BACKGROUND:Prognostication for people with Alzheimer's disease (AD) at the point of care could improve clinical management. METHODS:In this retrospective cohort study using the electronic health record (EHR) data from a large healthcare system (2011-2022), we applied an unsupervised latent factor clustering approach guided by knowledge graph embeddings to stratify AD patients into two groups at diagnosis (baseline) using clinical features in the two years preceding diagnosis. We prognosticated the risk of AD-related outcomes (nursing home admission and mortality) for the clusters in survival analyses adjusted for baseline confounders (age, gender, race, ethnicity, healthcare utilization, and comorbidities). To reflect real-world evolution in clinical trajectories, we updated patient stratification for patients remaining at risk one year post-diagnosis and repeated prognostication. RESULTS:We stratify 16,411 AD patients into two groups at baseline (41% Group 1, 59% Group 2). Baseline Group 2 has a significantly lower risk of nursing home admission (HR [95% CI] = 0.804 [0.765, 0.844], p < .001) but comparable mortality risk to baseline Group 1 (HR [95% CI] = 1.008 [0.963, 1.056], p = 0.733). We re-stratify the 12,606 patients remaining at risk one year post-diagnosis (46% Group 1, 54% Group 2). Consistent with baseline, the updated Group 2 has a lower risk of nursing home admission (HR [95% CI] = 0.815 [0.766, 0.868], p < .001) but comparable mortality risk (HR [95% CI] = 0.977 [0.922, 1.035], p = .430) to Group 1. CONCLUSIONS:Patient stratification enables outcome prognosis for AD patients. While baseline prognostication can guide early treatment and tailored management, dynamic prognostication may inform more timely interventions to improve long-term outcomes.

BACKGROUND:Alzheimer's disease (AD) carries a high societal burden inequitably distributed across demographic groups. Using real-world electronic health record (EHR) data with accurate population identification, we examine demographic differences and potentially modifiable drivers of AD decline. METHODS:Leveraging EHR data (1994-2022) from two large independent healthcare systems, we applied an unsupervised phenotyping algorithm to predict AD diagnosis and validated using gold-standard chart-reviewed and registry-derived diagnosis labels. Among patients with ≥24 months of EHR data not living in nursing homes pre-AD diagnosis, we estimated the time-to-decline (nursing home admission, death) in healthcare system-specific covariate-adjusted competing risk survival analyses stratified by demographic groups. We then performed covariate-adjusted fixed-effects meta-analyses using inverse variance weighting. RESULTS:The algorithm demonstrates robust performance in identifying AD populations across healthcare systems and demographic groups (AUROC score range: 0.835-0.923). Of the 29,262 AD patients in both healthcare systems (61% women, 90% non-Hispanic White, 79.52 ± 9.39 years of age at AD diagnosis), 49% transition to nursing homes and 52% die during follow-up. In covariate-adjusted fixed-effects meta-analysis, women have higher nursing home admission risk (HR [95% CI] = 1.061 [1.024-1.100], p = 1.203×10-3) but lower death risk (HR [95% CI] = 0.856 [0.811-0.904], p = 2.434×10-8) than men. Non-Hispanic White individuals have similar nursing home risk (HR [95% CI] = 1.006 [0.952-1.063], p = 8.306×10-1) but higher death risk (HR [95% CI] = 1.376 [1.245-1.521], p = 4.084×10-10) than racial and ethnic minorities. Older age at AD diagnosis and greater comorbidity burden increase both nursing home admission and death risk. CONCLUSIONS:We provide real-world evidence of drivers of demographic differences in AD decline that could inform individual clinical management and public health policies.

The widespread adoption of electronic health records has created new opportunities for translational clinical research, yet this promise remains constrained by fragmented data across privacy-siloed institutions and substantial heterogeneity in local coding practices. While privacy-preserving collaborative learning allows institutions to work together without sharing patient-level data, it does not address inconsistencies in how clinical concepts are represented across sites. We introduce a graph-based framework that addresses this gap by treating data harmonization as a scalable representation learning problem. Rather than relying on fixed standards or manual mappings, the framework integrates institution-specific summary statistics from health records, curated biomedical knowledge graphs, and semantic information derived from large language models to learn a shared semantic space. This joint learning approach aligns diverse, site-specific vocabularies while preserving patient privacy. Evaluated across seven institutions and two languages, the framework provides a robust, data-centric foundation for training and deploying clinical models across heterogeneous healthcare systems.

Despite the growing availability of Electronic Health Record (EHR) data, researchers often face substantial barriers in effectively using these data for translational research due to their complexity, heterogeneity, and lack of standardized tools and documentation. To address this critical gap, we introduce PEHRT, a common pipeline for harmonizing EHR data for translational research. PEHRT is a comprehensive, ready-to-use resource that includes open-source code, visualization tools, and detailed documentation to streamline the process of preparing EHR data for analysis. The pipeline provides tools to harmonize structured and unstructured EHR data to standardized ontologies to ensure consistency across diverse coding systems. In the presence of unmapped or heterogeneous local codes, PEHRT further leverages representation learning and pre-trained language models to generate robust embeddings that capture semantic relationships across sites to mitigate heterogeneity and enable integrative downstream analyses. PEHRT also supports cross-institutional co-training through shared representations, allowing participating sites to collaboratively refine embeddings and enhance generalizability without sharing individual-level data. The framework is data model-agnostic and can be seamlessly deployed across diverse healthcare systems to produce interoperable, research-ready datasets. By lowering the technical barriers to EHR-based research, PEHRT empowers investigators to transform raw clinical data into reproducible, analysis-ready resources for discovery and innovation.

Alzheimer's disease (AD) carries a high societal burden that is inequitably distributed across demographic groups. The objective of this study is to examine differences in readily ascertainable outcomes of AD decline by race and ethnicity and by gender. Using electronic health record (EHR) data from two large healthcare systems spanning 1994 to 2022, we identified patients with at least one diagnosis code for AD or related dementia. We applied an unsupervised phenotyping algorithm at each site to predict AD diagnosis status, which was validated with gold-standard chart-reviewed and registry-derived diagnosis labels. After excluding patients with less than 24 months of data or who were admitted to nursing homes prior to AD diagnosis, we examined the time to two outcomes of AD decline, nursing home admission and death, in survival analyses stratified by demographic groups. We accounted for baseline covariates (age, gender, race, ethnicity, healthcare utilization, and comorbidities) in the survival analysis. We then performed a fixed-effects meta-analysis of the survival data from both healthcare systems. The phenotyping algorithm demonstrated high accuracy in identifying AD patients across both healthcare systems (AUROC score range: 0.835-0.923). Of the 34,181 AD patients included (62% women, 90% non-Hispanic White, 80.39±9.28 years of age at AD diagnosis), 32% were admitted to nursing homes and 50% died. In the fixed-effect meta-analysis, non-Hispanic White patients had a lower risk of nursing home admission (HR[95% CI]=0.825[0.776-0.877], p  <0.001) and higher risk of death (HR[95% CI]=1.381[1.283-1.487], p  < .0001) than racial and ethnic minorities. There was no difference between women and men in the risk of nursing home admission (HR[95% CI]=1.008[0.967-1.050], p  = .762), but women had a lower risk of death (HR[95% CI]=0.873[0.837-0.910], p  < .0001) than men. Racial and ethnic minorities with AD may have a higher risk of nursing home admission, whereas non-Hispanic White patients and men with AD may have a higher risk of death. Future investigation of demographic disparity in AD decline can inform clinical management and public health policies.

Electronic Health Records (EHR) offer rich real-world data for personalized medicine, providing insights into disease progression, treatment responses, and patient outcomes. However, their sparsity, heterogeneity, and high dimensionality make them difficult to model, while the lack of standardized ground truth further complicates predictive modeling. To address these challenges, we propose SCORE, a semi-supervised representation learning framework that captures multi-domain disease profiles through patient embeddings. SCORE employs a Poisson-Adapted Latent factor Mixture (PALM) Model with pre-trained code embeddings to characterize codified features and extract meaningful patient phenotypes and embeddings. To handle the computational challenges of large-scale data, it introduces a hybrid Expectation-Maximization (EM) and Gaussian Variational Approximation (GVA) algorithm, leveraging limited labeled data to refine estimates on a vast pool of unlabeled samples. We theoretically establish the convergence of this hybrid approach, quantify GVA errors, and derive SCORE's error rate under diverging embedding dimensions. Our analysis shows that incorporating unlabeled data enhances accuracy and reduces sensitivity to label scarcity. Extensive simulations confirm SCORE's superior finite-sample performance over existing methods. Finally, we apply SCORE to predict disability status for patients with multiple sclerosis (MS) using partially labeled EHR data, demonstrating that it produces more informative and predictive patient embeddings for multiple MS-related conditions compared to existing approaches.

OBJECTIVE:Rheumatoid arthritis (RA) is a heterogeneous disease, with patients experiencing varied disease courses and responses to treatment. The objective of this study was to apply topic modeling to RA patient electronic health record (EHR) data and determine (1) the clinical topics/subgroups in those with RA before initiation of a biologic/targeted synthetic disease-modifying antirheumatic drug (b/tsDMARD) and (2) whether the clinical topics were associated with subsequent RA treatment course. METHODS:We studied patients from a validated EHR-based RA cohort who initiated a b/tsDMARD between 2011 and 2019. Diagnoses codes, laboratory data, and medication prescriptions in the year before their first b/tsDMARD initiation were extracted. Latent Dirichlet allocation, a topic modeling method, was applied to define the underlying "topics" representing clinical subgroups. We used multinomial regression to test association between the clinical topic with four previously published treatment trajectories: tumor necrosis factor inhibitor (TNFi) persisters, TNFi to abatacept, and those prescribed multiple b/tsDMARDs enriched with tocilizumab or rituximab. RESULTS:From the data of 1,102 patients with RA, diagnoses codes, laboratory data, and prescriptions from the year before b/tsDMARD initiation resulted in four main clinical topics/subgroups: (A) RA codes/methotrexate (MTX), (B) arthralgia/osteoarthritis, (C) hypertension (HTN)/cardiovascular (CV) comorbidities, and (D) mood disorders. Those with RA codes/MTX topic were more likely to persist on TNFi. Conversely, those associated with the HTN/CV topic were more likely to cycle through multiple b/tsDMARDs. CONCLUSION:Clinical topics derived from the EHR data of patients with RA before b/tsDMARD differentiated future RA treatment course. HTN/CV comorbidities were associated with a future need for multiple b/tsDMARD therapies.

In many modern machine learning applications, changes in covariate distributions and difficulty in acquiring outcome information have posed challenges to robust model training and evaluation. Numerous transfer learning methods have been developed to robustly adapt the model itself to some unlabeled target populations using existing labeled data in a source population. However, there is a paucity of literature on transferring performance metrics, especially receiver operating characteristic (ROC) parameters, of a trained model. In this paper, we aim to evaluate the performance of a trained binary classifier on unlabeled target population based on ROC analysis. We proposed Semisupervised Transfer lEarning of Accuracy Measures (STEAM), an efficient three-step estimation procedure that employs (1) double-index modeling to construct calibrated density ratio weights and (2) robust imputation to leverage the large amount of unlabeled data to improve estimation efficiency. We establish the consistency and asymptotic normality of the proposed estimator under the correct specification of either the density ratio model or the outcome model. We also correct for potential overfitting bias in the estimators in finite samples with cross-validation. We compare our proposed estimators to existing methods and show reductions in bias and gains in efficiency through simulations. We illustrate the practical utility of the proposed method on evaluating prediction performance of a phenotyping model for rheumatoid arthritis (RA) on a temporally evolving EHR cohort.

We leveraged real-world data from electronic health records (EHR) to examine differences in Alzheimer's disease (AD) progression among gender, race, and ethnicity groups.

Though electronic health record (EHR) systems are a rich repository of clinical information with large potential, the use of EHR-based phenotyping algorithms is often hindered by inaccurate diagnostic records, the presence of many irrelevant features, and the requirement for a human-labeled training set. In this paper, we describe a knowledge-driven online multimodal automated phenotyping (KOMAP) system that i) generates a list of informative features by an online narrative and codified feature search engine (ONCE) and ii) enables the training of a multimodal phenotyping algorithm based on summary data. Powered by composite knowledge from multiple EHR sources, online article corpora, and a large language model, features selected by ONCE show high concordance with the state-of-the-art AI models (GPT4 and ChatGPT) and encourage large-scale phenotyping by providing a smaller but highly relevant feature set. Validation of the KOMAP system across four healthcare centers suggests that it can generate efficient phenotyping algorithms with robust performance. Compared to other methods requiring patient-level inputs and gold-standard labels, the fully online KOMAP provides a significant opportunity to enable multi-center collaboration.

Background: T2* relaxation times in the spinal cartilage endplate (CEP) measured using ultra-short echo time magnetic resonance imaging (UTE MRI) reflect aspects of biochemical composition that influence the CEP's permeability to nutrients. Deficits in CEP composition measured using T2* biomarkers from UTE MRI are associated with more severe intervertebral disc degeneration in patients with chronic low back pain (cLBP). The goal of this study was to develop an objective, accurate, and efficient deep-learning-based method for calculating biomarkers of CEP health using UTE images. Methods: Multi-echo UTE MRI of the lumbar spine was acquired from a prospectively enrolled crosssectional and consecutive cohort of 83 subjects spanning a wide range of ages and cLBP-related conditions. CEPs from the L4-S1 levels were manually segmented on 6,972 UTE images and used to train neural networks utilizing the u-net architecture. CEP segmentations and mean CEP T2* values derived from manually- and model-generated segmentations were compared using Dice scores, sensitivity, specificity, Bland-Altman, and receiver-operator characteristic (ROC) analysis. Signal-to-noise (SNR) and contrast-tonoise (CNR) ratios were calculated and related to model performance. Results: Compared with manual CEP segmentations, model-generated segmentations achieved sensitives of 0.80-0.91, specificities of 0.99, Dice scores of 0.77-0.85, area under the receiver-operating characteristic curve values of 0.99, and precision-recall (PR) AUC values of 0.56-0.77, depending on spinal level and sagittal image position. Mean CEP T2* values and principal CEP angles derived from the model-predicted segmentations had low bias in an unseen test dataset (T2* bias =0.33 +/- 2.37 ms, angle bias =0.36 +/- 2.65 degrees). To simulate a hypothetical clinical scenario, the predicted segmentations were used to stratify CEPs into high, medium, and low T2* groups. Group predictions had diagnostic sensitivities of 0.77-0.86 and specificities of Conclusions: The trained deep learning models enable accurate, automated CEP segmentations and T2*

Purpose The composition of the subchondral bone marrow and cartilage endplate (CEP) could affect intervertebral disc health by influencing vertebral perfusion and nutrient diffusion. However, the relative contributions of these factors to disc degeneration in patients with chronic low back pain (cLBP) have not been quantified. The goal of this study was to use compositional biomarkers derived from quantitative MRI to establish how CEP composition (surrogate for permeability) and vertebral bone marrow fat fraction (BMFF, surrogate for perfusion) relate to disc degeneration. Methods MRI data from 60 patients with cLBP were included in this prospective observational study (28 female, 32 male; age = 40.0 ± 11.9 years, 19–65 [mean ± SD, min–max]). Ultra-short echo-time MRI was used to calculate CEP T2* relaxation times (reflecting biochemical composition), water-fat MRI was used to calculate vertebral BMFF, and T1ρ MRI was used to calculate T1ρ relaxation times in the nucleus pulposus (NP T1ρ, reflecting proteoglycan content and degenerative grade). Univariate linear regression was used to assess the independent effects of CEP T2* and vertebral BMFF on NP T1ρ. Mixed effects multivariable linear regression accounting for age, sex, and BMI was used to assess the combined relationship between variables . Results CEP T2* and vertebral BMFF were independently associated with NP T1ρ ( p = 0.003 and 0.0001, respectively). After adjusting for age, sex, and BMI, NP T1ρ remained significantly associated with CEP T2* ( p = 0.0001) but not vertebral BMFF ( p = 0.43). Conclusion Poor CEP composition plays a significant role in disc degeneration severity and can affect disc health both with and without deficits in vertebral perfusion.

Opal is the first published example of a full-stack platform infrastructure for an implementation science designed for ML in anesthesia that solves the problem of leveraging ML for clinical decision support. Users interact with a secure online Opal web application to select a desired operating room (OR) case cohort for data extraction, visualize datasets with built-in graphing techniques, and run in-client ML or extract data for external use. Opal was used to obtain data from 29,004 unique OR cases from a single academic institution for pre-operative prediction of post-operative acute kidney injury (AKI) based on creatinine KDIGO criteria using predictors which included pre-operative demographic, past medical history, medications, and flowsheet information. To demonstrate utility with unsupervised learning, Opal was also used to extract intra-operative flowsheet data from 2995 unique OR cases and patients were clustered using PCA analysis and k-means clustering. A gradient boosting machine model was developed using an 80/20 train to test ratio and yielded an area under the receiver operating curve (ROC-AUC) of 0.85 with 95% CI [0.80–0.90]. At the default probability decision threshold of 0.5, the model sensitivity was 0.9 and the specificity was 0.8. K-means clustering was performed to partition the cases into two clusters and for hypothesis generation of potential groups of outcomes related to intraoperative vitals. Opal’s design has created streamlined ML functionality for researchers and clinicians in the perioperative setting and opens the door for many future clinical applications, including data mining, clinical simulation, high-frequency prediction, and quality improvement.

Cartilage endplate (CEP) biochemical composition may influence disc degeneration and regeneration. However, evaluating CEP composition in patients remains a challenge. We used T2* mapping from ultrashort echo-time (UTE) magnetic resonance imaging (MRI), which is sensitive to CEP hydration, to investigate spatial variations in CEP T2* values and to determine how CEP T2* values correlate with adjacent disc degeneration. Thirteen human cadavers (56.4 +/- 12.7 years) and seven volunteers (36.9 +/- 10.9 years) underwent 3T MRI, including UTE and T1 rho mapping sequences. Spatial mappings of T2* values in L4-S1 CEPs were generated from UTE images and compared between subregions. In the abutting discs, mean T1 rho values in the nucleus pulposus were compared between CEPs with high vs low T2* values. To assess in vivo repeatability, precision errors in mean T2* values, and intraclass correlation coefficients (ICC) were measured from repeat scans. Results showed that CEP T2* values were highest centrally and lowest posteriorly. In the youngest individuals (<50 years), who had mild-to-moderately degenerated Pfirrmann grade II-III discs, low CEP T2* values associated with severer disc degeneration: T1 rho values were 26.7% lower in subjects with low CEP T2* values (P = .025). In older individuals, CEP T2* values did not associate with disc degeneration (P = .39-.62). Precision errors in T2* ranged from 1.7 to 2.6 ms, and reliability was good-to-excellent (ICC = 0.89-0.94). These findings suggest that deficits in CEP composition, as indicated by low T2* values, associate with severer disc degeneration during the mild-to-moderate stages. Measuring CEP T2* values with UTE MRI may clarify the role of CEP composition in patients with mild-to-moderate disc degeneration.

Electron cryo-tomography allows for high-resolution imaging of stereocilia in their native state. Because their actin filaments have a higher degree of order, we imaged stereocilia from mice lacking the actin crosslinker plastin 1 (PLS1). We found that while stereocilia actin filaments run 13 nm apart in parallel for long distances, there were gaps of significant size that were stochastically distributed throughout the actin core. Actin crosslinkers were distributed through the stereocilium, but did not occupy all possible binding sites. At stereocilia tips, protein density extended beyond actin filaments, especially on the side of the tip where a tip link should anchor. Along the shaft, repeating density was observed that corresponds to actin-to-membrane connectors. In the taper region, most actin filaments terminated near the plasma membrane. The remaining filaments twisted together to make a tighter bundle than was present in the shaft region; the spacing between them decreased from 13 nm to 9 nm. Our models illustrate detailed features of distinct structural domains that are present within the stereocilium.