Implementing Trustworthy AI in VA High Reliability Health Care Organizations.

Background: Artificial intelligence (AI) has great potential to improve health care quality, safety, efficiency, and access. However, the widespread adoption of health care AI needs to catch up to other sectors. Challenges, including data limitations, misaligned incentives, and organizational obstacles, have hindered implementation. Strategic demonstrations, partnerships, aligned incentives, and continued investment are needed to enable responsible adoption of AI. High reliability health care organizations offer insights into safely implementing major initiatives through frameworks like the Patient Safety Adoption Framework, which provides practical guidance on leadership, culture, process, measurement, and person-centeredness to successfully adopt safety practices. High reliability health care organizations ensure consistently safe and high quality care through a culture focused on reliability, accountability, and learning from errors and near misses. Observations: The Veterans Health Administration applied a high reliability health care model to instill safety principles and improve outcomes. As the use of AI becomes more widespread, ensuring its ethical development is crucial to avoiding new risks and harm. The US Department of Veterans Affairs National AI Institute proposed a Trustworthy AI Framework tailored for federal health care with 6 principles: purposeful, effective and safe, secure and private, fair and equitable, transparent and explainable, and accountable and monitored. This aims to manage risks and build trust. Conclusions: Combining these AI principles with high reliability safety principles can enable successful, trustworthy AI that improves health care quality, safety, efficiency, and access. Overcoming AI adoption barriers will require strategic efforts, partnerships, and investment to implement AI responsibly, safely, and equitably based on the health care context.

Applications of ChatGPT and Large Language Models in Medicine and Health Care: Benefits and Pitfalls.

Background: The use of large language models like ChatGPT is becoming increasingly popular in health care settings. These artificial intelligence models are trained on vast amounts of data and can be used for various tasks, such as language translation, summarization, and answering questions. Observations: Large language models have the potential to revolutionize the industry by assisting medical professionals with administrative tasks, improving diagnostic accuracy, and engaging patients. However, pitfalls exist, such as its inability to distinguish between real and fake information and the need to comply with privacy, security, and transparency principles. Conclusions: Careful consideration is needed to ensure the responsible and ethical use of large language models in medicine and health care. The development of [artificial intelligence] is as fundamental as the creation of the microprocessor, the personal computer, the Internet, and the mobile phone. It will change the way people work, learn, travel, get health care, and communicate with each other. Bill Gates1.

Revolutionizing Digital Pathology With the Power of Generative Artificial Intelligence and Foundation Models.

Digital pathology has transformed the traditional pathology practice of analyzing tissue under a microscope into a computer vision workflow. Whole-slide imaging allows pathologists to view and analyze microscopic images on a computer monitor, enabling computational pathology. By leveraging artificial intelligence (AI) and machine learning (ML), computational pathology has emerged as a promising field in recent years. Recently, task-specific AI/ML (eg, convolutional neural networks) has risen to the forefront, achieving above-human performance in many image-processing and computer vision tasks. The performance of task-specific AI/ML models depends on the availability of many annotated training datasets, which presents a rate-limiting factor for AI/ML development in pathology. Task-specific AI/ML models cannot benefit from multimodal data and lack generalization, eg, the AI models often struggle to generalize to new datasets or unseen variations in image acquisition, staining techniques, or tissue types. The 2020s are witnessing the rise of foundation models and generative AI. A foundation model is a large AI model trained using sizable data, which is later adapted (or fine-tuned) to perform different tasks using a modest amount of task-specific annotated data. These AI models provide in-context learning, can self-correct mistakes, and promptly adjust to user feedback. In this review, we provide a brief overview of recent advances in computational pathology enabled by task-specific AI, their challenges and limitations, and then introduce various foundation models. We propose to create a pathology-specific generative AI based on multimodal foundation models and present its potentially transformative role in digital pathology. We describe different use cases, delineating how it could serve as an expert companion of pathologists and help them efficiently and objectively perform routine laboratory tasks, including quantifying image analysis, generating pathology reports, diagnosis, and prognosis. We also outline the potential role that foundation models and generative AI can play in standardizing the pathology laboratory workflow, education, and training.

Predicting ward transfer mortality with machine learning.

In order to address a long standing challenge for internal medicine physicians we developed artificial intelligence (AI) models to identify patients at risk of increased mortality. After querying 2,425 records of patients transferred from non-intensive care units to intensive care units from the Veteran Affairs Corporate Data Warehouse (CDW), we created two datasets. The former used 22 independent variables that included "Length of Hospital Stay" and "Days to Intensive Care Transfer," and the latter lacked these two variables. Since these two variables are unknown at the time of admission, the second set is more clinically relevant. We trained 16 machine learning models using both datasets. The best-performing models were fine-tuned and evaluated. The LightGBM model achieved the best results for both datasets. The model trained with 22 variables achieved a Receiver Operating Characteristics Curve-Area Under the Curve (ROC-AUC) of 0.89 and an accuracy of 0.72, with a sensitivity of 0.97 and a specificity of 0.68. The model trained with 20 variables achieved a ROC-AUC of 0.86 and an accuracy of 0.71, with a sensitivity of 0.94 and a specificity of 0.67. The top features for the former model included "Total length of Stay," "Admit to ICU Transfer Days," and "Lymphocyte Next Lab Value." For the latter model, the top features included "Lymphocyte First Lab Value," "Hemoglobin First Lab Value," and "Hemoglobin Next Lab Value." Our clinically relevant predictive mortality model can assist providers in optimizing resource utilization when managing large caseloads, particularly during shift changes.

Artificial Intelligence for Clinical Flow Cytometry.

In this review, the authors discuss the fundamental principles of machine learning. They explore recent studies and approaches in implementing machine learning into flow cytometry workflows. These applications are promising but not without their shortcomings. Explainability may be the biggest barrier to adoption, as they contain "black boxes" in which a complex network of mathematical processes learns features of data that are not translatable into real language. The authors discuss the current limitations of machine learning models and the possibility that, without a multiinstitutional development process, these applications could have poor generalizability. They also discuss widespread deployment of augmented decision-making.

Pilot Lightweight Denoising Algorithm for Multiple Sclerosis on Spine MRI.

Multiple sclerosis (MS) is a severely debilitating disease which requires accurate and timely diagnosis. MRI is the primary diagnostic vehicle; however, it is susceptible to noise and artifact which can limit diagnostic accuracy. A myriad of denoising algorithms have been developed over the years for medical imaging yet the models continue to become more complex. We developed a lightweight algorithm which utilizes the image's inherent noise via dictionary learning to improve image quality without high computational complexity or pretraining through a process known as orthogonal matching pursuit (OMP). Our algorithm is compared to existing traditional denoising algorithms to evaluate performance on real noise that would commonly be encountered in a clinical setting. Fifty patients with a history of MS who received 1.5 T MRI of the spine between the years of 2018 and 2022 were retrospectively identified in accordance with local IRB policies. Native resolution 5 mm sagittal images were selected from T2 weighted sequences for evaluation using various denoising techniques including our proposed OMP denoising algorithm. Peak signal to noise ratio (PSNR) and structural similarity index (SSIM) were measured. While wavelet denoising demonstrated an expected higher PSNR than other models, its SSIM was variable and consistently underperformed its comparators (0.94 ± 0.10). Our pilot OMP denoising algorithm provided superior performance with greater consistency in terms of SSIM (0.99 ± 0.01) with similar PSNR to non-local means filtering (NLM), both of which were superior to other comparators (OMP 37.6 ± 2.2, NLM 38.0 ± 1.8). The superior performance of our OMP denoising algorithm in comparison to traditional models is promising for clinical utility. Given its individualized and lightweight approach, implementation into PACS may be more easily incorporated. It is our hope that this technology will provide improved diagnostic accuracy and workflow optimization for Neurologists and Radiologists, as well as improved patient outcomes.

Establishing a Hospital Artificial Intelligence Committee to Improve Patient Care.

Background: The use of artificial intelligence (AI) in health care is increasing and has shown utility in many medical specialties, especially pathology, radiology, and oncology. Observations: Many barriers exist to successfully implement AI programs in the clinical setting. To address these barriers, a formal governing body, the hospital AI Committee, was created at James A. Haley Veterans' Hospital in Tampa, Florida. The AI committee reviews and assesses AI products based on their success at protecting human autonomy; promoting human well-being and safety and the public interest; ensuring transparency, explainability, and intelligibility; fostering responsibility and accountability; ensuring inclusiveness and equity; and promoting AI that is responsive and sustainable. Conclusions: Through the hospital AI Committee, we may overcome many obstacles to successfully implementing AI applications in the clinical setting.

Neuroimaging in the Era of Artificial Intelligence: Current Applications.

Background: Artificial intelligence (AI) in medicine has shown significant promise, particularly in neuroimaging. AI increases efficiency and reduces errors, making it a valuable resource for physicians. With the increasing amount of data processing and image interpretation required, the ability to use AI to augment and aid the radiologist could improve the quality of patient care. Observations: AI can predict patient wait times, which may allow more efficient patient scheduling. Additionally, AI can save time for repeat magnetic resonance neuroimaging and reduce the time spent during imaging. AI has the ability to read computed tomography, magnetic resonance imaging, and positron emission tomography with reduced or without contrast without significant loss in sensitivity for detecting lesions. Neuroimaging does raise important ethical considerations and is subject to bias. It is vital that users understand the practical and ethical considerations of the technology. Conclusions: The demonstrated applications of AI in neuroimaging are numerous and varied, and it is reasonable to assume that its implementation will increase as the technology matures. AI's use for detecting neurologic conditions holds promise in combatting ever increasing imaging volumes and providing timely diagnoses.

Molecular Testing of Atypical Thyroid Nodules with Corresponding Surgical Correlation: Five-Year Retrospective Review in Veterans Population.

Objective We report the results of a retrospective five-year study within a veteran population aimed at correlating abnormal thyroid fine-needle aspiration (FNA) diagnosis with associated molecular testing to the histology of the surgical resection. Methods A retrospective analysis of abnormal thyroid FNAs with associated molecular testing and surgical outcome was conducted from January 1, 2015 to December 31, 2020. Aspirates were classified using the Bethesda system for reporting thyroid cytopathology, including atypia of undetermined significance/follicular lesion of undetermined significance (AUS/FLUS), follicular neoplasm/suspicious for follicular neoplasm (FN/SFN), suspicious for malignancy (SM), and malignant. Pertinent data, including patient demographics, imaging, and ancillary testing were reviewed. A thyroid cancer mutation panel assessing the most common mutations and rearrangements associated with neoplasia was utilized. The results of molecular testing were directly compared and correlated with final cytological and histological diagnosis. Results A total of 1850 thyroid aspirates were performed, 200 of which were given an abnormal cytologic diagnosis. Thirty-six samples were submitted for molecular testing and subsequent surgical follow-up. Four were called malignant on cytology. 32 were placed in an indeterminate category (89%). Within indeterminate cases: 53% exhibited positive molecular mutations (n=17), 34% no mutation detected (n=11), and 13% insufficient quantity for testing (n=4). Upon surgical resection in the mutation-positive group: 18% had no malignancy (n=3), and the remaining 82% were positive for malignancy (n=14). Mutations in the histologically malignant group included: 57% BRAF (n=8), 21% NRAS (n=3), 7% HRAS (n=1), 7% KRAS (n=1), and 7% PAX8/PPAR gamma (n=1). In indeterminate cases with no mutation detected, 10 cases were found to be benign, and one case of malignancy was diagnosed. The probability of indeterminate diagnosis in combination with no mutation yielded a 91% chance of benign entity and 9% chance of malignancy. We demonstrated 93% sensitivity and 91% negative predictive value (NPV) for the risk of malignancy in indeterminate cytology specimens with ancillary molecular testing. There was 77% specificity and 82% positive predictive value (PPV) for our data set. Conclusions In indeterminate samples, the detection of a mutation was highly predictive of malignancy and a strong indicating factor for surgery with a high sensitivity and NPV. Molecular testing refined or established the diagnosis in 89% of the cases. Our results indicate that molecular testing of thyroid nodules enhances the accuracy of FNA cytology and the subsequent surgical outcome.

Cervical Cancer Screening in South Florida Veteran Population, 2014 to 2020: Cytology and High-Risk Human Papillomavirus Correlation and Epidemiology.

Objective This project aims to use our robust women's health patient data to analyze the correlation between cytology and high-risk human papillomavirus (Hr-HPV) testing, study the performance of Hr-HPV testing for detecting cytology lesions, and examine epidemiologic measures of human papillomavirus (HPV) infections in the women's veteran population. Methods We collected patient data from 2014 to 2020 from our computerized patient record system. We performed HPV assays using the cobas® 4800 system (Roche Diagnostics, Basel, Switzerland). The cobas HPV assay detects HPV 16, HPV 18, and 12 other HPV types (31, 33, 35, 39, 45, 51, 56, 58, 59, 66, and 68). We organized cytology results and Hr-HPV assays with Microsoft Access and Microsoft Excel (Microsoft Corporation, Washington, USA) for analysis. Results A total of 9437 cervical specimens were co-tested. High-grade cytology lesions - high-grade intraepithelial lesion (HSIL) or higher and atypical squamous cells, cannot exclude HSIL (ASC-H) - were overwhelmingly positive for Hr-HPV (94.1% and 87.2%, respectively). Low-grade cytology lesions - low-grade squamous intraepithelial lesion ((LSIL) and atypical squamous cells of undetermined significance (ASC-US) - were positive for Hr-HPV in lower percentages (72.6% and 54.9%, respectively). Hr-HPV testing had a sensitivity of 91.3%, a specificity of 93.1%, a positive predictive value of 16.4%, and a negative predictive value of 99.8% for detecting high-grade cytology lesions. Hr-HPV testing had a lower performance for detecting low-grade cytology lesions. Ten cases had high-grade cytology and negative Hr-HPV test. Out of 10 such patients, nine showed no dysplasia (six) or low-grade dysplasia (three) on subsequent biopsy. Overall, 14.4% of tests were positive for Hr-HPV. The highest positive Hr-HPV test rates were in the third and eighth decades of life, 25.1% and 22.0%, respectively. However, the eighth decade consisted of a small sample of only 50 women. In women over 30 years of age with Hr-HPV infections, HPV types 16 and 18 were present in 11.7% and 6.4% of tests, respectively. Other HPV types were present in 82.3% of tests. Conclusions Hr-HPV testing has a high performance in detecting high-grade cytology lesions and a lower performance for detecting low-grade cytology lesions. However, studies show that LSIL rarely progresses to cervical intraepithelial neoplasia grade 3 or higher (CIN3+), suggesting minimal to no impact on cervical cancer screening. We believe our findings are in accordance with recent studies and affirm the guidelines that recommend primary Hr-HPV testing as the preferred screening method. The percentage of positive Hr-HPV tests and rates for age and HPV types 16 and 18 in our women's veteran population suggest similar HPV prevalence to that of the general US population.

Artificial Intelligence: Review of Current and Future Applications in Medicine.

BACKGROUND: The role of artificial intelligence (AI) in health care is expanding rapidly. Currently, there are at least 29 US Food and Drug Administration-approved AI health care devices that apply to numerous medical specialties and many more are in development. OBSERVATIONS: With increasing expectations for all health care sectors to deliver timely, fiscally-responsible, high-quality health care, AI has potential utility in numerous areas, such as image analysis, improved workflow and efficiency, public health, and epidemiology, to aid in processing large volumes of patient and medical data. In this review, we describe basic terminology, principles, and general AI applications relating to health care. We then discuss current and future applications for a variety of medical specialties. Finally, we discuss the future potential of AI along with the potential risks and limitations of current AI technology. CONCLUSIONS: AI can improve diagnostic accuracy, increase patient safety, assist with patient triage, monitor disease progression, and assist with treatment decisions.

Using Artificial Intelligence for COVID-19 Chest X-ray Diagnosis.

BACKGROUND: Coronavirus disease-19 (COVID-19), caused by a novel member of the coronavirus family, is a respiratory disease that rapidly reached pandemic proportions with high morbidity and mortality. In only a few months, it has had a dramatic impact on society and world economies. COVID-19 has presented numerous challenges to all aspects of health care, including reliable methods for diagnosis, treatment, and prevention. Initial efforts to contain the spread of the virus were hampered by the time required to develop reliable diagnostic methods. Artificial intelligence (AI) is a rapidly growing field of computer science with many applications for health care. Machine learning is a subset of AI that uses deep learning with neural network algorithms. It can recognize patterns and achieve complex computational tasks often far quicker and with increased precision than can humans. METHODS: In this article, we explore the potential for the simple and widely available chest X-ray (CXR) to be used with AI to diagnose COVID-19 reliably. Microsoft CustomVision is an automated image classification and object detection system that is a part of Microsoft Azure Cognitive Services. We utilized publicly available CXR images for patients with COVID-19 pneumonia, pneumonia from other etiologies, and normal CXRs as a dataset to train Microsoft CustomVision. RESULTS: Our trained model overall demonstrated 92.9% sensitivity (recall) and positive predictive value (precision), with results for each label showing sensitivity and positive predictive value at 94.8% and 98.9% for COVID-19 pneumonia, 89% and 91.8% for non-COVID-19 pneumonia, 95% and 88.8% for normal lung. We then validated the program using CXRs of patients from our institution with confirmed COVID-19 diagnoses along with non-COVID-19 pneumonia and normal CXRs. Our model performed with 100% sensitivity, 95% specificity, 97% accuracy, 91% positive predictive value, and 100% negative predictive value. CONCLUSIONS: We have used a readily available, commercial platform to demonstrate the potential of AI to assist in the successful diagnosis of COVID-19 pneumonia on CXR images. The findings have implications for screening and triage, initial diagnosis, monitoring disease progression, and identifying patients at increased risk of morbidity and mortality. Based on the data, a website was created to demonstrate how such technologies could be shared and distributed to others to combat entities such as COVID-19 moving forward.

Comparing Artificial Intelligence Platforms for Histopathologic Cancer Diagnosis.

Two machine learning platforms were successfully used to provide diagnostic guidance in the differentiation between common cancer conditions in veteran populations.

IL-1 Receptor-Knockout Mice Develop Epidermal Cysts and Show an Altered Innate Immune Response after Exposure to UVB Radiation.

In this study, we observed that mice lacking the IL-1 receptor (IL-1R) (IL1r-/-) or deficient in IL1-ß developed multiple epidermal cysts after chronic UVB exposure. Cysts that developed in IL1r-/- mice were characterized by the presence of the hair follicle marker Sox 9, keratins 10 and 14, and normal melanocyte distribution and retinoid X receptor-α expression. The increased incidence of cysts in IL1r-/- mice was associated with less skin inflammation as characterized by decreased recruitment of macrophages, and their skin also maintained epidermal barrier function compared with wild-type mice. Transcriptional analysis of the skin of IL1r-/- mice after UVB exposure showed decreased gene expression of proinflammatory cytokines such as tumor necrosis factor-α and IL-6. In vitro, primary keratinocytes derived from IL1r-/- mice were more resistant to UVB-triggered cell death compared with wild-type cells, and tumor necrosis factor-α release was completely blocked in the absence of IL-1R. These observations illustrate an unexpected yet prominent phenotype associated with the lack of IL-1R signaling in mice and support further investigation into the role of IL-1 ligands in epidermal repair and innate immune response after damaging UVB exposure.

Non-coding Double-stranded RNA and Antimicrobial Peptide LL-37 Induce Growth Factor Expression from Keratinocytes and Endothelial Cells.

A critical function for skin is that when damaged it must simultaneously identify the nature of the injury, repair barrier function, and limit the intrusion of pathogenic organisms. These needs are carried out through the detection of damage-associated molecular patterns (DAMPs) and a response that includes secretion of cytokines, chemokines, growth factors, and antimicrobial peptides (AMPs). In this study, we analyzed how non-coding double-stranded RNA (dsRNAs) act as a DAMP in the skin and how the human cathelicidin AMP LL-37 might influence growth factor production in response to this DAMP. dsRNA alone significantly increased the expression of multiple growth factors in keratinocytes, endothelial cells, and fibroblasts. Furthermore, RNA sequencing transcriptome analysis found that multiple growth factors increase when cells are exposed to both LL-37 and dsRNA, a condition that mimics normal wounding. Quantitative PCR and/or ELISA validated that growth factors expressed by keratinocytes in these conditions included, but were not limited to, basic fibroblast growth factor (FGF2), heparin-binding EGF-like growth factor (HBEGF), vascular endothelial growth factor C (VEGFC), betacellulin (BTC), EGF, epiregulin (EREG), and other members of the transforming growth factor ß superfamily. These results identify a novel role for DAMPs and AMPs in the stimulation of repair and highlight the complex interactions involved in the wound environment.

Molecular cartography of the human skin surface in 3D.

The human skin is an organ with a surface area of 1.5-2 m(2) that provides our interface with the environment. The molecular composition of this organ is derived from host cells, microbiota, and external molecules. The chemical makeup of the skin surface is largely undefined. Here we advance the technologies needed to explore the topographical distribution of skin molecules, using 3D mapping of mass spectrometry data and microbial 16S rRNA amplicon sequences. Our 3D maps reveal that the molecular composition of skin has diverse distributions and that the composition is defined not only by skin cells and microbes but also by our daily routines, including the application of hygiene products. The technological development of these maps lays a foundation for studying the spatial relationships of human skin with hygiene, the microbiota, and environment, with potential for developing predictive models of skin phenotypes tailored to individual health.

Toll-like receptor 3 activation is required for normal skin barrier repair following UV damage.

UV damage to the skin leads to the release of noncoding RNA (ncRNA) from necrotic keratinocytes that activates Toll-like receptor 3 (TLR3). This release of ncRNA triggers inflammation in the skin following UV damage. Recently, TLR3 activation was also shown to aid wound repair and increase the expression of genes associated with permeability barrier repair. Here, we sought to test whether skin barrier repair after UVB damage is dependent on the activation of TLR3. We observed that multiple ncRNAs induced expression of skin barrier repair genes, that the TLR3 ligand Poly (I:C) also induced expression and function of tight junctions, and that the ncRNA U1 acts in a TLR3-dependent manner to induce expression of skin barrier repair genes. These observations were shown to have functional relevance as Tlr3-/- mice displayed a delay in skin barrier repair following UVB damage. Combined, these data further validate the conclusion that recognition of endogenous RNA by TLR3 is an important step in the program of skin barrier repair.

Warfarin pharmacogenomics: recommendations with available patented clinical technologies.

Warfarin pharmacogenomic testing has become a prime example of the utility of personalized molecular testing in the modern clinical laboratory. Warfarin is a commonly used drug for the prevention and treatment of thromboembolic complications in a variety of clinical situations. However, a number of factors lead to a high interindividual variability in dose requirements. Among the primary factors in this variability are genetic polymorphisms in general patient populations, which can account for 35-50% of varying dose requirements among patients. In this review, we discuss the implications of polymorphisms in the cytochrome P-450 enzyme 2C9 (CYP2C9) and Vitamin K Epoxide Reductase Enzyme Complex subunit 1 (VKORC1) as they relate to therapeutic warfarin dosing. We discuss the clinical utility of pharmacogenomics testing as related to warfarin dosing, and propose a clinical model for the implementation of the pharmacogenomic test results. Finally, we provide a brief overview of the currently available commercial testing platforms with discussion of the complexities of utilizing patented methodologies in bringing genetic testing such as this to the clinical laboratory.

UVB radiation illuminates the role of TLR3 in the epidermis.

UV radiation poses a significant risk to human health. The mechanisms that help repair UV-damaged cells have recently been more clearly defined with the observation that Toll-like receptor 3 can sense self RNA released from necrotic keratinocytes following UV damage. TLR3 activation in the skin induces inflammation and increases the expression of genes involved in skin barrier repair. Activation of TLR2 in the skin by commensal microbial products prevents excessive inflammation by blocking downstream TLR3 signaling. This review highlights how UV damage-induced inflammation in the skin is propagated by host products and regulated by host inhabitants.

Translation of the prion protein mRNA is robust in astrocytes but does not amplify during reactive astrocytosis in the mouse brain.

Prion diseases induce neurodegeneration in specific brain areas for undetermined reasons. A thorough understanding of the localization of the disease-causing molecule, the prion protein (PrP), could inform on this issue but previous studies have generated conflicting conclusions. One of the more intriguing disagreements is whether PrP is synthesized by astrocytes. We developed a knock-in reporter mouse line in which the coding sequence of the PrP expressing gene (Prnp), was replaced with that for green fluorescent protein (GFP). Native GFP fluorescence intensity varied between and within brain regions. GFP was present in astrocytes but did not increase during reactive gliosis induced by scrapie prion infection. Therefore, reactive gliosis associated with prion diseases does not cause an acceleration of local PrP production. In addition to aiding in Prnp gene activity studies, this reporter mouse line will likely prove useful for analysis of chimeric animals produced by stem cell and tissue transplantation experiments.