AMIA 2024 Annual Symposium On Demand

The rapid integration of Generative Artificial Intelligence (GenAI) into daily living has sent shockwaves throughout the world. Healthcare has seen an upsurge in the development of AI-enabled algorithms that have led to improvements in medical education, research, and practice. For example, remote exam proctors, patient-facing chatbots, EHR search engines, and practice management tools have been adopted by major health systems and academic institutions across the United States. These innovations are compelling us to adapt, adopt, and regulate swiftly. Although the potential and promise of integrating GenAI in healthcare are immense, there are some inherent risks and challenges, which require awareness and regulations. This panel will discuss the various applications of Gen AI in dentistry, especially as it relates to dental education, research, and practice. Dr. Sepideh Banava will discuss the applications of GenAI in dental education and clinical practice. Dr. Jay Patel will discuss its applications in dental research, specifically, developing, testing, and validating prediction models for dental diseases using large electronic dental record datasets. Dr. Enihomo Obadan-Udoh will discuss its potential applications in dental public health. This engaging and informative panel will sketch a realistic picture of GenAI use in academia and its potential to improve population health while suggesting intelligent strategies to make informed decisions.

Learning Outcomes

• List GenAI uses in dental education and practice.
• Outline the risks and challenges with the adoption of GenAI in dental education and practice.
• Discuss ethical issues in using GenAI in dental education and practice.
• Describe the uses of AI in various aspects of dental public health. 5. Discuss the ethical considerations for using AI in dental public health.

Speakers

• Enihomo Obadan-Udoh, DDS, MPH, Dr. Med. Sc., University of California San Francisco
• Sepideh Banava, DDS, MSc, MBA, DABDPH, Nationwide Children's Hospital
   

This panel, led by the real-world evidence core from the NICHD-supported MPRINT Hub, will describe techniques for establishing mother-child linkages (MCL) across diverse health systems. With experience in developing, applying, and assessing MCL algorithms, the panelists will explore the generalizability of these algorithms. They will highlight their performance and adaptability in various healthcare environments, shedding light on the complexities of MCL algorithm development and showcasing how these tools can be considered for broader application. The insights provided aim to advance the application of MCL in enhancing maternal and child health outcomes by leveraging real-world data and emphasizing cross-site collaboration.

Speakers

• Lang Li, PhD, The Ohio State University
• Christopher Bartlett, PhD, Nationwide Children's Hospital
• Eneida Mendonca, MD, PhD, Cincinnati Children's Hospital / University of Cincinnati
• Colin Rogerson, MD, Indiana University

The CodeX FHIR Accelerator and its CardX cardiovascular domain have made great strides towards standardizing clinical data to support better patient care and research. During this panel, members of the CardX community will discuss their efforts to build common cardiovascular lexicon called mCARD, modeled after the mCODE cancer lexicon and advance cardiovascular interoperability through the CardX community. They will share lessons they have learned from this experience, and discuss opportunities to apply those lessons to other clinical specialties. The panel will also discuss a forthcoming “playbook” for the development of clinical specialty data standards, and preview key themes and concepts from this playbook. Panelists will also discuss how the work done under CodeX and CardX helps enable a broader vision for a “standard health record”: a computable, clinically applicable, set of health information available in every electronic health record for every person.

Learning Outcomes

• Be aware of CodeX, the clinical FHIR Accelerator, and CardX, its cardiovascular domain.
• Understand the concept of the standard health record and the value of standard data elements across multiple clinical domains, including cardiovascular health. 
• Participate in building the future for health data interoperability by supporting the creation and real-world use of HL7 FHIR and standard health records thought eh HL7 CodeX FHIR Accelerator.

Speakers

• Adam Kroetsch, MS, MITRE
• James Tcheng, MD, Duke University Health System
• John Windle, MD, University of Nebraska Medical Center
• Sutin Chen, MD

Climate change poses a critical challenge to global health, with its impacts—ranging from heat waves and wildfires to changing disease endemicity and disruptive weather events—growing in frequency and intensity. Currently, 3.6 billion people reside in highly susceptible areas, with climate change projected to cause an additional 250,000 deaths annually by 2030-2050 due to diseases and undernutrition. The most affected will be those in low- and middle-income countries with fragile health infrastructures. Biomedical informatics is an important tool for tackling these challenges by providing essential data for understanding, preventing, and mitigating climate change's effects on health. This panel will describe the potential for global health informatics to facilitate evidence-based decisions and improve preparedness, especially in vulnerable regions. Addressing data collection and sharing challenges is critical for harnessing informatics in the fight against climate change-induced health crises.

Learning Outcomes

• Describe the potential for global health informatics to facilitate evidence-based decisions and improve preparedness, especially in vulnerable regions.
• Think about how to address data collection and sharing challenges.  

Speakers

• Felix Holl, PhD, MPH, M.Sc., FAMIA, Neu-Ulm University of Applied Sciences
• James Tcheng, MD, Duke University Health System
• Elizabeth Campbell, MS, MSPH, PhD, Columbia University Department of Biomedical Informatics
• Farah Magrabi, PhD, Macquarie University, Australian Institute of Health Innovation

Deep Learning-based Time-to-event Analysis of Depression and Asthma using the All of Us Research Program

While there is a growing recognition of the association between depression and asthma, few studies have leveraged deep learning-based (DL-based) models in analysis and prediction on a large sample size retrospective cohort study. We analyzed the association between depression and asthma among 239,161 participants of the All of Us Research Program through DL-based, logistic regression, and Cox Proportional Hazards (CoxPH) models. We used SHAP values to help interpret DL-based models and c-index to help model performance. Results suggest a significant odds ratio for depression in asthma. The c-indices for the CoxPH, DeepSurv, and DeepHit were 0.619, 0.625, and 0.596, respectively. SHAP indicated a different set of important variables when compared with CoxPH. In conclusion, we provide strong evidence of a positive relationship between depression and asthma. DL-based models did not outperform the CoxPH model on the c-index. Sex and income may play important roles in depression in asthma patients.

Learning Outcomes

Describe the Cox Proportional Hazard model, Deep-learning based time-to-event models. Discuss the utility of SHAP values to interpret deep learning models. Design a time-to-event analysis study in the All of Us Research Program. Understand the positive association between depression and asthma.

Speakers

• Xueting Wang, Master of Public Health, Yale University

Identification and Validation of Common Respiratory Infections in All of Us

This study integrates billing codes, multiple methods of microbiological testing, and medication usage from the All of Us Research Program to identify pathogen-specific seasonal respiratory infections. We validate this phenotyping approach against CDC epidemiological trends to demonstrate accurate seasonality of detection and capture the impact of reduced transmission during the COVID-19 pandemic.

Learning Outcomes

• Compose an EHR computable phenotype to identify seasonal respiratory infections and validate these cohorts using comparison with the CDC national surveillance.

Speakers

• Bennett Waxse, MD, PhD, NIAID/CNH

Pregnancy Outcomes in Hidradenitis Suppurativa Patients

Hidradenitis suppurativa is an autoinflammatory condition resulting in painful cysts, nodules, and sinus tracts in areas of high skin on skin contact. The microenvironment of affected tissues is high in pro-inflammatory cytokines and T-helper 17 cells. Other auto-inflammatory diseases, like psoriasis, have an enhanced risk of systemic inflammation and an elevated risk of spontaneous abortion. A cohort of pregnant patients from Cerner Health Facts® was identified using a Python adaptation of a validated pregnancy identification and classification algorithm. The HS population was identified among the pregnant population and was shown to be statistically significantly associated with outcome type by Chi square. A multinomial logistic regression also indicated a statistically significant increase in the odds of a pregnant patient having a spontaneous abortion over a live birth when controlling for thyroid disease, polycystic ovarian syndrome, antiphospholipid syndrome, other inflammatory diseases, and advanced maternal age.

Learning Outcomes

• Understand the interpretation of multinomial logistic regression.

Speakers

• David Walsh, BS, UMKC

EHR Phenotyping Methods for Measuring Treatment Adherence Among People Living With HIV in All of Us: Towards Disparities and Inequalities in HIV Care Continuum

HIV treatment adherence is among the most important determinants of HIV outcomes. However, only 50% of people living with HIV in the US were retained in care. Measuring HIV treatment adherence in the clinical settings is feasible but when it comes to the growing number of multi-site Electronic Health Records (EHR), there has been a dearth of research for adequate informatics methods to handle EHR. We sought to address this gap by developing a cluster of metrics for measuring HIV treatment adherence via EHR phenotyping. Our methods were developed and tested in the All of Us research program. We also performed preliminary analyses to explore disparities in HIV treatment adherence and demographic factors contributing to poor adherence. This study paves the way for systematic data mining and analyses for the HIV care continuum, disparities, and inequality on All of Us and other EHR normalized with the OMOP Common Data Model.

Learning Outcomes

• Understand EHR phenotyping in the All of Us research program. Develop HIV treatment adherence measurements using EHR data. Explore health disparities and inequalities in the HIV care continuum using EHR data.

Speakers

• Tianchu Lyu, PhD, University of South Carolina

Association of Concurrent Secure Messages on Clinician Workload

Secure messaging applications are ubiquitously used in clinical settings for clinician communication. We investigated effect of concurrent secure messaging—engaging in multiple conversational threads of communication—on clinician workload. Based on a large-scale study involving more than 2,400 clinicians (physicians, APPs, and trainees) and over 280,000 secure messages, we found that having 2 or more concurrent conversations increased total EHR time spent per day. The effects were significantly higher with an increasing number of concurrent conversations.

Learning Outcome

• Recognize that conversational multitasking is associated with increased EHR time and a higher number of patient switches; increased clinician workload and cognitive burden.

Speakers

• Linlin Xia, PhD, Washington University

Toward Relieving Clinician Burden by Automatically Generating Progress Notes using Interim Hospital Data

Regular documentation of progress notes is one of the main contributors to clinician burden. The abundance of structured chart information in medical records further exacerbates the burden, however, it also presents an opportunity to automate the generation of progress notes. In this paper, we propose a task to automate progress note generation using structured or tabular information present in electronic health records. To this end, we present a novel framework and a large dataset, ChartPNG, for the task which contains 7089 annotation instances (each having a pair of progress notes and interim structured chart data) across 1616 patients. We establish baselines on the dataset using large language models from general and biomedical domains. We perform both automated (where the best performing Biomistral model achieved a BERTScore F1 of 80.53 and MEDCON score of 19.61) and manual (where we found that the model was able to leverage relevant structured data with 76.9% accuracy) analyses to identify the challenges with the proposed task and opportunities for future research.

Learning Outcome

• Explain the significance and tasks involved in automatically generating clinical progress notes.
• Describe a framework for generating progress notes using interim hospital data.
• Identify and describe various strategies for evaluating automatically generated clinical notes.

Speakers

• Sarvesh Soni, PhD, National Library of Medicine (NLM)

Evaluating the Impact of Billing Patient Messages as E-Visits on Clinician EHR Inbox Burden

Following the onset of the COVID-19 pandemic, patient-initiated secure messages to clinicians increased dramatically and has remained at an elevated level. Many health systems have sought solutions to address the resulting clinician EHR inbox burden, including billing for a sub-set of these messages as ""e-visits."" Using EHR metadata from all clinicians delivering outpatient care at UCSF Health, a large academic medical center that implemented e-visit billing in November 2021, from November 2020 to November 2022, we describe clinician adoption of e-visits and use a difference-in-differences framework to identify the impact of e-visit billing on clinician EHR inbox time. We found that physicians, and clinicians practicing in family medicine, dermatology, or internal medicine, billed the most e-visits. Importantly, we found that clinicians in the top quartile of e-visit billing reduced their monthly inbox time by 19.6 minutes, a roughly 5% reduction in overall EHR inbox time, compared to clinicians in the lowest quartile of e-visit billing. These results suggest that clinicians who adopt e-visits more intensively may realize durable reductions in EHR inbox burden, while those who bill fewer e-visits are unlikely to see the same reductions.

Speakers

• A J Holmgren, PhD, University of California, San Francisco

Digital Overload: A Comparison of Electronic Health Record Time and Inbox Volume among Advanced Practice Providers and Physicians

This study compares electronic health record (EHR) burden among advanced practice providers (APPs) and physicians in ambulatory care settings from a large academic medical center. Using EHR audit log data, our findings suggest that APPs spend more time in the EHR and are responsible for more EHR messages than physicians. Time in the EHR has been highly associated with burnout, and therefore, organizations should address strategies to mitigate burnout among these essential providers.

Learning Outcome

• Explore differences in ambulatory care provider electronic health record use a month advanced practice providers and physicians.

Speakers

• Magdalene Kuznia, BSN, MS-HCI, RN, University of California San Francisco

Triaging Inbox Work: An Interview Study with Primary Care Physicians

Through semi-structured interviews with nine physicians we identify four themes in how primary care teams triage inbox work: 1) EHR inboxes require constant triage, 2) team support for triaging and performing inbox work can vary due to having new or distributed team members, 3) the initial triage and later involvement of team members affects how PCPs triage inbox work, and 4) expectations for rapid responses to patient messages contribute to EHR use outside clinic hours.

Learning Outcome

• Understand how primary care physicians triage their EHR inboxes.

Speakers

• Adam Rule, PhD, University of Wisconsin - Madison

The Moderating Effect of Health Literacy on the Impact of a Mobile Remote Monitoring Intervention with Tailored Messages for Breast Cancer Survivors: A Post Hoc Analysis of a Randomized Controlled Trial

Women with breast cancer starting adjuvant endocrine therapy were randomized to an app-based remote monitoring intervention with and without tailored educational messages versus enhanced usual care. Among participants with low health literacy randomized to the remote monitoring app with tailored messages, 80% had high AET adherence over 12 months compared with 42.1% in enhanced usual care (p=0.01); there were no significant differences by study arm for those with high health literacy.

Learning Outcomes

• Understand the moderating effect of health literacy on the efficacy of a remote monitoring intervention on one-year adjuvant endocrine therapy adherence among women with early-stage breast cancer.

Speakers

• Ilana Graetz, Emory University

Key considerations regarding usability and effective mobile app integration into two electronic health record systems

Substitutable Medical Applications and Reusable Technology (SMART)® Applications (app) that are compatible with the Fast Healthcare Interoperability Resources (FHIR)® are the standard for integrating mobile apps and electronic health records (EHR). Our team developed a SMART on FHIR mobile application, Info Viz for Health®, to support clinical HIV-related communication with diverse persons with HIV. Unfortunately, relatively little is known regarding clinician perspectives and preferences of mobile app integration with EHRs, which if not obtained, could render apps integrated into EHRs useless. Our study objectives were to explore the perceptions of clinical EHR users from both a developed (United States (US)) and a developing (Dominican Republic (DR)) setting regarding the usability of our app and identify key factors that researchers and designers should consider when creating apps to integrate with EHRs. We conducted semi-structured in-depth interviews with n=26 clinicians (n=13 per site) who provide HIV-related health education. Interviews were led with rigorously developed guides that contained questions based on Davis’s technology acceptance model and questions to explore characteristics of effective app integration. Interviews were analyzed using qualitative content analysis. Findings indicated high perceived usability of the Info Viz for Health app and several important considerations for effective integration of apps with EHRs were identified. Namely, apps integrated with EHRs must be easy and intuitive to access/use, and must leverage commonly used features of EHRs. These findings will provide valuable information for researchers, organizations, and/or other professionals designing health-related apps for EHR integration.

Learning Outcomes

• Identify and describe several important considerations while designing mobile applications for future integrations with EHR.

Speakers

• Samantha Stonbraker, PhD, MPH, RNz, University of Colorado College of Nursing

Examining Barriers to the Adoption of a Digital Mental Health Intervention: A Mixed-Methods Study using Thematic Analysis and Machine Learning

In our mixed-methods study, we examine barriers to the adoption of Digital Mental Health intervention (DMH) among young adults. We identified digital literacy, access, in-person interaction, and a need for personalization as barriers to DMH adoption. Machine learning insights highlight significant factors influencing engagement, including illegal drug use, and suicide ideation. We recommend integrating AI and real-time support into DMH services, tailored to young adults’ socio-cultural context, to enhance engagement and effectiveness.

Learning Outcomes

• Understand the underlying reason for college students' early dropout on digital mental health intervention (DMHI).

Speakers

• Ha Na Cho, Ph.D, University of California, Irvine

Computationally-guided Qualitative Analysis of User-Generated Data for Different Models of Mobile-Personal Health Records Apps

Mobile Personal Health Records (mPHR) are smartphone apps granting patients portable and continuous access to their medical records on the go, thereby increasing their potential to play an active role in managing their healthcare. An extensive body of literature has focused on understanding user(s) experiences with web-based tethered PHRs (i.e., Patient Portals) offered by healthcare organizations. However, patients' opinions of smartphone-based PHRs have received less attention. Our study aims to understand this gap. We used a computationally-guided qualitative analysis approach to identify latent topics indicating dimensions of user experiences present in app reviews left on popular m-PHR apps available on Google Play and Apple app stores. After following a detailed app selection process, 10 m-PHR, including tethered (n=6) and interconnected (n=4) apps, were selected for analysis. Our findings show similarities in user experiences for HCO-tethered PHRs and HCO-independent interconnected PHRs, and we discuss the design implications concerning the differences.

Learning Outcomes

• Identify the three levels of PHR integration.

Speakers

• Zainab Balogun, University of Maryland Baltimore County

Getting people access to services is also getting them access to a phone: Clarifying digital divide dynamics and their consequences in Community Mental Health Care

Access to mental healthcare is increasingly technologically-mediated. People with low socioeconomic status (SES) and serious mental illness (SMI) face lower rates of tech ownership and may lack technological skills, called digital divides. Yet, little is known about how digital divides may impact mental healthcare access. Therefore, a qualitative study (ethnographic observations and interviews) was conducted with staff working with low-SES SMI patients using community mental health care (CMH) (N=14). Findings showed that consumers struggled to maintain consistent internet—and thus mental healthcare—access despite owning smartphones. Consumers frequently faced care disruptions due to broken, lost, or uncharged phones. Staff and patients created effortful but ad-hoc workarounds to restore access during technological access disruptions. These solutions frequently occurred after healthcare appointments were missed. Digital divide concepts should accommodate the work necessary to maintain technology access even after ownership and its impact on care access—especially among low-SES SMI patients.

Learning Outcomes

• Understand how technology access and technology skills impact care access for low SES individuals with SMI.
• Identify ways consumers and care workers can accommodate technological disruptions to enable care access.

Speakers

• Alicia Williamson, School of Information, University of Michigan

Barriers and Facilitators of Digital Health Use for Self-Management of Hypertensive Disorders by Black Pregnant Women

Digital health is popular for managing health conditions; however, these applications are often developed with few considerations of the differences across user populations. Tailoring such applications to include cultural considerations could lead to better adoption and adherence in such programs, but a reproducible framework is needed. This study aims to capture Black women’s barriers and facilitators in self-managing hypertensive disorders of pregnancy (HDP) using digital health products. One-on-one interviews were conducted with 17 Black pregnant women with HDP using a semi-structured interview guide. Qualitative data obtained was analyzed using grounded theory and 38 codes were mapped within the four levels of the socioecological model of health. Themes were created that identified barriers and facilitators of the women’s pregnancy experiences and used to influence the feature development of a digital health intervention. Future work will instantiate and validate a framework that provides theoretical constructs for developing culturally tailored digital health interventions.

Learning Outcomes

• Understand the importance of considering cultural differences in their digital health and informatics took development.

Speakers

• Morgan Foreman, PhD Candidate, UTHealth Houston McWilliams SBMI & IBM Research

Association Between Digital Health Use and Hard of Hearing Status in a National VA Sample: Examining Secure Messaging and Video-based Modalities

Despite elevated hearing loss and tinnitus rates in the Veteran population, little is known about digital health use among hard of hearing Veterans. The current project examined Veterans Health Administration secure messaging and video visit use among hard of hearing patients. Hard of hearing patients showed higher secure messaging (p<.05) and video visit (p<.05) use. These findings support allocating resources to optimize digital health use and experiences among hard of hearing persons.

Learning Outcomes

• Understand variations in video telehealth use and asynchronous secure messaging use between hard of hearing an non-hard of hearing veterans.

Speakers

• Taona Haderlein, PhD, OCHIN, Inc

Reducing the Stigma of Sexual and Reproductive Health Care Through Supportive and Protected Online Communities

In many cultures where discussions and care-seeking for sexual and reproductive health (SRH) are stigmatized, unmarried women often suffer silently, facing risks of sexually transmitted infections and gynecological complications. South Korea exemplifies this challenge, with SRH topics remaining stigmatized, potentially contributing to Korean women’s high incidence rates of cervical cancer. To address this problem, we designed and studied a protected online community for unmarried Korean women with 9 weeks of guided activities relating to SRH. We describe how these activities helped participants reflect on and discuss the typically taboo topics surrounding SRH. Results indicate that the online community effectively supported participants in initiating additional offline conversations about SRH with more people, and even encouraged some women to seek clinical care. This work sheds light on the potential of supportive and protective online communities to facilitate SRH, offering newfound options for supporting women in cultures where such care is stigmatized.

Learning Outcomes

• Understand the importance of creating supportive, protected online spaces for fostering reflection on stigmatized health experiences and connections among people facing stigma.

Speakers

• Hyeyoung Ryu, MS, University of Washington

Designing for Better Pre-hospital Communication: Participatory Design of a Telemedicine Application for Emergency Departments

Pre-hospital communication, which usually refers to the communication process between pre-hospital and hospital providers, is crucial for the effective management of critically injured or ill patients. Despite its importance, persistent challenges such as miscommunication have been significant barriers. Telemedicine systems have been proposed to overcome these challenges, yet existing research primarily focuses on using off-the-shelf systems to evaluate their feasibility and effectiveness of implementation without investigating users' needs and perceptions. To bridge this research gap, our study employed a user-centered design approach to co-create an integrated telemedicine system with emergency care providers to ensure that the system meets the specific needs of care providers and aligns with existing clinical workflows. We present the system design process, the features desired by users to address challenges in pre-hospital communication, and the socio-technical considerations for implementing telemedicine in the dynamic emergency care setting. We conclude the paper by discussing the design implications.

Learning Outcomes

• Understand the user needs and socio-technical considerations for developing a telemedicine system that improves the pre-hospital communication.

Speakers

• Enze Bai, Phd Candidate, Pace University

Supporting Personalized prEgnancy Care wIth Artificial intelligence (SPECIAL): An Acceptability Study of a Personalized Educational Platform

The SPECIAL project aims to support postpartum depression (PPD) prevention through personalized educational. This study assesses the acceptability of the platform among pregnant individuals. Utilizing the Unified Theory of Acceptance of Use of Technology framework, surveys were conducted with 41 participants. Results suggest potential associations between demographic factors and intention to use the platform. The study finding underscores the feasibility of personalized technology for PPD prevention, highlighting the need for further investigation into socio-demographic influences.

Learning Outcomes

• Understand the important factors influencing the technology acceptance for postpartum depression prevention among patients.

Speakers

• Ziwen Zhang, MS, Weill Cornell Medicine

MentalGPT: Harnessing AI for Compassionate Mental Health Support

This paper introduces MentalGPT, fine-tuned large language models (LLMs) designed to function as a compassionate therapist in the realm of mental health support. Through the application of efficient model fine-tuning techniques, we have created LLMs capable of providing comprehensive and empathetic responses, simulating human-like interactions while delivering personalized mental health guidance. Five open-sourced LLMs with a size of 7B parameters were instruction fine-tuned using the Quantized Low-Rank Adaptations (QLoRA) method on a GPT-generated synthetic dataset, a dataset curated from interview transcripts, and a combination of both datasets. The performance of the LLMs was judged and scored by Google Gemini Pro on seven devised metrics targeting important aspects of mental health support. All fine-tuned models outperform their base models and existing models tailored for mental health support. The Mistral-V0.1 7B model finetuned on the interview data scored the highest in all seven metrics. Our work highlights the potential for LLMs to play a valuable role in mental health support by offering accessible and non-judgmental platforms for users to seek guidance and share their concerns. By bridging AI and mental health, this research offers a promising avenue to expand support services and reduce the stigma associated with seeking help.

Learning Outcomes

• Understand the process and benefits of using Quantized Low-Rank Adaptations (QLoRA) for fine-tuning large language models (LLMs) to provide effective mental health support.
• Recognize the importance of combining synthetic and interview data in enhancing the performance of LLMs in therapeutic communication.
• Identify the comprehensive evaluation metrics used to assess the effectiveness of MentalGPT in delivering empathetic and ethical mental health support.
• Appreciate the role of MentalGPT in expanding access to mental health services and reducing the stigma associated with seeking help.

Speakers

• Jia Xu, M.S., University of Pennsylvania

MyPostDischargePal: Preliminary Pilot of an Interoperable App for Adverse Event Surveillance Post-Discharge

We used a user center designed approach to develop, iteratively refine, and pilot a real-time symptom and global health monitoring intervention to mitigate risk of adverse events (AEs) post-hospital discharge. Mixed method analyses of pilot data and participant interviews suggest acceptability among patients and clinicians with modest refinements. Next steps include conducting a randomized controlled trial to evaluate the impact of this type of intervention on post-discharge AEs for patients with multiple chronic conditions.

Learning Outcomes

• Understand the core functionalities most important to patients and clinicians of an interoperable app designed to help patients self-monitor symptoms and overall health.

Speakers

• Madeline Smith, MPH - Master of Public Health, Brigham and Women's Hospital

Integrated Hands-Free Electronic Patient Care Report (ePCR) Charting (IHeC): Designing the Architecture

The nature of paramedic workloads typically results in incomplete or lack of patient care reports on patient handover to emergency department staff. Patient information gaps can increase emergency department staff's workload, cause care delays, and increase risks of adverse events. An integrated hands-free electronic patient care report (ePCR) could eliminate this gap. We conducted an environmental scan of the available literature on technologies to improve paramedic documentation and current advanced paramedic charting systems. Two technologies, speech recognition documentation and live telemetry sharing systems, were identified as potential improvements. A theoretical architecture for an integrated hands-free ePCR charting (IHeC) system was developed by combining these technologies. The ePCR could be completed and available upon patient arrival at the hospital using speech recognition and vital sign-sharing technology. The IHeC system could solve the problem of patient information gaps and provide a platform for more advanced integration of paramedic services.

Learning Outcomes

• Describe challenges faced by paramedics in delivering accurate and timely documentation on patient handover.
• Explain the current landscape of paramedicine documentation in terms of improving care continuity and preventing adverse events using technology.

Speakers

• Desmond Hedderson, BSc, MSc student, University of Victoria, school of Health Information Science

Automating and Evaluating LLM-Generated ED Handoff Notes

It has been burdensome for physicians to review and document large unstructured clinical data. We develop a summarization system that automatically generates Emergency Medicine handoff notes using LLMs, specifically designed for text generation in the clinical domain. We also propose a novel clinical evaluation rubric focused on the quality and safety of generated texts. The evaluation results show the effectiveness of the proposed framework.

Learning Outcomes

• Appreciate the importance and limitations to current emergency medicine handoff process.
• Understand the context of how LLMs can automate emergency medicine handoff documentation at near-physician documentation levels
• Understand how our novel LLM evaluation framework measures quality, accuracy, and usefulness of the hand-off notes, in relationship to patient safety.
• Explain our LLM path-to-implementation plan to navigate AI safety and governance obligations.

Speakers

• Vince Hartman, MS, Information Systems, Abstractive Health

Standardized documentation of nursing communication with advanced providers identifies evident and occult hypoxemia

We hypothesized that nursing documentation may increase when hypoxemia is present, but undetected by the pulse oximeter, in events termed “occult hypoxemia.” Methods: We conducted a retrospective study of patients with COVID-19 at five hospitals in a healthcare system with paired SpO2 and SaO2 readings (measurements within 10 minutes of oxygen saturation levels in arterial blood, SaO2, and by pulse oximetry, SpO2). We applied multivariate logistic regression to assess if having any nursing documentation of provider notification in the four hours prior to a paired reading confirming occult hypoxemia was more likely compared to a paired reading confirming normal oxygen status. Results: Among the 1,910 patients with 44,972 paired readings, having any nursing documentation of provider notification was 46% more common in the 4 hours before an occult hypoxemia paired reading compared to a normal oxygen status paired reading (OR 1.46, 95% CI: 1.28-1.67), and 84% more common before an evident hypoxemia paired reading (OR 1.84, 95% CI: 1.62-2.09). Discussion This study finds that nursing documentation of provider notification significantly increases prior to confirmed occult hypoxemia, which has potential in proactively identifying occult hypoxemia and other clinical issues.

Learning Outcomes

• Describe how nurse documentation frequency may change in the setting of hypoxemia, even when the hypoxemia is not evident on the pulse oximeter.

Speakers

• Kelly Gleason, PhD, RN, Johns Hopkins University

Ambient AI Scribes: Utilization and Impact on Documentation Practice

This study examines the initial implementation and evaluation of an ambient digital scribe powered by large language models at Stanford Health Care. During the 3-month pilot period, the tool was utilized for over half of the encounters and resulted in decreased time spent on clinical documentation. Adoption of the tool and the resulting documentation practices varied between physicians, suggesting a need for further development of this technology to accommodate personal preferences to maximize its potential.

Learning Outcomes

• Understand the potential benefits of ambient AI scribes.

Speakers

• Stephen Ma, MD, PhD, Stanford University School of Medicine

A Web-based Interface for Visualizing and Documenting SEEG Strategic Planning (WISP): Development and Qualitative Evaluation

WISP stands as an efficacious solution to the challenges associated with Stereoelectroencephalography (SEEG) strategic planning, offering lightweight and interactive web interfaces for rendering multiple brain views. These interfaces facilitate collaborative engagement among care team members across various disciplines during patient case conferences and SEEG strategic planning sessions. Moreover, WISP incorporates a collaborative electrode and electrode group library, serving as a standardized repository of knowledge. The application enables seamless conversion of case conference outcomes and SEEG plans into images and PDF files, with transmission to Electronic Health Record (EHR) systems through a customized HL7 engine. The initial assessment findings demonstrate WISP provides good usability according to the System Usability Scale (SUS) score, with physicians exhibiting a clear preference for its utilization over conventional approaches to case conference documentation and SEEG planning. Furthermore, physicians have actively embraced WISP in their collaborative sessions, indicating its seamless integration into their clinical workflows.

Learning Outcomes

• Know how many brain views are implemented in WISP to render SEEG electrodes.

Speakers

Shiqiang Tao, PhD, The University of Texas Health Science Center at Houston

Assessing the Impact of EHR Documentation Burden on Health Information Exchange Use

While electronic health record (EHR) documentation burden is known to be associated with reduced clinician well-being and burnout, it may have even worse unintended consequences if documentation work also crowds out other high-value EHR tasks. We examine this novel question by assessing the relationship between documentation burden and a high-value but optional EHR task – use of health information exchange (HIE) to view patient records from outside organizations. Our study takes advantage of an exogenous shock to documentation time, appointment no-shows. We find that documentation time has a strong impact on HIE use, with each additional hour spent documenting resulting in a 7.1 percent reduction in the proportion of a patients with an outside record viewed by the physician seeing them that day. This crowd out effect may explain why the US has yet to realize broad benefits from HIE and could also be true for other high-value EHR and non-EHR tasks as busy physicians simply lack time to incorporate them into their workflows. Our results point to the urgent need for policymakers to do more to reduce documentation burden.

Speakers

• A J Holmgren, PhD, University of California, San Francisco

Exploring the Utilization of Synthetic Data in Unsupervised Clustering for Opioid Misuse Analysis

Privacy and security restrictions on medical data pose challenges to collaborative research, making synthetic data an increasingly attractive solution. Recent advancements in Generative AI technologies, like GAN models, have improved synthetic data generation. This study investigates the use of synthetic data in clustering models for opioid misuse analysis, generating a dataset that replicates real-world data from 2017 to 2019, including demographics and diagnosis codes. By maintaining patient privacy, we enable comprehensive analysis without compromising security. We developed unsupervised clustering models to identify opioid misuse patterns and assessed the effectiveness of synthetic data across four scenarios: training on real dataset and testing on real dataset (TRTR), training on real dataset and testing on synthetic dataset (TRTS), TSTR, and TSTS. Results demonstrate that synthetic data can replicate real data distributions and clustering characteristics as a training set, offering significant potential for collaborative model development and optimization without exposing privacy or security risks.

Learning Outcomes

• Describe why synthetic data demonstrates a promising pathway for advancing medical research under privacy constraints and data scarcity challenges.
• Explain how the overall alignment between TRTR and TSTR in demographic distributions an the ability to replicate key trends in clustering tasks affirms the potential of building training models on synthetic data and testing on real data.
• State the direction of future research based on the presentation.

Speakers

• Yili Zhang, PhD, Georgetown University

Generating Synthetic Test Data Using LLMs for Automated Testing of a Patient-Focused, Survey-Based System: Does Generative AI Live Up to the Hype?

The excitement around the possibilities of generative AI is infinite, but can it live up to the hype when the goal is to generate quality test data quickly and efficiently as part of an automated testing process? In the context of a patient-focused, survey-based system, we demonstrated the potential of generative AI to create custom synthetic data using 2 different large language models (GPT 3.5 and Flan T5-XL) in AWS and Azure environments. While we improved test coverage and efficiency by synthetically generating many test cases, the experience included technical and communication challenges as well as complexities associated with balancing the desire for high utility and realism in the data with the available testing resources. Recommendations range from defining and gaining consensus on evaluation metrics early in the process as it influences technical questions like persona creation and prompt-engineering to encouraging test teams to build flexible frameworks from the start.

Learning Outcomes

• Understand the differences in synthetic data generated by GPT 3.5 and Flan T5-XL for the purposes of testing a patient-focused, survey-based system.

Speakers

• Catherine Anderson, PhD, Accenture

Clinician Perceptions of Generative Artificial Intelligence Tools and Clinical Workflows: Potential Uses, Motivations for Adoption, and Sentiments on Impact

Successful integration of Generative Artificial Intelligence (AI) into healthcare requires understanding of health professionals’ perspectives, ideally through data-driven approaches. In this study, we use a semi-structured survey and mixed methods analyses to explore clinicians’ perceptions on the utility of generative AI for all types of clinical tasks, familiarity and competency with generative AI tools, and sentiments regarding the potential impact of generative AI on healthcare. Analysis of 116 clinician responses found differing perceptions regarding the usefulness of generative AI across clinical workflows, with information gathering from external sources rated highest and communication rated lowest. Clinician-generated prompt suggestions focused most often on clinician decision making and were of mixed quality, with participants more familiar with generative AI suggesting more high-quality prompts. Sentiments regarding the impact of generative AI varied, particularly regarding trustworthiness and impact on bias. Thematic analysis of open-ended comments highlighted concerns about patient care and the role of clinicians.

Learning Outcomes

• 1. Understand the perspectives of clinicians regarding the potential usefulness of generative AI tools in different types of clinical workflows.
• Understand the ability of clinicians to suggest quality prompts.
• Understand which factors would motivate clinicians to adopt a generative AI tool.
• Understand the distribution of sentiments on the potential impact of generative AI on clinical practice.

Speakers

• Elise Ruan, MD, MPH, Columbia University Department of Biomedical Informatics

A framework for evaluating the value of generative AI in healthcare

In response to the burgeoning integration of Large Language Models (LLMs) and generative AI (Gen AI) within healthcare, our study at the Children's Hospital of Philadelphia introduces a pioneering framework to assess the value of Gen AI applications. This framework, designed through our pilot within the Epic EHR system, transcends traditional evaluation metrics to encompass a holistic array of dimensions including efficiency, user experience, scalability, reliability, and relatability, alongside an in-depth cost analysis. Our pilot revealed notable insights: clinicians and support staff experienced efficiency gains with reduced response times by leveraging LLM-generated drafts for patient communication. Specifically, clinicians saw a 14-second reduction in response times, with a corresponding 17-second reduction observed among support staff. Despite these efficiencies, a nuanced cost-benefit analysis underscored the complexity of justifying Gen AI implementation based solely on cost savings. Our findings indicate the multifaceted value of Gen AI extends beyond immediate financial gains, enhancing healthcare delivery through improved interaction quality and enabling tasks previously unfeasible without Gen AI. This study underscores the imperative of a comprehensive, multidimensional approach to evaluating Gen AI in healthcare. By presenting a framework that captures the broad spectrum of Gen AI's value, we aim to foster a deeper understanding of its potential to transform healthcare delivery, encouraging further exploration and refinement in the assessment of Gen AI's true value in healthcare settings.

Learning Outcomes

• Identify and evaluate the key dimensions of value when implementing generative AI solutions in healthcare settings beyond traditional ROI metrics.

Speakers

• Stephon Proctor, PhD, ABPP, Children's Hospital of Philadelphia

Exploring the Impact of Explainable AI on Clinicians’ Acceptance of AI-Generated Results in Healthcare

The adoption of advanced artificial intelligence (AI) techniques is rapidly advancing in healthcare. Despite AI's expanding role, challenges associated with transparency issues became a significant barrier to its full adoption in clinical practice. Explainable AI (XAI) emerged as a solution to improve clinicians’ acceptance of AI produced results by clarifying the inference process. Yet, the impact of XAI methods on clinicians’ acceptance of AI results has not been fully explored. This study explored how XAI affected clinicians’ mental models of decision making, trustworthiness, and satisfaction towards AI-generated results.

Learning Outcomes

• Discuss the need for continual refinement of XAI explanations to align with clinician feedback and support their evolving understanding, ensuring effective integration of AI systems in clinical practice.

Speakers

• Jinsun Jung, Master, Seoul National University

Neural Mosaics: Detecting Aberrant Brain Interactions using Algebraic Topology and Generative Artificial Intelligence

Epilepsy is a neurological disorder affecting more than 50 million worldwide with up to 30% of patients remaining refractory to medications. Accurate seizure detection is crucial for surgical planning and successful outcomes. Neurophysiological signal-based seizure detection methods are complicated, computationally expensive, and time consuming due to the large volumes of data from long periods of seizure monitoring and the complex methods for feature extraction from signal data. Moreover, these methods fail to capture multifocal interactions between brain regions. Persistent homology offers robust representations of complex brain interaction patterns. We propose a novel approach to classifying persistent homology structures representing brain interaction dynamics in epilepsy using the Google Gemini Pro Vision large language model (LLM). Using intracranial electroencephalography (iEEG) from refractory epilepsy patients, we apply persistent homology to one-second epochs during seizure and non-seizure periods and generate persistence diagrams to visualize the results. We introduce new prompting template for Gemini 1.0 Pro-Vision model to classify these diagrams, distinguishing multifocal brain interactions from seizure and non-seizure activity. To our knowledge, this is the first study to use persistence diagrams as input to a foundational model for analyzing aberrant brain interaction dynamics. In contrast to traditional approaches of using machine learning algorithms for EEG classification that require hand crafted feature engineered data, large volume of representative training data, and brittle hyperparameter tuning, our approach is a robust method that combines recent advances in algebraic topology and LLMs to analyze large-scale EEG data for seizure detection without requiring large volumes of training data.

Learning Outcomes

• Understand promises and limitations of using in-context learning with Goodle's Gemini 1.0 Pro-Vision model to classify persistence diagrams generated from intracranial electroencephalography (iEEG) signals to differentiate multifocal brain interactions during seizure and non-seizure events.

Speakers

• Katrina Prantzalos, MS, Case Western Reserve University

The advent of large language models (LLMs) and generative AI, such as GPT-4, has marked the beginning of an era filled with highly efficient AI-enabled solutions for life science industry. This multi-stakeholder panel will explore the current landscape of AI in life sciences from various industrial perspectives, encompassing the health system, biotechnology, and pharmaceutical industries. We will discuss how we can rapidly learn from EHR and real-world data, leveraging industry-specific clinical AI to interpret vast datasets at an unprecedented pace, thereby accelerating drug development in life sciences. Our goal is to showcase the tangible impact of AI within life sciences, review the challenges of last-mile adoption, examine the stances of regulatory agencies on the use of AI applications, filter through the noise, and start to build a consensus on defining realistic, ethical, equitable, and efficient adoption of AI in life science applications, spanning from drug development to regulatory decisions.

Speakers

• Ganhui Lan, PhD, Pfizer
• Phil Lindemann, BS, Epic
• Ying Li, Ph.D, Regeneron Pharmaceuticals
• John Cai, MD, PhD, FAMIA, Merck

Robust Visual Identification of Under-resourced Dermatological Diagnoses with Classifier-Steered Background Masking

Collecting images of rare dermatological diseases for machine learning detection applications is a costly, laborious task. It is difficult to collect enough images of these diagnoses to avoid the risk of low accuracy "in the wild." One of the sources of bias in these networks is irrelevant background pixel data. These pixels necessarily have no clinical significance, yet Deep Neural Networks will make weak correlations based on that information. To reduce their ability to do this, we introduce a masking augmentation algorithm, InfoMax-Cutout. It employs unsupervised Information Maximization losses to mask out background pixels. InfoMax-Cutout increased accuracy on classifying 319 diagnoses by 0.76%. These features generalized to an unseen diagnosis task (Fitzpatrick 17k), improving accuracy over a baseline by 43.3% and reducing Gini inequality by 20.9%. This approach of learning to separate out background pixels can increase accuracy in detecting diseases in Lower and Middle Income Countries.

Learning Outcomes

• Describe how imaging ML models can fail to generalize due to irrelevant background information, and how to address these failures.

Speakers

• Miguel Dominguez, PhD, VisualDx

A Multi-Task Learning Approach for Segmentation of Breast Arterial Calcifications in Screening Mammograms

Screening mammogram is a standard imaging procedure to measure breast cancer risk among 45+ year old women. Quantifying breast arterial calcification (BAC) from screening mammograms is a non-invasive and cost-efficient approach to assess the future risk of adverse cardiovascular events among women, such as heart attack and stroke. However, segmentation of breast arterial calcification is an involved task and poses several technical challenges such as extremely small BAC finding, low breast arteries to breast area ratio in the mammogram images, tissue features such as breast folds and heterogeneous density, have very similar imaging appearance. In this work, we aim to address the shortcomings of existing SOTA methods, e.g., SCUNet, and analyze the comparative performance. Given the fact that we will not be able to simply resize mammogram to preserve the resolution, we adopted a patch-based methodology for segmentation using the original resolution which may hinder the model understanding of whole mammogram. We propose a multi-task learning approach for patch-based BAC segmentation by adding an auxiliary task of patch position prediction which force the model to learn breast anatomy to comprehend the locations where BAC will not occur, such as breast boundary. The proposed method achieves state-of-the-art performance compared to the baselines. To demonstrate the utility, we also validate our method on external data and provide survival analysis for CVD based on the BAC score and provide a comparison with CAC score.

Learning Outcomes

• Understand the methodology for patch-based segmentation for smaller image findings.
• Describe the role of novel BAC biomarkers for MACE risk assessment

Speakers

• Imon Banerjee, PhD, Arizona State U, Mayo Clinic

Project Elucidate: Web-Based Single Cell Annotation Tool For Building Deep Segmentation Models on Stimulated Raman Histology

Single-cell analysis of cancer histology offers crucial insights into the tumor microenvironment. In brain cancer research, an advanced imaging technique called Stimulated Raman Histology (SRH) allows for rapid digital imaging of brain tumor biopsies without requiring tissue staining. SRH microscopes combined with AI are currently being used for intraoperative tumor classification in neurosurgery. However, there is yet to be single-cell annotations for SRH. Project Elucidate proposes a collaborative, open-source web platform for building cell segmentation AI models in SRH.

Learning Outcomes

• Evaluate advancements in AI-based cell segmentation.
• Discuss Stimulated Raman Histology (SRH).
• Annotate cells on SRH using Elucidate.
• Train a deep segmentation model using cell annotations

Speakers

Abhishek Bhattacharya, M.D., NYU Langone

Variogram Modeling of Spatially Variant Early Response to Concurrent Chemo- and Immunotherapy for Metastatic Non-Small Cell Lung Cancer

Predicting response of metastatic non-small cell lung cancer (mNSCLC) to chemo-immunotherapy (chemoICI) by incorporating the spatial correlation structure of PET imaging has potential to support clinical decisions regarding patient- and lesion-level risk stratification. As a prelude to extending our previous framework, the “Voxel Forecast” multiscale regression for predicting spatially variant tumor response, we explored different variograms models of spatial correlation in the mNSCLC chemoICI response stetting.

Learning Outcomes

• State the significance of variogram models in analyzing the spatial correlation structure of FDG PET/CT imaging for metastatic non-small cell lung cancer (mNSCLC).
• Apply variogram models to predict treatment response in patients with mNSCLC based on FDG PET/CT imaging data.

Speakers

• Faisal Yaseen, PhD student

Large-scale Text Mining of Suicide Attempt improves Identification of Distinct Suicidal Events in Electronic Health Records

In this study, we explore a natural language processing (NLP) algorithm’s capacity to identify proximal but distinct suicide attempt (SA) events compared to diagnostic code-based approaches. This study used an NLP algorithm with high precision in identifying SA events, which processes clinical notes for suicide-related text expressions and generates SA outcome relevance scores on mentioned dates. We chart reviewed all SA visit pairs less than 15 days apart. Despite sample size limitations, our NLP method surpassed the code-based model's performance (0.85 [95% CI: 0.74 - 0.92] vs. 0.78 [95% CI: 0.56 - 0.92], p = 0.71). More importantly, NLP detected three times more SA visit pairs <15 days compared to the code-based approach (71 vs. 23), with only 3 overlaps. This study demonstrates NLP's efficacy in identifying distinct SA visit pairs. Given minimal overlap, we suggest leveraging both clinical notes and diagnostic codes for a comprehensive SA event detection.

Learning Outcomes

• Understand the challenges in identifying real suicide attempt events in EHR data and how clinical notes and natural language processing can address these challenges.

Speaker

• Hyunjoon Lee, MS, Vanderbilt University Department of Biomedical Informatics

Using Large Language Models for sentiment analysis of health-related social media data: empirical evaluation and practical tips

Health-related social media data generated by patients and the public provide valuable insights into patient experiences and opinions toward health issues such as vaccination and medical treatments. Using Natural Language Processing (NLP) methods to analyze such data, however, often requires high-quality annotations that are difficult to obtain. The recent emergence of Large Language Models (LLMs) such as the Generative Pre-trained Transformers (GPTs) has shown promising performance on a variety of NLP tasks in the health domain with little to no annotated data. However, their potential in analyzing health-related social media data remains underexplored. In this paper, we report empirical evaluations of LLMs (GPT-3.5-Turbo, FLAN-T5, and BERT-based models) on a common NLP task of health-related social media data: sentiment analysis for identifying opinions toward health issues. We explored how different prompting and fine-tuning strategies affect the performance of LLMs on social media datasets across diverse health topics, including Healthcare Reform, vaccination, mask wearing, and healthcare service quality. We found that LLMs outperformed VADER, a widely used off-the-shelf sentiment analysis tool, but are far from being able to produce accurate sentiment labels. However, their performance can be improved by data-specific prompts with information about the context, task, and targets. The highest performing LLMs are BERT-based models that were fine-tuned on aggregated data. We provided practical tips for researchers to use LLMs on health-related social media for optimal outcomes. We also discuss future work needed to continue to improve the performance of LLMs for analyzing health-related social media data with minimal annotations.

Learning Outcome

• Use emerging LLM to analyze the sentiments of health social media data.
• Improve the performance of LLM on health social media data.

Speaker

• Lu He, PhD, University of Wisconsin-Milwaukee

The 20% among Suicide Deaths died Without Warning Signs: Trend Analysis based on Suicide Decedents in the US from 2003-2020

Understanding who is less likely to reveal suicidal intentions is crucial for developing effective prevention strategies, as suicide rates increase in the US, notably among marginalized groups. Existing studies, limited by small, homogeneous samples, fail to thoroughly analyze various demographics and contexts. This study investigates trends in disclosed and non-disclosed suicide deaths in the US from 2003 to 2020, considering age, gender, race, ethnicity, methods of suicide, intended recipients of disclosures, and drug/substance categories. It utilizes cross-sectional data from 500,072 suicide decedents across 49 states, Puerto Rico, and the District of Columbia, sourced from the National Violent Death Reporting System's Restricted Access Database, with statistical analyses conducted between October 2023 and January 2024. The main outcomes measured were disclosures of suicidal intent within one month prior to death and the presence of suicide notes. Results show a consistent 80:20 ratio of non-disclosed to disclosed suicides. Specific groups, including older adults, both genders, certain racial groups, and those who died by specific methods or substances, displayed significantly lower odds of disclosing suicidal intentions. Notably, Black decedents disclosed at markedly lower rates than White decedents, with disparities more pronounced among females of these racial groups. The study emphasizes the need for targeted suicide prevention strategies, especially for racial minorities, older adults, males, and individuals utilizing certain suicide methods or substances. It highlights the necessity for increased public health efforts to normalize mental distress and enhance access to mental health services. 2

Speaker

• Yunyu Xiao, PhD, Weill Cornell Medicine, Population Health Sciences

Assessing demographic differences in psychological pain, hopelessness, connectedness, and capacity for suicide based on terminology-driven natural language processing of VHA clinical progress notes

Psychological pain, hopelessness, connectedness, and capacity for suicide are among the most important drivers of suicidal behavior. Scores based on terminology-driven natural language processing (NLP) of Veterans Health Administration (VHA) clinical progress notes showed meaningful change in these four factors before patients attempted or died by suicide. It is as yet unknown if these changes depend on sex, age group, race/ethnicity, and being involved in the criminal legal system (e.g., in court or incarcerated). We will present results of a repeated measures analysis of psychological pain, hopelessness, connectedness, and capacity for suicide with a between-subjects component for the listed demographic subgroups. Clinical progress notes entered between 2014 and September 2022 during eight weeks before patients attempted or died by suicide (n=43,581, female 7089, male 36,492) were pulled from the VHA corporate data warehouse. These notes were tagged and labelled for the four factors using a vocabulary of terms available from BioPortal. Using these labels, patient mean scores for the four factors were computed across the last four weeks (1-4) and across the four weeks (5-8) before that. Repeated measures analysis of variance was used to test the effect of Time, patient subgroup, and the Time by subgroup interaction. Results support the hypothesis that our terminology-driven NLP pipeline to determine psychological pain, hopelessness, connectedness, and capacity for suicide captures meaningful change in demographic subgroups prior to a suicidal event. Scores for these four factors may support clinical decision-making regarding suicide prevention.

Learning Outcome

• Recognize alarming patterns in psychological pain, hopelessness, connectedness, and capacity for suicide that may signal an imminent suicide attempt.
• Understand that the level of psychological pain, hopelessness, connectedness, and capacity for suicide before a suicide attempt differs by demographic subgroup.

Speaker

• Esther Meerwijk, Phd, MSN, Weill Cornell Medicine, Population Health Sciences

Exceptional Jo: A Semi-Automated and Scalable System to Share Personalized Patient Positive Feedback with Employees

The Exceptional Jo recognition program, initiated by Vanderbilt University Medical Center in 2020, utilizes an automated system to match employees with positive patient feedback, fostering a culture of recognition and addressing caregiver burnout. By cross-referencing patient feedback with employee access to medical records, personalized recognition emails are sent, boosting employee morale. With almost 70,000 emails sent to date, the initiative has garnered overwhelmingly positive responses, showcasing its effectiveness in enhancing workforce engagement and well-being.

Learning Outcomes

• Understand how the Exceptional Jo Recognition Program at Vanderbilt University Medical Center (UVMC) uses patient feedback to enhance employee engagement and reduce burnout, highlighting the impact of personalized recognition on workforce satisfaction.

Speaker

• Peyton Larson, MPA, Vanderbilt University Medical Center

Configure or Integrate? Tradeoffs for Remote Symptom Monitoring Innovation with Electronic Health Records

There are two competing approaches for innovation with electronic health records (EHR): “configure” leverages EHR’s existing capabilities as much as possible; “integrate” views the EHR as a platform for integrating third-party tools. We compared technical feasibility and user experience implications of these approaches when implementing an asthma symptom monitoring intervention in two different health systems. We found fewer technical challenges implementing user requirements with the integrate, and pros and cons of each for user experience.

Learning Outcomes

• Understand tradeoffs between two approaches for innovation with electronic health records

Speaker

• Robert Rudin, RAND Corporation

A Case Study of Digital Phenotyping in a Large Integrated Healthcare System: An Evaluation of Veterans Sharing Unsolicited Patient-Generated Health Data

The Veterans Health Administration (VHA) recently launched a new mobile health app, allowing patients to voluntarily share patient-generated health data with their care teams. We examined early users of this app, including how they compared to the general VHA population and common digital phenotypes shared. We found that users of the SMHD had higher annual health care costs than non-users, despite being younger in age and living in more urban and higher income zip codes.

Learning Outcomes

• Gain new knowledge about a patient-generated health data (PGHD) collection effort in a large integrated health care system, potential for future clinical applications, and to better understand how veterans who share PGHD may differ from veterans who do not share PGHD.

Speaker

• Mark Zocchi, PhD, Veterans Health Administration

Learning Interpretable, Temporal Health Status Phenotypes from Self-Tracked Patient Data

Endometriosis is a debilitating, systemic chronic illness where unpredictable week-to-week variations care. We hypothesize that unsupervised probabilistic phenotype approaches can enable meaningful, interpretable representations of health status over time in the context of self-tracked data, independently of an individual’s level of engagement with self-tracking. We generate and evaluate temporal phenotypes from self-tracking data to represent individuals’ illness states over time, which have the potential to support new tools for tracking and management.

Learning Outcomes

• Explore the use of machine learning alongside patient self-tracked data to facilitate analysis that can support personal informatics tools to support patients with chronic illness. The focus of this podium abstract is constructing a learned phenotype model that can be used to characterize the health status of a poorly understood chronic illness.

Speaker

• Adrienne Pichon, Columbia University, Department of Biomedical Informatics

Examining Oral Anti-Cancer Medication Continuation Using Questionnaires, Prescription Refills, and Structured Electronic Health Records

Medication persistence is essential for the efficacy of treatment and patient health outcomes. This study investigates the discontinuation of oral anticancer medications (capecitabine, ibrutinib, or sunitinib) in a cohort that is well-characterized by medication discontinuation questionnaires, prescription refill data, and structured electronic health records (EHRs). We categorized discontinuation reasons based on the questionnaire of patients taking medication, revealing that 38% of 257 patients completed therapy, while discontinuation was due primarily to no response to therapy and/or progression of disease leading to discontinuation (33%) and side effects/complication (9%). Survival analysis showed variable medication persistence, with capecitabine persistence decreasing significantly over time, to 0.08 in two years. A logistic regression model demonstrated strong capability (with an AUC of 0.835) to identify patients at risk for medication discontinuation. The study shows the complexities of medication persistence and emphasizes the importance of understanding medication discontinuation patterns and leveraging predictive analytics to inform future research and clinical monitoring in the treatment of cancer.

Learning Outcomes

• Understand the primary reasons for discontinuation of oral anticancer medications and learn how predictive models can help identify patients at risk for early medication cessation.

Speaker

• Congning Ni, Ph.D. student, Vanderbilt University

Revealing Patterns of Child Maltreatment Policy Differences and Demographic Dynamics using BERT-Networks and Clustering Approach

Examining child abuse and neglect policies is crucial for shaping child health outcomes. 411 policy items were organized using Siamese BERT-Networks. 52 U.S. territories were categorized into 4 clusters primarily by mandated reporting and differential response policies. Race, gender, and economic status show significant differences among the 4 clusters. Sub-analysis on fatality-related policies revealed significant impact of fatality definitions on outcomes. These findings underscore the necessity of precise policy formulation for improving child outcomes.

Learning Outcomes

• Describe the role of various data analysis methods in transforming policy data into actionable insights regarding child maltreatment outcomes.

Speakers

• Zhidi Luo, MS, Northwestern University

Acceptance and Perceptions of Electronic Health Record-based Clinical Decision Support for Obesity in Pediatric Primary Care

We surveyed 245 clinicians at 84 primary care practices within three US health systems in a cluster-randomized trial of a clinical decision support (CDS) intervention. Clinicians in intervention vs. control sites had higher odds of perceived ease of providing patient materials and subjective norms regarding CDS use and lower odds of intention to use future CDS tools. Our findings highlight opportunities and challenges of CDS to address clinicians’ preferences within healthcare and EHR system constraints.

Learning Outcomes

• Gain exposure to application of the technology Acceptance Model in evaluating the implementation of clinical decision support tools.

Speakers

• Mona Sharifi, MD. MPH, Yale School of Medicine

Development and multi-center validation of a pre-trained language model for predicting neonatal morbidities

We present work in developing, training, and validating NeonatalBERT, a pre-trained language model to automatically predict neonatal diseases at birth from unstructured clinical notes based on a large dataset with over 30,000 newborns. We perform both internal and external validation on a comprehensive list of neonatal morbidities and demonstrate strong performance across hospitals and patient populations. NeonatalBERT has a great degree of flexibility and paves the way for various future applications in neonatal care.

Learning Outcomes

• Gain exposure to application of the technology Acceptance Model in evaluating the implementation of clinical decision support tools.

Speakers

• Feng Xie

Using the Technology Acceptance Model to guide refinements to the Color Me Healthy symptom assessment app for children

We describe revisions to the Color Me Healthy app for children and evaluation of its usability. Fourteen children with cancer and their parents participated in cognitive walkthrough evaluations. Observations of children and parents’ ability to complete key tasks and analysis of qualitative data supported the app’s perceived ease of use and perceived usefulness. Future directions include incorporating Color Me Healthy in clinical care to support monitoring trends in children’s symptoms and facilitating timely interventions.

Learning Outcomes

• Describe application of the constructs within the technology Acceptance Model in evaluating the usability of the revised Color Me Healthy symptom assessment app by children with cancer and their parents. 
• Describe the use of cognitive walkthrough interviews as a strategy to evaluate the usability of digital health resources with targe end users.

Speakers

• Lauri Linder, PhD, APRN, CPON, FAAN, FAPHON, University of Utah, Primary Children's Hospital

Acceptability of pictographs as a novel patient identifier to improve patient safety in the neonatal intensive care unit

As part of a randomized controlled trial on the use of pictographs (images used in lieu of a patient photo) embedded in the electronic health record to reduce wrong-patient errors in the neonatal intensive care unit (NICU), we conducted a series of surveys of parents, providers and nurses in the NICU. Data from survey responses were thematically analyzed and categorized. We found that in all groups, there was very high awareness of the intended purpose of the pictographs; however, the perception of providers and nurses about the effectiveness of pictographs was not as strong. While several providers and nurses acknowledged that pictographs can or have helped them avoid wrong-patient errors when caring for multiple birth infants (such as twins), many nurses believed that their current practice of the use of two patient identifiers was sufficient, and pictographs were not useful. Parents reported that pictographs improved their experience of care.

Learning Outcomes

• Describe the challenges with patient identification in the neonatal intensive care setting.

Speakers

• Hojjat Salmasian, MD, MPH, PhD, FAMIA, Children's Hospital of Philadelphia

Probabilistic Graphical Models for Evaluating the Utility of Data-Driven ICD Code Categories in Pediatric Sepsis

Electronic health records (EHRs) are digitalized medical charts and the standard method of clinical data collection. They have emerged as valuable sources of data for outcomes research, offering vast repositories of patient information for analysis. Definitions for pediatric sepsis diagnosis are ambiguous, resulting in delayed diagnosis and treatment, highlighting the need for precise and efficient patient categorizing techniques. Nevertheless, the use of EHRs in research poses challenges. EHRs, although originally created to document patient encounters, are now primarily used to satisfy billing requirements. As a result, EHR data may lack granularity, potentially leading to misclassification and incomplete representation of patient conditions. We compared data-driven ICD code categories to chart review using probabilistic graphical models (PGMs) due to their ability to handle uncertainty and incorporate prior knowledge. Overall, this paper demonstrates the potential of using PGMs to address these challenges and improve the analysis of ICD codes for sepsis outcomes research.

Learning Outcomes

• Identify our approach to classifying and validating ICD codes for exploring the pediatric sepsis population.

Speaker

• Lourdes Valdez, Biomedical Informatics Department, University of Utah"

Following the onset of the COVID-19 pandemic, delivery of care via synchronous video or audio telemedicine exploded in popularity, and the proportion of ambulatory care delivered virtually has remained elevated relative to pre-pandemic levels through 2023. While many studies have assessed the quality of care delivered via telemedicine, comparatively few have evaluated the impact of virtual care delivery on clinician work and electronic health record (EHR) use. This panel will present the current state of the evidence on the relationship between virtual care delivery modalities and clinician work and EHR use with insights from leading researchers across the United States.

Panelists discuss their perspectives on the specific drivers of differences during telemedicine, assess potential interventions to improve the clinician experience of providing virtual care, and outline future research agenda. The panel will inform health system leaders, policymakers, and informaticians seeking to develop a sustainable, long-term telemedicine model that enables patient access to virtual care with supporting the clinicians delivering that care.

Learning Outcomes

• Understand how clinician EHR use changes with telemedicine
• Assess how different organizational approaches to telehealth impact how clinicians use EHRs for virtual care. 
• Develop evidence-based strategies for technology and organizations to better support virtual care delivery.

Speakers

• A J Holmgren, PhD, University of California, San Francisco
• Mary Reed, DrPH, Kaiser Permanente Division of Research
• Soumik Mandal, PhD, NYU Grossman School of Medicine
• Nate Apathy, PhD, University of Maryland
• Thomas Kannampallil, PhD, Washington University School of Medicine

Empowering Patient-Centric Data Management in Healthcare Using Blockchain-based Self-Sovereign Identity and Non-Fungible Tokens

Patient tokenization, leveraging non-fungible tokens and self-sovereign identity on blockchain technology, represents a transformative approach for secure, anonymous patient data linkage across diverse healthcare domains, including medical, dental, and beyond. This study demonstrates the feasibility of this innovative system through a case study involving over three million transactions, showcasing its potential to fundamentally reshape identity management and health information exchange in a patient-centric manner. This work showcases its transformative potential across various healthcare domains.

Learning Outcomes

• Explain how blockchain-based self-sovereign identity (SSI) and non-fungible tokens (NFTs) enhance patient control and data privacy in health information exchange.
• Describe the role of NFTs in securely linking patient records across multiple healthcare facilities while maintaining anonymity and compliance with privacy regulations.
• Analyze the potential challenges and opportunities of integrating blockchain-based patient tokenization with existing healthcare frameworks and interoperability standards.

Speakers

• Yan Zhuang, PhD, Indiana University

Enhancing Wearable Sensor Data Classification Through Novel Modified-Recurrent Plot-Based Image Representation and Mixup Augmentation

Deep learning advancements have revolutionized scalable classification in many domains including computer vision, healthcare and Natural Language Processing (NLP). However, when it comes to classification and domain adaptation based on wearables, it suffers from persistent underperformance, largely due to the scarcity of pre-trained deep learning models that are abundantly available for computer vision and NLP. This is primarily because wearable sensor data need sensor-specific preprocessing, architectural modification, and extensive data collection. We present a novel modified-recurrent plot-based image representation that seamlessly integrates both temporal and frequency domain information. We employ an efficient Fourier Transform-based frequency domain angular difference estimation scheme in conjunction with existing temporal recurrent plots. We validated proposed method in two different domains: accelerometer-based activity-recognition and real-time glucose level prediction from wearable sensors. Our findings demonstrated the method we developed not only improves accuracy at recognizing activity but also makes a big leap in glucose level prediction.

Learning Outcomes

• Understand how modified recurrence plots can improve wearable sensor data classification.
• Identify the advantages of combining temporal and frequency domain information in wearable data analysis.
• Recognize the potential of mix-up augmentation in enhancing classification accuracy.

Speakers

• Mohammad Arif Ul Alam, PhD, University of Massachusetts Lowell

Voice-Activated Self-Monitoring Application (VoiS): Perspectives from People with Diabetes and Hypertension

This paper describes the development and usability test processes of the voice-activated self-monitoring (VoiS) application which is purposed to support the self-management of people with both diabetes and hypertension. VoiS is an innovative, theory-driven mobile app on a smart speaker platform to support people with coexisting diabetes and hypertension to self-monitor blood pressures, glucose levels, and health behaviors routinely and conveniently, and to improve the quality of communication with healthcare providers. The prototype of VoiS includes voice interaction with Amazon Alexa and data representation using smartphones (iOS and Android). A total of 14 people with coexisting diabetes and hypertension participated in usability testing. After completing a range of tasks individually, testers participated in group interviews. We used a survey based on the Technology Acceptance Model to measure the ease of use and perceived usefulness of VoiS. All interviews were recorded and transcribed, and then common themes were extracted. Participants found VoiS to be easy to use and useful.

Learning Outcomes

• Describe what VoiS is.
• Recognize the purpose of VoiS.
• Demonstrate the process of usability test.

Speakers

• Li Yang, MSM, University of Wisconsin-Milwaukee

Integrating Remote Patient Monitoring Data into Machine Learning Models for Predicting Emergency Department Utilization

The integration of Remote Patient Monitoring (RPM) data into risk stratification models has emerged as a promising approach for improving healthcare delivery and patient outcomes. In this work, we explore the integration of RPM features – including at home monitoring of body weight, blood pressure, and blood oxygen – into a machine learning model that uses EHR data to predict the likelihood of emergency department (ED) visits or unplanned inpatient admissions within the next 30 days. Through exploratory data analysis, feature engineering, model training, and evaluation of a dataset with 913 patients, we found that RPM data has signal to predict unplanned utilization, and combining RPM data with EHR data improves the predictive power of the model, compared with either data source alone. We discuss the transformative potential of RPM data to augment predictive analytics capabilities in care management settings.

Learning Outcome

• Be able to integrate remote patient monitoring data (RPM) with EHR to predict emergency department visits and unplanned utilization.

Speakers

• Ashika Farzana, MS, Geisinger

Detection of Short-Form Video Addiction with Wearable Sensors via Temporally-Coherent Domain Adaptation

Short-form Video Addiction (SVA), a novel digital addiction of the modern world, proliferates among young adults and is not formally diagnosable. SVA detection from resulting bio-signals is crucial to prevent its adverse impacts. Existing formal methods involve large and expensive neuro-imaging devices in laboratory setups that are intrusive and not feasible to use in daily life. A possible non-intrusive solution can be using wearable sensors which is challenging due to the resulting noisy and faint signals. To address this problem, we investigate multi-modal wearable sensing technology to detect SVA in a non-intrusive fashion. However, fusing multi-modal sensors effectively presents different challenges due to the presence of signal heterogeneity. In this study, we propose a novel multi-modal temporally coherent domain adaptation method to effectively detect SVA using Electroencephalogram (EEG) and Electrodermal Activity (EDA) sensors. We also investigate the nature and properties of SVA with the help of different components of EEG and EDA signals. We evaluate our proposed method for SVA detection and fatigue assessment tasks. Experimental evaluation posits the proposed model's superior performance (10% accuracy) over state-of-art domain adaptation models.

Learning Outcomes

• Discuss the use of wearable sensors in detecting cognitive fatigues resulting from short-form video addiction.
• Explain the advantage of using multi-modal sensors with time-domain alignments

Speakers

Mahmudur Rahman, PhD, University of Wisconsin-Madison

“I worry we’ll blow right by it” Barriers to Uptake of the STRATIFY CDSS for ED Discharge in Acute Heart Failure

We recently implemented a clinical decision support system (CDSS) to identify patients in the emergency department (ED) with acute heart failure that may be safe for discharge instead of the typical costly hospitalization. Despite user-centered-design initial tool uptake was low. To explore barriers to use we interviewed 10 ED clinicians with a case-simulation. Usability issues around tool launch, instead of the tool itself, along with low familiarity of evidence supporting the CDSS drove low uptake.

Speakers

• Matthew Christensen, MD, Vanderbilt University Medical Center

All content was recorded live at AMIA’s Annual Symposium event November 9-13, 2024, in San Francisco, CA. Plan now to join us for the next Annual Symposium!

CME or CNE credit must be claimed no later than January 20, 2028 or within two years of your purchase date, whichever is sooner. No credit will be issued after January 20, 2028.

Yes! AMIA 2024 Annual Symposium On Demand is available for anyone to purchase. Become an AMIA member before you purchase to receive exclusive member discounts. Join AMIA today.

We’re glad you asked! AMIA offers a variety of membership options, all with exclusive benefits and abundant networking opportunities. Choose the membership that’s right for you.

Join us at the next Annual Symposium and engage with leaders from across the health informatics field. Learn more.

Yes! You can claim Self-Study credit when you complete AMIA 2024 Annual Symposium On Demand sessions, in addition to claiming Live credit for attending the live event. View the full details on self-study accreditation for this product.