Jennifer L. Wolff

Background: Use of secure messaging within the patient portal has grown rapidly over recent years, yet clinical workflows, staffing models, and billing practices have generally not been comprehensively adapted to support this increasing volume of messages. Little is known about the characteristics of patients who are high-volume message senders, and we lack a nuanced understanding of patient messaging intensity beyond measures accounting for sheer volume, limiting health system efforts to anticipate message-related workflow demands and allocate resources effectively. Objective: Our objective was to characterize older adult patients (aged ≥65 years) with high secure messaging volume, examining both patient characteristics and other aspects of their messaging intensity, such as messaging frequency, length, and messaging use relative to patient portal logins and health care encounters. Methods: We analyzed electronic medical record and patient portal data from a large academic health system, encompassing 16,023 older adults who sent 199,981 messages over a 12-month period. We developed 5 measures to account for secure messaging intensity. Our primary measure of messaging intensity was based on message volume; high-volume message senders were identified using outlier analysis based on patients’ aggregate number of messages sent during the observation period. Additional measures of messaging intensity included identifying individuals with concentrated periods of messaging, message length (character count), a ratio of messages to portal logins, and a ratio of messages to health care encounters. We compared sociodemographic characteristics, health status, and messaging intensity of high-volume secure messaging senders with those of other message senders. Results: Of 16,023 older adult patients who sent at least 1 message during the observation period, 1298 (8.1%) were classified as high-volume message senders; these patients accounted for 39.7% (79,475/199,981) of total messages. High-volume message senders, compared with all other message senders, were more likely to be White (1044/1298, 80.4% vs 10,670/14,725, 72.5%; P <.001), have higher comorbidity scores (2.6 vs 1.8; P <.001), and higher incidence of cancer (465/1298, 35.8% vs 3354/14,725, 22.8%; P <.001) and dementia (107/1298, 8.2% vs 889/14,725, 6.1%; P <.002). High-volume message senders were also more likely to be identified as having concentrated periods of messaging, to send longer messages, and to send more messages in relation to patient portal logins and health care encounters. Conclusions: High-volume message senders represent a small but distinct group of older patients who send a large share of messages to clinicians. Using multiple measures of messaging intensity can provide additional context about patient messaging behavior and help health systems redesign workflows and care delivery models to meet patient communication needs while facilitating sustainable workloads for clinicians.

Background: Advance care planning (ACP) is especially important in the context of cognitive impairment. Little is known about long-term effects of ACP on care partners. Aim: To determine the sustained effect of facilitated ACP conversations on care partner-reported quality of communication (QoC) about end-of-life (EoL) care and readiness to engage in ACP at 24-months. Design: A total of 273 participants were randomized 1:1 to a multicomponent ACP intervention (SHARE) or usual care. ACP was delivered between baseline and 6 months and dyads were followed through 24 months (N = 114). Outcomes included care partner responses to the 7-item QoC instrument and 6-item readiness to engage in ACP instrument. Group differences at 24 months were compared using ANCOVA, adjusting for baseline scores and cognitive impairment. Setting/participants: Cognitively impaired older adults age 80+ and family care partners were recruited from 8 primary care clinics. Results: Care partners were mostly women (91.2%), average 65.5 years old, and adult children (66.7%) of the patient. 114 (42.8%) care partners completed 24-month assessments; One fifth of patients died within 24 months (n = 61; 22.3%). Caregiver QoC scores at 24 months were similar for intervention and control groups (3.34 vs 3.18; P = 0.92). Similarly, readiness to engage in ACP was comparable for intervention and control groups (26.63 vs 25.64; P = 0.67). Conclusions: ACP conversations did not significantly affect QoC or readiness to engage in ACP at 24 months follow-up in this randomized trial. ACP may require an iterative approach for ongoing benefit in this population.

BACKGROUND The growing volume of secure messaging within the patient portal has imposed significant demands on clinicians and contributed to burnout. Little is known about the characteristics of patients who comprise high-volume message senders, and we lack a nuanced understanding of patient messaging intensity beyond measures accounting for sheer volume. OBJECTIVE Our objective was to characterize older adult patients (65+) with high secure messaging volume, examining both patient characteristics and other aspects of their messaging intensity such as messaging frequency, length, and messaging use relative to patient portal logins and healthcare encounters. METHODS We analyzed electronic medical record (EMR) and patient portal data from a large academic health system, encompassing 16,023 older adults who sent 199,952 messages over a 12-month period. We developed five measures to account for secure messaging intensity. Our primary measure of messaging intensity was based on message volume; high-volume message senders were identified using outlier analysis based on patients’ aggregate number of messages sent during the observation period. Additional measures of messaging intensity included identifying individuals with concentrated periods of messaging, message length (character count), a ratio of messages to portal logins and a ratio of messages to healthcare encounters. We compared sociodemographic characteristics, health status, and messaging intensity of high-volume secure messaging senders to other message senders. We also identified patients who were classified as high-intensity message senders based on all five measures of messaging intensity (‘super-senders’). RESULTS Of 16,023 older adult patients who sent at least one message during the observation period, 1,298 (8.1%) were classified as high-volume message senders; these patients accounted for 39.7% of total messages. High-volume message senders, compared to all other message senders, were more likely to be White (80.4% vs. 72.5%, p < 0.001), have higher comorbidity scores (2.6 vs. 1.8, p <0.001), and higher incidence of cancer (35.8% vs. 22.8%, p<0.001) and dementia (8.3% vs. 6.1%, p < 0.002). High-volume message senders were also more likely to be identified as having concentrated periods of messaging, to send longer messages, and to send more messages in relation to patient portal logins and healthcare encounters. A small subgroup of patients classified as high-volume senders were also classified as high-intensity across all four of the other measures of messaging intensity (59/1,298; 4.5%), the ‘super senders’. CONCLUSIONS High-volume message senders represent a small but distinct group of older patients who send a disproportionate share of messages to clinicians. Triangulating multiple measures of messaging intensity can help provide additional context about patient messaging behavior and help to identify patients that may most benefit from targeted outreach while potentially easing clinicians' inbox workload.

Importance: Older adults with disabilities rely on support from family caregivers, medical professionals, and paid caregivers. Unlike team-based interprofessional care, coordination of care work is not protocolized or well understood. Objective: To describe family caregiver experiences coordinating care with medical professionals, other family caregivers, and paid caregivers. Design, Setting, and Participants: This cross-sectional analysis of the 2023 National Health and Aging Trends Study and the linked National Study of Caregiving included data on 2811 family and unpaid caregivers to community-living adults aged 65 years or older with disabilities. Exposure: Caregiver characteristics, caregiving circumstances, negative consequences, supportive services use, and, for medical professionals, frequency of interactions and types of communication behaviors. Main Outcomes and Measures: Adapted relational coordination index (RCI; range, 1-5; greater values indicate higher relational coordination) reflecting frequency of any communication and problem-solving communication, as well as shared goals, shared knowledge, and mutual respect with medical professionals, paid caregivers, and other family caregivers. Results: Among a weighted sample of 2811 family and unpaid caregivers (1633 women [63.4%]; 1044 aged ≥65 years [41.5%]), most (2448 [87.6%]) engaged in care work that involved medical professionals (2061 [76.4%]), other family caregivers (1791 [62.8%]), or paid caregivers (822 [23.2%]). Relational coordination was reported to be weak with medical professionals (RCI, 3.10 [95% CI, 3.03-3.16]) and moderate for paid caregivers (RCI, 3.80 [95% CI, 3.71-3.89]) and other family caregivers (RCI, 4.20 [95% CI, 4.13-4.27]). Across all roles, relational coordination was higher when caregivers were related to the older adult, assisted an older adult with greater care needs, and engaged in higher-intensity caregiving. Use of supportive services was associated with higher relational coordination with medical professionals and paid caregivers but varied in magnitude by service type; the greatest differential was observed for family caregivers who did (vs did not) receive training when reflecting on medical professionals (3.63 [95% CI, 3.42-3.83] vs 3.05 [95% CI, 2.99-3.12]; P < .001). The RCI fluctuated from 2.33 [95% CI, 2.24-2.42] among caregivers who never interacted with medical professionals to 4.21 [95% CI, 4.11-4.31] among caregivers who interacted with medical professionals often and were always or usually asked about needed help. Conclusions and Relevance: This cross-sectional study of a national survey found that few family caregivers engage in care work alone. Further study is needed to determine if weak relational coordination with medical professionals may be remediated through training and supportive communication.

The growing use of AI to support patient portal message management requires rigorous preclinical evaluation. Directly testing AI within electronic health record (EHR) systems poses significant safety, workflow, and data-governance risks. Here, we present a technical feasibility report on a secure user interface (UI) sandbox designed to enable clinical and technical teams to experiment with AI for portal messaging before clinical integration. In this context, a “sandbox” refers to a controlled, nonproduction environment that allows safe testing, prompt iteration, and evaluation of AI outputs without impacting live EHR systems or patient care. We developed a web UI in Python 3 with a modular backend for data handling and AI task execution that operates entirely within the institutional firewall. The system runs in a secure research environment equipped with an NVIDIA GRID T4-1Q graphics processing unit (GPU) and institutional access controls. We designed a deidentification pipeline to remove or replace personal health identifiers and assessed its precision. The platform supports single-message and batch workflows and exposes example large language model (LLM)–enabled tasks such as authorship identification, message categorization, criticality flagging, and response drafting using zero-shot, one-shot, and few-shot prompting. The system successfully executed end-to-end workflows to ingest messages, run individual or batch AI analyses, and present outputs for review. Personal health information partial masking was applied across the corpus using a deidentification pipeline validated against 110 manually adjudicated entities (sensitivity 95.1%, precision 82.1%). We ran use cases with an institutional review board–approved corpus of a dementia-relevant subset of 6941 patient portal messages categorized as “medical advice requests” from 497 unique patients. With the support of the UI, we tested which prompting strategies yielded interpretable outputs for authorship identification, categorization, and criticality flagging, and whether response drafting produced editable clinician starting points. A token-based cost readout provided transparent operating estimates for LLM-backed tasks. This framework offers a practical, secure path to test AI behavior on real messages without affecting live EHR workflows and thus supports exploratory testing, prompt iteration, and comparative analyses, including LLM prompts versus baseline models, while preserving governance boundaries. We discuss design choices, safety controls, and the limits of a sandbox approach. A secure, UI-based sandbox enables health system teams to evaluate AI for patient portal messaging before clinical integration. The goal is not to assume benefit but to generate evidence about feasibility, risks, and fit to clinical needs in a controlled setting.

BACKGROUND The growing use of artificial intelligence (AI) to support patient portal message management requires rigorous pre-clinical evaluation. Directly testing AI within electronic health record (EHR) systems poses significant safety, workflow, and data-governance risks. We developed and describe a secure user interface (UI) that enables clinical and technical teams to experiment with AI for portal messaging on de-identified data before clinical integration. OBJECTIVE We developed a secure user interface (UI) that enables clinical and technical teams to experiment with AI for portal messaging on de-identified data before clinical integration, which we describe here for others to adapt. METHODS We developed a web UI in Python 3 with a modular backend for data handling, de-identification, and AI task execution. The system runs in a secure research environment equipped with an NVIDIA GRID T4-1Q GPU and institutional access controls. An IRB-approved corpus of patient-portal messages was cleaned to a dementia-relevant subset of 6,941 Medical Advice Request messages. We designed a de-identification pipeline to remove or replace personal health identifiers. The platform supports single-message and batch workflows and exposes exemplar LLM-enabled tasks such as authorship identification, message categorization, criticality flagging, and response drafting using zero-, one-, and few-shot prompting. Only de-identified text is sent to the model endpoint. RESULTS The system successfully executed end-to-end workflows: ingest de-identified messages, run individual or batch AI analyses, and present outputs for review. PHI masking was consistently applied across the corpus. Exemplar runs demonstrated prompting strategies to yield interpretable outputs for authorship identification, categorization, and criticality flagging, while response drafting produced editable clinician starting points. A token-based cost readout provided transparent operating estimates for LLM-backed tasks. CONCLUSIONS This framework offers a practical path to study AI behavior on real, de-identified messages without affecting live EHR workflows and thus supports exploratory testing, prompt iteration, and comparative analyses, including LLM prompts versus baseline models, while preserving governance boundaries. We discuss design choices, safety controls, and the limits of a sandbox approach. A secure, UI-based sandbox enables health-system teams to evaluate AI for patient-portal messaging before clinical integration.

Background: Serious illness experiences, i.e. experiencing critical illness personally or through close ones, often influence end-of-life goals of care. Little is known about how older adults with cognitive impairment respond to serious illness experience questions during advance care planning (ACP). Objectives: Our aim was to demonstrate the utility of serious illness experience questions as a strategy for eliciting goals of care during ACP among older adults with cognitive impairment and care partners. Methods: Thematic analysis of 92 transcribed ACP conversations involving older adults ≥80 years old who screened positive for cognitive impairment and their care partner. Trained facilitators used a structured ACP conversation guide that probed serious illness experiences. Results: Participants often described multiple serious illness experiences influencing goals of care, which fell on a spectrum between longevity and comfort. Experiences involving negative clinician interactions provided opportunities for participants to emphasize longevity in their goals of care. Experiences involving severe disability or iatrogenic harm provided opportunities to emphasize comfort in their goals of care. Conclusions: Older adults with cognitive impairment and their care partners were able to respond to serious illness experience questions, with responses eliciting a spectrum of goals of care. Integrating such questions into ACP for older adults with cognitive impairment may clarify goals of care and improve decision-making for patients and families.

Importance: Managed Long-Term Services and Supports (MLTSS) has been implemented by more than half of all state Medicaid programs, but knowledge of its scope and consequences for caregiving is limited.Objective: To describe observed trends in MLTSS program presence and care hours among older adults dually enrolled in Medicare and Medicaid between 2012 and 2022.Design, Setting, and Participants: This repeated cross-sectional study used data from the 2012-2022 National Health and Aging Trends Study (NHATS) linked with county-level measures of MLTSS program presence. Included in the study were adults aged 70 years or older dually enrolled in Medicare and Medicaid (hereafter older dual enrollees) who were living in the community and receiving assistance with self-care or mobility activities. Data analyses were conducted from December 2023 to June 2025.Exposure: Residence in an area with MLTSS program presence.Main Outcomes and Measures: The primary outcome was the total care hours per week received by older dual enrollees estimated as the mean (SD) care hours received per week overall and by caregiver type for each survey round. Percentages were weighted to account for the complex survey design.Results: Among 2549 participants, the mean (SD) age was 80.3 (8.5) years, 70.1% were female, and 26.5% lived alone. The weighted percentage of community-living, older dual enrollees receiving assistance who resided in areas with an MLTSS program increased from 39.4% in 2012 to 71.4% in 2022. Older dual enrollees receiving assistance were helped by a mean (SD) of 2.8 (1.6) caregivers and received 65.6 care hours per week (weighted percentage of total care hours, 80.9%). Fifty percent of participants received paid help. Most care hours were provided by unpaid family caregivers (mean [SD] care hours per week, 42.0 [74.0]). Mean (SD) weekly care hours for those residing in areas with continuous MLTSS presence were higher than those living in areas that added or had no MLTSS program presence (72.9 [82.8] vs 61.6 [71.7] and 60.0 [85.0] hours, respectively; P = .03). Among those residing in areas with continuous MLTSS, mean (SD) care hours provided by paid family caregivers increased steadily over the observation period from 2.0 (11.0) hours in 2012 to 23.8 (57.9) hours in 2022.Conclusions and Relevance: This cross-sectional study found increases in the percentage of older dual enrollees receiving assistance in the community and residing in areas with MLTSS program presence, with persistent heavy reliance on unpaid family caregivers. Systems are needed to monitor the care experiences of patients and their caregivers.

Introduction: Medicaid home and community‐based services (HCBS) support community living for older adults and individuals living with disabilities. Information sharing and effective communication among home care team members are critical to HCBS care coordination but are often fragmented, resulting in potentially avoidable consequences, particularly for persons living with complex health conditions, like dementia. Medicaid HCBS programs collect a range of data that could be leveraged in a learning health system (LHS) model to strengthen coordination between home care team members to improve outcomes and equity in HCBS care delivery. Methods: We leverage Friedman's Socio‐technical Infrastructure for LHS to consider information sharing capabilities and needs within Medicaid HCBS in Maryland. Results: Given longitudinal care complexities, significant costs, and data collection protocols, Medicaid HCBS is promising for LHS efforts. Recommendations for implementing an LHS in Medicaid HCBS highlight a socio‐technical infrastructure (i.e., people, technology, policies, processes), information cycles (e.g., existing research relationships and opportunities for expansion), and governance (e.g., ensuring ethical LHS implementation). Additional considerations for persons living with dementia should be made, such as the integration of dementia‐related policies into care delivery. Conclusions: Integrating LHS strategies into Medicaid HCBS holds promise for strengthening efficiency and equity in information sharing across state agencies, care teams (e.g., direct care workers, nurses), and families.