Jennifer Hadlock

While large language models are powerful generators of new text, forecasting disease progression from longitudinal health histories remains a challenging problem. We introduce GenEHR, an autoregressive generative model trained on electronic health records (EHRs) from millions of patients in diverse health systems. The model learns from real-world multimodal patient histories, including diagnoses, medications, lab values, and procedures while explicitly representing irregular intervals between visits to forecast future clinical events. We combine the general-purpose patient representation learned during foundational training with parameter-efficient supervised adaptation for the task of risk stratification for 17 cancer types. In five large EHR cohorts, supervised adaptation substantially improved prediction performance of a first cancer diagnosis within a five year horizon window. Our retrospective results support the prospective evaluation of GenEHR-CancerRisk as a clinical decision-support tool for prioritizing high-risk patients for screening and prevention of aggressive cancer types, such as pancreatic and ovarian cancer.

T-cells are central to SARS-CoV-2 clearance and immunological memory, yet their contribution to the persistence of post-acute sequelae of COVID-19 (PASC) remains poorly understood. The immunological features that distinguish individuals who develop PASC from those who recover fully are unresolved, in part due to the phenotypic heterogeneity of the condition and the likely multiplicity of its underlying mechanisms. Here, we profiled longitudinal bulk TCR β repertoires from 120 individuals in the INCOV cohort–71 with PASC and 49 without–sampled at two to three time points spanning the acute and post-acute phases of infection. Using robust statistical modeling of repertoire composition and clonal dynamics, we found that global statistics such as V, J gene usage and CDR3 length do not differ between groups, but that locally enriched sequence motifs and differentially dynamic clones reveal distinct T-cell signatures associated with PASC status. Clones contracting following the peak of the acute response were significantly enriched for SARS-CoV-2 specificity in both groups. Interestingly, Influenza A-specific TCRs were disproportionately enriched among contracting clones in PASC ⁺ repertoires, implicating viral co-infection as a potential contributor to early disease severity and, possibly, PASC pathogenesis. Rare public TCR clones were markedly enriched for SARS-CoV-2 specificity, with PASC ⁺ individuals harboring a modestly but significantly higher proportion than PASC ⁻ individuals. Together, we identified over 1,000 candidate TCR β receptors potentially discriminating PASC ⁺ from PASC ⁻ immune responses, opening a path toward the identification of disease-relevant T-cell specificities and the development of T-cell-based immunological biomarkers for long COVID.

Introduction: The complete blood count with differential (CBD) is one of the most commonly performed blood tests worldwide, used in nearly all areas of medicine. Although modern CBD analyzers generate flow-cytometry based single-cell measurements, the resultant CBD markers are limited to coarse summary features, such as total cell counts and average cell sizes. This means, the markers cannotdetect subtle cell population shifts that may signal early-stage pathogenesis. To test this, we evaluate whether AI-based analysis of the raw single-cell data underlying the CBD can be used to develop novel, clinically prognostic biomarkers, across patient settings. Method: We developed two complementary methods for biomarker discovery using CBD tests and evaluated them with longitudinal data from an academic medical center. To create interpretable biomarkers, we clustered cells into physiologically meaningful sub-populations and performed robust statistical summarization. In tandem, self-supervised autoencoders were developed to extract novel non-linear markers. We evaluated the utility of these clustering (CLS) and autoencoder (AE) markers for patient prognostication across a range of outcomes (mortality, inpatient admission, and future disease development). Results: Our study included 242,623 CBD samples from 127,545 patients. Both clustering and embedding approaches successfully generated hundreds of new clinical biomarkers. Many biomarkers showed strong prognostic associations for all-cause mortality, inpatient admission, and development of anemia, cancer, or cardiovascular disease, with associations remaining significant after adjustment for demographics and clinical CBD markers. A large subset of these prognostic markers also showed high novelty – having low correlations to existing CBD markers, while also exhibiting significant correlations with broader physiologic signals, such as inflammatory, hormonal, infectious, and coagulopathic markers. Conclusion: Collectively, these results demonstrate how modern AI techniques can allow for deeper phenotyping of routine clinical blood counts, generating novel biomarkers that capture more subtle physiologic signals than what are currently clinically utilized.

Background: This study evaluated the impact of aging on the frequency and prevalent symptoms of Long COVID, also termed post‐acute sequelae of SARS‐CoV‐2, using a previously developed Long COVID research index (LCRI) of 41 self‐reported symptoms in which those with 12 or more points were classified as likely to have Long COVID. Methods: We analyzed community‐dwelling participants ≥ 60 years old (2662 with prior infection, 461 controls) compared to participants 18–59 years (7549 infected, 728 controls) in the Researching COVID to Enhance Recovery adult (RECOVER‐Adult) cohort ≥ 135 days post‐onset. Results: Compared to the Age 18–39 group, the adjusted odds of LCRI ≥ 12 were higher for the Age 40–49 group (odds ratio [OR] = 1.40, 95% confidence intervals [CI] = 1.21–1.61, p < 0.001) and 50–59 group (OR = 1.31, CI = 1.14–1.51, p < 0.001), similar for the Age 60–69 group (OR = 1.09, CI = 0.93–1.27, p = 0.299), and lower for the ≥ 70 group (OR = 0.68, CI = 0.54–0.85, p < 0.001). Participants ≥ 70 years had smaller adjusted differences between infected and uninfected symptom prevalence rates than those aged 18–39 for the following symptoms: hearing loss, fatigue, pain (including joint, back, chest pain and headache), post‐exertional malaise, sleep disturbance, hair loss, palpitations, and sexual desire/capacity, making these symptoms less discriminating for Long COVID in older adults than in younger. Symptom clustering, as described in Thaweethai et al. (JAMA 2023) also exhibited age‐related shifts: clusters 1 (anosmia and ageusia) and 2 (gastrointestinal, chronic cough and palpitations, without anosmia, ageusia or brain fog) were more likely, and clusters 3 (brain fog, but no loss of smell or taste) and 4 (a mix of symptoms) less likely to be found in older adults (relative risk ratios for clusters 3–4 ranging from 0.10–0.34, p < 0.001 vs. 18–39 year‐olds). Conclusions: Within the limits of this observational study, we conclude that in community‐dwelling older adults, aging alters the prevalence and pattern of reported Long COVID.

The placenta is essential for pregnancy, and its dysfunction can harm both mother and fetus. To better understand placental physiology and its disruption in disease, we employ a multiomics approach (transcriptomics, metabolomics, and proteomics) combined with clinical data and histopathology from 321 placentas across conditions: severe fetal growth restriction (FGR), FGR with hypertension (FGR + HDP), severe preeclampsia (PE), and spontaneous preterm delivery (PTD). Cellular deconvolution reveals FGR + HDP placentas have more extravillous trophoblasts than controls (p < 0.0001). After adjusting for fetal sex and gestational age, we build condition-specific interomics networks and detect communities (a.k.a. subnetworks). In a control community, miR-365a-3p is the most connected node, whereas in FGR + HDP placentas, it is hypoxia-induced miR-210-3p. From this community, we identify a signature containing mRNAs implicated in placental dysfunction (e.g. FLT1, FSTL3, HTRA4, LEP, and PHYHIP), which distinguishes FGR + HDP placentas from those with other conditions, illustrating the power of interomics in understanding obstetric syndromes.

The growing availability of biomedical data offers vast potential to improve human health, but the complexity and lack of integration of these datasets often limit their utility. To address this, the Biomedical Data Translator Consortium has developed an open‐source knowledge graph–based system—Translator—designed to integrate, harmonize, and make inferences over diverse biomedical data sources. We announce here Translator's initial public release and provide an overview of its architecture, standards, user interface, and core features. Translator employs a scalable, federated, knowledge graph framework for the integration of clinical, genomic, pharmacological, and other biomedical knowledge sources, enabling query retrieval, inference, and hypothesis generation. Translator's user interface is designed to support the exploration of knowledge relationships and the generation of insights, without requiring deep technical expertise and gradually revealing more detailed evidence, provenance, and confidence information, as needed by a given user. To demonstrate Translator's application and impact, we highlight features of the user interface in the context of three real‐world use cases: suggesting potential therapeutics for patients with rare disease; explaining the mechanism of action of a pipeline drug; and screening and validating drug candidates in a model organism. We discuss strengths and limitations of reasoning within a largely federated system and the need for rich concept modeling and deep provenance tracking. Finally, we outline future directions for enhancing Translator's functionality and expanding its data sources. Translator represents a significant step forward in making complex biomedical knowledge more accessible and actionable, aiming to accelerate translational research and improve patient care.

SARS-CoV-2 infection has been associated with increased autoimmune disease risk. Past studies have not aligned regarding the most prevalent autoimmune diseases after infection, however. Furthermore, the relationship between infection severity and new autoimmune disease risk has not been well examined. We used RECOVER’s electronic health record (EHR) networks, N3C, PCORnet, and PEDSnet, to estimate types and frequency of autoimmune diseases arising after SARS-CoV-2 infection and assessed how infection severity related to autoimmune disease risk. We identified patients of any age with SARS-CoV-2 infection between April 1, 2020 and April 1, 2021, and assigned them to a World Health Organization COVID-19 severity category for adults or the PEDSnet acute COVID-19 illness severity classification system for children (30 days after SARS-CoV-2 infection index date and occurring ≥1 day apart. We calculated overall and infection severity-stratified incidence ratesper 1000 person-years for all autoimmune diseases. With least severe COVID-19 severity as reference, survival analyses examined incident autoimmune disease risk. The most common new-onset autoimmune diseases in all networks were thyroid disease, psoriasis/psoriatic arthritis, and inflammatory bowel disease. Among adults, inflammatory arthritis was the most common, and Sjögren’s disease also had high incidence. Incident type 1 diabetes and hematological autoimmune diseases were specifically found in children. Across networks, after adjustment, patients with highest COVID-19 severity had highest risk for new autoimmune disease vs. those with least severe disease (N3C: adjusted Hazard Ratio, (aHR) 1.47 (95%CI 1.33–1.66); PCORnet aHR 1.14 (95%CI 1.02–1.26); PEDSnet: aHR 3.14 (95%CI 2.42–4.07)]. Overall, severe acute COVID-19 was most strongly associated with autoimmune disease risk in three EHR networks.

Importance The opioid epidemic continues to grow, and while the adverse effects of opioids are well–known, their impact on short-term mortality in patients with dementia or mild cognitive impairment (MCI) remains understudied, despite the high vulnerability of this population. Objective To evaluate the short-term mortality risk associated with initiating opioids in patients diagnosed with dementia or MCI Design Retrospective observational cohort study Setting Multisite healthcare system in Northern California Participants Health records of 27,759 patients aged 50–100 with dementia or MCI, with encounters between January 1, 2015, and July 31, 2024. Exclusions included patients who died within 14 days of surgery, had fewer than three clinical encounters before and after diagnosis, or were first diagnosed with dementia/MCI at death. Exposures Initial opioid use following dementia or MCI onset. Patients were categorized as new users (no opioid use in the prior year) or consistent users (prior opioid exposure). Main Outcome and Measures Short–term mortality risk, defined as death within 14 days of first opioid exposure, with additional monitoring up to 60 days after opioid initiation. Hazard ratios were calculated using Cox proportional hazards regression, adjusting for demographics, comorbidities, and medication exposure. We used GPT.3.5–Turbo to identify possible causes of death from unstructured clinical documentation, supplemented by data from California public death records. Results Among 14,107 patients prescribed opioids following the onset of dementia/MCI onset, 9444 were new users and 4663 were consistent users. The cohort was predominantly female (56.0%), with a median age of 81 years (IQR:73–87). New users exhibited a 1.95–fold (95% CI, 1.55 – 2.46; P < 0.0001) increased risk of mortality within 14 days of initial opioid exposure compared to consistent users. Respiratory illnesses were more prevalent among new users who died within 14 days after opioid exposure (62% vs. 48%, PConclusion Opioid initiation in patients with dementia or MCI is linked to significantly higher short–term mortality, particularly in those with respiratory conditions. These findings highlight the need for cautious prescribing and close monitoring, given the increased mortality risk.

Background: Social determinants of health (SDoH) such as housing insecurity are known to be intricately linked to patients’ health status. More efficient methods for abstracting structured data on SDoH can help accelerate the inclusion of exposome variables in biomedical research and support health care systems in identifying patients who could benefit from proactive outreach. Large language models (LLMs) developed from Generative Pre-trained Transformers (GPTs) have shown potential for performing complex abstraction tasks on unstructured clinical notes. Objective: Here, we assess the performance of GPTs on identifying temporal aspects of housing insecurity and compare results between both original and deidentified notes. Methods: We compared the ability of GPT-3.5 and GPT-4 to identify instances of both current and past housing instability, as well as general housing status, from 25,217 notes from 795 pregnant women. Results were compared with manual abstraction, a named entity recognition model, and regular expressions. Results: Compared with GPT-3.5 and the named entity recognition model, GPT-4 had the highest performance and had a much higher recall (0.924) than human abstractors (0.702) in identifying patients experiencing current or past housing instability, although precision was lower (0.850) compared with human abstractors (0.971). GPT-4’s precision improved slightly (0.936 original, 0.939 deidentified) on deidentified versions of the same notes, while recall dropped (0.781 original, 0.704 deidentified). Conclusions: This work demonstrates that while manual abstraction is likely to yield slightly more accurate results overall, LLMs can provide a scalable, cost-effective solution with the advantage of greater recall. This could support semiautomated abstraction, but given the potential risk for harm, human review would be essential before using results for any patient engagement or care decisions. Furthermore, recall was lower when notes were deidentified prior to LLM abstraction.

As large clinical and multiomics datasets and knowledge resources accumulate, they need to be transformed into computable and actionable information to support automated reasoning. These datasets range from laboratory experiment results to electronic health records (EHRs). Barriers to accessibility and sharing of such datasets include diversity of content, size and privacy. Effective transformation of data into information requires harmonization of stakeholder goals, implementation, enforcement of standards regarding quality and completeness, and availability of resources for maintenance and updates. Systems such as the Biomedical Data Translator leverage knowledge graphs (KGs), structured and machine learning readable knowledge representation, to encode knowledge extracted through inference. We focus here on the transformation of data from multiomics datasets and EHRs into compact knowledge, represented in a KG data structure. We demonstrate this data transformation in the context of the Translator ecosystem, including clinical trials, drug approvals, cancer, wellness, and EHR data. These transformations preserve individual privacy. We provide access to the five resulting KGs through the Translator framework. We show examples of biomedical research questions supported by our KGs, and discuss issues arising from extracting biomedical knowledge from multiomics data.