Sauver, Jennifer L. St.

ABSTRACT Background Mobility functional status is inconsistently recorded in electronic health records. Recent advances in Large Language Models (LLMs) enable automated extraction of functional information from unstructured clinical notes. Leveraging a validated functional LLM with high performance in mobility extraction, we aimed to study the association between LLM‐derived mobility functional status and clinical outcomes, including emergency department (ED) visits, hospitalizations, and mortality. Methods Individuals aged 60 years and older enrolled in the Mayo Clinic Study of Aging, with documented clinical notes in 2022 were included. Individuals without any identifiable functional information were excluded. The LLM‐derived mobility functional status categorized mobility as impaired or unimpaired, across the five mobility domains as classified by International Classification of Functioning, Disability, and Health. Clinical outcomes in 2023 were extracted from the Rochester Epidemiology Project database. Logistic regression models assessed associations between mobility impairment and outcomes, adjusted for age, sex, Elixhauser comorbidity index, and education. Results Three thousand three hundred thirty‐seven individuals were screened, of whom three hundred forty‐three had no functional information. Two thousand nine hundred ninety‐four individuals were analyzed. 79.4% had at least one mobility impairment. Higher mobility impairment is associated with older age and greater Elixhauser comorbidity. Individuals with 3 or more impairments had an increased adjusted odds of ED visits (Adjusted Odds Ratio (aOR): 2.18 (1.64, 2.90)) and hospitalizations (aOR: 2.35 (1.51, 3.65)) compared to unimpaired individuals ( p < 0.001), while mortality did not demonstrate a significant association with mobility impairment. Conclusion LLM‐derived mobility functional status assessment was associated with ED visits and hospitalization, but not with all‐cause mortality after adjustment. Prospective validation and evaluation of incremental discrimination, calibration, and clinical utility are needed before clinical implementation.

Background Alzheimer's disease (AD) and related dementias are leading causes of death worldwide. While socioeconomic status (SES) is a critical determinant, its impact on mortality after diagnosis remains understudied. Objective To investigate the roles of neighborhood-level and individual-level SES in predicting mortality risk among older adults with incident dementia. Methods We analyzed data from 924 participants with incident dementia from the Mayo Clinic Study of Aging (N = 6909), a cohort with clinical follow-up visits every 15 months. We assessed mortality risk using Cox's time-varying survival regression, adjusting for time-fixed and time-varying covariates. Primary predictors were neighborhood-level SES (Area Deprivation Index [ADI]) and individual-level SES (housing-based SES index [HOUSES], education, and occupation). We also applied a random survival forest model to assess the predictive contribution of SES indicators and identify influential predictors. Results 857 participants (92.7%) died, with a median time from dementia diagnosis to death of 29 months (IQR: 12–54). Lower SES measured by HOUSES was associated with increased mortality risk (HR 1.28, 95% CI 1.07–1.52). Conversely, ADI, education, and occupation were not associated with mortality. Male sex, older age, congestive heart failure and diabetes were associated with higher mortality risk. In predictive analysis, HOUSES ranked among the top 10 contributors to mortality risk. Conclusions Among older adults with dementia, lower HOUSES was associated with higher mortality risk beyond neighborhood-level deprivation, education, and occupation. Integrating this housing-based individual-level SES measure into dementia care planning may help identify patients with greater socioeconomic vulnerability and inform mortality risk stratification.

Abstract Objective This study evaluated sequential multi-site fine-tuning of large language models (LLMs), simulating incremental deployment across institutions for extracting mobility functional status from unstructured clinical notes. Materials and Methods We assembled a corpus of 600 clinical notes from 3 institutions, yielding 3810 annotated note sections (1200 training; 2610 testing). Building upon our previously established LLM mobility extraction pipeline, which utilized optimized prompt engineering, we applied parameter-efficient low-rank adaptation (LoRA) fine-tuning. We evaluated sequential multi-site fine-tuning, where the model is trained locally and only the updated LoRA weights are transferred to the next site. We compared this against joint fine-tuning, which requires pooling all multi-site patient data centrally before training. Performance was measured by micro F1-scores for 2 distinct subtasks: mobility extraction and impairment classification. Results Sequential multi-site fine-tuning yielded comparable results to centralized joint fine-tuning for mobility extraction (F1, 0.861 vs 0.894) and impairment classification (F1, 0.912 vs 0.900). Both sequential and joint fine-tuning outperformed an untuned 70B model. Additionally, single-site LoRA models showed robust cross-site generalization. Discussion Sequential multi-site fine-tuning offers a resource-friendly approach to adapt LLMs to clinical environments. By retaining previously acquired knowledge without severe interference, this strategy allows for local fine-tuning at each institution without the need to pool or share sensitive patient data required for joint fine-tuning. Conclusion Sequential multi-site fine-tuning provides a viable solution for the scalable, incremental deployment of LLMs across institutions, showing promising generalizability for clinical information extraction from electronic health record narratives.

Background Area‐based deprivation indices have been linked to hypertension and diabetes outcomes, but limited information exists about their relative performance. We compared 2 area‐based deprivation measures (Area Deprivation Index and Yost Index) as risk factors for hypertension and diabetes control. Methods This cohort study identified adults with hypertension or diabetes in a 19‐county region of Southeast Minnesota covered by the Rochester Epidemiology Project. The deprivation indices were operationalized as quintiles for state and national ranks. We assessed agreement using the weighted kappa statistic. We used modified Poisson regression to evaluate associations of the deprivation indices with blood pressure or diabetes control in 2022. We compared relative model fit using the area under the receiver operating characteristic curve. Results We identified 67 386 individuals with hypertension and 30 435 with diabetes. There was moderate to substantial agreement between indices, but the agreement was higher for the state rank (weighted kappa 0.68 for both cohorts) than the national rank (0.54 for both). Although the indices were not associated with hypertension control, greater deprivation was associated with worse diabetes control for the Yost Index (relative risk [RR], 1.21 [95% CI, 1.07–1.38]) and Area Deprivation Index (RR, 1.39 [95% CI, 1.12–1.73]) national ranks. The full‐model area under the receiver operating characteristic curve was similar across indices but differed across cohorts with worse fit for hypertension (0.53 for both) than diabetes (0.63 for both). Conclusions Two common area‐based deprivation indices had similar strength of association with diabetes control, suggesting that either index would be appropriate for researching and developing interventions for diabetes control.

Introduction People most likely to be affected by Alzheimer’s disease and related dementias (ADRD) are underrepresented in clinical research. Identifying and addressing factors which may influence their research participation is important to advance science and promote diversity. We examined the influence of social determinants of health (SDOH) on research participation among people living with ADRD. Methods We included those age 55 years and older living with ADRD, followed in our community practice in Rochester, Minnesota, who completed a questionnaire assessing SDOH and had at least one clinic visit between June 1, 2019 and June 1, 2020. Demographic and SDOH factors were examined by research participation. Results Among 3273 people included, research participation was low overall (265/3273; 8.1%). People participating in research were more likely to be socially integrated and of higher educational attainment (p < 0.05). Conclusion Promoting social interaction is a potential area for focus to increase research participation in at risk populations.

Abstract Background Integrating advanced artificial intelligence (AI) into clinical decision-support often requires the sharing of sensitive patient data with external services, raising privacy concerns. Homomorphic encryption (HE) allows computing directly on encrypted data, without revealing the underlying patient information. Objectives To develop a large language model (LLM)-assisted diagnosis framework while preserving patient privacy in the clinical text analysis, by leveraging HE and using rare disease (RD) diagnosis as a representative application. To demonstrate HE does not hinder the system performance. Materials and Methods Texts from patient histories and a RD knowledge base were embedded by LLMs into vectors, then encrypted using HE to obscure private information while retaining the semantic nuances. Diagnostic recommendations were generated by computing and ranking the similarities between the patient history and RD vectors in the encrypted space. The system was evaluated using 50 synthetic case reports (5 RDs, each with 10 reports). Results Applying HE did protect private information from reverse-embedding attacks. HE imposed little disruption to the diagnostic accuracy, with normalized discounted cumulative gains (nDCG) of 0.6108 ± 0.3412 (encrypted) versus 0.6083 ± 0.3415 (unencrypted). The accuracy and computational performance were tunable and consistent, as demonstrated across five different LLMs. Discussion Our privacy-preserving framework opens tremendous opportunities toward hosting and serving powerful AI solutions across institution boundaries, which would remove the need for local deidentification and incentivize users to access secure external decision-support services. Conclusions Integrating HE with LLM retrieval can promote the dissemination of nonredundant, high-capacity AI services by preserving both privacy and accuracy.