This study analyzed 1556 clinical investigation plan approvals granted by the Ministry of Food and Drug Safety (MFDS) in the Republic of Korea by integrating regulatory databases and comparing the clinical investigation characteristics of traditional and digital medical devices (2003-mid 2024). Indicators included clinical investigation purpose, device classification, number of participating institutions, design characteristics, annual approval trends, major product categories, and time required from plan approval to marketing authorization. According to the Korean MFDS four-class medical device classification system (Classes 1-4), digital medical devices, predominantly concentrated in Class 2 and 3 software, exhibited a high proportion of Pivotal clinical investigations (68%); traditional medical devices, distributed among high-risk Class 3 and 4 devices, had a relatively higher proportion of exploratory investigations. The number of plan approvals for digital medical devices surpassed that of traditional devices for the first time in 2023; cognitive therapy and diagnostic assistance software constituted the major product categories. The mean time from approval to marketing authorization was significantly shorter for digital medical devices than for traditional medical devices.
Metastatic melanoma presents clinical challenges due to tumor heterogeneity and treatment resistance. Here, we report an integrative workflow combining AI-based digital pathology with spatial proteomics to support personalized treatment strategies in a case of a young patient with recurrent melanoma and multiple metastases. Our AI model trained on H&E images identified two spatially separated cell subpopulations (PT1 and PT2) within the primary lesion, along with metastatic areas and stromal components. MS-based proteomics was used to map the spatial proteome across the clinically relevant regions. Our findings indicate inter-tumor heterogeneity and increased kinases associated with target-drug resistance. Convergent morphological and proteomic signatures identified PT1 as an aggressive melanoma subtype and the likely metastatic driver. Augmented glycolytic signaling and mitochondrial metabolism were identified as drivers of melanoma progression in this patient. Our findings suggest that targeted therapies may provide limited benefit, while the combination with metabolic inhibitors could represent a more effective treatment option for the patient.
暂无摘要（点击查看详情）
Challenges and opportunities facing health systems have prompted a re-evaluation of long-standing assumptions about digital transformation. This editorial summarizes interviews with nurse leaders who share their perspectives on leveraging technology to drive digital transformation and improve healthcare. Key themes include smart hospitals, interoperability, and innovative care delivery methods. These insights underscore the critical role of nurse leaders in leveraging emerging technologies to enhance the care continuum and drive sustainable change.
Cue-induced craving is a core driver of addiction and relapse, and its significant heterogeneity represents a major barrier to precision intervention. Currently, there remains a lack of objective, quantifiable, and individualized neurobiological biomarkers. Here, we employed task-based electroencephalography (EEG) to capture the dynamic neural signatures underlying cue-induced craving in patients with heroin use disorder (HUD) and developed an individualized functional connectivity (FC)-based prediction model. We identified β-band power envelope connectivity (PEC) as a reliable biomarker capable of estimating subjective craving severity at the individual level. Notably, even after FC reconfiguration induced by intermittent theta burst stimulation (iTBS) over the left dorsolateral prefrontal cortex (L-DLPFC) or precuneus (PCu), the PEC-based framework's prediction of immediate craving levels following these perturbed states remained effective. Crucially, baseline β-band PEC demonstrated strong prognostic value for improvements in craving scores (L-DLPFC-iTBS: r = 0.856, P < 0.001; PCu-iTBS: r = 0.675, P = 0.008). This individualized predictive model was further validated in an independent dot-probe task dataset, demonstrating its generalizability across distinct cue-induced craving paradigms. Together, our study demonstrates that EEG FC features predict individual cue-induced craving levels and intervention outcomes, facilitating the advancement of digital biomarker-driven precision medicine.
Relapsed/refractory lymphoid malignancies lack effective treatment selection strategies. We evaluated high-throughput ex vivo drug profiling in 26 patients, achieving successful profiling in 22 cases. We found broad ex vivo resistance to conventional chemotherapy but sensitivity to BH3 mimetics. Notably, p53-aberrant samples showed increased sensitivity to dasatinib and PI3K inhibitors across subtypes. These findings demonstrate feasibility of functional precision medicine pipelines in lymphoid malignancies and identify actionable vulnerabilities warranting further investigation.
Mental health applications (MHAs) hold promise for expanding access to care, yet user satisfaction varies across cultural contexts. Drawing on value co-creation theory, this study integrates developer and user perspectives to examine cross-country differences between China and the United States. We analyzed platforms and reviews data from leading Chinese and US app markets. On the developer side, US-based applications demonstrated substantially greater engagement in clinical pipelines and public claims than Chinese applications; however, these indicators were not significantly associated with app store ratings in the US sample and could not be meaningfully estimated in the Chinese sample due to very limited disclosure. On the user side, satisfaction drivers differed markedly: Chinese users emphasized core service functions such as online consultation, whereas US users showed stronger preferences for experiential features such as meditation practice. Across both contexts, personalized services were consistently rated as insufficient. Future development should shift MHAs from "digital products" toward "evidence-based, user-centered health tools".
Distinguishing degenerative cervical myelopathy from natural aging-related functional decline remains a critical challenge in primary care. The standard 10-s grip-and-release test loses specificity in older adults due to aging-induced motor slowing, creating a diagnostic gray zone that confounds risk stratification. Here we present a smartphone-based computer vision framework designed to disentangle pathological motor deficits from physiological aging. In a multicenter study comprising 2340 participants, we identified eight demographically robust kinematic digital biomarkers, including maximum release velocities and inter-finger synchronization. These metrics serve as objective surrogates for corticospinal integrity and maintain biological stability despite muscle senescence. Our model achieved an area under the curve (AUC) of 0.896 (95% CI 0.882-0.909) in the development cohort (80.0% sensitivity, 83.5% specificity), significantly outperforming the conventional 20-cycle threshold (AUC 0.768; P < 0.001). Crucially, this discriminative accuracy remained robust within the diagnostic gray zone. In the external validation cohort, the model yielded an AUC of 0.856 (95% CI 0.818-0.895), achieving 83.1% sensitivity, 79.9% specificity, 64.1% positive predictive value (PPV), and 91.8% negative predictive value (NPV) at an observed 30% prevalence. By shifting the diagnostic paradigm from the quantity of movement to the quality of kinematic patterns, this tool provides a scalable and objective solution for DCM risk stratification in aging populations within primary care.
Electronic health record (EHR) systems are critical to modern healthcare delivery, yet the dynamic workflows that govern electronic order processing remain underexplored. Inefficiencies in these digital pathways can cause delays in care, repetitive workloads, and even patient harm. This study presents a discrete-event simulation framework used to reconstruct and evaluate EHR-based order workflows in a large integrated healthcare system. Using real-world data extracted from the Veterans Health Administration's Corporate Data Warehouse, the authors mapped order events to standardized state transitions and modeled their progression across different facilities of varying complexity levels. After being calibrated with empirical distributions of transition times and validated against observed time-in-system metrics, the simulation demonstrates close alignment with historical performance. Scenario analyses reveal that resource capacity constraints significantly amplify the impact of electronic order surges, which are reflected in the disproportionate growth in backlogs and processing delays. Adjustments in transition probabilities further increased recirculation and extended workflow paths. Network-based analysis identified Reserved, InProgress, and Completed as structurally critical states that function as hubs within the process network but the transitions in-between also act as major bottlenecks. These results showcased the effectiveness of simulation-based approaches in monitoring EHR order processing performance and evaluating consequences of workflow changes on healthcare network resources planning. The proposed simulation framework provides a scalable data-driven tool to support operational decision-making and improve the efficiency of electronic order management in complex healthcare environments.
Patient-reported outcome measures (PROMs) can extend follow-up after joint replacement beyond discharge. Using Germany as a case example, national scalability and its potential impact on cost and health outcomes of a trial-proven PROM-based digital monitoring intervention were assessed. A four-step extrapolation mapped trial tasks and time, removed study-only procedures to derive streamlined delivery models, projected national implementation costs (personnel, licence fees, integration), and modelled population-level effects using trial outcomes and insurance claims. Personnel time decreased from 62 min per patient (trial) to 21 min in a hybrid model that retained limited human touchpoints and to 10 min in an automated model that restricted human involvement to safety-critical interventions, with costs of €158, €133 and €127, respectively. Assuming constant effectiveness, estimated health gains were 0.023-0.025 QALYs per patient with €218-€249 savings, suggesting potential annual national savings of €75-€86 million and a surplus of more than 8200 QALYs. Realised value depends on licence pricing, scalable integration, minimal human touchpoints for engagement and safety, and the transferability of trial effects to streamlined delivery models.
Agile legislation adapts principles from agile software development to lawmaking, emphasizing iteration, multi-stakeholder feedback, and embedded revision. We outline this learning-oriented governance model using three case studies: Germany’s stepwise digital health legislation, the EU AI Act, and U.S. FDA user-fee reauthorization. These examples highlight legislative designs that enable structured generation of real-world data and evidence during implementation, informing regulatory interpretation and iterative refinement in rapidly evolving technological domains.
The growing integration of digital technologies into healthcare requires understanding the impact of different consultation modalities on patient stress, memory and perceived credibility. This study compared different consultation modalities (human physician, in person or via video call; AI-style physician, as a chatbot or avatar) in standardized simulated medical consultations involving the delivery of bad news. For the AI-styled physician consultations, a Wizard of Oz design was used, in which participants were told they interacted with an AI physician while a human controlled the interaction. 163 healthy participants experienced either a human or an AI physician. Stress was measured through ratings and salivary cortisol, alongside memory retrieval and situation credibility. Human-based consultations led to higher stress levels than AI-styled formats. Greater perceived credibility was associated with stronger stress responses. Memory retrieval was lowest in the AI-styled chatbot condition. These findings show that different consultation modalities are associated with varying levels of stress and memory in medical settings. AI-styled physician interactions may reduce stress in routine medical communication, but their use should be carefully considered for critical medical communication.
REM sleep behavior disorder (RBD) is a robust prodromal marker of α-synucleinopathies: idiopathic RBD carries a 10-15-year phenoconversion risk of 80-90% to Parkinson's disease (PD) and related disorders. In major depressive disorder (MDD), comorbid RBD marks a subgroup at elevated prodromal PD risk, yet is frequently missed in psychiatric practice. Here, we developed a multimodal AI framework to detect comorbid RBD in MDD. From a clinical cohort of 329 patients, we obtained 261 video clips in 31 patients during reading and spontaneous-speech tasks, including 19 patients with MDD-RBD and 12 demographically and medication-matched MDD-only controls that, to our knowledge, formed the largest cohort of its kind worldwide. We used a dual-stream multimodal model that learned facial dynamics from video and vocal features from speech, and then combined both signals to predict comorbid RBD. In 5-fold cross-validation, our best model achieved 80.5% accuracy and 0.848 F1-score. Explainability analysis highlighted lower-face tension and variability, together with brow lowering, as candidate biomarkers requiring further validation. Predicted risk correlated with RBDQ score (r = 0.53, p = 0.005) and weakly with UPDRS motor score (r = 0.34, p = 0.067). Out-of-distribution evaluation showed broadly similar patterns, supporting the promise of multimodal AI for predicting RBD in MDD and identifying interpretable potential digital markers of prodromal synucleinopathy.
This scoping review summarizes the progress of reinforcement learning (RL) in clinical decision-making for sepsis at the intersection of medicine and artificial intelligence (AI). All 72 included studies were retrospective, with the majority using the Medical Information Mart for Intensive Care (MIMIC) database (58 studies [80.6%]), and relatively few employing private datasets (10 studies [13.9%]). Study designs varied widely, especially in state representation, action space, reward definition, and choice of algorithms. Most research focused on vasopressor and intravenous fluid management, while fewer studies addressed antibiotics, corticosteroids, mechanical ventilation, heparin, or vasopressin. Although many studies reported RL-derived policies that outperformed clinicians, the reliability and validity of the evaluation methods remain uncertain. Future research should emphasize clinically guided design of states, actions, and rewards, develop rigorous and widely accepted evaluation tools, and explore a broader range of sepsis treatment strategies. Ultimately, advancing interpretability, generalizability, and safety will be critical to effectively integrating RL into routine clinical practice.
Older adults face elevated opioid-related risks driven by multimorbidity, altered pharmacokinetics, and polypharmacy. We present a hybrid digital health framework that integrates a Long Short-Term Memory (LSTM) network for temporal risk prediction with a Retrieval-Augmented Generation (RAG) module for evidence-grounded clinical explanation tailored to adults aged ≥65 years. Using de-identified Prescription Drug Monitoring Program (PDMP) data from 2016 to 2022, the system predicts opioid risk classification, Beers medication safety level, and potentially inappropriate medication (PIM) status, achieving macro-F1 scores of 0.72, 0.77, and 0.84, respectively. To support interpretability and safety, explanations are generated from a curated clinical knowledge base incorporating the Centre for Disease Control (CDC) opioid guidance, the American Geriatrics Society (AGS) Beers Criteria, and geriatric pharmacotherapy literature. Retrieval-grounded reasoning, uncertainty signaling, and clinician-review cautions are embedded to mitigate unsupported inferences and automation bias. The framework is designed to align with PDMP-style data flows and incorporates Social Vulnerability Index (SVI) context to account for area-level determinants relevant to opioid stewardship. By coupling temporal prediction with verifiable, guideline-grounded explanations, this work illustrates an operational, responsible-AI design approach for transparent and clinician-centered decision support in opioid management for ageing populations.
Deep learning approaches for cognitive impairment diagnosis have shown considerable promise, but their clinical translation remains limited by poor interpretability and weak linkage between model outputs and established medical evidence. Here we developed the Multimodal Evidence-Driven Reasoning Framework (MEDRF), which integrates a Multimodal Hierarchical Cascade (mHC) classifier with a retrieval-augmented large language model (RAG-LLM) for evidence-guided reasoning. MEDRF leverages routinely collected non-invasive data from clinical profiles and structural MRI to identify cognitive impairment stages and etiologies. Across 15 diagnostic labels, mHC outperformed flat multimodal baselines, supporting hierarchical diagnostic modeling. When the mHC was evaluated under progressive feature masking, performance declined with increasing missingness, whereas RAG-LLM correction mitigated this effect, especially under severe sparsity. In external validation on a heterogeneous cohort with primary labels, integrating RAG-LLM with the mHC improved all evaluation metrics, increasing overall accuracy from 0.706 ± 0.038 to 0.753 ± 0.032. The RAG module resolves ambiguous predictions by retrieving analogous cases and supporting textual evidence, enabling multi-hop reasoning across conflicting clinical cues. Physician review further indicated favorable quality and perceived usefulness of the generated reports. By synthesizing hierarchical prediction with evidence-grounded reasoning, MEDRF provides a robust interpretable framework for decision support, particularly in incomplete or diagnostically ambiguous presentations.
Pancreatic ductal adenocarcinoma (PDAC) is one of the most lethal malignancies worldwide, and accurate prognostic prediction remains highly challenging due to its marked biological heterogeneity and complex tumor microenvironment. To address this challenge, a histopathomics-based survival prediction system (HPSurv) was developed using histopathological whole-slide images (WSIs) for individualized overall survival (OS) prediction. Within this framework, pathological tissue classification, quantitative characterization of tumor spatial heterogeneity, and a survival Transformer were integrated to enable multi-level representation learning from histopathological data. The system was developed and evaluated in 1020 patients across five independent cohorts. Compared with conventional clinicopathological indicators, significantly improved prognostic performance was achieved across multicenter cohorts (p < 0.05), with a mean C-index of 0.761 and time-dependent AUCs of 0.936, 0.877, and 0.772 for predicting 6-month, 2-year, and 3-year survival, respectively. Subgroup analyses further supported its role as an independent prognostic factor and suggested its potential utility in stratifying patients with respect to ACT-related outcomes. In addition, significant associations with key PDAC molecular pathways were observed, providing biological insights into the model predictions and supporting interpretability. In the study, an interpretable and high-performing artificial intelligence (AI) framework for quantitative modeling of PDAC was established. Objective characterization of tumor heterogeneity and accurate postoperative survival prediction are enabled, with potential value for personalized management in PDAC.
Bleeding and thrombosis represent a dynamic continuum in critically ill patients, where overlapping etiologies and heterogeneous pathophysiological mechanisms give rise to mixed-pattern coagulopathy. Conventional risk models usually focus on static, single-timepoint predictions and fail to adapt to rapidly evolving clinical states. Here, we retrospectively analyzed 2537 unique patients with 10,851 longitudinal records, constructing cycles based on clinical practice and engineering pharmacokinetic features. We then developed Adaptive Coagulopathy RL with BiLSTM-Attention (ACRLA), a distributional offline reinforcement learning framework designed to guide dynamic, individualized management of coagulopathy under laboratory-guided monitoring. Model performance was validated against clinician-adjudicated phenotypes, and sensitivity analyses to ensure robustness. The model showed stable convergence, achieving a positive per-patient test reward (0.077 ± 0.1651, 95% CI: 0.046-0.107) with a mean cumulative reward of 1.317. Model-derived tendencies aligned with clinical phenotypes, and feature importance identified D-dimer and platelet count as dominant drivers. The model's therapeutic preference hierarchy mirrored clinical guidelines. These findings indicate that ACRLA effectively captures dynamic coagulopathy trajectories in a clinically interpretable state space and provides a data-driven tool for personalized, adaptive management of coagulopathy in critically ill patients.
Cognitive impairment (CI) is an emerging public health challenge in rural aging populations, where access to formal cognitive testing is limited. Using data from 4781 adults in the Rural Chinese Cohort Study (RCCS) with 10-year follow-up and the Shenzhen Aging-related Disorder Cohort (SADC) with baseline cognitive assessment, this study developed an interpretable machine learning model for long-term CI risk prediction and evaluated its cross-population transportability using eight routinely collected indicators. Among nine algorithms, the Random Forest model showed the most balanced performance, with AUCs of 0.721 (95% CI: 0.596, 0.846) in the RCCS internal validation set for 10-year incident CI prediction and 0.690 (95% CI: 0.664, 0.717) in the SADC baseline-CI evaluation, acceptable calibration, and positive net benefit across threshold probabilities of ~10-40%. SHAP identified age, education, and systolic blood pressure as the main predictive contributors. Subgroup analyses showed heterogeneous discrimination, while sensitivity analyses supported broadly consistent performance under alternative data partitioning and MICE imputation. Mediation analysis suggested a primarily direct association between baseline age and later CI. These findings indicate that a simple, interpretable model based on widely available clinical indicators may support preliminary risk stratification and targeted cognitive screening in resource-constrained rural settings.
Despite mental health apps having a rigorous evidence base, little is known about the proportion of efficacious apps that are publicly available. Our systematic review and meta-analysis identified self-guided mental health apps evaluated in controlled trials, determined their public availability, and examined how evidence quality and strength influence availability. A literature search identified 16,971 records, of which 110 studies, investigating 112 apps, met inclusion criteria. Of the 81 unique self-guided apps, 42 (52%) were publicly available. Apps were most commonly available through both Apple App and Google Play Stores (79%) at no cost (40%). Study quality assessed using Cochrane Risk of Bias 2 tool had no significant association with app availability (p = 0.851). Meta-analysis found no significant difference (p = 0.228) in effect size compared to control between available apps (g = 0.33) and unavailable apps (g = 0.45). Considerations apart from evidence quality or strength determine app translation highlighting a gap between evidence and availability.