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.).

The cardiac preparticipation sports evaluation.

PURPOSE OF REVIEW: Sudden cardiac death (SCD) is the leading cause of death in young athletes during sports participation. Preparticipation cardiovascular screening aims to identify those at an increased risk of SCD. This review aims to provide a background of SCD in young athletes, to discuss the various screening recommendations of major medical societies, and to review recent evidence and current practice. RECENT FINDINGS: Numerous studies have evaluated various preparticipation screening practices, particularly regarding the inclusion of ECG as part of an initial evaluation to identify conditions with an increased risk of SCD. Some analyses have shown ECG inclusion to provide increased screening sensitivity and specificity, though others have shown no benefit when compared with evaluation with history and physical examination alone. Furthermore, in countries for which more extensive screening protocols have been employed, postimplementation statistics have not shown a significant decrease in SCD. SUMMARY: SCD in young athletes primarily results from underlying cardiac disease. Various preparticipation screening recommendations exist globally, with the common goal of decreasing the rates of SCD by identifying youth at risk during sports participation. Current guidelines in the United States support universal preparticipation evaluation using history and physical examination, with cardiology referral if abnormalities are identified.

Almanac: Retrieval-Augmented Language Models for Clinical Medicine.

Large-language models have recently demonstrated impressive zero-shot capabilities in a variety of natural language tasks such as summarization, dialogue generation, and question-answering. Despite many promising applications in clinical medicine, adoption of these models in real-world settings has been largely limited by their tendency to generate incorrect and sometimes even toxic statements. In this study, we develop Almanac, a large language model framework augmented with retrieval capabilities for medical guideline and treatment recommendations. Performance on a novel dataset of clinical scenarios (n= 130) evaluated by a panel of 5 board-certified and resident physicians demonstrates significant increases in factuality (mean of 18% at p-value < 0.05) across all specialties, with improvements in completeness and safety. Our results demonstrate the potential for large language models to be effective tools in the clinical decision-making process, while also emphasizing the importance of careful testing and deployment to mitigate their shortcomings.

Epidemiology of Lower Respiratory Tract Infections and Community-Acquired Respiratory Viruses in Patients with Bronchiolitis Obliterans Syndrome after Hematopoietic Cell Transplantation: A Retrospective Cohort Study.

BACKGROUND: Bronchiolitis obliterans syndrome (BOS)-chronic graft-versus-host disease (cGVHD) affecting the lungs-is an uncommon complication of allogeneic hematopoietic cell transplant (HCT). The epidemiology and complications of lower respiratory tract infections (LRTIs) and community-acquired respiratory viruses (CARVs) in these patients are poorly understood. OBJECTIVES: We aim to characterize the epidemiology of LRTIs in patients with BOS complicating HCT. We also aim to explore the association of LRTIs and CARV detection on lung function in BOS patients. STUDY DESIGN: Adult patients with BOS at Stanford Health Care between January 2010 and December 2019 were included in this retrospective cohort study. LRTI diagnosis was based on combined clinical, microbiologic, and radiographic criteria, using consensus criteria where available. RESULTS: Fifty-five patients with BOS were included. BOS was diagnosed at a median of 19.2 (IQR 12.5-24.7) months after HCT, and patients were followed for a median of 29.3 (IQR 9.9-53.2) months from BOS diagnosis. Twenty-two (40%) patients died after BOS diagnosis; 17 patients died from complications of cGVHD (including respiratory failure and infection) and 5 died from relapsed disease. Thirty-four (61.8%) patients developed at least one LRTI. Viral LRTIs were most common, occurring in 29 (52.7%) patients, primarily due to rhinovirus. Bacterial LRTIs-excluding Nocardia and non-tuberculous mycobacteria (NTM)-were the second most common, occurring in 21 (38.2%) patients, mostly due to Pseudomonas aeruginosa. Fungal LRTIs, NTM, and nocardiosis occurred in 14 (25.5%), 10 (18.2%), and 4 (7.3%) patients, respectively. Median time to development of the first LRTI after BOS diagnosis was 15.3 (4.7-44.7) months. Twenty-six (76.5%) of the 34 patients who developed LRTIs had infections due to more than one type of organism-fungi, viruses, Nocardia, NTM, and other bacteria-over the observation period. Patients with at least one LRTI had significantly lower forced expiratory volume in one second percent predicted (FEV1%) (37% vs. 53%, p = 0.0096) and diffusing capacity of carbon monoxide (DLCO) (45.5% predicted vs. 69% predicted, p = 0.0001). Patients with at least one LRTI trended toward lower overall survival (OS) (p = 0.0899) and higher non-relapse mortality (NRM) (p = 0.2707). Patients with a CARV detected or LRTI diagnosed after BOS-compared to those without any CARV detected or LRTI diagnosed-were more likely to have a sustained drop in FEV1% from baseline of at least 10% (21 [61.8%] versus 7 [33.3%]) and a sustained drop in FEV1% of at least 30% (12 [36.4%] versus 2 [9.5%]). CONCLUSIONS: LRTIs are common in BOS and associated with lower FEV1%, lower DLCO, and a trend toward decreased OS and higher NRM. Patients with LRTIs or CARVs (even absent lower respiratory tract involvement) were more likely to have substantial declines in FEV1% over time than those without. The array of organisms-including P. aeruginosa, mold, Nocardia, NTM, and CARVs-seen in BOS reflects the unique pathophysiology of this form of cGVHD, involving both systemic immunodeficiency and structural lung disease. These patterns of LRTIs and their outcomes can be used to guide clinical decisions and inform future research.

COVID-19 in hospitalized lung and non-lung solid organ transplant recipients: A comparative analysis from a multicenter study.

Lung transplant recipients (LTR) with coronavirus disease 2019 (COVID-19) may have higher mortality than non-lung solid organ transplant recipients (SOTR), but direct comparisons are limited. Risk factors for mortality specifically in LTR have not been explored. We performed a multicenter cohort study of adult SOTR with COVID-19 to compare mortality by 28 days between hospitalized LTR and non-lung SOTR. Multivariable logistic regression models were used to assess comorbidity-adjusted mortality among LTR vs. non-lung SOTR and to determine risk factors for death in LTR. Of 1,616 SOTR with COVID-19, 1,081 (66%) were hospitalized including 120/159 (75%) LTR and 961/1457 (66%) non-lung SOTR (p = .02). Mortality was higher among LTR compared to non-lung SOTR (24% vs. 16%, respectively, p = .032), and lung transplant was independently associated with death after adjusting for age and comorbidities (aOR 1.7, 95% CI 1.0-2.6, p = .04). Among LTR, chronic lung allograft dysfunction (aOR 3.3, 95% CI 1.0-11.3, p = .05) was the only independent risk factor for mortality and age >65 years, heart failure and obesity were not independently associated with death. Among SOTR hospitalized for COVID-19, LTR had higher mortality than non-lung SOTR. In LTR, chronic allograft dysfunction was independently associated with mortality.

Donor-Derived Coccidioides immitis Endocarditis and Disseminated Infection in the Setting of Solid Organ Transplantation.

Background. Endocarditis is a rare manifestation of infection with Coccidioides. This is the first reported case of donor-derived Coccidioides endocarditis obtained from a heart transplant. Methods. We present a unique case of donor-derived Coccidioides immitis endocarditis and disseminated infection in a heart transplant patient. We also conducted a review of the literature to identify other cases of donor-derived coccidioidomycosis in solid organ transplant recipients and reviewed their clinical characteristics. Results. Fifteen prior cases of donor-derived coccidioidomycosis were identified. A majority of these cases were diagnosed by positive culture (83%). Mortality was high at 58%. Conclusions. Clinicians should maintain a high index of suspicion for disseminated coccidioidomycosis in patients who received transplants with organs from donors with a history of residing in endemic regions.

Estimating the impacts of conservation on ecosystem services and poverty by integrating modeling and evaluation.

Scholars have made great advances in modeling and mapping ecosystem services, and in assigning economic values to these services. This modeling and valuation scholarship is often disconnected from evidence about how actual conservation programs have affected ecosystem services, however. Without a stronger evidence base, decision makers find it difficult to use the insights from modeling and valuation to design effective policies and programs. To strengthen the evidence base, scholars have advanced our understanding of the causal pathways between conservation actions and environmental outcomes, but their studies measure impacts on imperfect proxies for ecosystem services (e.g., avoidance of deforestation). To be useful to decision makers, these impacts must be translated into changes in ecosystem services and values. To illustrate how this translation can be done, we estimated the impacts of protected areas in Brazil, Costa Rica, Indonesia, and Thailand on carbon storage in forests. We found that protected areas in these conservation hotspots have stored at least an additional 1,000 Mt of CO2 in forests and have delivered ecosystem services worth at least $5 billion. This aggregate impact masks important spatial heterogeneity, however. Moreover, the spatial variability of impacts on carbon storage is the not the same as the spatial variability of impacts on avoided deforestation. These findings lead us to describe a research program that extends our framework to study other ecosystem services, to uncover the mechanisms by which ecosystem protection benefits humans, and to tie cost-benefit analyses to conservation planning so that we can obtain the greatest return on scarce conservation funds.

Encephalitis caused by Chikungunya virus in a traveler from the Kingdom of Tonga.

Febrile travelers from countries with unique endemic pathogens pose a significant diagnostic challenge. In this report, we describe the case of a Tongan man presenting with fever, rash, and altered mental status. The diagnosis of Chikungunya encephalitis was made using a laboratory-developed real-time RT-PCR and serologic testing.

Salt marsh as a coastal filter for the oceans: changes in function with experimental increases in nitrogen loading and sea-level rise.

Coastal salt marshes are among Earth's most productive ecosystems and provide a number of ecosystem services, including interception of watershed-derived nitrogen (N) before it reaches nearshore oceans. Nitrogen pollution and climate change are two dominant drivers of global-change impacts on ecosystems, yet their interacting effects at the land-sea interface are poorly understood. We addressed how sea-level rise and anthropogenic N additions affect the salt marsh ecosystem process of nitrogen uptake using a field-based, manipulative experiment. We crossed simulated sea-level change and ammonium-nitrate (NH(4)NO(3))-addition treatments in a fully factorial design to examine their potentially interacting effects on emergent marsh plants in a central California estuary. We measured above- and belowground biomass and tissue nutrient concentrations seasonally and found that N-addition had a significant, positive effect on a) aboveground biomass, b) plant tissue N concentrations, c) N stock sequestered in plants, and d) shoot:root ratios in summer. Relative sea-level rise did not significantly affect biomass, with the exception of the most extreme sea-level-rise simulation, in which all plants died by the summer of the second year. Although there was a strong response to N-addition treatments, salt marsh responses varied by season. Our results suggest that in our site at Coyote Marsh, Elkhorn Slough, coastal salt marsh plants serve as a robust N trap and coastal filter; this function is not saturated by high background annual N inputs from upstream agriculture. However, if the marsh is drowned by rising seas, as in our most extreme sea-level rise treatment, marsh plants will no longer provide the ecosystem service of buffering the coastal ocean from eutrophication.

Policy strategies to address sustainability of Alaskan boreal forests in response to a directionally changing climate.

Human activities are altering many factors that determine the fundamental properties of ecological and social systems. Is sustainability a realistic goal in a world in which many key process controls are directionally changing? To address this issue, we integrate several disparate sources of theory to address sustainability in directionally changing social-ecological systems, apply this framework to climate-warming impacts in Interior Alaska, and describe a suite of policy strategies that emerge from these analyses. Climate warming in Interior Alaska has profoundly affected factors that influence landscape processes (climate regulation and disturbance spread) and natural hazards, but has only indirectly influenced ecosystem goods such as food, water, and wood that receive most management attention. Warming has reduced cultural services provided by ecosystems, leading to some of the few institutional responses that directly address the causes of climate warming, e.g., indigenous initiatives to the Arctic Council. Four broad policy strategies emerge: (i) enhancing human adaptability through learning and innovation in the context of changes occurring at multiple scales; (ii) increasing resilience by strengthening negative (stabilizing) feedbacks that buffer the system from change and increasing options for adaptation through biological, cultural, and economic diversity; (iii) reducing vulnerability by strengthening institutions that link the high-latitude impacts of climate warming to their low-latitude causes; and (iv) facilitating transformation to new, potentially more beneficial states by taking advantage of opportunities created by crisis. Each strategy provides societal benefits, and we suggest that all of them be pursued simultaneously.