National American University

The COVID-19 pandemic has reshaped global health in ways that extend beyond acute illness. As nations recover, growing evidence points to a concerning long-term consequence: individuals infected with SARS-CoV-2 face a heightened risk of developing diabetes. This post-viral metabolic disruption affects individuals across all age groups, including those without prior glycaemic abnormalities. Cohort studies from the United States, Europe, and Asia consistently demonstrate an increased prevalence of diabetes among COVID-19 survivors. Proposed mechanisms include direct pancreatic injury via ACE2 receptor pathways, immune-mediated β-cell dysfunction, systemic inflammation, and the secondary effects of therapy, stress, and physical inactivity. Despite these findings, post-COVID diabetes remains under-recognized in both clinical practice and public health planning. We contend that this phenomenon represents an impending silent epidemic that could substantially exacerbate the global burden of non-communicable diseases, particularly in low- and middle-income countries with limited screening infrastructure. The absence of targeted surveillance, clinical guidelines, and coordinated research efforts reflects a missed opportunity for early intervention. To prevent a slow-moving public health crisis, post-COVID diabetes must be swiftly integrated into national and global health strategies—through routine metabolic screening in long COVID care, improved data systems, public awareness campaigns, and equitable access to care. Current preventive and policy actions will determine the metabolic health outcomes of future generations.

Background: Healthcare databases in the United States, including electronic health record systems, claims repositories, imaging archives and research data warehouses, have become high-value targets for ransomware, credential theft, insider misuse and privacy attacks. Artificial intelligence and machine learning are widely promoted as the basis for adaptive security that learns from system behaviour, yet the supporting evidence and the fitness of the governing regulatory environment remain uncertain. Objectives: This review critically appraises evidence on security approaches for healthcare data that are enabled by artificial intelligence and machine learning, examines how the contemporary threat landscape shapes design requirements, and identifies policy gaps specific to the United States healthcare system. Methods: A critical narrative review was conducted using structured searches of biomedical, multidisciplinary and open scholarly indexes, supplemented by federal regulatory, standards and oversight sources and by citation searching. Evidence was appraised for study design, realism of evaluation data, external validity and relevance to United States healthcare settings, and was synthesised thematically. Principal Findings: Empirical evidence is strongest for the scale and operational consequences of ransomware and for the inadequacy of de-identification as a stand-alone safeguard. Machine learning methods for detecting inappropriate record access and network intrusions show promising discrimination in retrospective and testbed evaluations, but prospective, multi-site and adversarially tested deployments are rare, and benchmark data often represent clinical environments poorly. Federated learning, differential privacy and cryptographic computation reduce specific exposures without eliminating leakage, and recent work indicates that privacy risk is distributed unevenly across patient groups. Artificial intelligence systems themselves introduce poisoning, prompt injection and membership inference risks. Federal governance remains fragmented: the Security Rule issued under the Health Insurance Portability and Accountability Act is technology-neutral, its proposed modernisation had not been finalised by the end of the review period, and guidance on artificial intelligence is largely voluntary and subject to rapid policy change. Conclusions: Adaptive security for healthcare databases is technically plausible but empirically immature. Progress will depend on realistic evaluation standards, security assurance for the defensive models themselves, sustained support for under-resourced organisations and clearer regulatory expectations for artificial intelligence that processes or protects health data.

Early detection of chronic diseases such as diabetes and cardiovascular conditions significantly improves patient prognosis and reduces healthcare costs. While multimodal deep learning-combining medical imaging and Electronic Health Records (EHR)-has shown superior diagnostic accuracy, current state-of-the-art models are often computationally expensive, hindering deployment in resource-constrained clinical settings or wearable edge devices. To address this, we propose LMFNet, a Lightweight Multimodal Fusion Network. Our architecture integrates a modified MobileNetV3 for visual feature extraction and a distilled Transformer for textual clinical notes, fused via a novel Gated Cross-Modality Attention (GCMA) mechanism. This approach dynamically weighs the importance of visual versus textual data while minimizing floating-point operations (FLOPs). We evaluated our method on the MIMIC-IV and a composite diabetic retinopathy dataset. Experimental results demonstrate that LMFNet achieves an accuracy of 94.2% and an F1-score of 92.7%, comparable to heavy ensemble models, while reducing inference time by 65% and parameter count by 70%. This study presents a scalable, efficient solution for real-time chronic disease screening.

The analysis of gigapixel Whole-Slide Images (WSIs) in digital pathology is the gold standard for cancer diagnosis, but it presents a formidable challenge for AI models due to extreme data scale. Current state-of-the-art methods rely on Multiple Instance Learning (MIL), treating WSIs as an unordered “bag-of-patches.” This approach critically discards the spatial microarchitecture of tissue, which is a primary source of diagnostic information for pathologists. To overcome this limitation, we propose Path-HGT, a novel Pathology Hierarchical Graph Transformer framework. Path-HGT first models a WSI as a spatially-explicit graph of patches. It then employs a hierarchical architecture composed of stacked Graph Transformer (GATv2) layers and self-attention graph pooling (SAGPool) modules. This design enables the model to learn multi-scale representations, capturing both local cellular interactions at the patch level and global tissue-region architectures at coarser graph levels. We validate Path-HGT on the Camelyon16 and TCGA-RCC benchmarks. Our framework achieves state-of-the-art performance, attaining an AUC of 0.965 on Camelyon16 and an accuracy of 0.946 on TCGA-RCC, significantly outperforming existing MIL, Transformer, and “flat” GNN baselines. Our work demonstrates that modeling explicit spatial hierarchy is crucial for robust and interpretable gigapixel WSI analysis.

The exorbitant cost and high attrition rates of drug discovery necessitate the development of high-precision computational models for identifying Drug-Target Interactions (DTIs). While Graph Neural Networks (GNNs) have become the de facto standard for modeling molecular structures, existing approaches largely rely on pairwise graph topologies, failing to capture highorder biochemical dependencies such as pharmacophore clusters and protein binding pockets. Furthermore, the scarcity of labeled interaction data often leads to poor generalization in “cold-start” scenarios. To address these limitations, we propose $\mathbf{H}^{\mathbf{2}}$ CL-DTI, a novel framework integrating Hierarchical Hypergraph Convolutional Networks with Cross-View Contrastive Learning. We model drugs and proteins as hypergraphs to encapsulate multiscale structural motifs and employ a dual-view encoder to align topological representations with semantic embeddings from pretrained language models. Extensive experiments on the Davis and KIBA benchmarks demonstrate that $\mathbf{H}^{\mathbf{2}} \mathbf{C L}$-DTI achieves state-of-the-art performance, reducing Mean Squared Error (MSE) to 0.196 and 0.129 respectively, significantly outperforming current baselines like DrugBAN and GraphDTA. Our results confirm that modeling high-order correlations via hypergraphs enhances both predictive accuracy and robustness against data sparsity.