Almanac - Retrieval-Augmented Language Models for Clinical Medicine.

BACKGROUND: Large language models (LLMs) have recently shown impressive zero-shot capabilities, whereby they can use auxiliary data, without the availability of task-specific training examples, to complete a variety of natural language tasks, such as summarization, dialogue generation, and question answering. However, despite many promising applications of LLMs in clinical medicine, adoption of these models has been limited by their tendency to generate incorrect and sometimes even harmful statements. METHODS: We tasked a panel of eight board-certified clinicians and two health care practitioners with evaluating Almanac, an LLM framework augmented with retrieval capabilities from curated medical resources for medical guideline and treatment recommendations. The panel compared responses from Almanac and standard LLMs (ChatGPT-4, Bing, and Bard) versus a novel data set of 314 clinical questions spanning nine medical specialties. RESULTS: Almanac showed a significant improvement in performance compared with the standard LLMs across axes of factuality, completeness, user preference, and adversarial safety. CONCLUSIONS: Our results show the potential for LLMs with access to domain-specific corpora to be effective in clinical decision-making. The findings also underscore the importance of carefully testing LLMs before deployment to mitigate their shortcomings. (Funded by the National Institutes of Health, National Heart, Lung, and Blood Institute.).

Machine Learning Multicenter Risk Model to Predict Right Ventricular Failure After Mechanical Circulatory Support: The STOP-RVF Score.

Importance: The existing models predicting right ventricular failure (RVF) after durable left ventricular assist device (LVAD) support might be limited, partly due to lack of external validation, marginal predictive power, and absence of intraoperative characteristics. Objective: To derive and validate a risk model to predict RVF after LVAD implantation. Design, Setting, and Participants: This was a hybrid prospective-retrospective multicenter cohort study conducted from April 2008 to July 2019 of patients with advanced heart failure (HF) requiring continuous-flow LVAD. The derivation cohort included patients enrolled at 5 institutions. The external validation cohort included patients enrolled at a sixth institution within the same period. Study data were analyzed October 2022 to August 2023. Exposures: Study participants underwent chronic continuous-flow LVAD support. Main Outcome and Measures: The primary outcome was RVF incidence, defined as the need for RV assist device or intravenous inotropes for greater than 14 days. Bootstrap imputation and adaptive least absolute shrinkage and selection operator variable selection techniques were used to derive a predictive model. An RVF risk calculator (STOP-RVF) was then developed and subsequently externally validated, which can provide personalized quantification of the risk for LVAD candidates. Its predictive accuracy was compared with previously published RVF scores. Results: The derivation cohort included 798 patients (mean [SE] age, 56.1 [13.2] years; 668 male [83.7%]). The external validation cohort included 327 patients. RVF developed in 193 of 798 patients (24.2%) in the derivation cohort and 107 of 327 patients (32.7%) in the validation cohort. Preimplant variables associated with postoperative RVF included nonischemic cardiomyopathy, intra-aortic balloon pump, microaxial percutaneous left ventricular assist device/venoarterial extracorporeal membrane oxygenation, LVAD configuration, Interagency Registry for Mechanically Assisted Circulatory Support profiles 1 to 2, right atrial/pulmonary capillary wedge pressure ratio, use of angiotensin-converting enzyme inhibitors, platelet count, and serum sodium, albumin, and creatinine levels. Inclusion of intraoperative characteristics did not improve model performance. The calculator achieved a C statistic of 0.75 (95% CI, 0.71-0.79) in the derivation cohort and 0.73 (95% CI, 0.67-0.80) in the validation cohort. Cumulative survival was higher in patients composing the low-risk group (estimated <20% RVF risk) compared with those in the higher-risk groups. The STOP-RVF risk calculator exhibited a significantly better performance than commonly used risk scores proposed by Kormos et al (C statistic, 0.58; 95% CI, 0.53-0.63) and Drakos et al (C statistic, 0.62; 95% CI, 0.57-0.67). Conclusions and Relevance: Implementing routine clinical data, this multicenter cohort study derived and validated the STOP-RVF calculator as a personalized risk assessment tool for the prediction of RVF and RVF-associated all-cause mortality.

A value-based approach to optimize red blood cell transfusion in patients receiving extracorporeal membrane oxygenation.

INTRODUCTION: The risk, cost, and adverse outcomes associated with packed red blood cell (RBC) transfusions in patients with cardiopulmonary failure requiring extracorporeal membrane oxygenation (ECMO) have raised concerns regarding the overutilization of RBC products. It is, therefore, necessary to establish optimal transfusion criteria and protocols for patients supported with ECMO. The goal of this study was to establish specific criteria for RBC transfusions in patients undergoing ECMO. METHODS: This was a retrospective cohort study conducted at Stanford University Hospital. Data on RBC utilization during the entire hospital stay were obtained, which included patients aged ≥18 years who received ECMO support between 1 January 2017, and 30 June 2020 (n = 281). The primary outcome was in-hospital mortality. RESULTS: Hemoglobin (HGB) levels >10 g/dL before transfusion did not improve in-hospital survival. Therefore, we revised the HGB threshold to ≤10 g/dL to guide transfusion in patients undergoing ECMO. To validate this intervention, we prospectively compared the pre- and post-intervention cohorts for in-hospital mortality. Post-intervention analyses found 100% compliance for all eligible records and a decrease in the requirement for RBC transfusion by 1.2 units per patient without affecting the mortality. CONCLUSIONS: As an institution-driven value-based approach to guide transfusion in patients undergoing ECMO, we lowered the threshold HGB level. Validation of this revised intervention demonstrated excellent compliance and reduced the need for RBC transfusion while maintaining the clinical outcome. Our findings can help reform value-based healthcare in this cohort while maintaining the outcome.

A flexible electronic strain sensor for the real-time monitoring of tumor regression.

Assessing the efficacy of cancer therapeutics in mouse models is a critical step in treatment development. However, low-resolution measurement tools and small sample sizes make determining drug efficacy in vivo a difficult and time-intensive task. Here, we present a commercially scalable wearable electronic strain sensor that automates the in vivo testing of cancer therapeutics by continuously monitoring the micrometer-scale progression or regression of subcutaneously implanted tumors at the minute time scale. In two in vivo cancer mouse models, our sensor discerned differences in tumor volume dynamics between drug- and vehicle-treated tumors within 5 hours following therapy initiation. These short-term regression measurements were validated through histology, and caliper and bioluminescence measurements taken over weeklong treatment periods demonstrated the correlation with longer-term treatment response. We anticipate that real-time tumor regression datasets could help expedite and automate the process of screening cancer therapies in vivo.

Human Coronary Plaque T Cells Are Clonal and Cross-React to Virus and Self.

BACKGROUND: Coronary artery disease is an incurable, life-threatening disease that was once considered primarily a disorder of lipid deposition. Coronary artery disease is now also characterized by chronic inflammation' notable for the buildup of atherosclerotic plaques containing immune cells in various states of activation and differentiation. Understanding how these immune cells contribute to disease progression may lead to the development of novel therapeutic strategies. METHODS: We used single-cell technology and in vitro assays to interrogate the immune microenvironment of human coronary atherosclerotic plaque at different stages of maturity. RESULTS: In addition to macrophages, we found a high proportion of αß T cells in the coronary plaques. Most of these T cells lack high expression of CCR7 and L-selectin, indicating that they are primarily antigen-experienced memory cells. Notably, nearly one-third of these cells express the HLA-DRA surface marker, signifying activation through their TCRs (T-cell receptors). Consistent with this, TCR repertoire analysis confirmed the presence of activated αß T cells (CD4

Computational Fluid Dynamics Simulations to Predict False Lumen Enlargement After Surgical Repair of Type-A Aortic Dissection.

We aim to use computational fluid dynamics to investigate the hemodynamic conditions that may predispose to false lumen enlargement in this patient population. Nine patients who received surgical repairs of their type-A aortic dissections between 2017-2018 were retrospectively identified. Multiple contrast-enhanced post-operative CT scans were used to construct 3D models of aortic geometries. Computational fluid dynamics simulations of the models were run on a high-performance computing cluster using SimVascular - an open-source simulation package. Physiological pulsatile flow conditions (4.9 L/min) were used at the aortic true lumen inlet, and physiological vascular resistances were applied at the distal vascular ends. Exploratory analyses showed no correlation between rate of false lumen growth and blood pressure, immediate post-op aortic diameter, or the number of fenestrations (p = 0.2). 1-year post-operative CT scans showed a median false lumen growth rate of 4.31 (3.66, 14.67) mm/year Median (Interquartile range) peak systolic, mid-diastolic, and late diastolic velocity magnitudes were 0.90 (1.40); 0.10 (0.16); and 0.06 (0.06) cm/s respectively. Spearman's ranked correlations between fenestration velocity and 1-year false lumen growth rates were found to be statistically significant: Velocity magnitude at peak systolic (p = 0.025; rho = 0.75), mid diastolic (p = 0.025; rho = 0.75) and late diastolic phases of the cardiac cycle (p = 0.006; rho = 0.85). We have shown that false lumen growth is strongly correlated to fenestration flow velocity, which has potential implications for post-operative surveillance and risk stratification.

Altered Cardiac Energetics and Mitochondrial Dysfunction in Hypertrophic Cardiomyopathy.

BACKGROUND: Hypertrophic cardiomyopathy (HCM) is a complex disease partly explained by the effects of individual gene variants on sarcomeric protein biomechanics. At the cellular level, HCM mutations most commonly enhance force production, leading to higher energy demands. Despite significant advances in elucidating sarcomeric structure-function relationships, there is still much to be learned about the mechanisms that link altered cardiac energetics to HCM phenotypes. In this work, we test the hypothesis that changes in cardiac energetics represent a common pathophysiologic pathway in HCM. METHODS: We performed a comprehensive multiomics profile of the molecular (transcripts, metabolites, and complex lipids), ultrastructural, and functional components of HCM energetics using myocardial samples from 27 HCM patients and 13 normal controls (donor hearts). RESULTS: Integrated omics analysis revealed alterations in a wide array of biochemical pathways with major dysregulation in fatty acid metabolism, reduction of acylcarnitines, and accumulation of free fatty acids. HCM hearts showed evidence of global energetic decompensation manifested by a decrease in high energy phosphate metabolites (ATP, ADP, and phosphocreatine) and a reduction in mitochondrial genes involved in creatine kinase and ATP synthesis. Accompanying these metabolic derangements, electron microscopy showed an increased fraction of severely damaged mitochondria with reduced cristae density, coinciding with reduced citrate synthase activity and mitochondrial oxidative respiration. These mitochondrial abnormalities were associated with elevated reactive oxygen species and reduced antioxidant defenses. However, despite significant mitochondrial injury, HCM hearts failed to upregulate mitophagic clearance. CONCLUSIONS: Overall, our findings suggest that perturbed metabolic signaling and mitochondrial dysfunction are common pathogenic mechanisms in patients with HCM. These results highlight potential new drug targets for attenuation of the clinical disease through improving metabolic function and reducing mitochondrial injury.

Predicting post-operative right ventricular failure using video-based deep learning.

Despite progressive improvements over the decades, the rich temporally resolved data in an echocardiogram remain underutilized. Human assessments reduce the complex patterns of cardiac wall motion, to a small list of measurements of heart function. All modern echocardiography artificial intelligence (AI) systems are similarly limited by design - automating measurements of the same reductionist metrics rather than utilizing the embedded wealth of data. This underutilization is most evident where clinical decision making is guided by subjective assessments of disease acuity. Predicting the likelihood of developing post-operative right ventricular failure (RV failure) in the setting of mechanical circulatory support is one such example. Here we describe a video AI system trained to predict post-operative RV failure using the full spatiotemporal density of information in pre-operative echocardiography. We achieve an AUC of 0.729, and show that this ML system significantly outperforms a team of human experts at the same task on independent evaluation.

Long-term survival in patients with post-LVAD right ventricular failure: multi-state modelling with competing outcomes of heart transplant.

BACKGROUND: Multicenter data on long term survival following LVAD implantation that make use of contemporary definitions of RV failure are limited. Furthermore, traditional survival analyses censor patients who receive a bridge to heart transplant. Here we compare the outcomes of LVAD patients who develop post-operative RV failure accounting for the transitional probability of receiving an interim heart transplantation. METHODS: We use a retrospective cohort of LVAD patients sourced from multiple high-volume centers based in the United States. Five- and ten-year survival accounting for transition probabilities of receiving a heart transplant were calculated using a multi-state Aalen Johansen survival model. RESULTS: Of the 897 patients included in the study, 238 (26.5%) developed post-operative RV failure at index hospitalization. At 10 years the probability of death with post-op RV failure was 79.28% vs 61.70% in patients without (HR 2.10; 95% CI 1.72 - 2.57; p = < .001). Though not significant, patients with RV failure were less likely to be bridged to a heart transplant (HR 0.87, p = .4). Once transplanted the risk of death between both patient groups remained equivalent; the probability of death after a heart transplant was 3.97% in those with post-operative RV failure shortly after index LVAD implant, as compared to 14.71% in those without. CONCLUSIONS AND RELEVANCE: Long-term durable mechanical circulatory support is associated with significantly higher mortality in patients who develop post-operative RV failure. Improving outcomes may necessitate expeditious bridge to heart transplant wherever appropriate, along with critical reassessment of organ allocation policies.

Patient-Specific Computational Fluid Dynamics Reveal Localized Flow Patterns Predictive of Post-Left Ventricular Assist Device Aortic Incompetence.

BACKGROUND: Progressive aortic valve disease has remained a persistent cause of concern in patients with left ventricular assist devices. Aortic incompetence (AI) is a known predictor of both mortality and readmissions in this patient population and remains a challenging clinical problem. METHODS: Ten left ventricular assist device patients with de novo aortic regurgitation and 19 control left ventricular assist device patients were identified. Three-dimensional models of patients' aortas were created from their computed tomography scans, following which large-scale patient-specific computational fluid dynamics simulations were performed with physiologically accurate boundary conditions using the SimVascular flow solver. RESULTS: The spatial distributions of time-averaged wall shear stress and oscillatory shear index show no significant differences in the aortic root in patients with and without AI (mean difference, 0.67 dyne/cm2 [95% CI, -0.51 to 1.85]; P=0.23). Oscillatory shear index was also not significantly different between both groups of patients (mean difference, 0.03 [95% CI, -0.07 to 0.019]; P=0.22). The localized wall shear stress on the leaflet tips was significantly higher in the AI group than the non-AI group (1.62 versus 1.35 dyne/cm2; mean difference [95% CI, 0.15-0.39]; P<0.001), whereas oscillatory shear index was not significantly different between both groups (95% CI, -0.009 to 0.001; P=0.17). CONCLUSIONS: Computational fluid dynamics serves a unique role in studying the hemodynamic features in left ventricular assist device patients where 4-dimensional magnetic resonance imaging remains unfeasible. Contrary to the widely accepted notions of highly disturbed flow, in this study, we demonstrate that the aortic root is a region of relatively stagnant flow. We further identified localized hemodynamic features in the aortic root that challenge our understanding of how AI develops in this patient population.

Use of patient-specific computational models for optimization of aortic insufficiency after implantation of left ventricular assist device.

OBJECTIVE: Aortic incompetence (AI) is observed to be accelerated in the continuous-flow left ventricular assist device (LVAD) population and is related to increased mortality. Using computational fluid dynamics (CFD), we investigated the hemodynamic conditions related to the orientation of the LVAD outflow in these patients. METHOD: We identified 10 patients with new aortic regurgitation, and 20 who did not, after LVAD implantation between 2009 and 2018. Three-dimensional models of patients' aortas were created from their computed tomography scans. The geometry of the LVAD outflow graft in relation to the aorta was quantified using azimuth angles (AA), polar angles (PAs), and distance from aortic root. The models were used to run CFD simulations, which calculated the pressures and wall shear stress (rWSS) exerted on the aortic root. RESULTS: The AA and PA were found to be similar. However, for combinations of high values of AA and low values of PA, there were no patients with AI. The distance from aortic root to the outflow graft was also smaller in patients who developed AI (3.39 ± 0.7 vs 4.07 ± 0.77 cm, P = .04). There was no significant difference in aortic root pressures in the 2 groups. The rWSS was greater in AI patients (4.60 ± 5.70 vs 2.37 ± 1.20 dyne/cm2, P < .001). Qualitatively, we observed a trend of greater perturbations, regions of high rWSS, and flow eddies in the AI group. CONCLUSIONS: Using CFD simulations, we demonstrated that patients who developed de novo AI have greater rWSS at the aortic root, and their outflow grafts were placed closer to the aortic roots than those patients without de novo AI.

A Planarian Motility Assay to Gauge the Biomodulating Properties of Natural Products.

A straightforward, controllable means of using the non-parasitic planarian, Dugesia tigrina, a free-living aquatic flatworm, to study the stimulant and withdrawal properties of natural products is described. Experimental assays benefitting from unique aspects of planarian physiology have been applied to studies on wound healing, regeneration, and tumorigenesis. In addition, because planarians exhibit sensitivity to a variety of environmental stimuli and are capable of learning and developing conditioned responses, they can be used in behavioral studies examining learning and memory. Planarians possess a basic bilateral symmetry and a central nervous system that uses neurotransmitter systems amenable to studies examining the effects of neuromuscular biomodulators. Consequently, experimental systems monitoring planarian movement and motility have been developed to examine substance addiction and withdrawal. Because planarian motility offers the potential for a sensitive, easily standardized motility assay system to monitor the effect of stimuli, the planarian locomotor velocity (pLmV) test was adapted to monitor both stimulation and withdrawal behaviors by planarians through the determination of the number of grid lines crossed by the animals with time. Here, the technique and its application are demonstrated and explained.

Atheroprotective roles of smooth muscle cell phenotypic modulation and the TCF21 disease gene as revealed by single-cell analysis.

In response to various stimuli, vascular smooth muscle cells (SMCs) can de-differentiate, proliferate and migrate in a process known as phenotypic modulation. However, the phenotype of modulated SMCs in vivo during atherosclerosis and the influence of this process on coronary artery disease (CAD) risk have not been clearly established. Using single-cell RNA sequencing, we comprehensively characterized the transcriptomic phenotype of modulated SMCs in vivo in atherosclerotic lesions of both mouse and human arteries and found that these cells transform into unique fibroblast-like cells, termed 'fibromyocytes', rather than into a classical macrophage phenotype. SMC-specific knockout of TCF21-a causal CAD gene-markedly inhibited SMC phenotypic modulation in mice, leading to the presence of fewer fibromyocytes within lesions as well as within the protective fibrous cap of the lesions. Moreover, TCF21 expression was strongly associated with SMC phenotypic modulation in diseased human coronary arteries, and higher levels of TCF21 expression were associated with decreased CAD risk in human CAD-relevant tissues. These results establish a protective role for both TCF21 and SMC phenotypic modulation in this disease.

Modeling conduit choice for valve-sparing aortic root replacement on biomechanics with a 3-dimensional-printed heart simulator.

OBJECTIVE: The optimal conduit for valve-sparing aortic root replacement is still debated, with several conduit variations available, ranging from straight tubular grafts to Valsalva grafts. Benefits of neosinus reconstruction include enhanced flow profiles and improved hemodynamics. Curiously, however, some clinical data suggest that straight grafts may have greater long-term durability. In this study, we hypothesized that straight tubular grafts may help maintain the native cylindrical position of the aortic valve commissures radially, resulting in preserved leaflet coaptation, reduced stresses, and potentially improved valve performance. METHODS: Using 3D printing, a left heart simulator with a valve-sparing root replacement model and a physiologic coronary circulation was constructed. Aortic valves were dissected from fresh porcine hearts and reimplanted into either straight tubular grafts (n = 6) or Valsalva grafts (n = 6). Conduits were mounted into the heart simulator and hemodynamic, echocardiographic, and high-speed videometric data were collected. RESULTS: Hemodynamic parameters and coronary blood flow were similar between straight and Valsalva grafts, although the former were associated with lower regurgitant fractions, less peak intercommissural radial separation, preserved leaflet coaptation, decreased leaflet velocities, and lower relative leaflet forces compared with Valsalva grafts. CONCLUSIONS: Valsalva grafts and straight grafts perform equally well in terms of gross hemodyanics and coronary blood flow. Interestingly, however, the biomechanics of these 2 conduits differ considerably, with straight grafts providing increased radial commissural stability and leaflet coaptation. Further investigation into how these parameters influence clinical outcomes is warranted.