Medicine & Health

468 items

Journal of general internal medicine Sep 15, 2026
In this retrospective cohort study at a single tertiary academic medical center, large language models predicted same-day discharge from clinical documentation in the 30 hours preceding a 06:00 index time for adult inpatients admitted in 2024 with a length of stay between 2 and 14 days, using a randomly generated validation set (n = 860) and test set (n = 886); the baseline GPT-5 prompt achieved an F1 score of 0.48 and sensitivity of 0.
In this retrospective cohort study at a single tertiary academic medical center, large language models predicted same-day discharge from clinical documentation in the 30 hours preceding a 06:00 index time for adult inpatients admitted in 2024 with a length of stay between 2 and 14 days, using a randomly generated validation set (n = 860) and test set (n = 886); the baseline GPT-5 prompt achieved an F1 score of 0.48 and sensitivity of 0.
Optimizing Large Language Models for Hospital Discharge Prediction: A Retrospective Cohort Study with Inference-Time Optimization Strategies
In this retrospective cohort study at a single tertiary academic medical center, large language models predicted same-day discharge from clinical documentation in the 30 hours preceding a 06:00 index time for adult inpatients admitted in 2024 with a length of stay between 2 and 14 days, using a randomly generated validation set (n = 860) and test set (n = 886); the baseline GPT-5 prompt achieved an F1 score of 0.48 and sensitivity of 0.
In this retrospective cohort study at a single tertiary academic medical center, large language models predicted same-day discharge from clinical documentation in the 30 hours preceding a 06:00 index time for adult inpatients admitted in 2024 with a length of stay between 2 and 14 days, using a randomly generated validation set (n = 860) and test set (n = 886); the baseline GPT-5 prompt achieved an F1 score of 0.48 and sensitivity of 0.
Chinese medical journal Sep 15, 2026
This review systematically evaluates established and emerging non-invasive tests for detecting early-stage liver fibrosis (F1–F2), spanning serum biomarkers, molecular imaging probes, AI-based analysis of conventional imaging, alternative biofluids such as urine, exhaled breath and saliva, gut microbiota-derived biomarkers, and multiomics-integrated models, concluding that these approaches show diagnostic promise but remain constrained by limited validation, absent standardization, and unclear integration into clinical pathways, and proposing a tiered diagnostic framework as a near-term implementation route.
This review systematically evaluates established and emerging non-invasive tests for detecting early-stage liver fibrosis (F1–F2), spanning serum biomarkers, molecular imaging probes, AI-based analysis of conventional imaging, alternative biofluids such as urine, exhaled breath and saliva, gut microbiota-derived biomarkers, and multiomics-integrated models, concluding that these approaches show diagnostic promise but remain constrained by limited validation, absent standardization, and unclear integration into clinical pathways, and proposing a tiered diagnostic framework as a near-term implementation route.
Non-invasive tests for early-stage liver fibrosis: current advances and challenges
This review systematically evaluates established and emerging non-invasive tests for detecting early-stage liver fibrosis (F1–F2), spanning serum biomarkers, molecular imaging probes, AI-based analysis of conventional imaging, alternative biofluids such as urine, exhaled breath and saliva, gut microbiota-derived biomarkers, and multiomics-integrated models, concluding that these approaches show diagnostic promise but remain constrained by limited validation, absent standardization, and unclear integration into clinical pathways, and proposing a tiered diagnostic framework as a near-term implementation route.
This review systematically evaluates established and emerging non-invasive tests for detecting early-stage liver fibrosis (F1–F2), spanning serum biomarkers, molecular imaging probes, AI-based analysis of conventional imaging, alternative biofluids such as urine, exhaled breath and saliva, gut microbiota-derived biomarkers, and multiomics-integrated models, concluding that these approaches show diagnostic promise but remain constrained by limited validation, absent standardization, and unclear integration into clinical pathways, and proposing a tiered diagnostic framework as a near-term implementation route.
Research Square Sep 15, 2026
This paper proposes an evidence-informed conceptual framework that links multimodal clinical data (electronic health records, laboratory data, imaging, physiological monitoring, surgical video, device telemetry) through a modular AI architecture, domain adaptation, uncertainty estimation, and safety controls to a clinician-facing decision-support interface, explicitly separating the proposed architecture from evidence already reported in the literature and noting that clinical effectiveness of surgical AI remains to be established prospectively.
This paper proposes an evidence-informed conceptual framework that links multimodal clinical data (electronic health records, laboratory data, imaging, physiological monitoring, surgical video, device telemetry) through a modular AI architecture, domain adaptation, uncertainty estimation, and safety controls to a clinician-facing decision-support interface, explicitly separating the proposed architecture from evidence already reported in the literature and noting that clinical effectiveness of surgical AI remains to be established prospectively.
Medical Artificial Intelligence: A Multimodal, Human-Centered, Domain-Adaptive Framework for Surgical Decision Support and Patient Safety
This paper proposes an evidence-informed conceptual framework that links multimodal clinical data (electronic health records, laboratory data, imaging, physiological monitoring, surgical video, device telemetry) through a modular AI architecture, domain adaptation, uncertainty estimation, and safety controls to a clinician-facing decision-support interface, explicitly separating the proposed architecture from evidence already reported in the literature and noting that clinical effectiveness of surgical AI remains to be established prospectively.
This paper proposes an evidence-informed conceptual framework that links multimodal clinical data (electronic health records, laboratory data, imaging, physiological monitoring, surgical video, device telemetry) through a modular AI architecture, domain adaptation, uncertainty estimation, and safety controls to a clinician-facing decision-support interface, explicitly separating the proposed architecture from evidence already reported in the literature and noting that clinical effectiveness of surgical AI remains to be established prospectively.
Experimental hematology & oncology Sep 15, 2026
This study integrated six public single-cell RNA sequencing datasets, used inferCNV to infer copy number variations at the single-cell level and identify a high-CNV (HCNV) malignant subpopulation, built a multimodal deep-learning AI prognostic model on routine H&E-stained sections with HCNV activity as the biological anchor for pathology feature selection, and through multiomics screening plus in vitro and in vivo experiments identified RTN3 as the core driver gene, showing that RTN3 activates JAK2/STAT3 to transcriptionally upregulate glycolytic enzymes PKM2, GLUT1 and LDHA, driving glycolytic metabolic reprogramming and conferring gemcitabine resistance in bladder cancer.
This study integrated six public single-cell RNA sequencing datasets, used inferCNV to infer copy number variations at the single-cell level and identify a high-CNV (HCNV) malignant subpopulation, built a multimodal deep-learning AI prognostic model on routine H&E-stained sections with HCNV activity as the biological anchor for pathology feature selection, and through multiomics screening plus in vitro and in vivo experiments identified RTN3 as the core driver gene, showing that RTN3 activates JAK2/STAT3 to transcriptionally upregulate glycolytic enzymes PKM2, GLUT1 and LDHA, driving glycolytic metabolic reprogramming and conferring gemcitabine resistance in bladder cancer.
Multimodal AI and single-cell transcriptomics integrate to construct a histopathological prognostic model for bladder cancer, revealing the RTN3-glycolysis axis in chemoresistance
This study integrated six public single-cell RNA sequencing datasets, used inferCNV to infer copy number variations at the single-cell level and identify a high-CNV (HCNV) malignant subpopulation, built a multimodal deep-learning AI prognostic model on routine H&E-stained sections with HCNV activity as the biological anchor for pathology feature selection, and through multiomics screening plus in vitro and in vivo experiments identified RTN3 as the core driver gene, showing that RTN3 activates JAK2/STAT3 to transcriptionally upregulate glycolytic enzymes PKM2, GLUT1 and LDHA, driving glycolytic metabolic reprogramming and conferring gemcitabine resistance in bladder cancer.
This study integrated six public single-cell RNA sequencing datasets, used inferCNV to infer copy number variations at the single-cell level and identify a high-CNV (HCNV) malignant subpopulation, built a multimodal deep-learning AI prognostic model on routine H&E-stained sections with HCNV activity as the biological anchor for pathology feature selection, and through multiomics screening plus in vitro and in vivo experiments identified RTN3 as the core driver gene, showing that RTN3 activates JAK2/STAT3 to transcriptionally upregulate glycolytic enzymes PKM2, GLUT1 and LDHA, driving glycolytic metabolic reprogramming and conferring gemcitabine resistance in bladder cancer.
Zhonghua er bi yan hou tou jing wai ke za zhi = Chinese journal of otorhinolaryngology head and neck surgery Sep 15, 2026
In this retrospective two-center study of 2 365 patients (1 562 from the First Affiliated Hospital of Sun Yat-sen University and 803 from Ruijin Hospital, Shanghai Jiao Tong University School of Medicine), the authors developed ENDOVISTA-ENT, an integrated AI system with Model 1 for inside/outside-body image determination, Model 2 for recognition of 11 standard anatomical sites, and Model 3 for lesion localization and benign/malignant classification, reporting internal/external accuracies of 99.44%/99.84% and 96.09%/94.73%, malignant-lesion AUCs of 0.986/0.968, early nasopharyngeal, laryngeal, and hypopharyngeal cancer AUCs of 0.878-0.932, an increase in six physicians' overall interpretation accuracy on 200 pathologically confirmed cases from 78.50% to 88.20% with AI assistance (χ²=40.
In this retrospective two-center study of 2 365 patients (1 562 from the First Affiliated Hospital of Sun Yat-sen University and 803 from Ruijin Hospital, Shanghai Jiao Tong University School of Medicine), the authors developed ENDOVISTA-ENT, an integrated AI system with Model 1 for inside/outside-body image determination, Model 2 for recognition of 11 standard anatomical sites, and Model 3 for lesion localization and benign/malignant classification, reporting internal/external accuracies of 99.44%/99.84% and 96.09%/94.73%, malignant-lesion AUCs of 0.986/0.968, early nasopharyngeal, laryngeal, and hypopharyngeal cancer AUCs of 0.878-0.932, an increase in six physicians' overall interpretation accuracy on 200 pathologically confirmed cases from 78.50% to 88.20% with AI assistance (χ²=40.
ENDOVISTA-ENT: Development and Preliminary Validation of an Integrated Artificial Intelligence System for Quality-Control Assistance and Lesion Recognition During Nasopharyngolaryngoscopy
In this retrospective two-center study of 2 365 patients (1 562 from the First Affiliated Hospital of Sun Yat-sen University and 803 from Ruijin Hospital, Shanghai Jiao Tong University School of Medicine), the authors developed ENDOVISTA-ENT, an integrated AI system with Model 1 for inside/outside-body image determination, Model 2 for recognition of 11 standard anatomical sites, and Model 3 for lesion localization and benign/malignant classification, reporting internal/external accuracies of 99.44%/99.84% and 96.09%/94.73%, malignant-lesion AUCs of 0.986/0.968, early nasopharyngeal, laryngeal, and hypopharyngeal cancer AUCs of 0.878-0.932, an increase in six physicians' overall interpretation accuracy on 200 pathologically confirmed cases from 78.50% to 88.20% with AI assistance (χ²=40.
In this retrospective two-center study of 2 365 patients (1 562 from the First Affiliated Hospital of Sun Yat-sen University and 803 from Ruijin Hospital, Shanghai Jiao Tong University School of Medicine), the authors developed ENDOVISTA-ENT, an integrated AI system with Model 1 for inside/outside-body image determination, Model 2 for recognition of 11 standard anatomical sites, and Model 3 for lesion localization and benign/malignant classification, reporting internal/external accuracies of 99.44%/99.84% and 96.09%/94.73%, malignant-lesion AUCs of 0.986/0.968, early nasopharyngeal, laryngeal, and hypopharyngeal cancer AUCs of 0.878-0.932, an increase in six physicians' overall interpretation accuracy on 200 pathologically confirmed cases from 78.50% to 88.20% with AI assistance (χ²=40.
Nature medicine Sep 15, 2026
This work developed and evaluated a fully on-premise clinical AI agent that couples local operational control with a multi-perspective reliability framework to support selective autonomy, achieving 90.04% accuracy on a seven-disease task and 83.8% on a four-disease task across two MIMIC-IV-derived benchmarks, and finding that diagnostic behavioral consistency best discriminated correctness (AUC = 0.860, and AUC = 0.875 under stress testing), with a consistency threshold of 0.90 retaining 49.4% of cases at 98.9% diagnostic accuracy.
This work developed and evaluated a fully on-premise clinical AI agent that couples local operational control with a multi-perspective reliability framework to support selective autonomy, achieving 90.04% accuracy on a seven-disease task and 83.8% on a four-disease task across two MIMIC-IV-derived benchmarks, and finding that diagnostic behavioral consistency best discriminated correctness (AUC = 0.860, and AUC = 0.875 under stress testing), with a consistency threshold of 0.90 retaining 49.4% of cases at 98.9% diagnostic accuracy.
On-premise medical AI agents for reliable clinical decision-making
This work developed and evaluated a fully on-premise clinical AI agent that couples local operational control with a multi-perspective reliability framework to support selective autonomy, achieving 90.04% accuracy on a seven-disease task and 83.8% on a four-disease task across two MIMIC-IV-derived benchmarks, and finding that diagnostic behavioral consistency best discriminated correctness (AUC = 0.860, and AUC = 0.875 under stress testing), with a consistency threshold of 0.90 retaining 49.4% of cases at 98.9% diagnostic accuracy.
This work developed and evaluated a fully on-premise clinical AI agent that couples local operational control with a multi-perspective reliability framework to support selective autonomy, achieving 90.04% accuracy on a seven-disease task and 83.8% on a four-disease task across two MIMIC-IV-derived benchmarks, and finding that diagnostic behavioral consistency best discriminated correctness (AUC = 0.860, and AUC = 0.875 under stress testing), with a consistency threshold of 0.90 retaining 49.4% of cases at 98.9% diagnostic accuracy.
Journal of minimally invasive surgery Sep 15, 2026
This single-center, repeated cross-sectional descriptive survey administered anonymous online questionnaires before and one month after installation of an AI-based computer-aided detection (AI-CADe) system, with 29 and 25 endoscopy unit staff completing the pre- and post-installation surveys respectively, and found that staff held generally favorable perceptions of AI-CADe, recognizing its potential value for adenoma detection rate improvement, lesion removal, patient and procedural satisfaction, and workflow, while also reporting concerns about false-positive detection, overdetection, unnecessary biopsies, and dependence on AI, with accuracy and sensitivity most frequently selected as adoption factors.
This single-center, repeated cross-sectional descriptive survey administered anonymous online questionnaires before and one month after installation of an AI-based computer-aided detection (AI-CADe) system, with 29 and 25 endoscopy unit staff completing the pre- and post-installation surveys respectively, and found that staff held generally favorable perceptions of AI-CADe, recognizing its potential value for adenoma detection rate improvement, lesion removal, patient and procedural satisfaction, and workflow, while also reporting concerns about false-positive detection, overdetection, unnecessary biopsies, and dependence on AI, with accuracy and sensitivity most frequently selected as adoption factors.
User perceptions of an artificial intelligence-based computer-aided detection system in colonoscopy: a single-center exploratory survey study
This single-center, repeated cross-sectional descriptive survey administered anonymous online questionnaires before and one month after installation of an AI-based computer-aided detection (AI-CADe) system, with 29 and 25 endoscopy unit staff completing the pre- and post-installation surveys respectively, and found that staff held generally favorable perceptions of AI-CADe, recognizing its potential value for adenoma detection rate improvement, lesion removal, patient and procedural satisfaction, and workflow, while also reporting concerns about false-positive detection, overdetection, unnecessary biopsies, and dependence on AI, with accuracy and sensitivity most frequently selected as adoption factors.
This single-center, repeated cross-sectional descriptive survey administered anonymous online questionnaires before and one month after installation of an AI-based computer-aided detection (AI-CADe) system, with 29 and 25 endoscopy unit staff completing the pre- and post-installation surveys respectively, and found that staff held generally favorable perceptions of AI-CADe, recognizing its potential value for adenoma detection rate improvement, lesion removal, patient and procedural satisfaction, and workflow, while also reporting concerns about false-positive detection, overdetection, unnecessary biopsies, and dependence on AI, with accuracy and sensitivity most frequently selected as adoption factors.
Chinese medical journal Sep 15, 2026
Using retrospective deidentified clinical data from 1370 inpatients with chronic pancreatitis who underwent extracorporeal shock wave lithotripsy (ESWL) at Changhai Hospital between May 31, 2016 and June 26, 2019, this study retained 55 of 109 variables and compared ten machine-learning and deep-learning algorithms (XGBoost, LightGBM, CatBoost, random forest, support vector machine, artificial neural network, multilayer perceptron, TabNet, Transformer, and Wide&Deep) across three modeling schemes (all preprocessed variables, 21 variables selected by univariate screening, and ADASYN-oversampled training data) using the F1-score as the primary selection metric; TabNet achieved the best held-out test performance in the first two schemes (F1 of 0.
Using retrospective deidentified clinical data from 1370 inpatients with chronic pancreatitis who underwent extracorporeal shock wave lithotripsy (ESWL) at Changhai Hospital between May 31, 2016 and June 26, 2019, this study retained 55 of 109 variables and compared ten machine-learning and deep-learning algorithms (XGBoost, LightGBM, CatBoost, random forest, support vector machine, artificial neural network, multilayer perceptron, TabNet, Transformer, and Wide&Deep) across three modeling schemes (all preprocessed variables, 21 variables selected by univariate screening, and ADASYN-oversampled training data) using the F1-score as the primary selection metric; TabNet achieved the best held-out test performance in the first two schemes (F1 of 0.
Machine learning for early post-ESWL risk stratification of pancreatitis in chronic pancreatitis
Using retrospective deidentified clinical data from 1370 inpatients with chronic pancreatitis who underwent extracorporeal shock wave lithotripsy (ESWL) at Changhai Hospital between May 31, 2016 and June 26, 2019, this study retained 55 of 109 variables and compared ten machine-learning and deep-learning algorithms (XGBoost, LightGBM, CatBoost, random forest, support vector machine, artificial neural network, multilayer perceptron, TabNet, Transformer, and Wide&Deep) across three modeling schemes (all preprocessed variables, 21 variables selected by univariate screening, and ADASYN-oversampled training data) using the F1-score as the primary selection metric; TabNet achieved the best held-out test performance in the first two schemes (F1 of 0.
Using retrospective deidentified clinical data from 1370 inpatients with chronic pancreatitis who underwent extracorporeal shock wave lithotripsy (ESWL) at Changhai Hospital between May 31, 2016 and June 26, 2019, this study retained 55 of 109 variables and compared ten machine-learning and deep-learning algorithms (XGBoost, LightGBM, CatBoost, random forest, support vector machine, artificial neural network, multilayer perceptron, TabNet, Transformer, and Wide&Deep) across three modeling schemes (all preprocessed variables, 21 variables selected by univariate screening, and ADASYN-oversampled training data) using the F1-score as the primary selection metric; TabNet achieved the best held-out test performance in the first two schemes (F1 of 0.
bioRxiv Sep 15, 2026
This work presents the Hierarchical Temporal Transformer (HTT), a two-level architecture in which level 1 encodes each report with BiomedBERT adapted by LoRA and level 2 is a temporal transformer that reads a patient's full report sequence using a continuous-time positional encoding built from the measured number of days between visits plus learnable cancer-type embeddings; on a controlled synthetic corpus of sequential radiology reports it reaches validation AUROC 0.942 versus 0.881 for a single-report baseline and transfers to held-out pancreatic cancer at 0.995 versus 0.949, and on 4,786 real pathology reports from TCGA-Reports spanning 14 cancer types it predicts tumor grade for three types withheld entirely from training with AUROC 1.000 on thyroid carcinoma, 0.960 on sarcoma and 0.
This work presents the Hierarchical Temporal Transformer (HTT), a two-level architecture in which level 1 encodes each report with BiomedBERT adapted by LoRA and level 2 is a temporal transformer that reads a patient's full report sequence using a continuous-time positional encoding built from the measured number of days between visits plus learnable cancer-type embeddings; on a controlled synthetic corpus of sequential radiology reports it reaches validation AUROC 0.942 versus 0.881 for a single-report baseline and transfers to held-out pancreatic cancer at 0.995 versus 0.949, and on 4,786 real pathology reports from TCGA-Reports spanning 14 cancer types it predicts tumor grade for three types withheld entirely from training with AUROC 1.000 on thyroid carcinoma, 0.960 on sarcoma and 0.
Hierarchical Temporal Transformer for Cancer Grade Prediction and Cross-Cancer Transfer Learning from Pathology Reports
This work presents the Hierarchical Temporal Transformer (HTT), a two-level architecture in which level 1 encodes each report with BiomedBERT adapted by LoRA and level 2 is a temporal transformer that reads a patient's full report sequence using a continuous-time positional encoding built from the measured number of days between visits plus learnable cancer-type embeddings; on a controlled synthetic corpus of sequential radiology reports it reaches validation AUROC 0.942 versus 0.881 for a single-report baseline and transfers to held-out pancreatic cancer at 0.995 versus 0.949, and on 4,786 real pathology reports from TCGA-Reports spanning 14 cancer types it predicts tumor grade for three types withheld entirely from training with AUROC 1.000 on thyroid carcinoma, 0.960 on sarcoma and 0.
This work presents the Hierarchical Temporal Transformer (HTT), a two-level architecture in which level 1 encodes each report with BiomedBERT adapted by LoRA and level 2 is a temporal transformer that reads a patient's full report sequence using a continuous-time positional encoding built from the measured number of days between visits plus learnable cancer-type embeddings; on a controlled synthetic corpus of sequential radiology reports it reaches validation AUROC 0.942 versus 0.881 for a single-report baseline and transfers to held-out pancreatic cancer at 0.995 versus 0.949, and on 4,786 real pathology reports from TCGA-Reports spanning 14 cancer types it predicts tumor grade for three types withheld entirely from training with AUROC 1.000 on thyroid carcinoma, 0.960 on sarcoma and 0.
Ear, nose, & throat journal Sep 15, 2026
This study prompted ChatGPT to generate zero-shot appeal letters for cochlear implant insurance denials in asymmetric hearing loss or single-sided deafness across 30 prompt variations, had three cochlear implant providers score them against American Cochlear Implant Alliance guidelines, and checked citation accuracy; 96.6% of responses listed multiple benefits of cochlear implants and 79.3% partially aligned with the guidelines, but 51.7% contained hallucinated benefits and only six of 96 citations (6.3%) accurately referenced peer-reviewed sources, with the rest hallucinated (57.3%), erroneous (24%), or irrelevant (10%), indicating that human verification is needed before clinical or advocacy use.
This study prompted ChatGPT to generate zero-shot appeal letters for cochlear implant insurance denials in asymmetric hearing loss or single-sided deafness across 30 prompt variations, had three cochlear implant providers score them against American Cochlear Implant Alliance guidelines, and checked citation accuracy; 96.6% of responses listed multiple benefits of cochlear implants and 79.3% partially aligned with the guidelines, but 51.7% contained hallucinated benefits and only six of 96 citations (6.3%) accurately referenced peer-reviewed sources, with the rest hallucinated (57.3%), erroneous (24%), or irrelevant (10%), indicating that human verification is needed before clinical or advocacy use.
Using an AI Chatbot to Generate Cochlear Implant Insurance Appeal Letters: An Assessment of Accuracy and Citation Reliability
This study prompted ChatGPT to generate zero-shot appeal letters for cochlear implant insurance denials in asymmetric hearing loss or single-sided deafness across 30 prompt variations, had three cochlear implant providers score them against American Cochlear Implant Alliance guidelines, and checked citation accuracy; 96.6% of responses listed multiple benefits of cochlear implants and 79.3% partially aligned with the guidelines, but 51.7% contained hallucinated benefits and only six of 96 citations (6.3%) accurately referenced peer-reviewed sources, with the rest hallucinated (57.3%), erroneous (24%), or irrelevant (10%), indicating that human verification is needed before clinical or advocacy use.
This study prompted ChatGPT to generate zero-shot appeal letters for cochlear implant insurance denials in asymmetric hearing loss or single-sided deafness across 30 prompt variations, had three cochlear implant providers score them against American Cochlear Implant Alliance guidelines, and checked citation accuracy; 96.6% of responses listed multiple benefits of cochlear implants and 79.3% partially aligned with the guidelines, but 51.7% contained hallucinated benefits and only six of 96 citations (6.3%) accurately referenced peer-reviewed sources, with the rest hallucinated (57.3%), erroneous (24%), or irrelevant (10%), indicating that human verification is needed before clinical or advocacy use.

Page 39 · showing 10