In Alzheimer's Disease and Related Dementia (ADRD) prediction, Electronic Health Records (EHRs) provide rich but fragmented information. Without a coherent clinical narrative, models tend to rely on a few dominant signals (e.g., age-related patterns) rather than capturing the underlying clinical mechanisms. Such reliance becomes problematic in younger-onset cases, where these signals are less informative. To address this challenge, we propose RaiLED-AD, a dual-encoder teacher-student framework where the student learns from serialized EHR data and the teacher leverages LLM-generated narratives that capture temporal and relational patterns. A hybrid objective with soft-label supervision and hierarchical contrastive alignment transfers these reasoning signals to the student, which operates independently at inference. On a real-world EHR cohort, RaiLED-AD consistently improves ADRD prediction over baselines and achieves substantial gains in the challenging younger-onset subgroup (index age$<65$). These results highlight the potential of integrating LLM-derived reasoning signals with structured EHR models for early-stage ADRD risk prediction.

OBJECTIVE: Health care providers increasingly rely upon predictive algorithms when making important treatment decisions, however, evidence indicates that these tools can lead to inequitable outcomes across racial and socio-economic groups. In this study, we introduce a bias evaluation checklist that allows model developers and health care providers a means to systematically appraise a model's potential to introduce bias. MATERIALS AND METHODS: Our methods include developing a bias evaluation checklist, a scoping literature review to identify 30-day hospital readmission prediction models, and assessing the selected models using the checklist. RESULTS: We selected 4 models for evaluation: LACE, HOSPITAL, Johns Hopkins ACG, and HATRIX. Our assessment identified critical ways in which these algorithms can perpetuate health care inequalities. We found that LACE and HOSPITAL have the greatest potential for introducing bias, Johns Hopkins ACG has the most areas of uncertainty, and HATRIX has the fewest causes for concern. DISCUSSION: Our approach gives model developers and health care providers a practical and systematic method for evaluating bias in predictive models. Traditional bias identification methods do not elucidate sources of bias and are thus insufficient for mitigation efforts. With our checklist, bias can be addressed and eliminated before a model is fully developed or deployed. CONCLUSION: The potential for algorithms to perpetuate biased outcomes is not isolated to readmission prediction models; rather, we believe our results have implications for predictive models across health care. We offer a systematic method for evaluating potential bias with sufficient flexibility to be utilized across models and applications.

RE: Barriers to Hospital Electronic Public Health Reporting and Implications for the COVID-19 Pandemic, JAMIA, https://doi.org/10.1093/jamia/ocaa112 Dear Dr. Bakken (editor of JAMIA) and A Jay Holmgren, Nate C Apathy, and Julia Adler-Milstein (authors), We applaud efforts to address concerns about information sharing between hospitals and public health systems, particularly during the COVID-19 pandemic. However, we have concerns about the validity and usefulness of the findings reported by Holmgren, et al and the characterization of the ability of public health agencies (PHAs) to receive electronic data. First, the study findings do not match the situation “on the ground”: In 2018, PHAs were able to receive electronic data, particularly lab and immunization data. All relevant (50 states and 6 large cities, such as Los Angeles and New York City) PHAs were receiving laboratory data for electronic laboratory reporting (ELR) (Personal communication: Jason Hall, CDC, 2020) Similarly, 96% of these PHAs were receiving data for their immunization registries.1 In fact, these registries also provide data back to electronic health records and can report timely information about underimmunized populations at risk for outbreaks.2 The authors reference a lack of local health department capacity in the text and Figure 2; but in reality, it is typical for state or large city PHAs to receive electronic data from hospitals on behalf of local agencies. State or large city PHAs host the IT infrastructure for electronic surveillance and manage interfaces with clinical systems, while staff from smaller PHAs access the hosted system. Second, the authors overgeneralized the survey findings: The authors conflate syndromic surveillance with the larger concept of “electronic surveillance,” as if syndromic surveillance were the important data source for controlling the COVID-19 pandemic. Syndromic surveillance is useful for early detection and population-level monitoring, but it does not include patient-level data needed for case investigation and outbreak management. The authors have overinterpreted the responses to the 2018 American Hospital Association (AHA) Annual survey question, leading to possible bias. Hospital CEOs or their designees were asked: “What are some of the challenges your hospital has experienced when trying to submit health information to public health agencies to meet meaningful use (MU) requirements?”3 Four in 10 CEOs (41%) selected the barrier “Public health agencies lacked the capacity (eg, technical, staffing) to electronically receive data.”3 A hospital CEO or delegate may not be aware of their local and state PHA’s capacity to receive electronic health information, or even their own organization’s involvement with electronic reporting of lab results, immunizations, case reports or other data. In addition, the response option does not provide clarity as to which data, when, and in what form. The AHA survey is important, but results should be interpreted in context. We are concerned the publication could lead to incorrect assumptions at a time when clinical and public health systems need to communicate more than ever and may discourage health care providers, leaders, and health IT vendors from engaging with public health agencies to avail themselves of existing data exchange capabilities. Given the critical need for both ELR and case reporting to manage the COVID-19 outbreak, major efforts are underway to expand electronic case reporting (eCR) (https://cdc.gov/ecr) and reduce the burden for health systems.4 As of July 6, 2020, the APHL Informatics Messaging Service (AIMS), a national resource for ELR and eCR reporting, had received 803 239 COVID-19 case reports from over 2000 facilities in 20 health care organizations which were shared with PHAs from 47 jurisdictions; and all but 2 state PHAs can receive eCR messages from the AIMS platform, and enhancements are underway to automatically integrate data into surveillance systems (Personal communication: Laura Conn, CDC, 2020). Public health authorities describe reluctance from providers and health systems to implement electronic reporting on the grounds that implementation is too burdensome. More engagement is needed. There is no question that inadequate resources have been a limiting factor for public health agencies to receive data from health systems. This problem is exacerbated by the many-to-one (hospitals-to-public health agency) nature of population health activities, the variable nature of hospital data contributions, and the resources required to onboard and manage interfaces with multiple health systems. We encourage clinical partners to work with public health agencies to improve surveillance of both clinical and public health outcomes and leverage information exchange to benefit communities.5 We recommend increasing support for public health agencies to enhance their ability to exchange (both receive and send) information while health care systems receive support to send data. CJS wrote the first draft. All authors provided input and revisions. All authors approved final submission. We thank Erin Holt for her input on this letter. None declared.

OBJECTIVE: Geriatric syndromes such as functional disability and lack of social support are often not encoded in electronic health records (EHRs), thus obscuring the identification of vulnerable older adults in need of additional medical and social services. In this study, we automatically identify vulnerable older adult patients with geriatric syndrome based on clinical notes extracted from an EHR system, and demonstrate how contextual information can improve the process. MATERIALS AND METHODS: We propose a novel end-to-end neural architecture to identify sentences that contain geriatric syndromes. Our model learns a representation of the sentence and augments it with contextual information: surrounding sentences, the entire clinical document, and the diagnosis codes associated with the document. We trained our system on annotated notes from 85 patients, tuned the model on another 50 patients, and evaluated its performance on the rest, 50 patients. RESULTS: Contextual information improved classification, with the most effective context coming from the surrounding sentences. At sentence level, our best performing model achieved a micro-F1 of 0.605, significantly outperforming context-free baselines. At patient level, our best model achieved a micro-F1 of 0.843. DISCUSSION: Our solution can be used to expand the identification of vulnerable older adults with geriatric syndromes. Since functional and social factors are often not captured by diagnosis codes in EHRs, the automatic identification of the geriatric syndrome can reduce disparities by ensuring consistent care across the older adult population. CONCLUSION: EHR free-text can be used to identify vulnerable older adults with a range of geriatric syndromes.

OBJECTIVES: The study sought to identify collaborative electronic health record (EHR) usage patterns for pediatric trauma patients and determine how the usage patterns are related to patient outcomes. MATERIALS AND METHODS: A process mining-based network analysis was applied to EHR metadata and trauma registry data for a cohort of pediatric trauma patients with minor injuries at a Level I pediatric trauma center. The EHR metadata were processed into an event log that was segmented based on gaps in the temporal continuity of events. A usage pattern was constructed for each encounter by creating edges among functional roles that were captured within the same event log segment. These patterns were classified into groups using graph kernel and unsupervised spectral clustering methods. Demographics, clinical and network characteristics, and emergency department (ED) length of stay (LOS) of the groups were compared. RESULTS: Three distinct usage patterns that differed by network density were discovered: fully connected (clique), partially connected, and disconnected (isolated). Compared with the fully connected pattern, encounters with the partially connected pattern had an adjusted median ED LOS that was significantly longer (242.6 [95% confidence interval, 236.9-246.0] minutes vs 295.2 [95% confidence, 289.2-297.8] minutes), more frequently seen among day shift and weekday arrivals, and involved otolaryngology, ophthalmology services, and child life specialists. DISCUSSION: The clique-like usage pattern was associated with decreased ED LOS for the study cohort, suggesting greater degree of collaboration resulted in shorter stay. CONCLUSIONS: Further investigation to understand and address causal factors can lead to improvement in multidisciplinary collaboration.

OBJECTIVE: The Johns Hopkins Center for Population Health IT hosted a 1-day symposium sponsored by the National Library of Medicine to help develop a national research and development (R&D) agenda for the emerging field of population health informatics (PopHI). MATERIAL AND METHODS: The symposium provided a venue for national experts to brainstorm, identify, discuss, and prioritize the top challenges and opportunities in the PopHI field, as well as R&D areas to address these. RESULTS: This manuscript summarizes the findings of the PopHI symposium. The symposium participants' recommendations have been categorized into 13 overarching themes, including policy alignment, data governance, sustainability and incentives, and standards/interoperability. DISCUSSION: The proposed consensus-based national agenda for PopHI consisted of 18 priority recommendations grouped into 4 broad goals: (1) Developing a standardized collaborative framework and infrastructure, (2) Advancing technical tools and methods, (3) Developing a scientific evidence and knowledge base, and (4) Developing an appropriate framework for policy, privacy, and sustainability. There was a substantial amount of agreement between all the participants on the challenges and opportunities for PopHI as well as on the actions that needed to be taken to address these. CONCLUSION: PopHI is a rapidly growing field that has emerged to address the population dimension of the Triple Aim. The proposed PopHI R&D agenda is comprehensive and timely, but should be considered only a starting-point, given that ongoing developments in health policy, population health management, and informatics are very dynamic, suggesting that the agenda will require constant monitoring and updating.