95 percent. One study combining XGBoost, SVM, and AdaBoost with SHAP analysis reached 93 percent accuracy, with interpretable features including IQ, sex, and specific EEG wave patterns. But the review’s authors urge caution: some of the highest reported figures come from small, homogeneous datasets vulnerable to data leakage, and performance that looks implausibly perfect in cross-validation rarely survives contact with external cohorts. Meta-analytic estimates are more sobering, with pooled MRI-based classification showing sensitivity of about 79 percent and specificity of about 80 percent.

Some of the most original work moves beyond the brain scan entirely. In the Italian SPECTRA project, researchers are combining speech, emotion, and locomotion data to identify treatment-resistant schizophrenia, a condition affecting roughly 30 percent of patients that is often diagnosed too late for clozapine to be started promptly. Using multi-camera motion capture and a dual-transformer architecture to reconstruct three-dimensional skeletal movement, the team identified 16 quantifiable gait markers associated with the illness. A SHAP-based visualization renders the model’s reasoning as a color-coded matrix, with red regions marking body joints and video frames that push the prediction toward schizophrenia and blue regions marking those that pull against it, allowing clinicians to inspect exactly which movements the algorithm found telling. Features such as head posture, walking speed, arm sway, and step-length variability, drawn from prior clinical work, anchor the analysis in measurable behavior.

Language and handwriting offer another window. Formal thought disorder is a core feature of schizophrenia, and recent findings suggest that specific PANSS items such as conceptual disorganization and unusual thought content predict treatment resistance. A new 16-item screening instrument for schizophrenic features in writing, validated in 157 participants, produced markedly higher scores in patients than controls and reached 93 percent sensitivity and specificity at a cutoff of 2, with discriminant analysis classifying cases with roughly 90 percent accuracy. On the treatment side, researchers combined the Personalized Advantage Index with Bayesian Rule Lists, an XAI method that generates transparent if-then rules with explicit cutoffs. Applied to a randomized trial of paliperidone, the framework identified a treatment-indicated subgroup whose effect size reached a Cohen’s d of 1.51, far exceeding the 0.82 seen in the overall sample, while consistently naming disturbance of volition and uncooperativeness as key predictors.

Where might AI contribute most immediately? The review argues that continuous monitoring, rather than one-time diagnosis, is the field’s most realistic near-term value. Smartphone-based digital phenotyping has demonstrated that statistically significant behavioral anomalies can precede relapse, and a 2025 study using the CrossCheck dataset forecast mental states hours to days in advance from passive sensor data using XGBoost and LSTM models. Digital medicine systems have already shifted real outcomes: in a phase 3b mirror-image trial, aripiprazole tablets with ingestible sensors were associated with reduced psychiatric hospitalization and a 26.5 percentage-point improvement in medication coverage during the first three months. Yet the evidence is mixed elsewhere. The Actissist trial, a powered randomized study of a CBT-informed app for early psychosis, failed to beat digital symptom monitoring, a reminder that digital delivery alone does not guarantee clinical benefit.

The authors are unsparing about the field’s weaknesses. Risk-of-bias analysis of a representative subset found high overall risk in eight of thirteen studies, largely from overfitting and inadequate validation. Diagnostic models mostly separate patients from healthy controls, an easier problem than distinguishing schizophrenia from overlapping psychiatric syndromes. Regulatory space remains nearly empty: FDA materials note that more than 1,200 AI-enabled medical devices have been authorized, but none for mental health indications, and generative AI chatbots raise specific concerns that they may validate or amplify delusional content in vulnerable users. Patient perspectives on trust, privacy, and acceptance are almost entirely absent from the literature.

The review closes with a concrete roadmap. Within one to three years, the field needs standardized multicenter datasets and clinically meaningful explainability metrics such as explanation fidelity, stability, and actionability. Within three to five years, prospective studies should test workflow integration and clinician trust against reporting standards including CONSORT-AI and STARD-AI, while within a decade, diagnostic systems should demonstrate externally validated AUCs of at least 0.90, clinician agreement with explanations above 80 percent, a 15 to 20 percent improvement in relapse prediction, and measurable gains in patient-centered outcomes such as adherence and quality of life, all under the transparency and oversight requirements of the EU AI Act. The verdict on explainable AI in schizophrenia is clear-eyed: scientifically promising, particularly for multimodal prediction and continuous monitoring, but clinically immature. The next decade will be judged not by headline accuracy figures but by whether these transparent systems reduce relapse, improve functioning, and earn the trust of clinicians and patients alike, as an adjunct to expert psychiatric care rather than a replacement for it.

Subject of Research: Explainable artificial intelligence for transparent clinical decision-making in schizophrenia

Article Title: Explainable artificial intelligence (XAI) in schizophrenia: enhancing transparency in clinical decision-making

Article References: Srivastava, K., & Srivastava, R. (2026). Explainable artificial intelligence (XAI) in schizophrenia: enhancing transparency in clinical decision-making. Frontiers in Systems Biology, 6, Article 1898853. https://doi.org/10.3389/fsysb.2026.1898853

Image Credits: AI Generated

DOI: 10.3389/fsysb.2026.1898853

Keywords: explainable AI, schizophrenia, machine learning, SHAP, LIME, Grad-CAM, neuroimaging, EEG, digital phenotyping, precision psychiatry, treatment-resistant schizophrenia, clinical decision support

The new study, published in Frontiers in Medical Technology, explored whether three-dimensional facial imaging could serve as a source of digital biomarkers in rare diseases, with a particular focus on hereditary angioedema, or HAE. HAE is a genetic disorder caused by a defect in the SERPING1 protein, which regulates inflammation and fluid balance in blood vessels. The result is intermittent swelling of soft tissues, most often around the eyes and mouth. In most cases the swelling is painful and disfiguring, but when it involves the airways it can become life-threatening within hours, demanding immediate medical care.

The problem, the researchers explain, is that HAE is frequently missed or diagnosed late, and even after a diagnosis is secured, there is no objective way to measure a flare-up. Treatment decisions rest on subjective patient reports and clinical impressions, which can delay therapy and heighten anxiety. The team, led by Saumya Jamuar of KK Women’s and Children’s Hospital in Singapore and Gareth Baynam of Perth Children’s Hospital in Australia, set out to determine whether the face itself could provide measurable, quantifiable evidence of disease activity.

The study, conducted over four years and eight months across three Singapore hospitals, had two complementary components. In the quantitative arm, 20 participants of Chinese genetic ancestry with rare diseases characterized by facial swelling or asymmetry were photographed with a Vectra M3 3D camera. The cohort included individuals with mucopolysaccharidosis, Noonan syndrome, incontinentia pigmenti, velocardiofacial syndrome, and achondroplasia, as well as five with HAE. Their facial scans were then processed with Cliniface, an artificial intelligence-powered software platform developed in Western Australia, which extracts facial landmarks and 58 distinct measurements from each image.

The analytical approach was deliberately rigorous. From the 58 base measurements, the team generated 108 separate measurement variables, each expressed as a z-score comparing the affected participant to a regressed growth curve derived from a demographically matched, disease-free reference population spanning ages zero to 70. Outliers beyond three standard deviations were removed before regression, and the resulting z-scores were tested for normality using Monte Carlo hypothesis testing. Two-tailed t-tests were then applied, and because so many variables were tested simultaneously, the team applied the Holm-Bonferroni correction to control the family-wise error rate, a conservative method that does not assume the measurements are statistically independent.

After this gauntlet of statistical filtering, a single measurement survived: the Outer Canthal, Nasal Angle, which retained a corrected p-value of 0.01, indicating a significant deviation from the ethnically matched norm. Ten other measurements, including several related to facial asymmetry, showed uncorrected p-values below 0.05 but did not survive the multiple-comparison adjustment. The researchers caution that the small sample size and the heterogeneity of the test cohort mean these findings require independent replication. Still, the result suggests that 3D facial imaging may be capable of detecting subtle, measurable differences in people who experience facial swelling even between overt episodes.

The qualitative arm of the study provided the more visually striking findings. Two adults with HAE were photographed serially over extended periods, including during acute flare-ups. The first participant, a man of Malay genetic ancestry, was imaged at seven time points over 687 days, including immediately after a facial oedema flare. Procrustes analysis of his scans revealed asymmetric periorbital swelling, with an increase of more than three millimeters around one eye following the flare. The swelling diminished over subsequent visits, and a later scan taken during a period when the participant had stopped taking his medication showed residual swelling that had not fully subsided, hinting that the technique might even detect subclinical exacerbations or treatment non-compliance.

The second participant, a woman of Indian genetic ancestry, was photographed twice, 261 days apart, with the first image captured immediately after an acute episode. Her scans showed asymmetric periorbital swelling, more pronounced on one side, along with asymmetric swelling of the cheeks and lower jaw. The consistency of these findings between two unrelated individuals, each showing swelling and asymmetry during flares, even on different sides of the face, led the team to propose periorbital swelling and asymmetry as a working hypothesis for a digital biomarker in HAE. Notably, they point out that purely landmark-based approaches that ignore symmetry could miss critical indicators of disease activity.

The researchers are careful to frame their conclusions as preliminary. The quantitative cohort contained only five individuals with HAE, the qualitative analysis was limited to two participants, and the reference growth curves did not account for sex differences. Singapore’s small population and the rarity of these conditions made a larger cohort impractical, but the team acknowledges that broader studies across different genetic ancestries, ages, and sexes will be essential before the approach can be validated for clinical use. They also note that while digital phenotyping’s diagnostic capabilities may be limited, its ability to identify digital biomarkers could still facilitate disease monitoring even when it cannot replace clinical diagnosis.

If validated, the implications extend well beyond HAE. Rare disease clinical trials are chronically hampered by small, fragmented patient populations, subjective outcome measures, and symptoms that fluctuate over time, contributing to higher failure rates than in trials for common conditions. An objective, non-invasive biomarker derived from a simple facial photograph could provide the precise, repeatable endpoints that such trials desperately need. The team hopes that ongoing data collection from participants of different ancestries and disease conditions will identify additional digital biomarkers, moving the field closer to a future where a face scan helps clinicians detect, monitor, and treat rare diseases with greater confidence and speed.

Subject of Research: Utility of 3D facial analysis as a biomarker in rare diseases exploration with hereditary angioedema

Article Title: Utility of 3D facial analysis as a biomarker in rare diseases exploration with hereditary angioedema

Article References: Jamuar, S., Palmer, R., Lee, H. Y., Chia, F. L.-A., Goh, C. B., Lee, S., Helmholz, P., Chan, S., & Baynam, G. (2026). Utility of 3D facial analysis as a biomarker in rare diseases exploration with hereditary angioedema. Frontiers in Medical Technology, 8, Article 1857918. https://doi.org/10.3389/fmedt.2026.1857918

Image Credits: AI Generated

DOI: 10.3389/fmedt.2026.1857918

Keywords: 3D facial imaging, digital biomarkers, hereditary angioedema, rare diseases, digital phenotyping, diagnostic odyssey, Cliniface, periorbital swelling, precision medicine, SERPING1, clinical trials, Singapore

The analysis drew on data from 120 generally healthy, middle-aged adults enrolled in the Alpha-ketoglutarate supplementation and BiologicaL agE in middle-aged adults trial, known as ABLE. The study was a double-blind, randomized, placebo-controlled trial registered as NCT05706389, designed to test whether six months of daily oral supplementation with one gram of alpha-ketoglutarate, a naturally occurring metabolite implicated in cellular metabolism and previously studied in aging research, would influence biological age measures in people in midlife. Participants, who had a mean age of 50.6 years with a standard deviation of 6.0 years and were evenly split between women and men, were assigned to either the supplement arm, with 59 participants, or the placebo arm, with 61 participants, and followed for six months with an additional three-month follow-up period.

At each of four study visits, participants were instructed to wear an ActiGraph wGT3X-BT accelerometer for seven consecutive days. These devices are small, lightweight triaxial accelerometers widely used in physical activity research, capable of recording movement patterns around the clock and generating objective estimates of activity intensity, sleep timing, and circadian rhythm. In the context of a gerotherapeutic trial, such continuous data streams serve as digital phenotypes: objective, high-resolution markers of physical behavior that may change as a person’s biological aging trajectory shifts. The research team, led by Jessica K. Lu and colleagues including Brian K. Kennedy and Andrea B. Maier, set out to determine whether this level of monitoring was genuinely workable for a population of adults in their late forties and fifties, a group often overlooked in aging trials that tend to focus on either younger volunteers or frail older adults.

The adherence results were striking. The median accelerometer wear time was 24 hours per day, meaning that the typical participant kept the device on essentially continuously throughout each seven-day monitoring window, removing it only briefly if at all. More than 80 percent of participants achieved at least four valid days of wear per visit, a threshold commonly used in the field to define an adequate monitoring window for physical activity analysis. Supplement use adherence exceeded 90 percent across the six-month intervention, indicating that participants managed both the passive monitoring and the active intervention components of the trial simultaneously without meaningful compromise on either front. Device-related operational concerns, such as malfunctioning units, charging difficulties, or discomfort, were minimal across all four visits, suggesting that the ActiGraph platform and the study’s implementation protocols were well matched to the needs and habits of middle-aged participants.

Equally important was what the researchers learned about who stayed in the study. Overall dropout across the trial was 17 percent, a figure that compares favorably with many longitudinal intervention studies but still represents a meaningful loss of statistical power and resources. To understand what predicted attrition, the team turned to survival analysis, specifically Cox proportional hazards models with study visit as the time scale. This approach allowed the investigators to account for when during the trial each participant dropped out, rather than simply tallying who finished and who did not, thereby using the full longitudinal structure of the data. The models tested whether the amount of time participants wore the accelerometer, or their overall adherence to the device, was associated with their risk of leaving the study before completion.

The results of those models carried a provocative implication: every additional minute of accelerometer wear time was associated with a lower hazard of dropout, with a hazard ratio of 0.998 and a 95 percent confidence interval of 0.997 to 0.999. While a hazard ratio per minute may sound small, the effect compounds across the thousands of minutes in a seven-day wear window, and the confidence interval excludes the null value of 1.0, indicating a statistically reliable association. In practical terms, participants who kept their devices on longer were meaningfully less likely to abandon the trial. The authors suggest that wear time may function as a behavioral marker of engagement, a real-time signal that trial coordinators could monitor to identify participants at elevated risk of dropping out and intervene before they disengage entirely.

Sex emerged as the other significant predictor. Being male was associated with a substantially lower risk of dropout, with a hazard ratio of 0.208 and a 95 percent confidence interval of 0.065 to 0.663. In other words, men in this cohort were roughly one-fifth as likely to leave the study as women, a large and statistically robust difference. The finding raises questions about whether middle-aged women face different burdens or motivations in aging intervention trials, whether the supplement or monitoring protocol interacted differently with women’s daily routines, or whether other unmeasured factors drove the disparity. The authors note that understanding and addressing this gap will be important for designing future trials that retain diverse participants and produce generalizable results.

Beyond the statistical associations, the study offers a practical playbook for trial designers considering wearable-based digital phenotyping. The near-universal 24-hour wear pattern suggests that middle-aged adults are willing to treat research accelerometers much like consumer fitness trackers, integrating them into daily life with little prompting. The high rate of achieving at least four valid days per visit indicates that a seven-day monitoring window provides a comfortable buffer, allowing participants to accumulate sufficient data even with occasional removals for bathing or charging. The minimal rate of device-related operational concerns implies that established research-grade hardware, paired with clear instructions and structured visit schedules, can avoid the technical pitfalls that sometimes plague wearable deployments, such as data loss, initialization errors, or participant confusion about when to wear the device.

The implications extend to the growing field of gerotherapeutics, where interventions aim not to treat a single disease but to slow, halt, or reverse the biological processes of aging themselves. Trials in this space increasingly rely on multidimensional readouts, from epigenetic clocks and blood biomarkers to continuous sensor data, to capture the subtle, systemic changes that aging interventions are designed to produce. Accelerometer-derived phenotypes, including total activity volume, activity fragmentation, and sleep-wake rhythms, are attractive because they are objective, repeatable, and sensitive to functional change. The ABLE analysis demonstrates that such data can be collected at high fidelity in a placebo-controlled supplement trial without overburdening participants or inflating dropout, removing a key practical objection to including wearables in future gerotherapeutic study designs.

The authors are careful to frame the work as exploratory, drawn from a single trial with a relatively homogeneous cohort of generally healthy middle-aged adults, so the findings may not generalize to older, frailer, or more clinically diverse populations, where device tolerance and dropout dynamics could differ substantially. Still, the central lessons are actionable: monitor wear time as an early-warning indicator of disengagement, attend to sex differences in retention, and design monitoring protocols around the demonstrated willingness of middle-aged adults to wear accelerometers continuously. As interest in alpha-ketoglutarate and other candidate gerotherapies accelerates, the operational insights from ABLE arrive at an opportune moment, offering trialists a data-driven foundation for building studies that keep participants engaged, keep sensors on wrists, and keep the digital phenotype data flowing from the first visit to the last.

Subject of Research: Feasibility and adherence of accelerometer-based digital phenotyping in a gerotherapeutic supplementation trial in middle-aged adults

Article Title: Optimizing accelerometer implementation in a gerotherapeutic trial: Feasibility, adherence, and operational insights of ABLE

Article References: Lu, J. K., Wang, W., Chew, Y. E., Guan, L., Kennedy, B. K., & Maier, A. B. (2026). Optimizing accelerometer implementation in a gerotherapeutic trial: Feasibility, adherence, and operational insights of ABLE. PLOS Digital Health, 5(9), e0001691. https://doi.org/10.1371/journal.pdig.0001691

Image Credits: AI Generated

DOI: 10.1371/journal.pdig.0001691

Keywords: accelerometers, digital phenotyping, gerotherapeutics, alpha-ketoglutarate, clinical trial adherence, study dropout, middle-aged adults, wearable technology, ActiGraph, Cox proportional hazards, aging intervention, PLOS Digital Health

The study tapped the Northern Finland Birth Cohort 1986, one of the world’s longest-running population-based longitudinal studies, which has followed thousands of individuals since before birth. When participants reached 33 to 35 years of age, they were asked to wear an Oura Ring, a commercially available smart ring, for two weeks. To ensure the data were robust, the researchers only included participants who provided at least one weekday and one weekend day with no less than fifteen hours of wear time. That filtering step left 1,290 participants with valid recordings, a remarkably large sample for research involving continuous physiological monitoring.

During those two weeks, the ring captured an unusually rich set of digital metrics. It recorded sleep architecture, meaning the distribution of sleep stages across the night, including rapid eye movement (REM) sleep, deep sleep, and light sleep. It measured nocturnal heart rate using photoplethysmography, an optical technique that tracks blood volume pulses beneath the skin. It derived heart rate variability, indexed by the root mean square of successive differences between normal heartbeats, a statistic known as RMSSD that reflects the fine-tuned activity of the autonomic nervous system. And it logged minute-by-minute movement intensity throughout the day, providing an objective picture of physical activity levels.

Mental health symptoms were assessed with two well-validated screening instruments. The Generalized Anxiety Disorder 7-item scale, or GAD-7, quantified anxiety symptoms, allowing participants to be grouped into no, mild, and moderate-to-severe symptom categories. The Hopkins Symptom Checklist-25, or HSCL-25, captured a broader blend of depression and anxiety symptoms, dividing participants into those with and without clinically relevant symptom burden. The researchers then compared the wearable-derived metrics across these groups using analysis of covariance, a statistical technique that controls for potential confounding variables, followed by Tukey’s post hoc test to pinpoint which group differences were statistically meaningful.

The results were striking in their consistency. Participants with moderate-to-severe anxiety on the GAD-7, and those flagged by the HSCL-25 as having depression and anxiety symptoms, spent a lower percentage of the night in REM sleep and deep sleep, and a higher percentage in light sleep, compared with participants who reported no depression or anxiety. REM sleep is the stage most closely associated with emotional memory processing and mood regulation, while deep sleep supports physical restoration and cognitive function. A shift away from these restorative stages toward lighter, more fragmented sleep is a well-documented feature of mood and anxiety disorders, but it had rarely been captured so cleanly in a free-living population using an unobtrusive consumer device.

The physiological differences extended beyond sleep staging. The symptomatic groups showed higher nocturnal heart rates and lower RMSSD during sleep, a combination that points to sustained sympathetic arousal, the fight-or-flight branch of the autonomic nervous system remaining more active than usual even during rest. Elevated resting heart rate and suppressed heart rate variability are classic correlates of chronic stress and have been linked in prior research to cardiovascular risk, making their presence in young adults with mental health symptoms particularly noteworthy. In parallel, the same groups recorded lower levels of physical activity during the day, consistent with the fatigue, anhedonia, and behavioral withdrawal that often accompany depression and anxiety.

Notably, the picture was less clear-cut for people with mild anxiety. The differences between the mild anxiety group and both the symptom-free and moderate-to-severe groups were overall less apparent on average, suggesting that the wearable metrics scale with symptom severity rather than merely detecting the presence of any symptoms at all. This dose-response-like pattern strengthens the argument that the ring’s data carry genuine signal about mental state, rather than simply flagging anyone who differs from the population average. It also hints at a potential role for such devices in tracking symptom trajectories over time, since worsening physiology might accompany worsening symptoms.

The researchers describe these wearable-derived measures as digital biomarkers, objective physiological and behavioral signals that can be collected continuously and passively in everyday life. The concept of digital phenotyping, using data from personal devices to characterize health states, has gained momentum as smartphones and wearables have become ubiquitous. What distinguishes this study is its population-based design and its use of a consumer-grade device rather than research-grade laboratory equipment. Polysomnography, the gold standard for sleep measurement, requires an overnight clinic stay with electrodes glued to the scalp and chest, an experience so intrusive that a single night’s recording may not represent typical sleep. A ring, by contrast, collects weeks of data in the participant’s own bed.

The implications reach into clinical practice and public health alike. Depression and anxiety are among the most common mental health conditions worldwide, yet they are frequently underdiagnosed, and treatment monitoring typically relies on periodic self-report questionnaires that are vulnerable to recall bias and stigma. If wearable metrics can complement these tools, clinicians might one day spot early warning signs between appointments, tailor interventions to individual physiological profiles, and evaluate treatment response with far greater temporal resolution. The study’s authors suggest that consumer wearables may be a promising tool for enhancing symptom tracking and informing personalized care strategies, though they stop short of claiming diagnostic capability.

Important caveats remain. The study was cross-sectional, capturing a snapshot in time, so it cannot determine whether the physiological differences cause the symptoms, result from them, or arise from shared underlying factors. The cohort, born in northern Finland in 1986, is predominantly young and Finnish, and generalization to other ages and populations will require further study. Screening questionnaires are not clinical diagnoses, and the effect sizes, while statistically significant, describe group averages rather than individual predictions. Still, the scale and consistency of the findings mark a milestone for the field. As wearables spread to hundreds of millions of wrists and fingers, the quiet data they gather each night may prove to be one of the most valuable untapped resources in mental health research, turning an ordinary consumer gadget into a sentinel for the mind.

Subject of Research: Digital biomarkers of depression and anxiety severity measured with a consumer wearable ring

Article Title: Severity of depression and anxiety symptoms is reflected in physiological and behavioral metrics collected from a consumer-grade wearable ring

Article References: Azadifar, S., Sameh, A., Nauha, L., Kärmeniemi, M., Niemelä, M., & Farrahi, V. (2026). Severity of depression and anxiety symptoms is reflected in physiological and behavioral metrics collected from a consumer-grade wearable ring. BMC Medicine, 24(1), Article 532. https://doi.org/10.1186/s12916-026-05276-y

Image Credits: AI Generated

DOI: 10.1186/s12916-026-05276-y

Keywords: wearable technology, digital biomarkers, depression, anxiety, Oura ring, sleep architecture, heart rate variability, mental health, digital phenotyping, Northern Finland Birth Cohort, BMC Medicine, population health

The study was motivated by a gap in the existing literature. Most research on digital media and cognition has focused on total screen time, treating hours of use as the key exposure variable. The authors argue that this approach misses the structure of modern digital behaviour. Generation Z university students, often described as digital natives, do not simply use their phones a lot; they use them in a particular pattern, dominated by brief sessions, frequent pickups and a substantial share of passive consumption such as scrolling through content without active engagement. To capture that structure, the researchers set out to characterize the entire digital footprint, including duration, fragmentation and content type, and to test how each dimension relates to cognitive control.

The methodology combined ecological realism with statistical rigour. Participants were health sciences students whose smartphones recorded seven days of objective usage logs. Rather than bringing students into a laboratory to perform a computerized task on unfamiliar equipment, the team deployed a custom mobile Stroop task that participants completed on their own devices, maximizing ecological validity. In the Stroop paradigm, people must name the ink colour of a word while ignoring the word itself; incongruent trials, where the word and colour conflict, tax the brain’s conflict-resolution machinery, while congruent trials, where word and colour match, can be resolved largely by automatic processing. A clinical battery also assessed sleep quality, attention deficit hyperactivity disorder symptoms, anxiety, headache, neck disability and computer vision syndrome, allowing the researchers to separate digital habits from somatic and psychiatric burden.

The objective data painted a portrait of intensive, highly fragmented use. The median participant logged 6.6 hours of daily screen time, picked up the phone 142 times per day, and averaged just 2.9 minutes per session, with roughly 40 percent of consumption being passive. These numbers quantify what many people intuitively recognize: the modern smartphone day is not a few long engagements but a continuous drip of micro-sessions, each one an interruption of whatever came before. The researchers also developed a Cumulative Digital Load Score, a self-report measure intended to summarize this burden, and compared it against the device logs.

Analyses proceeded on two levels. First, theory-driven family-based correlation analyses, corrected for multiple comparisons across 13 tests using the false discovery rate procedure, examined how screen patterns related to Stroop performance. These analyses showed that fragmentation metrics were significantly associated with conflict resolution costs, meaning that students whose phone use was most broken up performed worse when faced with incongruent, conflicting information. Second, trial-level linear mixed-effects models were fitted across 12,035 individual Stroop trials, allowing the researchers to isolate the momentary effects of different predictors on accuracy trial by trial, rather than relying on coarse summary scores.

The results of the modelling revealed two distinct profiles of impairment. High digital attention fragmentation was linked to a selective weakening of proactive cognitive control: it increased errors specifically on incongruent trials, the trials that require maintaining an intention and applying it against interference. In contrast, clinical and somatic symptom burden, along with ADHD symptomatology, was associated with a broader accuracy cost that extended even to highly automated congruent trials. Particularly for somatic pain burdens, the deficit was not confined to effortful conflict resolution but spilled over into tasks that normally run on autopilot. This dissociation is analytically important, because it suggests that fragmented media use and clinical symptom load degrade cognition through different mechanisms rather than a single generic impairment.

Passive consumption told its own story. The study found that passive use was associated with block-dependent processing delays, slowing responses without affecting performance on highly automatic cognitive tasks. The authors are careful to flag a limitation here: because the two task blocks were administered in a fixed order, this contrast is confounded with task order and practice effects, so the passive consumption finding should be interpreted with caution. Even so, the pattern fits a model in which passive scrolling dulls momentary processing speed while leaving automated routines intact, whereas fragmentation erodes the more strategic, anticipatory layer of executive control.

One of the study’s most sobering results concerns measurement itself. The convergence between the self-reported Cumulative Digital Load Score and the objective device logs was weak, with correlations of roughly 0.14 to 0.18. In other words, what students believed about their digital load and what their phones actually recorded captured related but distinct aspects of behaviour, not interchangeable estimates of the same construct. This has practical implications far beyond the study: much of the existing literature on screen time and mental health rests on self-report, and this work adds to growing evidence that subjective estimates of device use can diverge substantially from reality.

The authors’ conclusion resists a simple moral panic. Digital exposure, they argue, does not produce a single, homogeneous cognitive deficit. Instead, the data support a more nuanced account in which the pattern of use matters more than the volume. Highly fragmented use, defined by constant interruptions and short session durations, is associated with poorer proactive cognitive control and a specific accuracy cost on demanding trials, while passive consumption is linked to processing delays that spare automatic cognition. Because the study is cross-sectional, it cannot establish causation; it remains possible that people with weaker proactive control are also more prone to fragmenting their attention, rather than fragmentation causing the deficit. Longitudinal and interventional work will be needed to settle the direction of the relationship.

Even with that caveat, the practical message is actionable. If fragmentation, not duration, is the cognitive culprit, then cognitive health in the digital age may benefit from targeted digital hygiene strategies that prioritize minimizing attention fragmentation, such as batching notifications, lengthening uninterrupted periods and reducing the sheer number of pickups, rather than fixating solely on cutting total screen time. For a generation whose phones register more than a hundred pickups a day, the difference between six and four hours of screen time may matter far less than whether those hours arrive as sustained engagement or as a storm of two-minute fragments. The study, funded in part by TÜBA and approved by the Gazi University Ethics Commission, offers a technically sophisticated template for asking that question with objective data, and it reframes the debate about screens from how much to how.

Subject of Research: How smartphone use fragmentation and passive consumption relate to cognitive control in Generation Z university students.

Article Title: The neurocognitive cost of media multitasking: fragmented screens, fragmented minds!

Article References: Ince, M. S., Guzel, I., Bahcelioglu, M., & Bolay, H. (2026). The neurocognitive cost of media multitasking: fragmented screens, fragmented minds!. BMC Neuroscience. https://doi.org/10.1186/s12868-026-01043-0

Image Credits: AI Generated

DOI: 10.1186/s12868-026-01043-0

Keywords: media multitasking, cognitive control, digital footprint, screen fragmentation, passive consumption, proactive control, Stroop task, digital phenotyping, smartphone use, ADHD symptoms, digital hygiene, BMC Neuroscience