Developing a novel causal inference algorithm for personalized biomedical causal graph learning using meta machine learning.

BACKGROUND: Modeling causality through graphs, referred to as causal graph learning, offers an appropriate description of the dynamics of causality. The majority of current machine learning models in clinical decision support systems only predict associations between variables, whereas causal graph learning models causality dynamics through graphs. However, building personalized causal graphs for each individual is challenging due to the limited amount of data available for each patient. METHOD: In this study, we present a new algorithmic framework using meta-learning for learning personalized causal graphs in biomedicine. Our framework extracts common patterns from multiple patient graphs and applies this information to develop individualized graphs. In multi-task causal graph learning, the proposed optimized initial guess of shared commonality enables the rapid adoption of knowledge to new tasks for efficient causal graph learning. RESULTS: Experiments on one real-world biomedical causal graph learning benchmark data and four synthetic benchmarks show that our algorithm outperformed the baseline methods. Our algorithm can better understand the underlying patterns in the data, leading to more accurate predictions of the causal graph. Specifically, we reduce the structural hamming distance by 50-75%, indicating an improvement in graph prediction accuracy. Additionally, the false discovery rate is decreased by 20-30%, demonstrating that our algorithm made fewer incorrect predictions compared to the baseline algorithms. CONCLUSION: To the best of our knowledge, this is the first study to demonstrate the effectiveness of meta-learning in personalized causal graph learning and cause inference modeling for biomedicine. In addition, the proposed algorithm can also be generalized to transnational research areas where integrated analysis is necessary for various distributions of datasets, including different clinical institutions.

Exploring the role of blood pressure in the black-white disparity in cardiovascular disease mortality: a causal mediation analysis.

BACKGROUND: Cardiovascular diseases (CVDs) are the leading cause of death in the USA, and high blood pressure is a major risk factor for CVD. Despite the overall declining rates of CVD mortality in the USA in recent years, marked disparities between racial and ethnic groups persist, with black adults having a higher mortality rate than white adults. We investigated the extent to which blood pressure mediated the black-white disparity in CVD mortality. METHODS: Data came from the Multi-Ethnic Study of Atherosclerosis, a diverse longitudinal cohort. We included 1325 black and 2256 white community-based adults aged 45-80 years free of clinical CVD at baseline and followed for 14 years. We used causal mediation analysis to estimate the effect of race on CVD mortality that was mediated through blood pressure. RESULTS: Black participants had a higher hazard of dying from CVD compared with white participants (adjusted hazard ratio (HR): 1.28 (95% CI 0.88, 1.88)), though estimates were imprecise. Systolic blood pressure mediated 27% (HR: 1.02, 95% CI 1.00, 1.06) and diastolic blood pressure mediated 55% (HR: 1.07, 95% CI 1.01, 1.10) of the racial disparities in CVD mortality between white and black participants. Mediation effects were present in men but not in women. CONCLUSIONS: We found that black-white differences in blood pressure partially explain the observed black-white disparity in CVD mortality, particularly among men. Our findings suggest that public health interventions targeting high blood pressure prevention and management could be important strategies for reducing racial disparities in CVD mortality.

Examination of community-level interventions addressing early childhood obesity in Los Angeles County.

Objective: To help inform decisions regarding the equitable implementation of obesity interventions, we examined whether interventions were equitably reaching the most vulnerable communities, identified communities that received fewer interventions than expected, and estimated the effect of 'dose' of interventions on obesity prevalence. Methods: We created a database to identify and characterize obesity-related interventions implemented in Los Angeles County from 2005 to 2015 linked to community-level sociodemographic and obesity prevalence data. We ran generalized linear models with a Gamma distribution and log link to determine if interventions were directed toward vulnerable communities and to identify communities that received fewer interventions than expected. We ran fixed-effects models to estimate the association between obesity prevalence and intervention strategy count among preschool-aged children enrolled in the Special Supplemental Nutrition Assistance Program for Women Infants and Children. Results: We found that interventions targeted vulnerable communities with high poverty rates and percentages of minority residents. The small cluster of communities that received fewer interventions than expected tended to have poor socioeconomic profiles. Communities which received more intervention strategies saw greater declines in obesity prevalence (ß = -0.023; 95 % CI: -0.031, -0.016). Conclusions: It is important to determine if interventions are equitably reaching vulnerable populations as resources to tackle childhood obesity become available. Evaluating the population impact of multiple interventions implemented simultaneously presents methodological challenges in measuring intervention dose and identifying cost-effective strategies. Addressing these challenges must be an important research priority as community-wide interventions involve multiple intervention strategies to reduce health disparities.

Social Isolation, Brain Food Cue Processing, Eating Behaviors, and Mental Health Symptoms.

Importance: Perceived social isolation is associated with negative health outcomes, including increased risk for altered eating behaviors, obesity, and psychological symptoms. However, the underlying neural mechanisms of these pathways are unknown. Objective: To investigate the association of perceived social isolation with brain reactivity to food cues, altered eating behaviors, obesity, and mental health symptoms. Design, Setting, and Participants: This cross-sectional, single-center study recruited healthy, premenopausal female participants from the Los Angeles, California, community from September 7, 2021, through February 27, 2023. Exposure: Participants underwent functional magnetic resonance imaging while performing a food cue viewing task. Main Outcomes and Measures: The main outcomes included brain reactivity to food cues, body composition, self-reported eating behaviors (food cravings, reward-based eating, food addiction, and maladaptive eating behaviors), and mental health symptoms (anxiety, depression, positive and negative affect, and psychological resilience). Results: The study included 93 participants (mean [SD] age, 25.38 [7.07] years). Participants with higher perceived social isolation reported higher fat mass percentage, lower diet quality, increased maladaptive eating behaviors (cravings, reward-based eating, uncontrolled eating, and food addiction), and poor mental health (anxiety, depression, and psychological resilience). In whole-brain comparisons, the higher social isolation group showed altered brain reactivity to food cues in regions of the default mode, executive control, and visual attention networks. Isolation-related neural changes in response to sweet foods correlated with various altered eating behaviors and psychological symptoms. These altered brain responses mediated the connection between social isolation and maladaptive eating behaviors (ß for indirect effect, 0.111; 95% CI, 0.013-0.210; P = .03), increased body fat composition (ß, -0.141; 95% CI, -0.260 to -0.021; P = .02), and diminished positive affect (ß, -0.089; 95% CI, -0.188 to 0.011; P = .09). Conclusions and Relevance: These findings suggest that social isolation is associated with altered neural reactivity to food cues within specific brain regions responsible for processing internal appetite-related states and compromised executive control and attentional bias and motivation toward external food cues. These neural responses toward specific foods were associated with an increased risk for higher body fat composition, worsened maladaptive eating behaviors, and compromised mental health. These findings underscore the need for holistic mind-body-directed interventions that may mitigate the adverse health consequences of social isolation.

Lessons Learned From TranslateCovid, a Multilingual Online Resource Hub for Asian American and Pacific Islander Communities and Beyond.

The COVID-19 pandemic exacerbated health disparities among immigrant communities. Delivering accurate information and addressing misinformation on protective measures and vaccination to linguistically disadvantaged groups was critical for mitigating the effects of the pandemic. One group that was especially vulnerable to miscommunication about COVID-19 was non-native English-speaking immigrants. To address these disparities, the Asian American Studies Center and the Fielding School of Public Health at the University of California, Los Angeles, partnered to create a multilingual resource hub, TranslateCovid.org, to disseminate credible and reliable information about COVID-19 safety measures, the science behind the vaccines, and vaccine safety. We identified >1300 verified resources in 60 languages from government, academic, and nonprofit organizations and reposted them on the TranslateCovid website. We also developed public service announcement videos on handwashing, use of face masks, and social distancing in 10 languages and a fact sheet for frequently asked questions in 20 languages. We used a participatory approach to develop strategies for disseminating these resources. We discuss lessons learned, including strategies for forming government, community, and academic partnerships to support the timely development and dissemination of information. We conclude with a discussion on the unique role of universities in promoting equitable access to public health resources among immigrant communities in times of crisis.

Downregulation of the kidney glucagon receptor, essential for renal function and systemic homeostasis, contributes to chronic kidney disease.

The glucagon receptor (GCGR) in the kidney is expressed in nephron tubules. In humans and animal models with chronic kidney disease, renal GCGR expression is reduced. However, the role of kidney GCGR in normal renal function and in disease development has not been addressed. Here, we examined its role by analyzing mice with constitutive or conditional kidney-specific loss of the Gcgr. Adult renal Gcgr knockout mice exhibit metabolic dysregulation and a functional impairment of the kidneys. These mice exhibit hyperaminoacidemia associated with reduced kidney glucose output, oxidative stress, enhanced inflammasome activity, and excess lipid accumulation in the kidney. Upon a lipid challenge, they display maladaptive responses with acute hypertriglyceridemia and chronic proinflammatory and profibrotic activation. In aged mice, kidney Gcgr ablation elicits widespread renal deposition of collagen and fibronectin, indicative of fibrosis. Taken together, our findings demonstrate an essential role of the renal GCGR in normal kidney metabolic and homeostatic functions. Importantly, mice deficient for kidney Gcgr recapitulate some of the key pathophysiological features of chronic kidney disease.

The role of breastfeeding support in racial/ethnic disparities in breastfeeding practices.

BACKGROUND: Persistent racial/ethnic disparities in breastfeeding practices in the United States are well documented but the underlying causes remain unclear. While racial/ethnic disparities are often intertwined with socioeconomic disparities in breastfeeding, studies suggest that lack of breastfeeding support from family, health care organizations and workplaces may contribute to racial/ethnic disparities in breastfeeding rates. No studies have investigated the extent to which racial/ethnic disparities in breastfeeding practices can be explained by breastfeeding support. METHODS: We used survey data from participants of a federal nutrition assistance program in Los Angeles County, the most populous county in the United States, to examine causal mechanisms underlying racial/ethnic disparities in breastfeeding in five groups: Spanish-speaking Latina, English-speaking Latina, Non-Hispanic White (NHW), Non-Hispanic Black (NHB) and Non-Hispanic Asian (NHA). Applying causal mediation analysis, this study estimated the proportion of racial/ethnic differences in breastfeeding ('any' breastfeeding, i.e., partial or exclusive) rates at 6 months that could be explained by differential access to breastfeeding support from family, birth hospitals and workplaces. RESULTS: NHB and English-speaking Latina mothers were less likely, and Spanish-speaking Latina mothers more likely to breastfeed through 6 months than NHW mothers. Lack of breastfeeding support from family, hospitals and workplaces accounted for approximately 68% of the difference in any breastfeeding rates at 6 months between NHW and NHB mothers and 36% of the difference between NHW and English-speaking Latina mothers. CONCLUSION: These findings highlight the importance of improving support from family, hospitals and workplaces for breastfeeding mothers to reduce racial/ethnic disparities in breastfeeding.

Leptin Reduction as a Required Component for Weight Loss.

Partial leptin reduction can induce significant weight loss, while weight loss contributes to partial leptin reduction. The cause-and-effect relationship between leptin reduction and weight loss remains to be further elucidated. Here, we show that FGF21 and the glucagon-like peptide 1 receptor (GLP-1R) agonist liraglutide rapidly induced a reduction in leptin. This leptin reduction contributed to the beneficial effects of GLP-1R agonism in metabolic health, as transgenically maintaining leptin levels during treatment partially curtailed the beneficial effects seen with these agonists. Moreover, a higher degree of leptin reduction during treatment, induced by including a leptin neutralizing antibody with either FGF21 or liraglutide, synergistically induced greater weight loss and better glucose tolerance in diet-induced obese mice. Furthermore, upon cessation of either liraglutide or FGF21 treatment, the expected immediate weight regain was observed, associated with a rapid increase in circulating leptin levels. Prevention of this leptin surge with leptin neutralizing antibodies slowed down weight gain and preserved better glucose tolerance. Mechanistically, a significant reduction in leptin induced a higher degree of leptin sensitivity in hypothalamic neurons. Our observations support a model that postulates that a reduction of leptin levels is a necessary prerequisite for substantial weight loss, and partial leptin reduction is a viable strategy to treat obesity and its associated insulin resistance.

Integrating Multi-Omics Data With EHR for Precision Medicine Using Advanced Artificial Intelligence.

With the recent advancement of novel biomedical technologies such as high-throughput sequencing and wearable devices, multi-modal biomedical data ranging from multi-omics molecular data to real-time continuous bio-signals are generated at an unprecedented speed and scale every day. For the first time, these multi-modal biomedical data are able to make precision medicine close to a reality. However, due to data volume and the complexity, making good use of these multi-modal biomedical data requires major effort. Researchers and clinicians are actively developing artificial intelligence (AI) approaches for data-driven knowledge discovery and causal inference using a variety of biomedical data modalities. These AI-based approaches have demonstrated promising results in various biomedical and healthcare applications. In this review paper, we summarize the state-of-the-art AI models for integrating multi-omics data and electronic health records (EHRs) for precision medicine. We discuss the challenges and opportunities in integrating multi-omics data with EHRs and future directions. We hope this review can inspire future research and developing in integrating multi-omics data with EHRs for precision medicine.

Discrimination exposure impacts unhealthy processing of food cues: crosstalk between the brain and gut.

Experiences of discrimination are associated with adverse health outcomes, including obesity. However, the mechanisms by which discrimination leads to obesity remain unclear. Utilizing multi-omics analyses of neuroimaging and fecal metabolites, we investigated the impact of discrimination exposure on brain reactivity to food images and associated dysregulations in the brain-gut-microbiome system. We show that discrimination is associated with increased food-cue reactivity in frontal-striatal regions involved in reward, motivation and executive control; altered glutamate-pathway metabolites involved in oxidative stress and inflammation as well as preference for unhealthy foods. Associations between discrimination-related brain and gut signatures were skewed towards unhealthy sweet foods after adjusting for age, diet, body mass index, race and socioeconomic status. Discrimination, as a stressor, may contribute to enhanced food-cue reactivity and brain-gut-microbiome disruptions that can promote unhealthy eating behaviors, leading to increased risk for obesity. Treatments that normalize these alterations may benefit individuals who experience discrimination-related stress.

Improving explainable AI with patch perturbation-based evaluation pipeline: a COVID-19 X-ray image analysis case study.

Recent advances in artificial intelligence (AI) have sparked interest in developing explainable AI (XAI) methods for clinical decision support systems, especially in translational research. Although using XAI methods may enhance trust in black-box models, evaluating their effectiveness has been challenging, primarily due to the absence of human (expert) intervention, additional annotations, and automated strategies. In order to conduct a thorough assessment, we propose a patch perturbation-based approach to automatically evaluate the quality of explanations in medical imaging analysis. To eliminate the need for human efforts in conventional evaluation methods, our approach executes poisoning attacks during model retraining by generating both static and dynamic triggers. We then propose a comprehensive set of evaluation metrics during the model inference stage to facilitate the evaluation from multiple perspectives, covering a wide range of correctness, completeness, consistency, and complexity. In addition, we include an extensive case study to showcase the proposed evaluation strategy by applying widely-used XAI methods on COVID-19 X-ray imaging classification tasks, as well as a thorough review of existing XAI methods in medical imaging analysis with evaluation availability. The proposed patch perturbation-based workflow offers model developers an automated and generalizable evaluation strategy to identify potential pitfalls and optimize their proposed explainable solutions, while also aiding end-users in comparing and selecting appropriate XAI methods that meet specific clinical needs in real-world clinical research and practice.

Early and fair COVID-19 outcome risk assessment using robust feature selection.

Personalized medicine plays an important role in treatment optimization for COVID-19 patient management. Early treatment in patients at high risk of severe complications is vital to prevent death and ventilator use. Predicting COVID-19 clinical outcomes using machine learning may provide a fast and data-driven solution for optimizing patient care by estimating the need for early treatment. In addition, it is essential to accurately predict risk across demographic groups, particularly those underrepresented in existing models. Unfortunately, there is a lack of studies demonstrating the equitable performance of machine learning models across patient demographics. To overcome this existing limitation, we generate a robust machine learning model to predict patient-specific risk of death or ventilator use in COVID-19 positive patients using features available at the time of diagnosis. We establish the value of our solution across patient demographics, including gender and race. In addition, we improve clinical trust in our automated predictions by generating interpretable patient clustering, patient-level clinical feature importance, and global clinical feature importance within our large real-world COVID-19 positive patient dataset. We achieved 89.38% area under receiver operating curve (AUROC) performance for severe outcomes prediction and our robust feature ranking approach identified the presence of dementia as a key indicator for worse patient outcomes. We also demonstrated that our deep-learning clustering approach outperforms traditional clustering in separating patients by severity of outcome based on mutual information performance. Finally, we developed an application for automated and fair patient risk assessment with minimal manual data entry using existing data exchange standards.

Hyperleptinemia contributes to antipsychotic drug-associated obesity and metabolic disorders.

Despite their high degree of effectiveness in the management of psychiatric conditions, exposure to antipsychotic drugs, including olanzapine and risperidone, is frequently associated with substantial weight gain and the development of diabetes. Even before weight gain, a rapid rise in circulating leptin concentrations can be observed in most patients taking antipsychotic drugs. To date, the contribution of this hyperleptinemia to weight gain and metabolic deterioration has not been defined. Here, with an established mouse model that recapitulates antipsychotic drug-induced obesity and insulin resistance, we not only confirm that hyperleptinemia occurs before weight gain but also demonstrate that hyperleptinemia contributes directly to the development of obesity and associated metabolic disorders. By suppressing the rise in leptin through the use of a monoclonal leptin-neutralizing antibody, we effectively prevented weight gain, restored glucose tolerance, and preserved adipose tissue and liver function in antipsychotic drug-treated mice. Mechanistically, suppressing excess leptin resolved local tissue and systemic inflammation typically associated with antipsychotic drug treatment. We conclude that hyperleptinemia is a key contributor to antipsychotic drug-associated weight gain and metabolic deterioration. Leptin suppression may be an effective approach to reducing the undesirable side effects of antipsychotic drugs.

Protective roles of adiponectin and molecular signatures of HNF4α and PPARα as downstream targets of adiponectin in pancreatic ß cells.

The disease progression of the metabolic syndrome is associated with prolonged hyperlipidemia and insulin resistance, eventually giving rise to impaired insulin secretion, often concomitant with hypoadiponectinemia. As an adipose tissue derived hormone, adiponectin is beneficial for insulin secretion and ß cell health and differentiation. However, the down-stream pathway of adiponectin in the pancreatic islets has not been studied extensively. Here, along with the overall reduction of endocrine pancreatic function in islets from adiponectin KO mice, we examine PPARα and HNF4α as additional down-regulated transcription factors during a prolonged metabolic challenge. To elucidate the function of ß cell-specific PPARα and HNF4α expression, we developed doxycycline inducible pancreatic ß cell-specific PPARα (ß-PPARα) and HNF4α (ß-HNF4α) overexpression mice. ß-PPARα mice exhibited improved protection from lipotoxicity, but elevated ß-oxidative damage in the islets, and also displayed lowered phospholipid levels and impaired glucose-stimulated insulin secretion. ß-HNF4α mice showed a more severe phenotype when compared to ß-PPARα mice, characterized by lower body weight, small islet mass and impaired insulin secretion. RNA-sequencing of the islets of these models highlights overlapping yet unique roles of ß-PPARα and ß-HNF4α. Given that ß-HNF4α potently induces PPARα expression, we define a novel adiponectin-HNF4α-PPARα cascade. We further analyzed downstream genes consistently regulated by this axis. Among them, the islet amyloid polypeptide (IAPP) gene is an important target and accumulates in adiponectin KO mice. We propose a new mechanism of IAPP aggregation in type 2 diabetes through reduced adiponectin action.

Endogenous renal adiponectin drives gluconeogenesis through enhancing pyruvate and fatty acid utilization.

Adiponectin is a secretory protein, primarily produced in adipocytes. However, low but detectable expression of adiponectin can be observed in cell types beyond adipocytes, particularly in kidney tubular cells, but its local renal role is unknown. We assessed the impact of renal adiponectin by utilizing male inducible kidney tubular cell-specific adiponectin overexpression or knockout mice. Kidney-specific adiponectin overexpression induces a doubling of phosphoenolpyruvate carboxylase expression and enhanced pyruvate-mediated glucose production, tricarboxylic acid cycle intermediates and an upregulation of fatty acid oxidation (FAO). Inhibition of FAO reduces the adiponectin-induced enhancement of glucose production, highlighting the role of FAO in the induction of renal gluconeogenesis. In contrast, mice lacking adiponectin in the kidney exhibit enhanced glucose tolerance, lower utilization and greater accumulation of lipid species. Hence, renal adiponectin is an inducer of gluconeogenesis by driving enhanced local FAO and further underlines the important systemic contribution of renal gluconeogenesis.

GPR84-mediated signal transduction affects metabolic function by promoting brown adipocyte activity.

The G protein-coupled receptor 84 (GPR84), a medium-chain fatty acid receptor, has garnered attention because of its potential involvement in a range of metabolic conditions. However, the precise mechanisms underlying this effect remain elusive. Our study has shed light on the pivotal role of GPR84, revealing its robust expression and functional significance within brown adipose tissue (BAT). Mice lacking GPR84 exhibited increased lipid accumulation in BAT, rendering them more susceptible to cold exposure and displaying reduced BAT activity compared with their WT counterparts. Our in vitro experiments with primary brown adipocytes from GPR84-KO mice revealed diminished expression of thermogenic genes and reduced O2 consumption. Furthermore, the application of the GPR84 agonist 6-n-octylaminouracil (6-OAU) counteracted these effects, effectively reinstating the brown adipocyte activity. These compelling in vivo and in vitro findings converge to highlight mitochondrial dysfunction as the primary cause of BAT anomalies in GPR84-KO mice. The activation of GPR84 induced an increase in intracellular Ca2+ levels, which intricately influenced mitochondrial respiration. By modulating mitochondrial Ca2+ levels and respiration, GPR84 acts as a potent molecule involved in BAT activity. These findings suggest that GPR84 is a potential therapeutic target for invigorating BAT and ameliorating metabolic disorders.

Mediation of the association between disadvantaged neighborhoods and cortical microstructure by body mass index.

BACKGROUND: Living in a disadvantaged neighborhood is associated with worse health outcomes, including brain health, yet the underlying biological mechanisms are incompletely understood. We investigated the relationship between neighborhood disadvantage and cortical microstructure, assessed as the T1-weighted/T2-weighted ratio (T1w/T2w) on magnetic resonance imaging, and the potential mediating roles of body mass index (BMI) and stress, as well as the relationship between trans-fatty acid intake and cortical microstructure. METHODS: Participants comprised 92 adults (27 men; 65 women) who underwent neuroimaging and provided residential address information. Neighborhood disadvantage was assessed as the 2020 California State area deprivation index (ADI). The T1w/T2w ratio was calculated at four cortical ribbon levels (deep, lower-middle, upper-middle, and superficial). Perceived stress and BMI were assessed as potential mediating factors. Dietary data was collected in 81 participants. RESULTS: Here, we show that worse ADI is positively correlated with BMI (r = 0.27, p = .01) and perceived stress (r = 0.22, p = .04); decreased T1w/T2w ratio in middle/deep cortex in supramarginal, temporal, and primary motor regions (p < .001); and increased T1w/T2w ratio in superficial cortex in medial prefrontal and cingulate regions (p < .001). Increased BMI partially mediates the relationship between worse ADI and observed T1w/T2w ratio increases (p = .02). Further, trans-fatty acid intake (high in fried fast foods and obesogenic) is correlated with these T1w/T2w ratio increases (p = .03). CONCLUSIONS: Obesogenic aspects of neighborhood disadvantage, including poor dietary quality, may disrupt information processing flexibility in regions involved in reward, emotion regulation, and cognition. These data further suggest ramifications of living in a disadvantaged neighborhood on brain health.

Neighborhood disadvantage (a combination of low average income, more people leaving education earlier, crowding, lack of complete plumbing, etc.) is known to impact the health of people's brains. We evaluated whether neighborhood disadvantage was associated with differences in the structure of people's brains, and whether any differences were related to an unhealthily high weight and a high intake of trans-fatty acids, a component of fried fast food, on the structure of people's brains. Based on our results, regions of the brain that are involved in reward, emotion and gaining knowledge and understanding might be affected by aspects of neighborhood disadvantage that contribute to obesity, such as poor dietary quality. This suggests that it might be important to make healthier food more readily available in disadvantaged neighborhoods to improve the health of people's brains.

Use of Natural Language Processing of Patient-Initiated Electronic Health Record Messages to Identify Patients With COVID-19 Infection.

Importance: Natural language processing (NLP) has the potential to enable faster treatment access by reducing clinician response time and improving electronic health record (EHR) efficiency. Objective: To develop an NLP model that can accurately classify patient-initiated EHR messages and triage COVID-19 cases to reduce clinician response time and improve access to antiviral treatment. Design, Setting, and Participants: This retrospective cohort study assessed development of a novel NLP framework to classify patient-initiated EHR messages and subsequently evaluate the model's accuracy. Included patients sent messages via the EHR patient portal from 5 Atlanta, Georgia, hospitals between March 30 and September 1, 2022. Assessment of the model's accuracy consisted of manual review of message contents to confirm the classification label by a team of physicians, nurses, and medical students, followed by retrospective propensity score-matched clinical outcomes analysis. Exposure: Prescription of antiviral treatment for COVID-19. Main Outcomes and Measures: The 2 primary outcomes were (1) physician-validated evaluation of the NLP model's message classification accuracy and (2) analysis of the model's potential clinical effect via increased patient access to treatment. The model classified messages into COVID-19-other (pertaining to COVID-19 but not reporting a positive test), COVID-19-positive (reporting a positive at-home COVID-19 test result), and non-COVID-19 (not pertaining to COVID-19). Results: Among 10 172 patients whose messages were included in analyses, the mean (SD) age was 58 (17) years; 6509 patients (64.0%) were women and 3663 (36.0%) were men. In terms of race and ethnicity, 2544 patients (25.0%) were African American or Black, 20 (0.2%) were American Indian or Alaska Native, 1508 (14.8%) were Asian, 28 (0.3%) were Native Hawaiian or other Pacific Islander, 5980 (58.8%) were White, 91 (0.9%) were more than 1 race or ethnicity, and 1 (0.01%) chose not to answer. The NLP model had high accuracy and sensitivity, with a macro F1 score of 94% and sensitivity of 85% for COVID-19-other, 96% for COVID-19-positive, and 100% for non-COVID-19 messages. Among the 3048 patient-generated messages reporting positive SARS-CoV-2 test results, 2982 (97.8%) were not documented in structured EHR data. Mean (SD) message response time for COVID-19-positive patients who received treatment (364.10 [784.47] minutes) was faster than for those who did not (490.38 [1132.14] minutes; P = .03). Likelihood of antiviral prescription was inversely correlated with message response time (odds ratio, 0.99 [95% CI, 0.98-1.00]; P = .003). Conclusions and Relevance: In this cohort study of 2982 COVID-19-positive patients, a novel NLP model classified patient-initiated EHR messages reporting positive COVID-19 test results with high sensitivity. Furthermore, when responses to patient messages occurred faster, patients were more likely to receive antiviral medical prescription within the 5-day treatment window. Although additional analysis on the effect on clinical outcomes is needed, these findings represent a possible use case for integration of NLP algorithms into clinical care.

Dissociating instructive from permissive roles of brain circuits with reversible neural activity manipulations.

Neuroscientists rely on targeted perturbations and lesions to causally map functions in the brain1. Yet, since the brain is highly interconnected, manipulation of one area can impact behavior through indirect effects on many other brain regions, complicating the interpretation of such results2,3. On the other hand, the often-observed recovery of behavior performance after lesion can cast doubt on whether the lesioned area was ever directly involved4,5. Recent studies have highlighted how the results of acute and irreversible inactivation can directly conflict4-6, making it unclear whether a brain area is instructive or merely permissive in a specific brain function. To overcome this challenge, we developed a three-stage optogenetic approach which leverages the ability to precisely control the temporal period of regional inactivation with either brief or sustained illumination. Using a visual detection task, we found that acute optogenetic inactivation of the primary visual cortex (V1) suppressed task performance if cortical inactivation was intermittent across trials within each behavioral session. However, when we inactivated V1 for entire behavioral sessions, animals quickly recovered performance in just one to two days. Most importantly, after returning these recovered animals to intermittent cortical inactivation, they quickly reverted to failing on optogenetic inactivation trials. These data support a revised model where the cortex is the default circuit that instructs perceptual performance in basic sensory tasks. More generally, this novel, temporally controllable optogenetic perturbation paradigm can be broadly applied to brain circuits and specific cell types to assess whether they are instructive or merely permissive in a brain function or behavior.