Citations

Abstract Hypertensive disorders of pregnancy (HDP) are a leading cause of maternal and perinatal morbidity, yet modifiable environmental risk factors remain poorly characterized. Prior studies typically have only examined a limited number of exposures and have rarely distinguished HDP subtypes (ie, gestational hypertension, preeclampsia, eclampsia, and chronic hypertension with or without superimposed preeclampsia) or accounted for residential mobility during pregnancy. To address these gaps, we conducted a spatial and contextual exposome study of HDP using linked electronic health records (EHR) and vital statistics data in Florida. We analyzed 686 412 singleton pregnancies conceived between 2013 and 2018, using computable phenotyping to distinguish HDP subtypes. A total of 245 spatial and contextual exposome measures spanning natural, built, and social environments were linked to residential histories from conception through gestational week 19. Using a two-phase double machine learning (DML) framework with exposure-specific, directed acyclic graph-guided confounder adjustment, we conducted discovery and replication analyses, followed by multi-treatment DML to estimate effect sizes. In Phase 1, 26 exposome measures replicated for gestational hypertension and 34 for overall HDP. In Phase 2, 12 measures remained associated with gestational hypertension and 11 with overall HDP, including air toxicants, meteorological factors, ultraviolet radiation, neighborhood crime indicators, environmental noise, and proximity to coastline. No exposures passed multiple-comparison thresholds for preeclampsia or eclampsia. These findings demonstrate that the spatial and contextual exposome contributes to HDP in a subtype-specific manner. Integrating EHR-linked phenotyping, residential mobility, and causal machine-learning methods offers a scalable framework for identifying environmental factors relevant to HDP prevention.

Abstract Objectives To comprehensively evaluate the validity of International Classification of Diseases, Tenth Revision, Clinical Modification (ICD-10-CM) codes for both prevalent diagnoses and less common diseases, and to assess the performance of a large language model (LLM)-based system in validating these codes. Materials and Methods This retrospective study analyzed hospital admissions from the Medical Information Mart for Intensive Care IV (MIMIC-IV) database. We developed a validated LLM-based system using GPT-4o, refined through iterative prompt engineering, to assess ICD-10-CM code validity. We measured the positive predictive value (PPV) of ICD-10-CM codes, PPV of principal and secondary diagnoses, and the performance of an LLM-based system in code validation. Results Among 865 079 assigned codes, the PPV was 84.6% (95% CI, 84.5%-84.6%). Principal diagnoses had a PPV of 93.9% (95% CI, 93.7%-94.1%), while secondary diagnoses had a PPV of 83.8% (95% CI, 83.7%-83.9%). The LLM system demonstrated high performance in validating ICD codes, achieving 93.6% accuracy, 95.4% sensitivity, and 85.2% specificity. Among correctly assigned secondary diagnoses, the majority (67.9%) represented historical or baseline conditions, while 32.1% reflected active conditions that deviated from baseline status; 22.3% of these emerged after hospital admission. PPV decreases with later diagnosis positions, with the largest decline occurring between principal and secondary diagnoses. Discussion and Conclusion In this large-scale evaluation, ICD-10-CM codes exhibited generally high accuracy, though variability existed by position and condition type. A validated LLM system performed comparably to physician review and offers a scalable means to improve coding accuracy. These findings support the potential for integrating LLM-based auditing into routine workflows to strengthen the quality of administrative and research data.

ABSTRACT Objective To comprehensively evaluate the validity of ICD-10-CM codes for both prevalent diagnoses and less common diseases, and to assess the performance of a large language model (LLM)-based system in validating these codes. Materials and Methods This retrospective study analyzed hospital admissions from the Medical Information Mart for Intensive Care (MIMIC-IV) database. We developed a validated LLM-based system using GPT-4o, refined through iterative prompt engineering, to assess ICD-10-CM code validity. We measured the PPV of ICD-10-CM codes, PPV of principal and secondary diagnoses, and the performance of an LLM-based system in code validation. Results Among 865,079 assigned codes, the PPV was 84.6% (95% CI, 84.5%-84.6%). Principal diagnoses had a PPV of 93.9% (95% CI, 93.7%-94.1%), while secondary diagnoses had a PPV of 83.8% (95% CI, 83.7%-83.9%). The LLM system demonstrated high performance in validating ICD codes, achieving 93.6% accuracy, 95.4% sensitivity and 85.2% specificity. Among correctly assigned secondary diagnoses, the majority (67.9%) represented historical or baseline conditions, while 32.1% reflected active conditions that deviated from baseline status; 22.3% of these emerged after hospital admission. PPV decreases with later diagnosis positions, with the largest decline occurring between principal and secondary diagnoses. Discussion and Conclusion In this large-scale evaluation, ICD-10-CM codes exhibited generally high accuracy, though variability existed by position and condition type. A validated LLM system performed comparably to physician review and offers a scalable means to improve coding accuracy. These findings support the potential for integrating LLM-based auditing into routine workflows to strengthen the quality of administrative and research data.

Abstract Background Dynamically predicting patients at risk of Alzheimer's disease (AD) is crucial for timely treatment and care in precision medicine. Method We present a novel architecture, a Kolmogorov‐Arnold networks (KANs)‐based joint prediction model of longitudinal and survival data (JM‐KAN). Real‐life data of patients with Mild Cognitive Impairment (MCI) with no prior AD diagnosis from National Alzheimer's Coordinating Center (NACC) dataset was utilized for individualized predictions. We built joint models with landmark times every 1 year up to 4 years after the MCI diagnosis, and predicted survival outcomes one year after each landmark (i.e. 1, 2, 3, and 4 years). Integrated AUC (iAUC), and integrated Brier score (iBS) were measured for existing prediction models and our novel architecture. Result Out of 2711 unique patients with MCI, 821 patients progressed to develop AD. Mean time‐to‐event in years was 3.19 (standard deviation (SD): 2.29). The number of visits prior to AD diagnosis was 4.03 (SD: 2.41). With iAUC of 0.789, iBS of 0.118, JM‐KAN has both the highest iAUC and lowest iBS among joint models, which indicates superiority in discrimination and overall performance. For individualized predictions, the model successfully predicted both longitudinal risk factors and survival probability, as new information came in every year. Conclusion Using neural network‐based Artificial Intelligence (AI) model, we demonstrated that this architecture improves upon prior methods in dynamically updating predictions of AD as new longitudinal data becomes available.

Background: Generative AI and synthetic media have enabled realistic human Embodied Conversational Agents (ECAs) or avatars. A subset of this technology replicates faces and voices to create realistic likenesses. When combined with avatars, these methods enable the creation of “digital twins” of physicians, offering patients scalable, 24/7 clinical communication outside the immediate clinical environment. This study evaluated surgical patient perceptions of an AI-generated surgeon avatar for postoperative education. Methods: We conducted a pilot feasibility study with 30 plastic surgery patients at Mayo Clinic, USA (July–August 2025). A bespoke interactive surgeon avatar was developed in Python using the HeyGen IV model to reproduce the surgeon’s likeness. Patients interacted with the avatar through natural voice queries, which were mapped to predetermined, pre-recorded video responses covering ten common postoperative topics. Patient perceptions were assessed using validated scales of usability, engagement, trust, eeriness, and realism, supplemented by qualitative feedback. Results: The avatar system reliably answered 297 of 300 patient queries (99%). Usability was excellent (mean System Usability Scale score = 87.7 ± 11.5) and engagement high (mean 4.27 ± 0.23). Trust was the highest-rated domain, with all participants (100%) finding the avatar trustworthy and its information believable. Eeriness was minimal (mean = 1.57 ± 0.48), and 96.7% found the avatar visually pleasing. Most participants (86.6%) recognized the avatar as their surgeon, although many still identified it as artificial; voice resemblance was less convincing (70%). Interestingly, participants with prior exposure to deepfakes demonstrated consistently higher acceptance, rating usability, trust, and engagement 5–10% higher than those without prior exposure. Qualitative feedback highlighted clarity, efficiency, and convenience, while noting limitations in realism and conversational scope. Conclusions: The AI-generated physician avatar achieved high patient acceptance without triggering uncanny valley effects. Transparency about the synthetic nature of the technology enhanced, rather than diminished, trust. Familiarity with the physician and institutional credibility likely played a key role in the high trust scores observed. When implemented transparently and with appropriate safeguards, synthetic physician avatars may offer a scalable solution for postoperative education while preserving trust in clinical relationships.

Background: During postoperative recovery, patients and their caregivers often lack crucial information, leading to numerous repetitive inquiries that burden healthcare providers. Traditional discharge materials, including paper handouts and patient portals, are often static, overwhelming, or underutilized, leading to patient overwhelm and contributing to unnecessary ER visits and overall healthcare overutilization. Conversational chatbots offer a solution, but Natural Language Processing (NLP) systems are often inflexible and limited in understanding, while powerful Large Language Models (LLMs) are prone to generating “hallucinations”. Objective: To combine the deterministic framework of traditional NLP with the probabilistic capabilities of LLMs, we developed the AI Virtual Assistant (AIVA) Platform. This system utilizes a retrieval-augmented generation (RAG) architecture, integrating Gemini 2.0 Flash with a medically verified knowledge base via Google Vertex AI, to safely deliver dynamic, patient-facing postoperative guidance grounded in validated clinical content. Methods: The AIVA Platform was evaluated through 750 simulated patient interactions derived from 250 unique postoperative queries across 20 high-frequency recovery domains. Three blinded physician reviewers assessed formal system performance, evaluating classification metrics (accuracy, precision, recall, F1-score), relevance (SSI Index), completeness, and consistency (5-point Likert scale). Safety guardrails were tested with 120 out-of-scope queries and 30 emergency escalation scenarios. Additionally, groundedness, fluency, and readability were assessed using automated LLM metrics. Results: The system achieved 98.4% classification accuracy (precision 1.0, recall 0.98, F1-score 0.9899). Physician reviews showed high completeness (4.83/5), consistency (4.49/5), and relevance (SSI Index 2.68/3). Safety guardrails successfully identified 100% of out-of-scope and escalation scenarios. Groundedness evaluations demonstrated strong context precision (0.951), recall (0.910), and faithfulness (0.956), with 95.6% verification agreement. While fluency and semantic alignment were high (BERTScore F1 0.9013, ROUGE-1 0.8377), readability was 11th-grade level (Flesch–Kincaid 46.34). Conclusion: The simulated testing demonstrated strong technical accuracy, safety, and clinical relevance in simulated postoperative care. Its architecture effectively balances flexibility and safety, addressing key limitations of standalone NLP and LLMs. While readability remains a challenge, these findings establish a solid foundation, demonstrating readiness for clinical trials and real-world testing within surgical care pathways.

Abstract Alzheimer’s Disease (AD) 1 is a progressive neurodegenerative disorder with limited therapeutic options, driving interest in drug repurposing to accelerate treatment discovery. Drug repurposing has emerged as a promising strategy to accelerate therapeutic discovery by repositioning existing drugs for new clinical indications. Recent computational repurposing approaches, including knowledge graph reasoning, transcriptomic signature analysis, and integrative literature mining, have demonstrated strong predictive capabilities 2 . However, these methods often yield divergent drug rankings, which makes it difficult to decide which candidates to advance for experimental follow-up and results in substantial gaps between computational predictions and feasible in vivo validation 2 .To bridge this computational-experimental gap, we proposed an advanced prioritization framework leveraging large language models (LLMs). Our method systematically evaluated three state-of-the-art (SOTA) and representative computational methods (TxGNN 3 , Composition-based Graph Convolutional Network (CompGCN) 4 , and a regularized logistic regression (RLR) 5 , to analyze both their predictive performance and pharmaceutical class distributions. By integrating the strengths and divergences of these models, we generated a unified, streamlined list of 90 candidate drugs for further prioritization. We then utilized an LLM-based agent to perform evidence synthesis from biomedical literature abstracts for each candidate. This process mimics expert manual curation but significantly reduces human effort and time by efficiently distilling vast textual data into actionable insights. Applying consistent and transparent selection criteria, we obtained a refined and prioritized list of drug candidates suitable for subsequent in vivo experimental validation. The robustness and clinical relevance of our framework were validated using real-world data from Alzheimer’s patient cohorts, clinical trial registries, and expert pharmacological reviews. This comprehensive validation confirmed that our LLM-driven approach enhances efficiency, consistency, scalability, and generalizability. By integrating computational predictions with scalable evidence synthesis and multifaceted validation, our framework facilitated rapid and informed prioritization of repurposed drugs. Our framework can potentially accelerate the translational pathway toward viable AD therapeutics. Moreover, the versatility of our framework can also be applied to drug repurposing efforts for other diseases beyond AD.

Abstract Background Acupuncture, a key modality in traditional Chinese medicine, is gaining global recognition as a complementary therapy and a subject of increasing scientific interest. However, fragmented and unstructured acupuncture knowledge spread across diverse sources poses challenges for semantic retrieval, reasoning, and in-depth analysis. To address this gap, we developed AcuKG, a comprehensive knowledge graph that systematically organizes acupuncture-related knowledge to support sharing, discovery, and artificial intelligence–driven innovation in the field. Methods AcuKG integrates data from multiple sources, including online resources, guidelines, PubMed literature, ClinicalTrials.gov, and multiple ontologies (SNOMED CT, UBERON, and MeSH). We employed entity recognition, relation extraction, and ontology mapping to establish AcuKG, with human-in-the-loop to ensure data quality. Two cases evaluated AcuKG’s usability: (1) how AcuKG advances acupuncture research for obesity and (2) how AcuKG enhances large language model (LLM) application on acupuncture question-answering. Results AcuKG comprises 1839 entities and 11 527 relations, mapped to 1836 standard concepts in 3 ontologies. Two use cases demonstrated AcuKG’s effectiveness and potential in advancing acupuncture research and supporting LLM applications. In the obesity use case, AcuKG identified highly relevant acupoints (eg, ST25, ST36) and uncovered novel research insights based on evidence from clinical trials and literature. When applied to LLMs in answering acupuncture-related questions, integrating AcuKG with GPT-4o and LLaMA 3 significantly improved accuracy (GPT-4o: 46% → 54%, P = .03; LLaMA 3: 17% → 28%, P = .01). Conclusion AcuKG is an open dataset that provides a structured and computational framework for acupuncture applications, bridging traditional practices with acupuncture research and cutting-edge LLM technologies.