Year: 2026
Type: editorial
Abstract: Population impact begins before an application is built. A lifestyle needs assessment in communities in Ireland and Romania shows that barriers, motivations, and preferred forms of digital coaching differ across local settings. This finding complements a survey-based model of digital-health intention, in which expectations, health beliefs, and perceptions of persuasive design shaped willingness to adopt digital services. Both studies caution against treating access to a smartphone as equivalent to readiness, trust, or sustained use.The qualitative contributions deepen this point. A meta-synthesis of cardiovascular patients' experiences identifies reassurance, human oversight, age-friendly design, and culturally responsive connections among families, hospitals, and communities as recurring needs. A protocol focused on young adults with cardiovascular disease brings a life-course perspective and asks how the possibilities offered by a technology are understood and used by a population with distinctive social roles and expectations. Although the latter does not yet report findings, it highlights a neglected issue: the same interface may enable independence for one group while creating surveillance, anxiety, or inconvenience for another.These contributions place co-design, usability, and social context within the causal pathway of effectiveness. An intervention that is not reached, trusted, understood, or sustained cannot improve population health. Evaluations should therefore report who participates, who declines, who disengages, and why not only outcomes among active users.Mobile and wearable technologies can make behavior and physiology visible in everyday life. In a community sample, wearable-derived sleep duration showed a nonlinear association with hypertension, illustrating how objective longitudinal measures can refine risk characterization beyond coarse self-report. The Terrapino application similarly demonstrates the reach of mobile cognitive-health assessment and identifies modifiable factors associated with an Alzheimer's risk score among thousands of users.These studies show the epidemiologic potential of devices that people already carry or wear, but they also mark some important boundaries. Cross-sectional associations cannot establish that changing sleep will reduce hypertension, and predicting a questionnaire-derived risk score is not the same as predicting incident dementia.Voluntary app users may also differ systematically from the wider population. Digital measurement is therefore most useful when it is validated against meaningful outcomes, stable across devices and generalizable to different demographic groups, and connected to an intervention that can change the measured risk.Two technically adjacent studies reinforce the importance of measurement quality. A pilot classification framework for voice disorders found that agreement was stronger for broad categories than for specific and perceptually ambiguous diagnoses. A YOLOv8 study achieved promising local performance in detecting and staging pressure injuries from clinical images. Although neither study directly evaluates population-level chronic disease prevention, both expose issues that apply throughout AI-enabled health research: algorithms inherit uncertainty in clinical labels, and impressive internal accuracy does not guarantee performance across patients, devices, or settings.Several contributions explore AI as a means of stratifying risk or anticipating change.An electronic medical record model identified patients at risk of exercise intolerance after percutaneous coronary intervention, with strong internal discrimination and potentially useful negative predictive value. The Sugar Slay ecosystem combines glucose forecasting, wearable data, gamification, personalized feedback, and caregiver support for type 1 diabetes. These systems illustrate complementary uses of prediction: prioritizing rehabilitation assessment at the service level and supporting near-term decisions in daily self-management.Yet prediction is an intermediate outcome, not a health benefit. A model may perform well in a local dataset but lose calibration when prevalence, workflow, coding, or patient mix changes. A forecast may be accurate but arrive too late, demand an action that is unavailable, or increase cognitive burden. Before deployment, prediction systems need temporal and geographic external validation, subgroup assessment, prospective monitoring, and comparison with simpler rules or existing care. More importantly, studies must evaluate the complete chain from prediction to response: whether an alert was seen, whether the recommended action was appropriate and completed, and whether outcomes improved.The broader reviews in the Topic place these examples in context. Reviews of AIsupported chronic disease self-management and AI in type 2 diabetes describe functions spanning monitoring, risk prediction, conversational support, multi-omics, continuous glucose monitoring, and personalized treatment. Their scope shows how quickly the field is expanding, while also revealing that technical development is advancing faster than evidence on long-term clinical effectiveness, affordability, and equitable implementation.The strongest practical message across the collection is that digital information produces value when it changes care. In type 2 diabetes, an AI-and big-datasupported mobile program translated risk estimates into individualized diet, exercise, and glucose-monitoring plans and reported modest improvements in glycemia, selfcare, and quality of life compared with conventional nurse-led management. Because allocation was not randomized, the results require confirmation, but the intervention illustrates a closed loop in which data inform action and subsequent data allow support to adapt.Other interventions show that useful digital transformation does not always require complex AI. App-based follow-up after arteriovenous-fistula creation was associated with earlier cannulation and shorter dialysis-catheter use than telephone follow-up. In a randomized study, WeChat-and cloud-supported nursing improved early anticoagulation monitoring, knowledge, adherence, and satisfaction among patients receiving warfarin, although complications and readmissions were not reduced over six months. These studies suggest that timeliness, continuity, education, and access to professional review may be more important than the novelty of the delivery channel.A meta-analysis of digital interventions for obstructive sleep apnea sharpens this conclusion. Passive monitoring did not improve several objective outcomes overall, whereas more active, synchronous, or theory-informed approaches appeared more promising. The evidence base remains small, but the implication is important: measurement alone rarely changes health. Digital interventions must alter capability, opportunity, motivation, treatment, or coordination-and should identify which of these mechanisms actually produced the observed benefit.Behavior change is central to prevention, but engagement should not be confused with health benefit. A review of persuasive system design finds that chronic disease applications often emphasize reminders, tailoring, and task support, while using social support and credibility features less consistently. A scoping review of digital nudges organizes approaches across lifestyle domains and identifies gaps in population diversity and long-term evaluation. Together, these reviews support more explicit behavior-change theory and clearer reporting of active components.The Topic also shows why maximizing digital exposure is an unsafe objective. A prospective cohort of people with lumbar disc herniation links algorithm-driven exposure to heterogeneous disease-related short videos with poorer psychological and rehabilitation outcomes through proposed mediating pathways. Causal interpretation remains limited by nonrandom exposure and possible residual confounding, but the study expands the safety agenda beyond privacy and model error. Repetitive, alarming, contradictory, or poorly contextualized content may increase anxiety and undermine recovery.Conversational AI offers both promise and responsibility. A co-designed largelanguage-model prototype for HIV prevention in South Africa was experienced as usable, private, and less stigmatizing. Although outside chronic non-communicable disease prevention, it demonstrates how conversational systems may facilitate sensitive disclosure and prepare more focused encounters. Acceptability, however, is not diagnostic accuracy or clinical benefit. Such tools require evaluation of harmful outputs, uncertainty communication, escalation, human review, and downstream service uptake.Four broad reviews make clear that population impact depends on the surrounding system. Discussions of emerging information technologies, DeepSeek in Chinese primary care, and AI in acute and critical care describe opportunities in screening, decision support, follow-up, workflow, and resource allocation, while repeatedly identifying fragmented data, weak interoperability, uneven capacity, privacy concerns, "Human in the loop" is meaningful only when responsibility, response time, and authority to override are defined.The next phase of research should evaluate complete pathways rather than isolated devices or models. First, predictive performance should be followed by prospective external validation and impact evaluation. Trials and pragmatic studies should compare AI-enabled care with simpler risk rules and well-delivered usual care, so that the incremental contribution of AI can be distinguished from extra contact or better organization.Second, outcomes should extend beyond short-term engagement and process measures. Depending on the intervention, studies should assess sustained behavior change, disease incidence or progression, functional status, quality of life, adverse effects, caregiver burden, cost, and service use. Across the 23 contributions, the field's direction is encouraging, but the evidence remains uneven. We know that mobile and wearable technologies can extend observation beyond visits, AI can produce clinically relevant risk estimates, and digitally supported care can improve selected short-term outcomes (7,8). We know less about durability, transportability, cost-effectiveness, hard clinical outcomes, and the fair distribution of benefit and burden. The central challenge is therefore not to build more stand-alone tools. It is to create learning prevention systems in which trustworthy measurement leads to an effective action, that action is accessible to diverse populations, outcomes feed back into improvement, and governance keeps the system accountable. Judged by that standard, innovation is not the volume of data collected or the novelty of an algorithm, but the durable and equitable prevention it enables.
Source: Frontiers in Public Health
Authors: Han Feng, Yishi Jia, Yingshuo Liu, Xiang Li, Duo Yu
Language: English
Subfield: General Health Professions
Field: Health Professions
Domain: Health Sciences
Open Access status: gold
Estimated APC paid: $3,801
OpenAlex ID: W7223834831