Frontiers in digital health最新文献

Chronic Kidney Disease (CKD) remains a major contributor to global morbidity, often progressing unnoticed until advanced stages when treatment options become limited and costly. Recent advances in deep learning have reshaped early CKD assessment by enabling the analysis of complex imaging, clinical, and longitudinal laboratory datasets. This mini-review synthesizes findings from studies published between 2020 and 2025, highlighting models that report diagnostic accuracies ranging from 88% to 99.96%, AUC values reaching 0.93, and ensemble architectures capable of forecasting CKD 6 to12 months before clinical diagnosis with up to 99.31% accuracy. These systems spanning Convolutional Neural Networks (CNNs), Long Short-Term Memory networks (LSTMs), hybrid CNN-LSTM designs, and transfer-learning frameworks have demonstrated clear advantages over conventional diagnostic markers such as serum creatinine and eGFR. Despite impressive numerical performance, key limitations persist: class imbalance in early-stage CKD, restricted generalizability due to single-centre datasets, variability in imaging quality, and the limited interpretability of high-capacity neural networks. As deep learning continues to advance, robust external validation, transparent model explanations, and multi-institutional datasets will be essential to support safe and reliable clinical integration.

Background: Digital technology has led to innovations in healthcare, particularly in the field of nursing informatics. Although challenges such as resistance to technology, insufficient training, and security have been reported, comprehensive bibliometric analyses evaluating research trends and patterns are scarce. Therefore, this study aimed to examine trends and patterns in the application of digital technologies to nursing informatics by utilizing an integrative bibliometric analysis.

Methods: A comprehensive literature search was conducted on PubMed, CINAHL, MEDLINE, Embase, Web of Science, and Scopus for original articles published before 2024. This review followed the PRISMA guidelines, and 409 studies were included. VOSviewer and Excel 2019 were used to analyze the number of publications by journal, year, country, authors, citations, and keywords.

Results: Digital technology research began in 1985 and has increased significantly since 2015. The United States had the highest number of publications, whereas Computers, Informatics, and Nursing had the highest number of publications. Two authors were at the center of the collaboration network. The most frequently used keywords in these studies were virtual reality, nursing, and artificial intelligence. The primary research focus of the top 10 most-cited studies was intervention programs utilizing virtual reality.

Conclusion: This study highlights the growing importance of digital technology in nursing informatics, with research surging since 2015 due to advancements in artificial intelligence, virtual reality, and big data. Issues such as non-standardized nursing practices that utilize digital technologies and ethical considerations remain underexamined. Therefore, nursing professionals should focus on developing digital technology nursing standards in diverse nursing contexts, promote global collaboration, and strengthen digital competencies to maximize the benefits of digital innovation in the field of nursing informatics.

The integration of social media platforms and artificial intelligence (AI) has transformed the patient's role from that of a passive recipient to an active participant in healthcare navigation. The advent of short video platforms (such as TikTok and YouTube Shorts) has eliminated many conventional limitations related to location and time in medical education. Recent research suggests that sophisticated AI models (e.g., GPT-4) might outperform physicians in specific measurable aspects, such as diagnostic accuracy in controlled settings or empathy demonstrated through written communication. Nevertheless, physicians continue to be essential for coordinating complex care, resolving intricate ethical dilemmas, and maintaining the integrity of the physician-patient relationship. Consequently, although human participation remains essential, the digital environment is affected by integrity concerns. It is estimated that approximately 37% of medical social media posts contain misinformation, although this rate varies considerably among different health categories. To effectively resolve these challenges, we advocate for a collaborative stakeholder approach to governance. Through the implementation of formal platform certification, ongoing education for healthcare professionals, and AI-enabled filtering of user-generated content, we can improve the efficiency of medical resource allocations such as minimizing unnecessary inquiries-while laying a solid foundation for a sustainable, trust-based relationship between physicians and patients.

Introduction: This article presents the design and implementation of a digital health technology architecture focused on healthcare regulation in Brazil's National Health System (SUS). The objective was to develop an architectural model capable of optimizing resource allocation, increasing transparency, and integrating health information from different levels of care, with a focus on reducing inequalities in access.

Methods: Methodologically, a transdisciplinary applied research approach based on action research was adopted, with iterative development cycles in accordance with agile methodologies. The architecture was implemented in the states of Rio Grande do Norte, Espírito Santo, and Mato Grosso, respecting regional specificities and integrating international interoperability standards, as well as architectural principles and software engineering.

Results: The results point to flexibility, interoperability, real-time monitoring, queue management, and transparency, including direct access for control bodies and process auditability.

Discussion and conclusions: It can be concluded that the proposed architecture represents a significant advance for equity in access and could serve as a basis for solutions on a national and international scale.

Background: We explore the experiences of Artificial Intelligence (AI) innovators who had received funding to pilot their innovation in the English NHS, with the aim of understanding what hinders and supports, from their perspective, the sustainable implementation of their innovation beyond the funding period.

Methods: We first identified a list of companies that had received funding from two national schemes supporting AI innovations in the NHS, focusing on early rounds of these schemes. We then used personal contacts to identify key individuals from these companies, and used a snowball approach as well as LinkedIn contacts to increase our sample. We interviewed participants individually, using semi-structured interviews and analysed the data thematically.

Results: We interviewed 18 individuals from 11 AI companies, who had received funding from two national schemes. Our findings show that the funding offered the companies a unique opportunity to pilot their innovations, show early successes and grow recognition around AI and its potential. Yet, innovators faced several barriers in their effort to implement their AI innovations beyond the pilot phase, including misaligned expectations regarding the programmes' goal, fragmented adoption efforts with little national coordination, and inadequate evaluation mechanisms to generate the evidence needed for wider adoption.

Conclusion: The UK has set great ambitions for the adoption of AI in the NHS and has invested significantly in public funding to support its use. Our findings show that public investment alone is not sufficient to achieve this ambitious target. A better understanding of the implementation challenges of using AI innovation in practice is needed.

Background: Mycoplasma pneumoniae pneumonia represents a significant cause of community-acquired pneumonia in children, with clinical presentations ranging from mild to severe forms requiring respiratory support. Early identification of children at risk for oxygen therapy remains challenging using conventional clinical and laboratory parameters.

Methods: We conducted a multicenter retrospective study involving 206 pediatric patients (aged 1 month to 18 years) with confirmed Mycoplasma pneumoniae pneumonia admitted to three Italian hospitals between 2023 and 2025. Nine machine learning algorithms were developed and validated using routine admission data including demographics, clinical presentation, laboratory tests, and imaging findings. The primary outcome was the need for oxygen therapy during hospitalization. Model performance was evaluated using area under the curve (AUC), precision, recall, and F1-score metrics. Feature importance was assessed using SHAP (Shapley Additive Explanations) analysis.

Results: Among the 206 patients, 42 (20.4%) required oxygen therapy during hospitalization. The cohort had a mean age of approximately 4.6 years (SD ≈ 3.5), with comorbidities present in approximately 40% of cases. Support Vector Machine (SVM) achieved the highest performance with an AUC of 0.97, precision of 0.93, recall of 0.93, and F1-score of 0.92. Logistic Regression (AUC 0.95), XGBoost (AUC 0.94), and LightGBM (AUC 0.93) also demonstrated strong predictive performance. SHAP analysis consistently identified C-reactive protein (CRP), lactate dehydrogenase (LDH), neutrophil-to-lymphocyte ratio (NLR), neutrophil percentage, and respiratory distress as the most important predictive features across models.

Conclusion: Machine learning models using routine admission data can accurately predict oxygen therapy requirements in pediatric Mycoplasma pneumoniae pneumonia. The integration of interpretable artificial intelligence approaches may enable earlier risk stratification and improve clinical decision-making in pediatric respiratory infections.

Background: Despite the increasing number of systematic reviews on digital health interventions (DHIs), clear and robust evidence remains elusive due to methodological shortcomings in formulating research questions and conducting search and screening processes. The growing volume of reviews necessitates higher-level syntheses like umbrella reviews and evidence gap maps, requiring methods for rapid, systematic evidence assessment at the abstract level.

Objective: With the development of the PICO-based Assessment and Categorization of Evidence for Digital Health Interventions (PACE4DHI) framework we aim to enable the efficient structured screening of systematic reviews and meta-analyses at the level of abstracts for subsequent evidence and gap mapping (EGM).

Methods: A comprehensive literature search was performed across five databases, adhering to PRISMA guidelines, to capture systematic reviews and meta-analyses published between 2011 and October 2023. All categories of DHIs, populations, settings, and outcomes were considered. From 21,161 results, we screened 9,030 titles and abstracts post-de-duplication, with 2,528 remaining. To construct the framework, thematic analysis was conducted on a random sample of 250 studies. The framework's accuracy was validated on 138 open-access articles through full-text comparisons.

Results: The PACE4DHI framework encompasses 41 categories, spanning 11 problems (e.g., cardiovascular diseases), 13 DHIs (e.g., telemedicine), 6 comparative care settings (e.g., outpatient care), 7 outcome dimensions (e.g., effectiveness), and 4 evidence classification levels. The PICO-categorization and evidence classification was confirmed with varying accuracy and largely consistent results at both abstract and full-text levels. Variability in the accuracy reflects that abstracts provided more detail on problems and interventions than they did for the comparator and outcomes. The likelihood of conclusive evidence was more accurately predicted for cardinal classes (high and low) than for inconclusiveness.

Conclusions: The PACE4DHI framework provides a systematic and pragmatic methodology, with potential to enhance structured access to existing evidence. The framework may also inform the research questions and the search and screening strategies of future systematic reviews. The application in EGM has potential to optimize evidence-based decision-making, while also enabling precise identification of research gaps. Its use with artificial intelligence tools may facilitate efficient ongoing evidence screening and synthesis, ultimately supporting a searchable evidence database.

Introduction: Predicting Parkinson's disease (PD) motor progression remains challenging despite advances in neuroimaging. Blood-based transcriptomic profiling offers a more accessible and cost-effective alternative. This study aimed to develop and validate a machine learning approach using blood-based transcriptomic data to predict 12-month motor severity in PD and to identify the transcriptomic features and biological pathways most strongly associated with progression.

Methods: A Stacking Regressor ensemble model combining three gradient boosting algorithms (XGBoost, LightGBM, CatBoost) was developed using baseline Parkinson's Progression Markers Initiative (PPMI) data (n = 390), integrating blood RNA sequencing (RNA-seq) and clinical features to predict 12-month UPDRS Part III scores. SHapley Additive exPlanations (SHAP) analysis was applied to identify key prognostic features, evaluating seven PD risk genes (SNCA, LRRK2, GBA, PRKN, PINK1, PARK7, VPS35) and pathway scores for mitochondrial dysfunction, neuroinflammation, and autophagy.

Results: On an independent test set (n = 78), the model achieved a Coefficient of Determination (R²) of 0.551 and Mean Absolute Error (MAE) of 6.01. SHAP analysis identified the baseline UPDRS × PINK1 interaction (UPDRS_BL × PINK1) as the most influential feature (mean |SHAP| = 0.283). Among transcriptomic features, VPS35 (mean |SHAP| = 0.010), GBA, and LRRK2 were most prominent. Mitochondrial dysfunction showed the highest pathway contribution (mean |SHAP| = 0.008).

Discussion: The study establishes that machine learning integrating blood transcriptomics and clinical data effectively predicts motor progression in PD. Crucially, the interplay between initial clinical state and specific genetic backgrounds-particularly PINK1-is a more powerful prognostic indicator than any factor alone. This study provides systematic evidence that mitochondrial dysfunction is a dominant prognostic signal for disease progression, nominating key genes and pathways for future mechanistic and therapeutic investigation.

Introduction: The Prescription Drug Monitoring Program (PDMP) is a digital tool that can mitigate controlled substance misuse in the United States; however, it remains underutilized and end-users lack sufficient training. Thus, the purpose of this study was to assess changes in factors that influence PDMP utilization before and after an interprofessional educational workshop.

Materials and methods: Fifteen two-hour interprofessional workshops were conducted from July 2022 to April 2025. Healthcare providers and law enforcement personnel in Alabama were recruited to participate via email, and data were collected at pre- and post-workshop via an anonymous online survey informed by the Unified Theory of Acceptance and Use of Technology (UTAUT). Measures included: 1) perceived usefulness; 2) ease of use; 3) social influence; 4) resources; 5) concerns; and 6) intentions regarding PDMP utilization. Differences in mean UTAUT scale scores from pre- to post-workshop were analyzed using Wilcoxon signed-rank tests, and predictors of PDMP utilization intention were analyzed using generalized estimating equations (GEE) with normal distribution and identify link function.

Results: Overall (N = 199), mean perceived usefulness, ease of use, social factors, resources, and intentions to use the PDMP all increased (p < 0.001) from pre- to post-workshop, while concerns decreased (p = 0.007). Perceived availability of resources (β=0.165, 95%CI = 0.023, 0.307; p = 0.023) positively predicted and concerns (β = -0.137, 95%CI = -0.223, -0.051; p = 0.002) negatively predicted PDMP utilization intentions.

Conclusion: Findings supports the utility of interprofessional educational interventions to increase PDMP engagement. Future studies may promote resources and alleviate concerns as key leverage points to enhance PMDP utilization.

Introduction: Walking speed is a key measure of health and mobility across a wide range of diseases. Traditional gait assessments in clinical settings may not accurately reflect real-world mobility patterns. Wearable sensors offer an ecologically valid alternative by capturing every movement in daily life, but there are few robust, validated reports. We aimed to identify evidence on real-world gait speed measurements that have received or are seeking regulatory approval from agencies such as the European Medicines Agency and the U.S. Food and Drug Administration.

Method: We conducted a systematic review following a comprehensive search strategy using the Ovid platform, guided by pre-defined selection criteria and in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses statement. We also manually searched the websites of key regulatory agencies and the ClinicalTrials.gov database.

Results: Our search identified 503 records, of which 10 met the inclusion criteria. Most studies were part of large-scale initiatives, including the qualification of the Stride Velocity 95th Centile and the MOBILISE-D project. No device or outcome measure that assesses walking speed in real-world conditions has been fully validated by the FDA. We found four letters of intent on the FDA website related to this concept. One outcome, the 95th centile of stride velocity, has been approved by the EMA as a primary endpoint for assessing ambulant patients with Duchenne Muscular Dystrophy.

Conclusion: Despite the potential of wearable devices to enhance drug development and clinical decision-making, real-world walking speed remains insufficiently validated across most conditions because data is missing. The widespread adoption of digital outcomes to assess ambulation will require extensive validation efforts, regulatory pathway adaptations, and improved standardization of devices, algorithms, and study methodologies.

Systematic review registration: https://www.crd.york.ac.uk/PROSPERO/view/CRD42025633578, PROSPERO CRD42025633578.