PLOS digital health最新文献

Diabetes technology generates vital health data, but healthcare professionals (HCP) and patients must navigate multiple platforms to access it. We developed a digital health platform, co-designed with patients and families living with type 1 diabetes (T1D) and their HCPs, that aim to support a collaborative care experience through shared access to diabetes data, clinical recommendations, and resources. We describe caregivers' views on the platform's impact on clinic visits and child self-management in children with T1D. A six-month observational pilot study at BC Children's Hospital Diabetes Clinic in British Columbia, Canada, gathered data through surveys and interviews. Surveys were administered to caregivers and HCPs at different time points throughout the study; 18 qualitative interviews were conducted with caregivers at the conclusion of the study. Quantitative data were summarized descriptively. Interview data were transcribed, coded using open and systematic coding, and subsequent inductive thematic analysis. Eighteen caregivers completed the surveys, and 11 HCP participants submitted 41 surveys (approximately 3-4 each) after using the platform. Most caregivers (61%; 11/18) found the platform helpful, and 56% (10/18) reported that using the platform made their clinical visits and recommendations more personalized. Nearly all HCPs (90%; 37/41) were satisfied with the platform's ability to support clinical visits. Themes identified from caregiver qualitative interviews revealed that (1) the platform provided a convenient connection that improved preparedness and empowered caregivers in managing their child's T1D; (2) the platform's value was driven by the healthcare team's usage of it; and (3) caregivers felt hopeful that the platform could better support their child's T1D management. The platform could foster a collaborative and personalized care experience that enables caregivers to engage in diabetes self-management and feel connected to their healthcare team. These results will guide the future development, evaluation, and implementation of the platform.

[This corrects the article DOI: 10.1371/journal.pdig.0000926.].

At the emergency department, it is important to quickly and accurately identify patients at risk of acute myocardial infarction (AMI). One of the main tools for detecting AMI is the electrocardiogram (ECG), which can be difficult to interpret manually. There is a long history of applying machine learning algorithms to ECGs, but such algorithms are quite data hungry, and correctly labeled high-quality ECGs are difficult to obtain. Transfer learning has been a successful strategy for mitigating data requirements in other applications, but the benefits for predicting AMI are understudied. Here we show that a straightforward application of transfer learning leads to large improvements also in this domain. We pre-train models to classify sex and age using a collection of 840 k ECGs from non-chest-pain patients, and fine-tune the resulting models to predict AMI using 44 k ECGs from chest-pain patients. The results are compared with models trained without transfer learning. We find a considerable improvement from transfer learning, consistent across multiple state-of-the-art ResNet architectures and data sizes, with the best performing model improving from 0.79 AUC to 0.85 AUC. This suggests that even a simple form of transfer learning from a moderately sized dataset of non-chest-pain ECGs can lead to major improvements in predicting AMI.

Dental age (DA) estimation is a key diagnostic tool in pediatric dentistry, particularly when birth records are unavailable or unreliable. It guides decisions on growth assessment, orthodontic planning, and timing of interventions such as space maintenance or extractions. Unlike skeletal maturity, dental development is less affected by nutritional and environmental factors, making it a reliable marker of biological age. Conventional methods require expert interpretation and are prone to variability. There is growing interest in automated, objective approaches to streamline this process and enhance clinical utility. A total of 550 panoramic radiographs from children aged 3-14 years were labeled into 11 dental age groups based on the AAPD reference chart by two experienced pediatric dentists. Images with poor quality were excluded. The dataset was divided into training (80%) and validation (20%) sets, with data augmentation applied to the training set. The YOLOv11n-cls model, consisting of 86 layers and 1.54 million parameters, was trained for 30 epochs using the Ultralytics engine and AdamW optimizer. Model performance was evaluated using Top-1 and Top-5 accuracy on the validation set and tested on an independent set of 203 images. Grad-CAM was used for model interpretability. The model achieved 92.6% Top-1 and 99.5% Top-5 accuracy on the validation set. Performance on the test set remained high, with most misclassifications occurring between adjacent age groups. Grad-CAM visualizations showed attention to clinically relevant areas like erupting molars and root development. The findings support the high performance of DL, through YOLOv11 for pediatric age prediction. The AI tool enabled fast, accurate, and interpretable DA classification, making it a strong candidate for clinical integration as an adjunct tool into pediatric dental practice.

The availability of medical imaging data is indispensable for medical advancements such as the development of new diagnostic tools, improved surgical navigation systems, and profiling for personalized medicine through imaging biomarkers. A central challenge in data governance is balancing the need to protect patient privacy with the necessity of promoting scientific innovation. Restrictive data governance policies could limit access to the large, high-quality datasets needed for such advancements. Conversely, lenient policies could compromise patient trust and lead to potential misuse of sensitive information. We call for a deliberate and well-considered approach to data governance, highlighting important factors that patients and healthcare organizations should consider when making imaging data governance decisions around data sharing.

Food preferences differ among individuals, and these variations reflect underlying personalities or mental tendencies. However, capturing and predicting these individual differences remains challenging. Here, we propose a novel method to predict individual food preferences by using CLIP (Contrastive Language-Image Pre-Training), which can capture both visual and semantic features of food images. By applying this method to food image rating data obtained from human subjects, we demonstrated our method's prediction capability, which achieved better scores compared to methods using pixel-based embeddings or label text-based embeddings. Our method can also be used to characterize individual traits as characteristic vectors in the embedding space. By analyzing these individual trait vectors, we captured the tendency of the trait vectors of the high picky-eater group. In contrast, the group with relatively high levels of general psychopathology did not show any bias in the distribution of trait vectors, but their preferences were significantly less well-represented by a single trait vector for each individual. Our results demonstrate that CLIP embeddings, which integrate both visual and semantic features, not only effectively predict food image preferences but also provide valuable representations of individual trait characteristics, suggesting potential applications for understanding and addressing food preference patterns in both research and clinical contexts.

Earthquakes and other disasters often cause substantial damage to health facilities, impacting short-term response capacity and long-term health system needs. Identifying health facility damage following disasters is therefore crucial for coordinating response, but ground-based evaluations require substantial time and labor. Artificial intelligence (AI) models trained on satellite imagery can estimate building damage and could be used to generate rapid health facility damage reports. There is little published about methods of generating these estimates, testing real-world accuracy, or exploring error. This study presents a novel method of overlaying model damage outputs with health facility location data to generate health facility damage estimates following the February 2023 earthquake in Turkey. Two models were compared for agreement, accuracy, and errors. Building-level damage estimates were obtained for Model A (Microsoft neural network model), and Model B (Google AI model), and overlaid with health facility location data to identify facilities with significant damage. Model agreement, sensitivity and specificity for damage detection were calculated. A descriptive review of common error sources based on selected satellite imagery was conducted. A spatially aggregated damage estimation, based on proportion of buildings damaged in a 0.125km2 area, was also generated and assessed for each model. Twenty-five hospitals, 13 dialysis facilities, and 454 pharmacies were evaluated across three cities. Estimated damage was higher for Model A (10.4%) than Model B (4.3%). Cohen's kappa was 0.32, indicating fair agreement. Sensitivity was low for both models at 42.9%, while specificity was high (A:93.6%, B:96.8%). Agreement and sensitivity were best for hospitals. Common errors included building identification and underestimation of damage for destroyed buildings. Spatially aggregated damage estimates yielded higher sensitivity (A:71.4%, B:57.1%) and agreement (Cohen's kappa 0.38). Leveraging remote-sensing models for health facility damage assessment is feasible but currently lacks the sensitivity to replace ground evaluations. Improving building identification, damage detection for destroyed buildings, and spatially aggregating results may improve the performance and utility of these models for use in disaster response settings.

Studying for final exams is often regarded as difficult for nursing students, therefore, activities using game-based learning methods may increase student satisfaction. Therefore, this study aimed to understand the feasibility of a game-based learning activity on nursing students' learning and revision processes. A one-group pre and post-questionnaire design was undertaken to evaluate the effectiveness of a game-based learning activity. All nursing students found the game-based learning activity valuable when preparing for written exams. The learning activity increased the levels of knowledge retention and the final grades. Although two students found the activity somewhat distracting, most students believed that game-based learning should be embedded into the nursing curriculum. The game-based learning activity was well-accepted when revising for written exams in nursing. However, research at a larger scale is required to confirm the effectiveness of the activity on students' knowledge, grades and long-term retention.

Over the past few years, health delivery organizations (HDOs) have been adopting and integrating AI tools, including clinical tools for tasks like predicting risk of inpatient mortality and operational tools for clinical documentation, scheduling and revenue cycle management, to fulfill the quintuple aim. The expertise and resources to do so is often concentrated in academic medical centers, leaving patients and providers in lower-resource settings unable to fully realize the benefits of AI tools. There is a growing divide in HDO ability to conduct AI product lifecycle management, due to a gap in resources and capabilities (e.g., technical expertise, funding, data infrastructure) to do so. In previous technological shifts in the United States including electronic health record and telehealth adoption, there were similar disparities in rates of adoption between higher and lower-resource settings. The government responded to these disparities successfully by creating centers of excellence to provide technical assistance to HDOs in rural and underserved communities. Similarly, a hub-and-spoke network, connecting HDOs with technical, regulatory, and legal support services from vendors, law firms, other HDOs with more AI capabilities, etc. can enable all settings to be well equipped to adopt AI tools. Health AI Partnership (HAIP) is a multi-stakeholder collaborative seeking to promote the safe and effective use of AI in healthcare. HAIP has launched a pilot program implementing a hub-and-spoke network, but targeted public investment is needed to enable capacity building nationwide. As more HDOs are striving to utilize AI tools to improve care delivery, federal and state governments should support the development of hub-and-spoke networks to promote widespread, meaningful adoption of AI across diverse settings. This effort requires coordination among all entities in the health AI ecosystem to ensure these tools are implemented safely and effectively and that all HDOs realize the benefits of these tools.