Long COVID affects a substantial proportion of the over 778 million individuals infected with SARS-CoV-2, yet predictive models remain limited in scope. While existing efforts, such as the National COVID Cohort Collaborative (N3C), have leveraged electronic health record (EHR) data for risk prediction, accumulating evidence points to additional contributions from social, behavioral, and genetic factors. Using a diverse cohort of SARS-CoV-2-infected individuals (n>17,200) from the NIH All of Us Research Program, we investigated whether integrating EHR data with survey-based and genomic information improves model performance. Our multi-scale approach outperformed EHR-only models original AUROC 0.736 (95% CI: 0.730, 0.741), achieving an AUROC of 0.748 (0.741,0.755). Among the top predictors, active-duty service status, self-reported fatigue, and chr19:4719431:G:A_A were among the most informative survey and genetic features. These findings highlight the importance of incorporating multi-scale data to improve risk stratification and inform personalized interventions for long COVID.

BACKGROUND:To enhance biological understanding of ARDS, pneumonia, and sepsis and to accelerate therapeutic development in these areas, the National Institutes of Health developed the ARDS, Pneumonia, and Sepsis (APS) Consortium. RESEARCH QUESTION:Is the APS Consortium study rapidly generating data and biospecimens from a large cohort of critically ill adults with ARDS, pneumonia, and sepsis that will facilitate phenotyping of these syndromes? STUDY DESIGN AND METHODS:The APS Consortium Phenotyping Study is a multicenter, longitudinal, prospective, observational cohort study aimed at enrolling 4,000 critically ill adults with ARDS, pneumonia, sepsis, or a combination thereof over 4 years. Data and biospecimens are collected to characterize many aspects of each participant's chronic health, acute illness, and long-term recovery to facilitate phenotyping, that is, subclassifying ARDS, pneumonia, and sepsis into precise biologically based subsets with shared pathophysiologic characteristics. Feasibility of the study was assessed by evaluating the first 1,000 participants in terms of recruitment pace, participant characteristics, biospecimen collection, and proportion with confirmed ARDS, pneumonia, and sepsis based on expert adjudication. RESULTS:The first 1,000 participants were recruited ahead of schedule in < 13 months. Median age was 64 years, 75% received vasopressors, 50% received invasive mechanical ventilation, and 25% died in the hospital within 4 weeks of enrollment. Biospecimen collection rates were high, with 99% of participants with blood samples, 98% with upper respiratory swabs, 37% with lower respiratory samples, 80% with urine samples, and 65% with gastrointestinal samples. Expert adjudication resulted in 40% classified with ARDS, 52% classified with pneumonia, and 89% classified with sepsis. INTERPRETATION:The APS Consortium Phenotyping Study is producing a cohort of critically ill adults with ARDS, pneumonia, and sepsis with high severity of disease and a rich set of data and biospecimens. The study will continue to full enrollment of 4,000 participants. CLINICAL TRIAL REGISTRATION:ClinicalTrials.gov; No.: NCT06521502; URL: www. CLINICALTRIALS:gov.

Rationale:Sepsis is a life-threatening syndrome causing significant morbidity and mortality especially in the aging population. Clonal hematopoiesis of indeterminate potential (CHIP) is an age-related condition of clonal expansion of hematopoietic stem cells harboring somatic mutations associated with increased incidence of chronic illness and all-cause mortality. Objective:Evaluate the association of pre-illness CHIP with mortality and morbidity in patients admitted to the ICU with sepsis. Methods:We performed a retrospective study using a de-identified electronic health record linked with a DNA biorepository. We identified adult patients with sepsis who had DNA collected prior to ICU admission. We tested the association between CHIP status, determined from whole-genome sequencing, and ICU mortality, organ support-free days, and long-term survival adjusting for age, sex, race and Sequential Organ Failure Assessment (SOFA) score on ICU admission. Measurements and Main Results:Pre-illness CHIP was associated with increased sepsis mortality (OR = 1.54, 95% CI 1.13 to 2.07, P = 0.005) and fewer days alive and free of organ support (-1.7 days, 95% CI -3.2 to -0.2, P = 0.028) after adjusting for age, sex, race, and SOFA score. In sepsis survivors, CHIP was also associated with increased long-term mortality after discharge (HR 1.40, 95% CI 1.01 to 1.93, P = 0.041). Conclusions:Pre-illness CHIP was independently associated with increased mortality and morbidity in critically-ill adults with sepsis. These findings suggest that CHIP is a risk factor for sepsis severity. Elucidating the mechanism underlying this association could uncover new therapeutic interventions for sepsis.

BACKGROUND:Conservative fluid management improves outcomes in mechanically ventilated adults with acute respiratory distress syndrome, but its impact across the broader spectrum of acute respiratory failure is less clear. The severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) pandemic provided an opportunity to assess the relationship between daily fluid balance and respiratory status across varying degrees of respiratory impairment with a relatively uniform lung injury mechanism. METHODS:Retrospective study of adults hospitalized with SARS-CoV-2 within ±7 days of hospital admission in an academic medical system (March 2020-July 2022). The primary outcome was a modified World Health Organization Clinical Scale from 1 (Hospitalized, no respiratory support) to 5 (Death or discharge to hospice). Associations between daily net fluid balance and next-day modified World Health Organization Clinical Scale were tested using cumulative logistic regression and adjacent-categories logistic regression, adjusting for demographics, comorbidities, hospital type, vasopressors, inotropes, renal replacement therapy, and unmeasured urine outputs. RESULTS:Among 4,254 patients (median age 58 [41-70]; 52.9% male; median hospital stay 4.5 [2.5-8.1] days), 2,860 (67.2%) required respiratory support, 539 (12.7%) received invasive ventilation, and 571 (13.4%) died. In critically ill adults, positive fluid balance was associated with higher odds of subsequent mechanical ventilation (odds ratio [OR] per L 1.48, 95% confidence interval [CI] 1.41-1.55) or death (OR 1.15, 95% CI 1.08-1.23). Among noncritically ill patients on low-flow oxygen, positive fluid balance was associated with improved respiratory status (OR 0.89, 95% CI 0.86-0.92). CONCLUSIONS:Positive fluid balance had divergent effects by illness severity, suggesting distinct physiologic or clinical mechanisms across the spectrum of SARS-CoV-2 respiratory failure.

Immune dysfunction is a major driver of morbidity and mortality in critical illness syndromes including sepsis. Specifically, CD8+ T cell dysfunction has been linked to organ failure and death. To characterize the immune substructure of circulating CD8+ T cells in critical illness at high dimension, we used single-cell RNA sequencing of peripheral blood CD8+ T cells from 38 critically ill patients and 9 healthy controls. We annotated seven CD8+ T cell clusters, which included a CD8+ effector subset, termed T effector state 2 (TEff-2), that was only present in critically ill patients and associated with more severe respiratory failure and higher mortality. TEff-2 showed effector activation and inflammatory stress conditioning yet had markedly reduced metabolic transcripts without canonical features of exhaustion. Trajectory analyses positioned TEff-2 as a terminal CD8+ T effector cell fate driven in part by DDIT4 and DUSP1, which negatively regulate mTOR and MAPK signaling, respectively. Interestingly, this transcriptional program was indistinguishable by classical protein cytometry methods. These results, including the mortality association, were validated in a larger (n=91) independent external cohort of critically ill patients with sepsis. In summary, TEff-2 represents a latent transcriptional program that delineates a clinically high-risk CD8+ T cell state in critical illness.

Metabolic and immunologic dysfunction, including pathological CD4+ T cell immunosuppression, are archetypal in critical illness, but whether these factors are mechanistically linked remains incompletely defined. Here we characterized the metabolic properties of human CD4+ T cells from critically ill patients with and without sepsis and healthy adults. CD4+ T cells in critical illness showed subset-specific metabolic plasticity, with regulatory T (Treg) cells preferentially acquiring glycolytic capacity that associated with sustained cellular fitness and worsened clinical illness. Adapted Treg cells were more metabolically flexible and stabilized suppressive markers FOXP3 and TIGIT under mitochondrial stress. Single-cell transcriptomics suggested reactive oxygen species (ROS) and kynurenine metabolism as drivers of Treg cell remodeling. Subsequent inhibition of ROS and kynurenine metabolism attenuated glycolytic adaptation and suppressive rewiring, respectively, in Treg cells. These findings indicate that metabolic dysfunction was a contributor to CD4+ T cell remodeling in critical illness and suggest avenues to restore effective immunity.

RATIONALE: Correctly recording causes of death is essential for accurate national mortality statistics, yet errors in death certificates are common. We evaluated the ability of a large language model (LLM) to identify ranked causes of death in a well-phenotyped cohort of critically ill adults. METHODS: We studied 278 critically ill patients enrolled in the Validating Acute Lung Injury Markers for Diagnosis (VALID) study and who died following ICU admission. We uploaded death summaries into a HIPPA-compliant institutional Azure OpenAI GPT-3.5 to identify ranked causes of death (CODs) using two zero-shot prompts. Prompt-1 identified a primary cause with secondary causes ranked by impact, while Prompt-2 generated a contributing diagnosis in order of relevance without explicitly defining a primary cause. All available CODs were mapped to ICD-10-CM and National Center for Health Statistics (NCHS) codes capturing varying levels of granularity. We compared LLM results with ranked CODs adjudicated by an expert physician-scientist reviewer according to CDC death certificate instructions as a gold-standard. We also compared investigator-adjudicated CODs with official death certificates. Evaluation was performed using Rank Bias Overlap (RBO), which emphasizes the top-ranked CODs at various stopping probabilities (p-parameter) and depths (k-value), and Kendall's Tau, which measures the overall correlation across the entire ranking. RESULTS: Prompt-1 and Prompt-2 ranked the primary COD identified by the expert adjudicator as first or second in 66.67% and 63.33% of cases. We observed strong internal consistency between Prompts for identifying CODs (RBO range 0.53-0.71; Kendall's Tau: 0.45). Both Prompts demonstrated moderate alignment with investigator-adjudicated CODs: for Prompt-1 mean RBO ranged from 0.24-0.32 when using ICD codes and 0.23-0.47 for NCHS codes, while for Prompt-2 RBO ranged from 0.21-0.36 (ICD) and from 0.27-0.53 (NCHS). This was comparable to the observed agreement between investigator-adjudicated and death certificate CODs (RBO range 0.25-0.46; Kendall's Tau: 0.21 using NCHS codes). Manual review showed that the LLM misidentified death mechanisms such as “cardiac arrest” or “respiratory arrest” as the primary COD (Prompt-1: 4 patients; Prompt-2: 2 patients) suggesting difficulty with distinguishing mechanisms from causes when both types are recorded in clinical documents. CONCLUSION: This study highlights the challenges in consistently identifying ranked CODs. The moderate rank-order consistency of the LLMs with investigator-adjudicated CODs demonstrates potential to assist with identifying CODs from clinical documentation. Future work will focus on prompt engineering to reduce reporting of death mechanisms as CODs, compare results with updated GPT versions, and explore additional administrative and research use cases.

Rationale: Although conservative fluid management improves clinical outcomes in mechanically ventilated ARDS, its impact across the wider spectrum of acute respiratory failure (ARF) is less well characterized. SARS-CoV-2 provides unique opportunities to examine fluid balance in ARF with high numbers of affected patients and comparatively uniform mechanisms of lung injury. We tested the association between fluid balance and ARF trajectories in adults hospitalized with SARS-CoV-2 across the full spectrum of respiratory impairment. Methods: We performed a retrospective observational cohort study of adults admitted to an inpatient ward or ICU with a positive SARS-CoV-2 PCR within ±7 days of admission at five Vanderbilt-affiliated hospitals from March-2020 to July-2022. We excluded psychiatric admissions and transfers from non-affiliated hospitals. The primary outcome was a modified WHO Clinical Scale (mWHO) ranging from 1 (Hospitalized; no respiratory support) to 5 (Death). We tested the association between daily net fluid balance and mWHO score on the subsequent hospital day using cumulative logistic regression and adjacent-categories regression to test non-proportionality across different mWHO levels. Covariates included demographics, hospital characteristics (academic vs community affiliate), heart failure, renal disease, dialysis, Elixhauser comorbidity score, and urine output events with missing volume measurements to capture lower-intensity fluid monitoring. Results: We identified 4,329 patients meeting inclusion and exclusion criteria. Median [IQR] age was 58 [41-70], hospitalization length was 4.5 [2.5-8.0] days; 2,287 (52.8%) were male; 2,874 (66.4%) received any respiratory support; 540 (12.5%) received invasive ventilation; 594 (13.7%) died. We observed substantial non-proportional associations between fluid balance and mWHO category. In cumulative logistic analysis, positive fluid balance was associated with progressively increasing odds of transitioning to a worse mWHO category the following day among patients with more severe respiratory impairment (Figure). In adjacent-categories analysis, positive fluid balance was associated with decreased odds of transitioning to a worse category in patients receiving no respiratory support (mWHO=1) or low-flow oxygen (mWHO=2), whereas it was associated with increased odds of transitioning to a worse category for high-flow oxygen / noninvasive ventilation (mWHO=3) or invasive ventilation (mWHO=4, Figure). Subgroup analyses by SARS-CoV-2 waves (March-2020 to June-2021; July-November 2021; and December-2021 to July-2022) demonstrated similar associations across time. Conclusion: In adults hospitalized with SARS-CoV-2, positive fluid balance was associated with differential effects on respiratory trajectory depending on clinical status. These findings may reflect clinical behaviors such as increased fluid administration surrounding clinical deterioration, or biological effects including pulmonary epithelial or endothelial barrier dysfunction in severely ill patients.

BACKGROUND:Large population-based DNA biobanks linked to electronic health records (EHRs) may provide novel opportunities to identify genetic drivers of ARDS. RESEARCH QUESTION:Can a computerized algorithm identify ARDS in a large EHR biobank database, and can this be used to identify ARDS genetic risk factors? STUDY DESIGN AND METHODS:We developed a classifier algorithm to identify a diagnosis of ARDS as identified from the electronic health record (EHR-ARDS) using diagnostic billing codes, laboratory test results, and chest radiography report text. The classifier model performance was evaluated against investigator-adjudicated ARDS using standard classification metrics including sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and the Cohen κ value. After confirming acceptable classifier performance, we evaluated the association between EHR-ARDS and the MUC5B promoter polymorphism rs35705950 in 2 parallel genotyped cohorts: a prospective biomarker cohort of critically ill adults (Validating Acute Lung Injury Biomarkers for Diagnosis [VALID]) and a retrospective cohort from our institution's de-identified EHR biobank, BioVU. RESULTS:We included 2,795 patients from VALID and 9,025 hospitalized participants from BioVU. EHR-ARDS showed moderate agreement with investigator-adjudicated ARDS (VALID: sensitivity, 0.86; specificity, 0.70; PPV, 0.49; NPV, 0.93; and k, 0.45; BioVU: sensitivity, 0.94; specificity, 0.81; PPV, 0.66; NPV, 0.97; and k, 0.67). We observed a significant age-gene interaction effect for EHR-ARDS in VALID: among older patients, rs35705950 was associated with increased EHR-ARDS risk (OR, 1.37; 95% CI, 1.05-1.78; P = .019), whereas among younger patients, this association was absent (OR, 0.92; 95% CI, 0.70-1.21; P = .55). In BioVU, rs35705950 was associated with EHR-ARDS among all participants (OR, 1.20; 95% CI, 1.01-1.43; P = .043); however, this effect did not vary by age. INTERPRETATION:The MUC5B promoter polymorphism was associated with EHR-ARDS in 2 parallel cohorts of at-risk adults. An age-gene effect modification was observed in VALID, whereas BioVU identified a consistent association between MUC5B and EHR-ARDS regardless of age. Our study highlights the potential for EHR biobanks to enable precision medicine ARDS studies.

Topic modeling utilizes unsupervised machine learning to detect underlying themes within texts and has been deployed routinely to analyze social media for insights into healthcare issues. However, the inherent messiness of social media hinders the full realization of this technique's potential. As such, we hypothesized that restricting medical concepts in social media texts to specific related semantic types and applying topic modeling to these concepts could be a feasible approach to overcome the challenge of traditional topic modeling for social media texts. Therefore, we developed a semantic-type-based topic modeling pipeline to discover self-reported health-related topics. This pipeline integrated semantic type information and Systematized Medical Nomenclature for Medicine (SNOMED) precoordinated expressions into a traditional topic modeling approach to enhance effectiveness in clustering meaningful, distinct topics. Using social media texts regarding statins for illustration, we evaluated the efficacy of this new approach and validated a newly identified topic using real-world clinical data. Based on expert evaluations, this approach resulted in more novel, distinguishable, and meaningful health-related topics compared to traditional topic modeling. In addition, our electronic health record validation for a newly identified topic in two real-world clinical databases indicated that statin users had a higher prevalence of depression or anxiety compared to matched non-users. Our results indicate that this new topic modeling pipeline can improve the extraction of themes from noisy online discussions, thereby contributing to deeper insights for healthcare research.

BACKGROUND:Acute respiratory distress syndrome (ARDS) is a severe inflammatory process of the lung, often due to sepsis, and poses significant mortality burden in intensive care units. Here we conducted a genome-wide association study (GWAS) of ARDS to identify genetic risk loci that can help guide the development of new therapeutic options. METHODS:We performed a case-control GWAS in 716 cases with ARDS, mainly associated with severe infections, and 4399 at-risk controls from three independent studies. Results were meta-analysed across the three studies, with significance set at p < 5 × 10-8. Suggestive associations were declared for variants exhibiting consistent direction of effects, likely to replicate and nominal significance (p < 0.05) in all three studies. Prioritised loci were subjected to Bayesian fine mapping, in-silico functional assessments, and gene-based rare variant collapsing analysis using whole-exome sequencing data. Two independent studies with 430 ARDS cases and 1398 at-risk controls served as validation samples. FINDINGS:We identified a variant near HMGCR that showed genome-wide significant association with ARDS and had been previously linked to cholesterol metabolism. This locus was associated with ANKDD1B expression in artery. The rare exonic variant analysis showed associations between HMGCR and ARDS at nominal level (p < 0.05). While no nominal significance was achieved in the two additional validation cohorts, this variant exhibited a consistent direction of effects across all 5 studies. INTERPRETATION:A common variant near HMGCR was associated with ARDS risk, suggesting a link between cholesterol metabolism and ARDS risk. Validation in independent studies is needed. FUNDING:Wellcome Trust, National Institute for Health Research Leicester Biomedical Research Centre, National Heart, Lung, and Blood Institute, ATS Research Program, Gobierno de Canarias, Fundación Canaria Instituto de Investigación Sanitaria de Canarias, Instituto Tecnológico y de Energías Renovables, Cabildo Insular de Tenerife, Instituto de Salud Carlos III, Agencia Estatal de Investigación, German Ministry of Education and Research, Thuringian Ministry of Education, Science and Culture, the Thuringian Foundation for Technology, Innovation, and Research, German Sepsis Society.

INTRODUCTION: Hyper- and hypoinflammatory phenotypes have consistently been identified by latent class analysis in numerous cohorts of patients with ARDS and sepsis. In retrospective analysis of clinical trials of simvastatin in ARDS, corticosteroids in COVID-19 and activated protein C in severe sepsis, enrichment of treatment effect was observed in the hyperinflammatory phenotype. We hypothesized that the treatment effect of acetaminophen would be enriched in the hyperinflammatory phenotype among sepsis patients with respiratory or circulatory organ dysfunction enrolled in a recently completed trial of acetaminophen for sepsis (ASTER, Ware et al. JAMA 2024). METHODS: Patients enrolled in ASTER were retrospectively classified into the hyper- and hypoinflammatory phenotypes using a previously validated parsimonious classifier that included baseline plasma IL-8, sTNFR1, serum bicarbonate, and need for vasopressors. Multivariate logistic regression models were fit with key trial endpoints (days alive and free of respiratory, cardiovascular and renal organ support to 28-days, 28-day and 90-day mortality, and 28-day ventilator-free days) as the dependent variable and the interaction term of treatment group and phenotype as the independent variable. Treatment effect for acetaminophen was also evaluated separately for the hyper- and hypoinflammatory phenotypes. RESULTS: Of the 447 patients enrolled in ASTER, 392 had biomarkers and clinical data available for parsimonious classification: 108 (28%) patients were classified as hyperinflammatory and 284 (72%) patients as hypoinflammatory. Consistent with prior studies, mortality was higher in the hyperinflammatory phenotype at both 28-days (34% vs 17%, p < 0.001) and 90-days (43% vs. 26%, p = 0.002). There was a significant interaction (p=0.050) between acetaminophen treatment effect and inflammatory phenotype for days alive and free of organ support, but contrary to our hypothesis, the acetaminophen treatment effect was evident only in the hypoinflammatory group (Table). Although not statistically significant, the acetaminophen treatment effect was also larger in the hypoinflammatory phenotype across other the clinical outcomes tested (Table). CONCLUSIONS: In contrast to prior studies that have shown enrichment for treatment effect of simvastatin, corticosteroids and activated protein C in sepsis and ARDS patients with the hyperinflammatory phenotype, we did not identify an enrichment of treatment effect in the hyperinflammatory phenotype in the ASTER clinical trial. The treatment effect of acetaminophen appeared to be more beneficial in the hypoinflammatory phenotype, comprising the majority of patients enrolled. These findings indicate that future clinical trials of acetaminophen in sepsis should not prospectively enrich for the hyperinflammatory phenotype.

Rationale Latent class analysis (LCA) of patients with sepsis and ARDS has consistently identified two phenotypes with distinct plasma proinflammatory cytokine profiles and differences in clinical outcomes and treatment responses. In previous work, we identified an unexpected association between premorbid metabolic syndrome and the hypoinflammatory phenotype (Bogart, ATS 2023). To elucidate metabolic pathways that contribute to this relationship, we determined the association of plasma metabolites with metabolic syndrome in critically ill adults and those with the hypoinflammatory phenotype. Methods We utilized untargeted plasma metabolomic data (746 metabolites) from 182 septic adults with or without ARDS enrolled between 2008-2016 in a prospective observational cohort (Early Assessment of Renal and Lung Injury, EARLI) in which LCA phenotype assignment had previously been performed. MetaboAnalyst 6.0 was used to identify pathways that were overrepresented in the cohort and to identify metabolites of interest (False Discovery Rate, FDR <0.1). The association between number of component diagnoses for metabolic syndrome (obesity, hypertension, or diabetes; dyslipidemia data were not available) and levels of plasma metabolites identified in the pathway analysis was tested using the Jonckheere-Terpstra trend test with FDR <0.05 to control for multiple comparisons. We also assessed the relationship between circulating metabolite concentrations and LCA phenotype by the Mann-Whitney test. Results 114 patients were classified as hypoinflammatory and 68 as hyperinflammatory. 58 subjects had zero component diagnoses of metabolic syndrome, 81 had one, and 43 had two or more. Of the 746 measured metabolites, 175 (representing 18 pathways) were significantly enriched in the full cohort. There were 13 metabolites significantly associated with the number of metabolic syndrome component diagnoses, 6 of which were also significantly different between LCA phenotypes. All 6 showed similar directional association, decreasing with both number of component diagnoses for metabolic syndrome and the hypoinflammatory phenotype. Interestingly, 3 of these 6 identified metabolites, including phenyllactate (Figure), are involved in amino acid metabolism, specifically of phenylalanine, tyrosine, alanine and aspartate. Conclusions Decreased circulating levels of amino acid metabolites are associated with both more metabolic syndrome component diagnoses and the hypoinflammatory phenotype. Given that immune cell functions, including of CD4+ T-cell subsets and macrophage subtypes, are influenced by amino acid metabolism, these data suggest a potential mechanistic link between premorbid metabolic syndrome and critical illness phenotypes. Overall, our findings enhance the current understanding of the biologic drivers of phenotype divergence and indicate that pre-existing medical conditions may play a critical role in phenotype pathogenesis.

Myocardial injury is common in acute respiratory distress syndrome (ARDS) and sepsis and associated with increased mortality. Two latent class analysis derived subphenotypes are associated with differential risk of mortality in these populations, though the association of troponin-I with mortality within each subphenotype is unknown. The derivation (n = 597 in EARLI) and validation (n = 452 in VALID) cohorts consisted of patients with sepsis or ARDS admitted to the ICU and enrolled in two separate prospective observational studies. Patients with troponin-I measured between hospital presentation and within 24 h of ICU admission were included. A parsimonious classifier model using interleukin-8, soluble tumor necrosis factor receptor-1, and vasopressor use assigned patients to subphenotype. Association between peak troponin-I concentration and 60-day in-hospital mortality within each subphenotype was assessed through logistic regression adjusting for age, admission laboratory values, vasopressor use, invasive ventilation use, and cardiac comorbidities. Median peak troponin-I was significantly higher in the hyperinflammatory vs hypoinflammatory subphenotype in both cohorts (0.07 vs 0.04 ng/mL and 0.17 vs 0.07 ng/mL, both p < 0.05). The association between peak troponin-I and mortality differed between inflammatory subphenotypes (p-interaction 0.004, EARLI). In EARLI, each doubling of peak troponin-I was associated with increased adjusted odds of 60-day mortality (aOR 1.14, 95

OBJECTIVES:Hyperinflammatory and hypoinflammatory molecular subphenotypes in sepsis and acute respiratory distress syndrome have divergent mortality and treatment responses in secondary analyses of randomized controlled trials. However, the prevalence of immunocompromise is low in these populations, and how preexisting immunocompromise contributes to subphenotypes is unknown. We studied two observational sepsis cohorts to test associations between immunocompromise and the hyperinflammatory subphenotype and to assess whether the prognostic relevance of molecular subphenotypes is generalizable to immunocompromised populations. DESIGN:Observational cohort study. SETTING:Prospective data from two ICU cohorts in the United States. PATIENTS:We included 1826 patients from two combined sepsis cohorts. INTERVENTIONS:None. MEASUREMENTS AND MAIN RESULTS:We defined immunocompromise as a history of solid organ transplant, AIDS, hematologic malignancy, solid malignancy on chemotherapy, or immunosuppressive medication use. Subphenotype was previously assigned using latent class analysis. We used logistic regression to investigate associations between type of immunocompromise and hyperinflammatory subphenotype. Models were repeated with individual covariates known or hypothesized to be associated with the hyperinflammatory subphenotype. Kaplan-Meier survival plots were used to assess mortality differences by subphenotype. Hematologic malignancy was strongly associated with the hyperinflammatory subphenotype (odds ratio [OR], 4.3; p < 0.0001), an association that persisted after adjustment for identified pathogen, presence of bacteremia, or illness severity. History of solid organ transplantation was also associated with the hyperinflammatory subphenotype (OR, 1.6; p = 0.02) but was no longer significant after accounting for bacteremia. Hyperinflammatory classification was associated with a decreased likelihood of survival in hematologic malignancy, but not in organ transplant or solid malignancy populations. CONCLUSIONS:Preexisting immune status is associated with subphenotype assignment and may influence its prognostic utility.

BackgroundOnly a subset of patients at risk for acute respiratory distress syndrome (ARDS) go on to develop it, and the contribution of preexisting comorbidities, such as diabetes, to ARDS risk is not well understood. Prior studies of the association between diabetes and ARDS yielded conflicting results.Research QuestionDoes assessing ARDS risk based on hemoglobin A1c (HbA1c) as a marker of long-term blood glucose levels, rather than a charted diagnosis of diabetes, clarify the relationship between diabetes and ARDS?Study Design and MethodsUsing data from two prospective observational cohorts of critically ill adults (VALID and EARLI), we analyzed the association between clinical HbA1c category and development of ARDS in patients with a risk factor for ARDS and at least one clinical HbA1c measurement within the 180 days prior through 14 days after enrollment.Results599 patients in VALID and 276 in EARLI met inclusion criteria, of whom 164 and 58 developed ARDS, respectively. Patients with a charted diagnosis of diabetes were not more likely to develop ARDS (VALID: 24.6% ARDS in non-diabetics vs. 30.0% in diabetics, p=0.14; EARLI: 19.6% vs. 22.8%, p=0.55). However, in VALID, patients categorized as diabetic with inadequate glycemic control based on their HbA1c had an increased risk of developing ARDS compared to those with a non-diabetic HbA1c (20.9% vs. 34.0%; p=0.0073), a finding that persisted in multivariable analysis (OR for diabetic with inadequate glycemic control vs. non-diabetic range HbA1c = 1.25, 95% CI 1.01-1.57). These findings were not reproduced in the smaller EARLI cohort, but were appreciated when the cohorts were combined for analysis.InterpretationElevated HbA1c may be associated with risk of developing ARDS, independent of clinical diagnosis of diabetes, but prospective validation is needed. If confirmed, these findings suggest that inadequate glycemic control could be an unrecognized risk factor for ARDS.

"Host Response to Infection: Not All Lymphopenia Is Created Equal in SARS-CoV-2." American Journal of Respiratory and Critical Care Medicine, 0(ja), pp.

Objectives:Phenotyping is a core task in observational health research utilizing electronic health records (EHRs). Developing an accurate algorithm demands substantial input from domain experts, involving extensive literature review and evidence synthesis. This burdensome process limits scalability and delays knowledge discovery. We investigate the potential for leveraging large language models (LLMs) to enhance the efficiency of EHR phenotyping by generating high-quality algorithm drafts. Materials and Methods:We prompted four LLMs-GPT-4 and GPT-3.5 of ChatGPT, Claude 2, and Bard-in October 2023, asking them to generate executable phenotyping algorithms in the form of SQL queries adhering to a common data model (CDM) for three phenotypes (i.e., type 2 diabetes mellitus, dementia, and hypothyroidism). Three phenotyping experts evaluated the returned algorithms across several critical metrics. We further implemented the top-rated algorithms and compared them against clinician-validated phenotyping algorithms from the Electronic Medical Records and Genomics (eMERGE) network. Results:GPT-4 and GPT-3.5 exhibited significantly higher overall expert evaluation scores in instruction following, algorithmic logic, and SQL executability, when compared to Claude 2 and Bard. Although GPT-4 and GPT-3.5 effectively identified relevant clinical concepts, they exhibited immature capability in organizing phenotyping criteria with the proper logic, leading to phenotyping algorithms that were either excessively restrictive (with low recall) or overly broad (with low positive predictive values). Conclusion:GPT versions 3.5 and 4 are capable of drafting phenotyping algorithms by identifying relevant clinical criteria aligned with a CDM. However, expertise in informatics and clinical experience is still required to assess and further refine generated algorithms.