Research is necessary to generate evidence to influence change in health care. However, research alone is insufficient to drive change. The Betty Irene Moore Fellowship for Nurse Leaders and Innovators prepares nurses to translate research into change-inducing equitable practices, policies, and sustainable products. This paper illustrates how several Moore fellows are applying their research and findings to develop actionable guidance, best practices, and innovative skills that nurses in all settings can adopt. The first exemplar highlights impactful policy work to promote health and safety of a marginalized population. The second example focuses on barriers and facilitators of research-to-practice translation among people living with serious illnesses. The third example recognizes nurses’ role in translating research innovations into commercialized products and/or sustainable businesses. The findings and exemplars presented in this paper inform a call to action that underscores the critical role of nurses in translating research into practice, policies, and sustainable products.

Large language models (LLMs) are increasingly utilized for named entity recognition (NER) in health care, with significant potential to enhance symptom detection within electronic health records (EHRs). This study explores the application of LLMs to identify symptoms of anxiety and nausea/vomiting documented in the clinical notes of patients with cancer. We analyzed clinical notes from 8,490 patients diagnosed with various cancer types. Bio Clinical BERT and Bio GPT models were further pretrained on clinical text from this dataset. Two modeling strategies, fine-tuning and prompt-based learning, were implemented using Symptom-BERT and Symptom-GPT frameworks. Model performance was evaluated using F1 scores, emphasizing recognizing psychological symptoms (anxiety) and physical symptoms (nausea/vomiting). Fine-tuning with Symptom-BERT achieved the highest F1 scores, 0.989 for nausea/vomiting and 0.912 for anxiety, significantly outperforming Symptom-GPT in detection accuracy. While prompt-based learning with Symptom-GPT surpassed that of a few-shot learning, it remained less effective than fine-tuning. Fine-tuning excelled in identifying well-documented symptoms, particularly physical ones like nausea/vomiting. Using named entity recognition (NER), the study analyzed the entire dataset, detecting anxiety in 2,436 patients (28.69%) and nausea/vomiting in 3,338 patients (39.31%). While both fine-tuning and prompt-based learning approaches offer utility, fine-tuning demonstrates superior accuracy in recognizing symptoms from clinical narratives, particularly physical ones. LLM-based symptom detection can support oncology nurses and care teams by enabling earlier recognition of patient-reported symptoms documented in narrative notes. These tools offer practical value in improving symptom monitoring, care planning, and timely intervention, thereby enhancing patient-centered care in oncology settings.

Background Surrogate decision makers (SDM) are considered the spokesperson for individuals when they lose decision-making capacity. Despite their critical importance at the end-of-life, there is a subset of patients who do not have a documented SDM. Objective To identify predictors of missing SDM documentation among older decedents using electronic health record (EHR) data. Methods EHR data from 7,360 decedents aged ≥65 years with cancer, dementia, or heart failure. Demographic factors (age, race, sex, language, religion, marital status), palliative care (PC) referrals, and ten chronic conditions were explored as potential predictors of missing SDM documentation. Descriptive statistics were calculated, and group differences between decedents with and without a documented SDM were assessed. Weighted generalized linear models with logistic regression were used to identify independent predictors. Results The mean age of the sample was 79.58 years, primarily male (56.85 %), and majority white, non-Hispanic (94.73%). Among 7,360 decedents, 326 (4.4%) lacked SDMs. For univariate comparison, religion, marital status, and PC referral were statistically different between the two groups (p < 0.05). In the fully adjusted model of the whole sample (Pseudo R²=0.21), non-English language (OR=0.23; 95% CI: 0.17,0.33), religious affiliation (OR=0.77; 95% CI: 0.68,0.86), unmarried status (OR=0.74; 95% CI: 0.66,0.82) and PC referral (OR=0.17; 95% CI: 0.15,0.19) were associated with significantly higher odds of having SDM documentation. Chronic conditions of cerebrovascular disease, dementia, kidney disease, diabetes, heart failure, and COPD were also associated with documented SDM in the model with ORs ranging from .49 to .77 with significant CIs. Conclusions Lack of documented SDMs was disproportionately found in decedents without PC involvement and few/no documented chronic conditions. The strong protective effect of non-English language warrants further investigation. These findings highlight the need for systematic screening protocols and proactive interventions to ensure SDM documentation occurs upstream, prior to acute decline at the end-of-life.

Background Functional performance scales guide critical decisions in oncology, including prognostication and palliative care referrals [1]. Early identification of functional decline is critical for timely supportive care and improved quality of life, yet inconsistent documentation and limited visibility into decline patterns hinder proactive management [2]. Scalable, AI-enabled extraction from clinical notes may help address these gaps and support needs-based prognostication [3]. Objectives To characterize functional status documentation patterns by clinical service and identify trajectories of decline in lung cancer patients using automated extraction from clinical notes. Methods We applied a validated AI-based extraction system (FuncStatAI, accuracy=83%) to 87,006 clinical notes from 2,342 lung cancer patients, extracting ECOG, KPS, and PPS scores from 56% of notes. We calculated patient-level decline trajectories using linear regression, measured time to severe impairment, and analyzed scale preference by clinical service. Results The cohort included 2,342 patients (mean age 63, 45% female, 41% stage IV). Among 911 patients with trajectory data (median 17 assessments over 1.2 years), 9% showed rapid decline (>0.5 points/day on normalized 0–100 scale), losing median 50 points over 51 days. Among 1,068 patients progressing to severe status, 25% reached this threshold within 29 days (median 174 days). Overall, 47% experienced decline, with 28% declining from independence to severe impairment. Most patients (83%) had multiple scales documented; among 1,522 with longitudinal data, 58% showed consistent trajectories across scales (median slope difference: 0.08 points/day). Scale preference varied by specialty (p< 0.0001): palliative care predominantly used PPS (88%), medical oncology used ECOG (82%), and radiation oncology used KPS (96%). Implications Moderate trajectory consistency (58%) across scales supports normalized comparisons despite specialty-specific documentation preferences. Early identification of rapid decline (9% of patients) and narrow windows to severe impairment (25% within one month) underscore the need for automated trajectory-based surveillance to enable timely, proactive palliative care referrals.

Background While early palliative care (PC) improves outcomes [1, 2], real-world referral timing often appears reactive and crisis-driven rather than proactive [3, 4]. Objectives To identify diagnosis-specific patterns in PC referral timing, with focus on distinguishing acute event triggers from chronic condition trajectories. Methods Retrospective cohort study of 3,349 older adult decedents (age ≥65) who received PC at a large academic medical center (2010-2023). Primary outcome was late PC referral (≤30 days before death). We used multivariable logistic regression to identify independent predictors of late referral, adjusting for demographics and individual chronic conditions. Multimorbidity burden was quantified as the total count of 10 common chronic conditions (hypertension, hyperlipidemia, diabetes, COPD, dementia, cancer, coronary heart disease, cerebrovascular disease, arthritis, congestive heart failure) identified via ICD codes. Results The cohort had a mean age at death of 75.1±6.0 years, was 54.5% male and 94.5% White, with a mean of 4.5±2.1 chronic conditions. PC was initiated profoundly late, with a median of just 25 days before death (IQR 8-96) and a majority of patients (54.4%) receiving a late referral. Cerebrovascular disease was the strongest predictor of late referral (OR 1.40, 95% CI 1.20-1.63, p< 0.001), with median referral timing of only 19 days before death. In contrast, dementia predicted earlier referral (OR 0.74, 95% CI 0.62-0.88, median 55 days before death). Each additional year of age increased odds of late referral (OR 1.01, 95% CI 1.00-1.03, p=0.017). Overall multimorbidity burden did not independently predict timing when specific diagnoses were considered. Implications PC referral follows a crisis-driven model, with acute cerebrovascular events triggering late consultations while chronic progressive conditions like dementia enable earlier, trajectory-based referrals. Current systems react to acute deterioration rather than anticipating predictable decline. Systematic redesign should create dual pathways: rapid-response protocols for acute events and proactive monitoring triggers for chronic trajectories.

The symptom representations (i.e., beliefs and attitudes) that people with cancer hold about their symptom experience can impact how they self-manage their symptoms. Having two or more chronic conditions (multimorbidity) can complicate illness representations. Little is known about symptom representations in people with cancer and multimorbidity. This qualitative descriptive study was conducted with a sample of adults with a diagnosis of cancer and at least one additional chronic condition. Semi-structured interviews were conducted to understand their symptom representations. Leventhal’s Common-Sense Model of Illness Representations (i.e., identity, consequences, cure/control, timeline, and cause) provided the guiding framework. A qualitative thematic analysis was used to identify codes, themes, and subthemes. The mean age of the participants (n = 17) was 62.1 years and primary cancer sites were gastrointestinal, thoracic, or head/neck. Five themes were identified: (1) perceiving and living with symptoms, (2) being unable to do things, (3) self-management behaviors, (4) domino theory, and (5) a side effect of conditions. These themes aligned with Leventhal’s Common-Sense Model dimensions. The interaction among diagnoses and multimorbidity was identified by a minority of participants. People with cancer and multimorbidity described symptom representations primarily in the context of cancer. Consistent with previous research, symptoms negatively impacted their lives, and their representations include an understanding of how symptoms interact. Few participants described their symptoms within the larger context of multimorbidity. Future research is needed to determine how symptom representations impact their communication patterns with providers and coping behaviors.

Accurate survival prediction for older adults with lung cancer is critical for guiding supportive care, yet standard models often underutilize valuable information in clinical notes. The objective of this study was to determine if a prognostic model integrating features from unstructured clinical notes could more accurately predict survival in older adults with lung cancer than a model using structured data alone. We studied 1,391 patients (≥60 years at time of diagnosis) with lung cancer, analyzing 19,377 clinical encounters. We compared a baseline survival model using structured electronic health record (EHR) data to an enriched model that integrated previously validated natural language processing (NLP)-derived features, including functional status. Models were evaluated for 12-month mortality using AUC and C-Statistic. Thirty-eight percent of patients presented with metastatic disease. The baseline prediction model achieved good performance (AUC: 0.740; C-Statistic: 0.697). The enriched model demonstrated statistically significant improvement (AUC: 0.768 vs. 0.740; C-Statistic: 0.717 vs. 0.697; DeLong’s test p<.0001) with well-calibrated predictions and effective risk stratification into distinct prognostic groups. Feature analysis confirmed NLP-derived functional status and longitudinal chemotherapy timing were key predictors, with functional status ranking among the top three most important features. Clinical notes contain crucial prognostic information not captured in structured data. Leveraging NLP to extract this narrative data offers a powerful pathway to more accurate survival prediction, which could enable better identification of high-risk patients to improve supportive care interventions.

Electronic health records (EHRs) contain valuable patient information, yet certain aspects of care remain infrequently documented and difficult to extract. Identifying these rarely documented elements requires advanced informatics approaches to uncover clinical documentation patterns that would otherwise remain inaccessible for research and quality improvement.This study developed and validated an informatics approach using natural language processing (NLP) to detect and characterize rarely documented elements in EHRs, using spiritual care documentation as an exemplar case.Using EHR data from a Midwestern US hospital (2010-2023), we fine-tuned Spiritual-BERT, an NLP model based on Bio-Clinical-BERT. The model was trained on 80% of a manually annotated, gold-standard corpus of EHR notes, and its performance was validated using the remaining 20% of the corpus, alongside 150 synthetic notes generated by GPT-4 and curated by clinical experts. We applied Spiritual-BERT to identify spiritual care documentation and analyzed patterns across diverse patient populations, provider roles, and clinical services.Spiritual-BERT demonstrated high accuracy in capturing spiritual care documentation (F1-scores: 0.938 internal validation, 0.832 external validation). Analysis of nearly 3.6 million EHR notes from 14,729 older adults revealed that 2% of clinical notes contained spiritual care references, while 73% of patients had spiritual care documented in at least one note. Significant variations were observed across provider types: chaplains documented spiritual care in 99.4% of their notes, compared to 1.7% for nurses and 1.2% for physicians. Documentation patterns also varied based on ethnicity, language, and medical diagnosis.This study demonstrates how advanced NLP techniques can effectively identify and characterize rarely documented elements in EHRs that would be challenging to detect through traditional methods. This approach revealed distinct documentation patterns across provider types, clinical settings, and patient characteristics, with promise for analyzing other under-documented clinical information.

BackgroundIndividuals with type 2 diabetes (T2D) experience epigenetic age acceleration (EAA), as described by DNA methylation based epigenetic clocks. It is critical to examine relationships between social determinants of health (SDOH) and EAA in people living with T2D to understand mechanisms interconnecting social and biologic drivers of health disparities. The purpose of this systematic review was to describe the role of SDOH as factors influencing EAA in T2D.MethodsPubMed, CINAHL, and Embase were comprehensively searched. Research reports were independently screened and abstracted; quality was assessed using JBI checklists. The Healthy People 2030 SDOH Framework guided this study. Domains of SDOH were benchmarked against the framework to identify roles and gaps.ResultsOf 25 included research reports which evaluated epigenetic aging in T2D, 64% describe at least one SDOH. SDOHs within education access and quality (44%), neighborhood and built environment (40%), and economic stability (36%) domains are the most well represented; however, the depth and breadth of conceptual understanding were limited. Concepts representative of SDOH including childhood low socioeconomic status and victimization, trauma, and lower education and income were positively associated with EAA in T2D.DiscussionWhile SDOH are increasingly included in studies of EAA in people with T2D, critical gaps in understanding the roles and relationships between SDOH and EAA were revealed. Findings support the need to move further than socioeconomic status to comprehensively explore SDOH domains influencing EAA in individuals living with T2D, which will provide a framework for identifying health inequities.

Nurse Ruth, an AI-driven assistant, is designed to support obstetric nursing in resource-limited environments and for non-specialist healthcare providers. To develop and validate Nurse Ruth, we introduced novel evaluation metrics—Semantic Transparency Metric (STM) and Semantic Understanding Metric (SUM)—to assess response accuracy, contextual relevance, and robustness against conventional and adversarial clinical queries. Through iterative refinement and targeted knowledge integration, Nurse Ruth surpassed the 80% threshold for STM and SUM, reinforcing its ability to provide clear, evidence-based, and contextually precise clinical guidance. While excelling in response clarity and contextual accuracy, further improvements are needed to enhance recall in complex, multi-domain obstetric scenarios. A comparative evaluation against leading AI models (GPT-4o, GPT-4, and GPT-o1) for semantic validation demonstrated Nurse Ruth’s superiority. It achieved 100% accuracy on obstetric challenge queries, outperforming general-purpose AI models in both precision and efficiency. Unlike these models, Nurse Ruth delivered concise, rapid responses, making it the most effective system for real-world clinical applications. These findings validate Nurse Ruth’s semantic understanding and establish a replicable framework for AI-driven decision support in specialized medical fields. Future work will focus on refining recall in multi-faceted obstetric cases and validating real-world clinical impact.

Outcomes 1. Identify the demographic and clinical factors associated with disparities in advance directive completion and recognize the association between advance directive completion and the life-sustaining treatment preference in older adults with chronic conditions.2. Discuss the use of novel methods like natural language processing (NLP) to extract patient-centered information including life-sustaining treatment preferences from clinical narratives and assess their implications for palliative care research. Key Message This study highlights disparities in advance directive completion and their association with life-sustaining treatment preferences among older adults with chronic conditions. It demonstrates how natural language processing systems like Care-BERT can extract treatment preferences from clinical narratives to generate real-world evidence at scale. Abstract Engaging older adults with chronic conditions in completing advance directives (AD) is essential to ensuring their treatment preferences, particularly regarding life-sustaining treatments, are respected. Despite the increasing recognition of AD importance, completion rates remain low, and disparities persist across different demographic groups. Objectives To examine disparities in AD completion and assess the association between AD completion and life-sustaining treatment choices among older adults with chronic conditions. Methods This retrospective study analyzed electronic health records (EHR) from 14,303 older adults with chronic conditions treated in a Midwestern US healthcare system. Logistic regression models were employed to assess the impact of demographic and clinical factors on AD completion and the choice of life-sustaining treatments. To extract treatment preferences from EHR narratives, we used Care-BERT, a natural language processing (NLP) model specifically tailored for this study. Results The mean age of the sample was 78.5±6.04 years. AD completion was less likely among older individuals (OR = 0.935; 95% CI: 0.930, 0.940), but more likely among those with a cancer diagnosis (OR = 1.521; 95% CI: 1.391, 1.664). Non-White individuals had significantly lower odds of completing an AD compared to White individuals (OR = 0.646; 95% CI: 0.538, 0.775). AD completion was associated with a higher likelihood of choosing life-sustaining treatments (OR = 1.822; 95% CI: 1.567, 2.117). Additionally, non-White individuals (OR = 2.283; 95% CI: 1.349, 3.861) and non-English speakers (OR = 3.492; 95% CI: 1.049, 11.624) were more likely to choose life-sustaining treatments. Conclusions These findings highlight disparities in AD completion and life-sustaining treatment preferences among older adults. By leveraging NLP tools like Care-BERT, EHR narratives can yield key insights into treatment preferences. Future efforts should focus on developing interventions that promote equitable access to advance care planning and enhance patient-centered care. References - Jimenez, G., Tan, W. S., Virk, A. K., Low, C. K., Car, J., & Ho, A. H. Y. (2018). Overview of Systematic Reviews of Advance Care Planning: Summary of Evidence and Global Lessons. Journal of Pain and Symptom Management, 56(3), 436-459.e425. https://doi.org/10.1016/j.jpainsymman.2018.05.016 - McMahan, R. D., Tellez, I., & Sudore, R. L. (2021). Deconstructing the Complexities of Advance Care Planning Outcomes What Do We Know and Where Do We Go? A Scoping Review. Journal of the American Geriatrics Society, 69(1), 234-244. https://doi.org/10.1111/jgs.16801 - Morin, L., & Onwuteaka-Philipsen, B. D. (2021). The promise of big data for palliative and end-of-life care research. Palliative Medicine, 35(9), 1638-1640. https://doi.org/10.1177/02692163211048307

Goal-concordant care is essential, yet disparities in advance care planning (ACP) persist. Electronic health record (EHRs) documentation of specific preferences, such as comfort and life-sustaining treatments (LST) and natural language processing (NLP) to extract free-text data, provides real-world information to explore patterns. To identify predictors associated with documentation of specific preferences in a cohort of older adults with chronic conditions. In a dataset of records from 14,729 older adults with heart failure, cancer, or dementia from a large academic medical center, we used a validated NLP model, Priorities-BERT (accuracy= 90.91%), to categorize documented care preferences from 3.6 million EHR notes into four groups: comfort-only, LST-only, mixed, or none. Logistic regression identified factors associated with each preference category (vs. no preference). The mean age of the sample was 78.67 years (SD = 6.05). Female patients (vs. male) (OR = 1.11, 95% CI: 1.04–1.19), heart failure (OR =1.59, 95% CI: 1.34–1.89), dementia (OR = 1.33, 95%CI: 1.25-1.43), were associated with higher odds of having any priorities of care documented. Increasing age was associated with lower odds of LST (OR = 0.94, 95%CI: 0.94-0.95). Patients with heart failure (OR = 1.50, 95%CI: 1.39-1.61) or dementia (OR = 1.56, 95% CI: 1.42-1.71) were more likely to have comfort preferences documented. Conversely, cancer patients were more likely to have LST documented (OR = 1.12, 95%CI: 1.03-1.21). Significant disparities existed across these three diagnoses in documentation ACP preferences. Targeted clinical and health-system interventions are needed to promote equitable and timely ACP conversations for patients likely to benefit from this care.

Background and ObjectivesThere is a need to understand how different factors influence health care utilization for patients with Huntington disease (HD) to maximize benefits of primary and specialty care while minimizing need for costly emergency visits or hospitalizations. The primary objective of this study was to characterize how settings where patients with HD in Northern America receive care change throughout the disease course and determine whether the likelihood of different types of service utilization is influenced by clinical, sociodemographic, and caregiver characteristics.MethodsData from the Enroll-HD study and joinpoint regression were used to assess trends in neurology visits, general practitioner visits, emergency department visits, and inpatient stays over the disease course and as a function of total functional capacity. Generalized estimating equation models were then used to identify factors associated with use of these different services in the 6 months before their study visit.ResultsVisits from 1,631 participants in the Northern America region from the Enroll-HD study were included in this study. Trends in neurology, emergency, and inpatient visits remained constant over most of the disease duration. For the general practitioner visits, there was an increasing trend in use throughout the course of disease. Clinical factors, such as psychiatric symptoms, functional ability, and comorbidities, were associated with use of multiple types of health care services. Sociodemographic and caregiver factors, such as race or ethnicity, urban or rural residence, and caregiver employment status, were also associated with use of multiple health care services.DiscussionClinical, sociodemographic, and caregiver-related factors were all associated with outpatient, emergent, and inpatient care. This work identifies multiple avenues for future research on how to improve access to and quality of care for patients with HD, specifically relating to reducing the need for emergency visits and inpatient stays and promoting collaboration among primary and specialty clinicians.

Background The integration of patient-reported outcomes (PROs) into clinical care, particularly in the context of cancer and multimorbidity, is crucial. While PROs have the potential to enhance patient-centered care and improve health outcomes through improved symptom assessment, they are not always adequately documented by the health care team. Objectives This study aimed to explore the concordance between patient-reported symptom occurrence and symptoms documented in electronic health records (EHRs) in people undergoing treatment for cancer in the context of multimorbidity. Methods We analyzed concordance between patient-reported symptom occurrence of 13 symptoms from the Memorial Symptom Assessment Scale and provider-documented symptoms extracted using NimbleMiner, a machine learning tool, from EHRs for 99 patients with various cancer diagnoses. Logistic regression guided with the Akaike Information Criterion was used to identify significant predictors of symptom concordance. Results Our findings revealed discrepancies in patient and provider reports, with itching showing the highest concordance (66%) and swelling showing the lowest concordance (40%). There was no statistically significant association between multimorbidity and high concordance, while lower concordance was observed for women, patients with advanced cancer stages, individuals with lower education levels, those who had partners, and patients undergoing highly emetogenic chemotherapy. Conclusion These results highlight the challenges in achieving accurate and complete symptom documentation in EHRs and the necessity for targeted interventions to improve the precision of clinical documentation. By addressing these gaps, health care providers can better understand and manage patient symptoms, ultimately contributing to more personalized and effective cancer care.