Citations

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.

Abstract Approximately 17% of people aged 65 years and older are anemic, and 10% of death certificates report anemia as a secondary cause of death in the United States. Nonetheless, anemia remains unexplained in 30%-50% of older adults. This unexplained anemia of aging (UAA) is a diagnosis of exclusion. The mechanism, impact, and progression of UAA remain unknown. At older ages, anemia adds to pre-existing co-morbidities with significant adverse health consequences, representing a compelling unmet clinical concern. The National Institute on Aging held a workshop in 2024 to discuss current knowledge and research opportunities. Topics included the epidemiology of anemia at older age and its clinical implications; probable mechanism(s) underlying UAA, that is, low-grade inflammation’s effects on erythropoiesis; the role of microbiota in iron regulation in bone marrow; the importance of ruling out a diagnosis of leukemic clonal hematopoiesis (CH), which is more prevalent in older age; the role of senescence and aging governing hematopoiesis; and the effects of sex hormones on hematopoietic stem cell aging. Understanding the roles of these factors could reduce the proportion of the older anemic population whose anemia remains unexplained and offer insights into new potential diagnostic and intervention strategies. Speakers reviewed previous clinical trials in patients with UAA and CH. They discussed lessons learned and future research priorities, including efforts to develop new diagnostic algorithms and potential uses of machine learning.

Abstract Background Delirium is often underdiagnosed in clinical practice and is not routinely coded for billing. While manual chart review can identify delirium, it is labor-intensive and impractical for large-scale studies. Natural language processing (NLP) can analyze unstructured text in electronic health records (EHRs) to extract meaningful clinical information. Methods To support national integration of NLP for EHR-based delirium identification across different institutions, we launched the Delirium Interest Group within the national Evolve to Next-Gen Accrual to Clinical Trials (ENACT) NLP Working Group. This paper outlines our initial efforts to standardize, evaluate, and translate an NLP-based delirium detection model into the i2b2/ENACT platform. Results Multisite contextual inquiry identified several key challenges, including variations in local screening practices (eg, tools used, documentation frequency, and quality control), the need for harmonized definitions in the context of EHRs, and the complexity of modeling temporal logic. Multisite NLP evaluation revealed variable performance degradation driven by differences in delirium screening practices, clinical documentation patterns and semantics, and note syntactic structures. Conclusion Our work represents an important first step toward enabling scalable and standardized NLP-based delirium detection across institutions. By engaging diverse institutions through the ENACT NLP Working Group, we identified shared challenges and site-specific variations that impact model implementation and performance. Our collaborative approach enabled the development of a more robust framework for delirium identification across heterogeneous EHR systems. Future efforts will build on this foundation to enhance the validity, usability, and translational impact of delirium detection.

Objectives: To investigate the association between hormone therapy use and severe outcomes due to COVID-19 in perimenopausal and postmenopausal women. Methods: Using the Rochester Epidemiology Project, we retrospectively searched for the records of women aged 40-65 years who had a documented case of COVID-19 between March 12 and September 29, 2020. We further identified those who were perimenopausal or postmenopausal. The rates of emergency department visits, hospitalizations, and deaths were compared by hormone therapy use. Logistic regression models were used to assess the odds of the combined outcome based on hormone therapy use. Results: Among the 689 patients included in this study, the median age was 56 years, and 77.5% were White. Compared with patients currently using hormone therapy, those with past/never use of hormone therapy were significantly older, had more risk factors for severe COVID-19, and had a higher incidence of diabetes. A total of 102 emergency department visits/hospitalizations/deaths occurred among both groups. Participants with current hormone therapy use had similar rates and risk of emergency department/hospitalization/death outcomes as those without hormone therapy, even after adjusting for age and severe COVID-19 risk factors. Conclusions: Severe COVID-19 outcomes were not significantly different between women with current versus past/never use of hormone therapy. Our study adds to the data suggesting no harm with hormone therapy use in women with severe symptoms of COVID-19.

Abstract With global aging, assessing functional status is vital for precision medicine. Electronic Health Records (EHRs), particularly unstructured data, hold abundant information on patient mobility. This study explores using Large Language Models (LLMs) to extract and standardize mobility status from unstructured EHR data (i.e., clinical notes). We annotated 600 clinical notes from three health care institutions located in southeastern Minnesota and west-central Wisconsin, focusing on expressions of mobility and associated impairment. Leveraging the open-source Llama 3 model, we tested various prompting strategies—including zero-shot, few-shot, and task decomposition—and evaluated their performance. Error analysis showed that while the model sometimes inferred impairments without explicit evidence, most errors were clinically reasonable, often reflecting borderline or ambiguous cases. While considering reasonable inference as correct, at the patient-level, Mobility Extraction achieves a micro-average accuracy of 0.952 with an F1-score of 0.962, and Impairment Classification produces a micro-average accuracy of 0.912 and an F1-score of 0.948. A local, deterministic setup improved trustworthiness by ensuring consistent outputs, safeguarding privacy, and demonstrating cross-institution generalizability. These findings highlight the feasibility of LLM-based solutions for extracting mobility functional status from unstructured EHR data, supporting both clinical applications and research.

ImportanceUptake of human papillomavirus (HPV) vaccination varies by characteristics, exposing some children to higher HPV cancer risks than others.ObjectiveTo examine whether the effectiveness of a multilevel intervention on HPV vaccination differed by race and ethnicity, rurality, and Area Deprivation Index (ADI) in children ages 11 to 12 years.Design, Setting, and ParticipantsA stepped-wedge cluster randomized trial was conducted from April 2018 to August 2022 among children at 6 Mayo Clinic primary care practices in Minnesota to improve HPV vaccination. This secondary analysis was performed from March to June 2024.InterventionA multilevel intervention that included parent reminder/recall letters, which alerted parents of children due or past due for vaccination, and health care professional audit/feedback reports, which alerted health care professionals of their own vaccination rates.Main Outcome and MeasureVaccine initiation (first dose of the 2-dose HPV vaccine) and vaccine completion (second dose) were the primary study outcomes. In this secondary analysis, the effect of the intervention on HPV vaccine initiation and completion by race and ethnicity, rurality, and ADI quartiles (Qs) was assessed.ResultsA total of 6232 children aged 11 to 12 years (3285 [52.7%] male; 3481 [55.9%] aged 11 years and 2751 [44.1%] aged 12 years) were included in the analysis. Of the study participants, 304 (4.9%) were Asian, 561 (9.0%) Black, 146 (2.3%) Hispanic, 4501 (72.2%) White, and 720 (11.6%) other, including American Indian or Alaskan Native, Native Hawaiian or Pacific Islander, Other Pacific Islander, Samoan, unable to provide, unknown, chose not to disclose, or other unspecified. A total of 5434 participants (87.2%) were urban residents, and 2794 (44.8%) resided in ADI Q2 areas. With usual care, HPV vaccine initiation and completion rates were significantly lower with each increasing ADI quartile (initiation: Cochran-Armitage test for trend [SE], −0.02 [0.01]; P < .001; completion: Cochran-Armitage test for trend [SE], −0.05 [0.01]; P < .001) but did not differ by children’s race and ethnicity or rurality. With the intervention, vaccine initiation increased significantly for most children (range of rates, 9.2% [95% CI, 5.2%-13.3%] to 24.0% [95% CI, 7.5%-40.6%]) except those with Black race, in rural settings, and in ADI Q4 (highest area deprivation); vaccine completion increased significantly for most children (range of rates, 19.4% [95% CI, 5.5%-33.3%] to 31.2% [95% CI, 12.1%-50.3%]) except for those in ADI Q4.Conclusions and RelevanceIn this secondary analysis of a cluster randomized trial, a multilevel intervention was associated with increased HPV vaccination for most children but had limited effect for those residing in areas of highest deprivation. Future research should explore other intervention strategies that would effectively promote HPV vaccination among families in socioeconomically disadvantaged areas to reduce HPV vaccination disparities.Trial RegistrationClinicalTrials.gov Identifier: NCT03501992

AbstractINTRODUCTIONIdentifying individuals with dementia is crucial for prevalence estimation and service planning, but reliable, scalable methods are lacking. We developed novel set algorithms using both structured and unstructured electronic health record (EHR) data, applying Diagnostic and Statistical Manual of Mental Disorders criteria for dementia case identification.METHODSOur cohort (n = 1082) included individuals aged ≥ 60 with dementia identified through specialist clinics and a comparison group without dementia. Clinicians from Australia and the United States informed predictor selection. We developed algorithms through a biostatistics stream for structured data and a natural language processing (NLP) stream for text, synthesizing results via logistic regression.RESULTSThe final structured model retained 16 variables (area under the receiver operating characteristic curve [AUC] 0.853, specificity 72.2%, sensitivity 80.6%). NLP classifiers (logistic regression, support vector machine, and random forest models) performed comparably. The final, combined model outperformed all others (AUC = 0.951, P < 0.001 for comparison to structured model).DISCUSSIONEmbedding text‐derived insights within algorithms trained on structured medical data significantly enhances dementia identification capacity.Highlights Algorithmic tools for detection of individuals with dementia are available; however, previous work has used heterogeneous case definitions which are not clinically meaningful, and has relied on proxies such as diagnostic codes or medications for case ascertainment. We used a novel, dual‐stream algorithmic development approach, simultaneously and separately modeling a clinically meaningful outcome (diagnosis of dementia according to specialized clinical impression) using structured and unstructured electronic health record datasets. Our clinically grounded case definition supported the inclusion of key structured variables (such as dementia International Classification of Disease codes and medications) as modeling predictors rather than outcomes. Our algorithms, published in detail to support validation and replication, represent a major step forward in the use of routinely collected data for detection of diagnosed dementia.

Abstract Historically, cancers diagnosed via the emergency department (ED) portend a poor prognosis. Recent data from the United States are sparse, and analyses of cancers detected in the years following ED visits are lacking. Thus, we analyzed data from nine rural U.S. Midwest counties included within the population-based Rochester Epidemiology Project (2015–2021). Participants without a history of cancer (N = 42,074) who did not receive ED care were matched 1:1 to ED participants on the date of ED visit, age, sex, race, ethnicity, and county of residence. Analyses were restricted to participants with records ≤2 years prior to ED or index visit and ≥30 days after. HRs and 95% confidence intervals (CI) comparing cancer incidence and deaths among ED and non-ED participants were estimated from Cox proportional hazards regression models, either unadjusted or adjusted for covariates. Cumulative cancer incidence curves accounting for competing risks of death and survival (all cause and cancer-specific) were estimated. The median follow-up was 6.3 years, with 2,719 (6.46%) cancers diagnosed among ED participants and 3,139 (7.46%) among non-ED participants. ED participants experienced lower cancer risk overall (HRAdjusted = 0.70; 95% CI, 0.66–0.74; P = 8.89 × 10−31), specifically for breast cancer, prostate cancer, melanoma, and secondary cancers. Cancer-specific mortality was higher among ED participants (HRAdjusted = 1.76; 95% CI, 1.49–2.08; P = 3.62 × 10−11). Compared with non-ED participants, ED participants experienced a lower incidence of cancer but higher overall cancer-specific mortality, suggesting that subsets of ED patients may benefit from postvisit preventive interventions. Prevention Relevance: This cohort analysis shows that cancer incidence over 6 years was lower among participants after an ED visit than among matched non-ED participants, whereas cancer-specific mortality was higher in the ED group (HRAdjusted = 1.76; 95% CI, 1.49–2.08; P = 3.62 × 10−11), suggesting the potential benefit of preventive interventions.

Abstract The impact of early-life traumatic experiences on late-life morbidity remains partly unknown. We tested the hypothesis that adverse childhood experiences (ACE) and specifically abuse (physical, verbal or emotional, or sexual) experienced during childhood or early adulthood are associated with a higher rate of accumulation of multi-morbidity in women. Here we show strong evidence in support of the hypothesis. We studied 1,026 women aged 21–45 years randomly selected from the general population in Olmsted County, Minnesota and used the Rochester Epidemiology Project medical records-linkage system to measure the rate of development of 18 chronic conditions. The women had a median age of 41 years at inclusion in the study and were followed historically for a median of 21 years. Women with an ACE score ≥ 2 had higher incidence of 10 of the 18 chronic conditions considered separately and an accelerated accumulation of multi-morbidity measured as a score compared to women with ACE score 0. In addition, women exposed specifically to abuse in childhood or early adulthood had accelerated accumulation of multi-morbidity. We excluded the possible confounding effect of socioeconomic status and explored a series of possible mediation events or characteristics. We discuss several possible biological and social or behavioral mechanisms underlying these associations.