Background: Parkinson disease (PD) is a pervasive neurodegenerative disorder globally, largely characterized by motor symptoms. Most existing artificial intelligence models for PD detection are trained on participants in well-resourced settings with confirmed clinical diagnoses. However, specialist-confirmed labels are often infeasible in low-resource settings. Objective: We developed a web platform for structured mouse data collection through pattern tracing tests. We sought to assess the feasibility of leveraging data from a community-recruited sample of participants with suspected but undiagnosed PD to train artificial intelligence models that achieve respectable performance in predicting diagnosed PD. We tested whether using weaker diagnostic labels that may be more feasible to collect in community or global health settings, where access to professional neurologists is sparse or nonexistent, can lead to models that learn predictive signals that are diagnostically useful. Methods: 261 participants (73 self-reported PD, 155 non-PD, and 33 suspected PD) were recruited from community organizations in Hawaii and completed 3 pattern tracing tasks on our custom web assessment: straight line, sine wave, and spiral wave. During each task, cursor positions, screen dimensions, and an in-target boolean flag were recorded. From these data, we engineered features and generated mouse trace images. We built 3 categories of classifiers: (1) a feed-forward neural network using engineered features, (2) fine-tuned computer vision deep learning models, and (3) multimodal models concatenating a feed-forward neural network with computer vision models. Performance was evaluated using 1 primary experiment and 2 secondary analyses. The primary experiment involved training on suspected PD versus non-PD and testing on self-reported PD versus non-PD. A secondary analysis evaluated the reverse direction by training on participants with self-reported PD and without PD and then testing on participants with suspected PD versus participants without PD. Additionally, a cross-validation analysis was conducted using participants with self-reported PD versus those without PD with 5-fold cross-validation to establish baseline performance under well-defined diagnostic labels. Results: The best-performing models included a multimodal Vision Transformer in the primary experiment (: mean 0.7619, SD 0.0535), a multimodal ResNet-50 in the secondary analysis (: mean 0.9353, SD 0.0334), and an image-based DenseNet-201 in the cross-validation analysis (: mean 0.9027, SD 0.0332). Training on patients with suspected PD yielded meaningful performance in predicting self-reported PD, supporting the feasibility of using lower-specificity labels for model development. Conclusions: This pilot feasibility study suggests that remotely collected mouse-tracing data can support PD screening models under data labeling conditions of low diagnostic specificity: models trained on suspected PD from a community sample may learn signals that can transfer to predicting actual PD. Future work may consider pretraining using weaker labels and then fine-tuning on stronger clinical labels.
<img src="https://jmir-production.s3.us-east-2.amazonaws.com/thumbs/591064dc81d94c027cbe8ed308f60713" />
<![CDATA[Expert shares ADHD medication sequencing tips and practical insomnia strategies, from stimulants to melatonin, for kids and adults.]]>
Adherence to a Digital Knee Rehabilitation Platform Among Patients With Knee Osteoarthritis and Anterior Cruciate Ligament Reconstruction in Hong Kong: Qualitative Study
Background: Exercise therapy is fundamental to rehabilitation for knee osteoarthritis and anterior cruciate ligament (ACL) reconstruction, yet adherence to prescribed exercise typically declines once clinical supervision ends. Digital rehabilitation platforms offer a promising means of supporting sustained exercise adherence, but qualitative evidence on how patients experience these platforms in real-world clinical practice remains limited, particularly in non-Western health care contexts. Objective: This study aimed to explore how patients with different knee conditions experienced the Healthy Knees digital rehabilitation platform in Hong Kong and to identify the factors shaping their platform engagement and exercise adherence. Methods: A qualitative design was adopted using reflexive thematic analysis. Fifteen adults (9 with ACL, 6 with osteoarthritis) who had been prescribed the Healthy Knees web-based platform at Prince of Wales Hospital participated in semistructured, in-person interviews (30‐45 min). Interviews were conducted in Cantonese or Mandarin, transcribed verbatim, translated into English, and analyzed inductively. Ethics approval was obtained from the Chinese University of Hong Kong and the University of New South Wales. Results: Participants were aged 21 to 79 years, with most being male (11/15). Younger participants were predominantly patients with postoperative ACL, while older participants were predominantly patients with preoperative osteoarthritis. Three interrelated themes were identified, collectively describing the fit between the platform and participants’ contexts. Content fit captured the alignment between exercise content and rehabilitation needs; participants across both groups perceived substantial overlap with existing physiotherapy, and content was often mismatched to their recovery stage. Motivational fit captured the alignment between platform support features and motivational needs; pain functioned as both a driver and a deterrent to exercise, and participants ranged from highly self-directed to reliant on external scaffolding, not following a simple age pattern. Access fit captured the alignment between the platform’s delivery mechanism and participants’ technological circumstances; QR code–dependent access, absence of a dedicated mobile app, and display issues created friction that led several participants to migrate to alternative resources, maintaining exercise adherence while abandoning platform engagement. Conclusions: Adherence to digital knee rehabilitation was shaped by the degree of fit between the platform and users’ contexts across content, motivational, and access dimensions. When access fit failed, participants often substituted alternative exercise resources rather than ceasing exercise entirely, highlighting a distinction between platform engagement and exercise adherence. As the sample’s clinical and demographic characteristics were closely linked, these findings should not be interpreted as diagnostic comparisons between ACL and osteoarthritis populations but as patterns shaped by the recovery phase and age. These findings suggest that digital rehabilitation platforms should incorporate adaptive content aligned with the recovery stage, integrated feedback mechanisms, and reduced access friction to sustain platform engagement within an ecosystem of competing alternatives.
<img src="https://jmir-production.s3.us-east-2.amazonaws.com/thumbs/7a14e974f70af5794765b21582feb228" />
Perspectives on Continuous Glucose Monitoring Among Adults with Type 2 Diabetes in the United Kingdom: Cross-Sectional Survey
<strong>Background:</strong> Type 2 diabetes (T2D) is one of the most common noncommunicable diseases, requiring ongoing lifestyle changes and continuous glucose management through medication, diet, and physical activity. Traditional self-monitoring of blood glucose can be burdensome, especially with frequent finger pricks. As continuous glucose monitoring (CGM) becomes more affordable and accessible, it offers benefits such as increased glucose awareness, behavioral modifications, and reduced anxiety. However, challenges remain, including cost, discomfort, skin reactions, and privacy concerns. In the United Kingdom, perceptions of CGM among people with T2D, including both users and nonusers, are not well understood, limiting insight into factors influencing adoption and sustained use. <strong>Objective:</strong> This study aims to explore how adults with T2D perceive the benefits and challenges of using CGM, including both current users and nonusers. <strong>Methods:</strong> This study used a cross-sectional, online survey using YouGov’s nationally representative panel to explore experiences of CGM among adults with T2D in the United Kingdom. A total of 531 participants were recruited from November to December 2024. Thematic analysis of responses to 2 open-ended questions identified key perceived benefits and challenges associated with CGM use. <strong>Results:</strong> A total of 531 adults with T2D completed the YouGov online survey. Over half were male (297/531, 55.9%) and aged 65 years and older (281/531, 52.9%). Two-thirds (347/531, 65.3%) had lived with T2D for more than 5 years, and 9.6% (51/531) use or had previously used a CGM. Overall, 50.8% (270/531) responded to at least one free-text question, with 49% (260/531) commenting on benefits and 33.1% (176/531) on challenges. Thematic analysis identified five key benefit themes: (1) reduced monitoring burden, described as eliminating frequent finger prick testing and simplifying daily routines; (2) lifestyle feedback, enabling participants to better understand how diet and physical activity influence glucose levels; (3) greater control, by supporting more informed decision-making and increasing confidence in self-management; (4) feeling safer, through alerts for hypo- and hyperglycemia; and (5) sharing data with clinicians, which facilitated communication and more collaborative care. The main challenges were (1) access barriers, including restrictive eligibility criteria and the high cost of self-funding; (2) device issues, such as discomfort, inconvenience, and practical difficulties wearing the sensor; (3) technology reliance, with concerns about depending on devices rather than listening to bodily cues; (4) emotional strain, including anxiety, over-monitoring, and increased preoccupation with glucose levels; and (5) data concerns, particularly regarding accuracy, interpretation, and privacy. <strong>Conclusions:</strong> Adults with T2D, including both users and nonusers, described CGM as a practical and empowering tool that improves understanding, safety, and collaboration with health care providers. Nevertheless, access barriers, usability issues, and emotional and data-related burdens remain major obstacles to equitable adoption. Addressing these through improved affordability, digital literacy support, and customized clinical guidance may support ongoing and inclusive CGM use in routine care.
Backed by $165M, Bionyra Pharma Launches to Advance Inflammatory Disease Biologics
Though he is trained as a gastroenterologist and scientist, Frédéric Marrache, MD, PhD, has always had something of an entrepreneurial itch. Following his post-doctoral program and a stint in management consulting, he made his way to Sanofi where he would work on early- to mid-stage drug development programs focused on immune-mediated diseases.
“This was right around the time when Sanofi, together with Regeneron, was finalizing the development of Dupixent,” a prescription biologic injection used to treat multiple inflammatory conditions, he told GEN. Those experiences gave him “meaningful insights” into patient care as well as about “how to develop therapies in this space.”
One of those insights was the scale of the unmet medical need in the immune-driven inflammatory disease space. Though some large pharma companies have developed products for the space already, “I had a few insights about what could be differentiated,” he said. That led him to engage with a team at Sofinnova Partners in early 2025. “I came in with my insights about patient needs, immunology, and target selection, and [my] view on right and wrong assets,” he said. “They came with experience in building companies” and “we mapped out the entire asset space specifically on the target and pathway of interest.”
Those discussions led to the launch of Bionyra Pharma, a clinical-stage biopharmaceutical company that is developing next-generation biologics for severe immunological and inflammatory diseases. The company emerged from stealth this week after raising $165 million in an oversubscribed Series A. The round was co-led by Jeito Capital and Sofinnova Partners with participation from Arkin Bio, Sanofi Ventures, Sixty Degree Capital, Vives Partners and Apollo Health Ventures.
Marrache serves as the co-founder and CEO of the company. In addition to the financing, Mehdi Ainouche, partner at Jeito Capital; Anta Gkelou, partner at Sofinnova Partners; Avital Adler, principal at Arkin Bio; and Laia Crespo, partner at Sanofi Ventures, will join Bionyra’s board of directors.
“When we co-founded Bionyra with Frédéric, our conviction in both the company and his leadership was grounded in his deep expertise in immune and inflammatory diseases,” said Sofinnova’s Gkelou. “Looking ahead, we are focused on advancing these programs with the aim of bringing meaningful new treatment options to patients.”
Specifically, the funds will support Bionyra’s efforts to advance mono and multispecific antibodies for various inflammatory conditions including atopic dermatitis and inflammatory bowel disease (IBD).
Right out of the gate, Bionyra is launching a pipeline of three clinical and near-clinical anti-inflammatory therapies, some of which are already in clinical trials. The company’s first asset, BYN-002 is a TL1A monoclonal antibody with the potential to treat IBD and other TL1A-relevant indications. This therapy is currently in a fully-enrolled Phase I study in healthy people. Its next candidate, BYN-003, is a TL1A*IL-23p19 bispecific antibody that is also in Phase I testing. Both assets have been improved with half-life extension (HLE) engineering to maximize efficacy and patient benefit.
Generally speaking, “TL1A is a game changer target right now in [immunology and inflammation] with great results in inflammatory disease,” he said. However, it is likely that this target will be relevant across multiple indications. To that end, Bionyra is keeping its options open in terms of what it will target with its TL1A assets. “Whether it’s going to be in the inflammatory bowel disease space, whether we go for another indication space or whether we decide to develop it in combination in any of these indications, that’s an option,” he said.
For now, the focus is on validating the safety and efficacy of both therapies in healthy volunteers. “That’s especially a question around the bispecific antibody” because there will likely be questions around the immunogenicity, he noted. “Our advantage here is that our bispecific is built on the backbone of our monospecific, so at least we have some level of early validation here, and we hope to present some results soon.”
A third candidate, BYN-001, is an IL-25 monoclonal antibody that has also benefited from HLE technology. It is currently in the IND-stage for atopic dermatitis and type 2 inflammation. While there are several assets in development that aim to target type 2 inflammation, once all of the me-too drugs are excluded, the field becomes narrower, Marrache said while explaining the rationale for choosing this particular drug candidate for Bionyra’s portfolio. “IL-25 has been known to be a strong driver of type 2 inflammation for some time,” he said.
Furthermore, some recently published early clinical data from a competitor, who are developing their own asset for IL-25, “clearly validated the pathway and suggested potential for differentiation.” At the time, Bionyra was already exploring the same target space so “we were able to move very quickly” and find what, Marrache believes, is the “most potent IL-25 antibody out there” with the “longest half life.”
Two of the assets BYN-002 and BYN-003 were licensed from TrueLab Biopharmaceutical. Under the terms of the agreement Bionyra was granted exclusive worldwide rights, excluding Greater China, to research, develop, manufacture and commercialize both therapies. TrueLab is eligible to receive up to $985 million in total consideration related to both assets, including the upfront payment as well as development, regulatory, and commercial milestone payments. The agreement also includes tiered royalties on future net sales. In addition, TrueLab has a single-digit equity stake in Bionyra Pharma following completion of its Series A financing.
For its part, BYN-001 was licensed from NovaRock Biotherapeutics. Bionyra is also progressing additional preclinical assets including some from TrueLab. It will support these efforts with some of the funds from the Series A.
The post Backed by $165M, Bionyra Pharma Launches to Advance Inflammatory Disease Biologics appeared first on GEN – Genetic Engineering and Biotechnology News.
VR Rehabilitation Improves Arm and Hand Movement After Stroke
A new rehabilitation platform combining virtual reality (VR) and nerve stimulation significantly improved the recovery of arm and hand function after a stroke compared to conventional rehabilitation approaches. Published today in Nature Medicine, results from a small-scale clinical study show early promise for a more effective and accessible rehabilitation approach that can be personalized to each patient’s needs.
Approximately 60% of stroke survivors develop long-term disability affecting their mobility. Even after extensive physiotherapy and occupational therapy, many continue to live with reduced arm and hand function, which severely impacts their ability to perform day to day tasks and live independently.
“Our aim was to go beyond mere movement training,” said Stanisa Raspopovic, PhD, professor of biomedical engineering at the Medical University of Vienna and senior author of the study. “After a stroke, patients often have difficulty not only moving the affected limb, but also feeling it and perceiving it correctly. MultiSensy was developed to reconnect movement, sensation and body awareness during rehabilitation.”
The MultiSensy rehabilitation platform combines immersive VR with electrical nerve stimulation. The VR goggles present users with interactive virtual tasks designed to train arm and hand functions such as reaching, grasping, pinching, and forearm rotation. Meanwhile, electrodes on the skin stimulate sensory nerves in real time to make patients feel virtual objects as if they were physically touching them.
The system was tested on a cohort of 34 patients who had suffered a stroke over three months before. Participants were divided into two groups who were treated either with MultiSensy or conventional rehabilitation including physiotherapy and occupational therapy. Both groups completed a total of 12 training sessions over the course of three weeks.
Patients who used the VR system saw a greater recovery of arm and hand movement compared to those in the control group, achieving nearly twice the improvement according to a standard assessment of motor impairment after stroke. In addition, MultiSensy was able to address body awareness and sensory deficits caused by stroke, which are often left aside by conventional rehabilitation strategies.
“After a stroke, some patients struggle to feel touch in their affected hand and may even perceive the arm as distorted in size, shape, or position,” said Valerio Aurucci, PhD, lead author of the study and former graduate student at ETH Zurich. “Participants treated with the new system showed improvements in their sense of touch and in perception of their affected arm.”
Another advantage of the MultiSensy platform is that each task can be adapted to the impairment level of the user, tailoring treatment to their unique needs. The VR system collects movement data during training, providing objective measurements of progression that clinicians can rely on to monitor a patient’s performance and recovery over time.
“The results provide early clinical evidence that immersive virtual reality combined with sensory nerve stimulation can support recovery after stroke, even after months from the event”, said Raspopovic. “The technology is still at the research stage, and larger clinical trials are needed to confirm its benefits. However, the study opens a promising perspective for future personalized and potentially home-based stroke rehabilitation.”
The post VR Rehabilitation Improves Arm and Hand Movement After Stroke appeared first on Inside Precision Medicine.
Toward a Digitally Informed Knitted Prosthetic Interface With Graded Stiffness to Enhance Comfort in Transtibial Amputees: Proof-of-Concept Case Study
Background: Despite considerable advancements in prosthetic technology, a substantial proportion of lower limb amputees reduce or discontinue prosthesis use, with reported nonuse rates ranging from 12% to 53%. This reflects the multifactorial challenges associated with long-term prosthetic use, among which comfort and skin health are consistently identified as key determinants. More specifically, studies point toward nonbreathable silicone liners trapping heat and sweat, leading to skin and hygiene problems. These persistent limitations underscore the need for alternative interface materials that offer improved breathability, moisture management, and tunable mechanical properties. Objective: This study aimed to introduce Flexoknit, a transtibial prosthetic liner that integrates user-specific digital skin strain analysis with computer numerical control multimaterial knitting to create a mechanically tuned, breathable, and anatomically customized interface. Using digital biomechanical data as the primary design driver—rather than clinician heuristics alone—Flexoknit aims to determine the feasibility and performance of a skin strain–guided, computer numerical control–knitted prosthetic interface in terms of material function, clinical performance, and user experience. Methods: Flexoknit uses programmable multimaterial knitting, incorporating thermal-reactive yarns that stiffen when heated to create structural support zones, alongside spandex yarns that provide elastic compression and breathable zones. Uniaxial tensile tests showed that yarn and stitch combinations can generate distinct stiffness grades, with nearly order-of-magnitude differences. The spatial layout of these graded zones aligns high-stiffness regions with the lines of nonextension, and low-stiffness regions with areas of greater skin strain. With the new prosthetic interface, a series of controlled tests was conducted to compare performance against the participant’s existing prosthesis with a conventional silicone liner. User testing was organized into 3 domains (ie, mobility, suspension, and comfort) using standardized quantitative assessments and structured qualitative data collection. Results: User testing demonstrated a 22.5% improvement in total range of motion, a 37.5% reduction in interface mass, and improved thermal regulation in hot, humid environments compared to that of a conventional silicone liner. The user walked unaided and performed sit-to-stand movements, reporting positive comfort and usability feedback. Conclusions: This work establishes Flexoknit as a promising direction for future prosthetic development—one that integrates principles of biomechanics, textile engineering, and digital fabrication to create user-centered interface solutions. The findings suggest that digitally engineered knitted interfaces can provide a highly customizable, breathable, and compliant alternative to conventional silicone liners, particularly for lower-activity amputees or individuals prioritizing comfort and ease of use.
<img src="https://jmir-production.s3.us-east-2.amazonaws.com/thumbs/1f123aec95a2fcf5815c3d0e97a3a02f" />
Clinical Outcomes of Individualized Electrostimulation Using a Wearable Electro Suit and Qualitative Feedback From a Mixed Cohort of Survivors of Stroke and Spinal Cord Injury With Spasticity: Case Series
Background: Various forms of electrical stimulation have been integrated into the multimodal management of spasticity. However, high-frequency electrical stimulation can potentially induce muscle fatigue. The Exopulse Mollii Suit (EMS) is a multichannel full-body garment that delivers low-frequency (20 Hz), low-amplitude (20 V), subthreshold sensory stimulation aimed at reducing spasticity. Objective: Primarily, we examined the effects of a single session of the EMS on spasticity in 7 participants with chronic stroke or cervical spinal cord injury (SCI), specifically those with upper or lower limb spasticity impacting function and gait who were able to walk with minimal or no assistance (Functional Ambulatory Category scores of 2‐5). We assessed the impact on gait and ambulatory function, as well as user perceptions of usability and acceptability. Methods: Participants wore the EMS for 60 minutes, consisting of 30 minutes of standardized goal-directed activities performed in two 15-minute blocks, interspersed with 15-minute rest breaks. Measurements included the Modified Tardieu Scale with surface electromyography for spasticity and functional mobility tests (Functional Ambulatory Category, 10-meter walk test, 5 times sit-to-stand test, and step test). Spatiotemporal gait parameters were quantified using a markerless vision-based motion capture system using the OpenPose BODY25 pose estimation model. Results: On the basis of the Modified Tardieu Scale and surface electromyography signals, improvements in spasticity were only observed in 2 participants. However, 4 participants demonstrated faster walking speeds. Improvements in the 5 times sit-to-stand test and step test were noted in 3 and 4 participants, respectively. Spatiotemporal gait parameters revealed improvements in gait symmetry in 6 participants. Qualitative feedback based on the Assistive Technology Usability Questionnaire for People With Neurological Diseases (NATU Quest) returned positive results in 3 participants. Overall outcomes, defined as meeting the individualized goals of each participant, were positive in 4 participants. Conclusions: This case series provides preliminary evidence that a single session with the EMS may offer benefits for functional mobility and gait quality for individuals with spasticity resulting from stroke or SCI. To our knowledge, this is the first report examining the effects of the EMS in participants with SCI and the first to include spatiotemporal gait parameters associated with its use. However, the small sample size, variable outcomes, and lack of a control group necessitate caution in interpreting these findings and preclude definitive conclusions regarding the efficacy of the EMS. Larger, controlled trials with repeated sessions of EMS use are required to establish the effectiveness and optimal application of the EMS for spasticity management.
<img src="https://jmir-production.s3.us-east-2.amazonaws.com/thumbs/be396b1bb6faa61e07efb5c94224db5d" />
Improving Models to Predict Care Utilization Using Machine Learning: Retrospective Observational Study
Background: The use of artificial intelligence and machine learning (ML) tools is now common in the advancement of health care services and clinical risk estimation. Legacy systems make use of highly informative feature sets developed from years of clinical expertise and research to estimate different outcomes, but only recently have they been tested against novel statistical approaches. One such system, the Johns Hopkins Adjusted Clinical Group (ACG) System, is a long-standing and widely used approach to categorizing clinical risk factors, and it is amenable to ML techniques. Objective: This study aims to test the ACG System using a contrasted area under the receiver operating characteristic (AUROC) and classification optimization strategy and compare its performance against traditional logistic regression methods. Assuming that selected ML algorithms can be tuned to enhance overall measures of performance, this would strengthen arguments for incorporating them into ACG-related workflows. Methods: Using a retrospective observational design, prospective year estimates of all-cause hospitalization and elevated total cost were modeled using a cross-validation framework. Patients with elevated costs were identified as those falling above the 95th percentile of total amounts billed, including pharmacy costs. Hyperparameter settings for XGBoost (Extreme Gradient Boosting), random forest, and elastic net were determined using average cross-validated performances for and AUROC in a grid search aimed at maximizing either statistic. Additional iterated cross-validation was used to compare point-estimated average AUROC and -scores between models, further decomposed by sensitivity, positive predictive value, and -beta statistics. Results: There were 350,463 patients selected in 2019 from the Johns Hopkins Health System. Model features identified by the ACG System for predicting prospective year hospitalization and total cost were included in these analyses. Findings suggest small but statistically significant improvements in cross-validated AUROC and -scores over logistic regression, using either optimization strategy and XGBoost. Logistic models achieved average receiver operating characteristic values of 0.886 and 0.841 for cost and hospitalization, respectively, whereas XGBoost achieved 0.891 and 0.849, respectively. optimization yielded similar findings, with logistic models achieving 0.367 and 0.341 on average for hospitalization and cost, respectively, but XGBoost exceeded values for cost but not for hospitalization (0.411 and 0.328, respectively). Conclusions: The clinical implications of these findings and the effect of class imbalance on model calibration are explored, along with the limitations of these data and approach. The core finding is that logistic regression remains very well-suited to these tasks, especially in situations where the efficiency or interpretability of models is critical. Under conditions of imbalance, regressions tended to yield high-precision estimates for the outnumbered class. Nevertheless, the findings also underscore a diversity of suitable models depending on clinical use cases, each having its own tradeoffs for evaluating performance. As such, health systems must clearly identify the needs and expectations of a model before calibrating one for use.
<img src="https://jmir-production.s3.us-east-2.amazonaws.com/thumbs/cc6a31773ec690198d6914d248fe33eb" />
Real-World Engagement With a Generative AI Conversational Agent for Mental Health Support: Retrospective Descriptive Study
Background: Generative artificial intelligence (GenAI) conversational agents are increasingly integrated within digital mental health interventions (DMHIs). However, empirical data on real-world engagement, usage patterns, and satisfaction with GenAI conversational agents remain limited. Objective: This study examined real-world engagement among users who interacted with the GenAI conversational agent within Mental, a DMHI designed to support mental health. We aimed to (1) characterize users engaging with Mental’s GenAI conversational agent, (2) examine real-world usage patterns, (3) examine satisfaction and user feedback following sessions, and (4) explore preliminary predictors of engagement with Mental’s GenAI conversational agent. Methods: This retrospective study analyzed naturalistic user data from 5082 paid subscribers who engaged with Mental’s GenAI conversational agent between October 2024 and March 2026. Users’ onboarding characteristics (ie, sex, mindset, distress level, desire for greater discipline, and primary stressors) and session satisfaction were collected via optional app-native items; session-level engagement metrics were captured through backend app usage data. Descriptive statistics were used to characterize user demographics and usage patterns. Session satisfaction was compared across temporal engagement variables using ANOVAs and independent-samples tests. As an exploratory aim, session-level mixed-effects logistic regression was used to estimate predictors of session-to-session return, with session satisfaction as the primary predictor and moderation by self-reported mindset. Results: Among users reporting onboarding data, 78.8% (2610/3312) identified as male and 90.0% (2667/2964) reported moderate-to-high distress on an app-native item. A total of 59,602 sessions were recorded (mean 11.8 sessions per user), most frequently occurring in the evening (17,206/59,602, 28.9%) and outside traditional business hours (37,181/59,602, 62.4%). Mean session satisfaction was high (mean 4.5, SD 0.9) and did not differ by time of day or day of the week. The most commonly selected session descriptors were “Insightful” (9236/19,883, 46.5%), “Felt seen” (7974/19,883, 40.1%), and “Good advice” (7510/19,883, 37.8%). The session-to-session return rate was 92.6%, and 69.4% (3528/5082) of users returned after their first session. In an exploratory analysis, session satisfaction was a significant predictor of return (odds ratio 1.35, 95% CI 1.14-1.60; <.001), although this finding should be interpreted as hypothesis-generating. Conclusions: Users engaged with a GenAI conversational agent within the Mental app outside of traditional care hours and presented with high return rates. Objective behavioral engagement data (eg, session frequency, timing, and session-to-session return rate) provide novel evidence that GenAI conversational agents may sustain real-world engagement, including among individuals who face barriers to traditional mental health services. Future research should determine whether these engagement patterns translate into clinically meaningful outcomes.
