Melissa McPheeters

Research Professor

Vanderbilt University Medical Center
Country flagUnited States

Research Interests

ApplyKite TwinLearns as you apply

Make your next move.

Prepare your application

Review drafts before you send them.

Explore related searches

Articles (10)

The National Institutes of Health’s All of Us Research Program (All of Us) aims to enhance precision medicine by collecting multimodal data from 1 million or more participants. Because All of Us prioritizes enrollment from populations for which there is limited data on health outcomes using nonprobability sampling methods, prevalence estimates may not reflect those of the general US population. This study examines the challenges of estimating electronic health record–based disease prevalence from All of Us and offers a framework and novel R package (waou) to help researchers consider these complex issues. We investigated the application of 3 weighting techniques to improve generalizability for dementia, type 2 diabetes, and depression prevalence estimates. Using data from All of Us alongside the National Health Interview Survey as a benchmark, we found that weighting approaches yielded more representative estimates for dementia and type 2 diabetes, yet amplified bias for depression. The waou is presented as a tool to facilitate the application of these methodologies, empowering researchers to critically evaluate the generalizability of their estimates. This work underscores the need for careful consideration of bias in epidemiological research when using the All of Us dataset for population-level inferences.

Year:

2026

This study evaluated death ascertainment from publicly available internet sources for patients in two large tertiary care US healthcare systems, Mass General Brigham (MGB) and Vanderbilt University Medical Center (VUMC), benchmarked against state and federal vital statistics data. Names, dates of birth, and dates of death were extracted from 8.1 million internet media records using previously developed natural language processing models. Internet records were matched to 78 848 deceased patients from MGB and VUMC on first name, last name, and date of birth. Dates of death were validated against state vital statistics databases or the National Death Index as reference standards. We calculated sensitivity and positive predicted values (PPV) of internet sources in identifying dates of death within 7 days of the reference standard. Exact matching of records between internet media and reference standards on first name, last name, and date of birth, resulted in 30 067 (38.8%) matches, which showed PPV for death identification (98.2%—MGB; 98.9%—VUMC) in internet media and increased sensitivity of death capture over EHR alone by 24% at MGB and 18% at VUMC. In conclusion, using internet sources to augment mortality data increased capture of death meaningfully over reliance on EHR records alone.

Year:

2026

Objectives To develop and validate machine learning (ML) models that predict probable cause of death (CoD) using structured electronic health record (EHR) data, unstructured clinical notes, and publicly available sources. Materials and Methods This multi-institutional retrospective study was conducted across Vanderbilt University Medical Center (VUMC) and Massachusetts General Brigham (MGB), including deceased patients with encounters between October 1, 2015, and January 1, 2021, and confirmed death records. The cohort included 13 708 patients from VUMC and 34 839 from MGB.The primary outcome was underlying CoD categorized into the top 15 National Center for Health Statistics rankable causes, with others grouped as “Other.” Performance was assessed using weighted area under the receiver operating characteristic curve (AUC) and F-measure. Results The XGBoost model using structured EHR data alone achieved weighted AUCs of 0.86 (95% CI, 0.84-0.88) at VUMC and 0.80 (95% CI, 0.79-0.80) at MGB. Adding unstructured notes improved performance, with weighted AUCs of 0.90 (95% CI, 0.88-0.93) at VUMC and 0.92 (95% CI, 0.91-0.92) at MGB. Adding publicly available data did not further improve performance. Cross-institutional validation revealed significant performance degradation. Discussion Models integrating structured and unstructured EHR data show strong within-institution performance but limited generalizability across healthcare systems, highlighting challenges related to institutional data heterogeneity. Conclusions Machine learning models combining structured and unstructured EHR data accurately predict CoD within institutions but perform poorly across sites. Health-care institutions may benefit from adopting robust processes for locally tailored models, and future research should focus on enhancing model generalizability while addressing unique institutional data environments.

Year:

2026

Year:

2026

Importance Alcohol use is a leading cause of morbidity and mortality worldwide. Growing evidence suggests that glucagon-like peptide-1 receptor agonists (GLP-1RAs) may represent a novel potential pharmacotherapeutic tool for alcohol use disorder (AUD). Objective To examine the association between GLP-1RA prescriptions and alcohol use. Design This cohort study used a cross-sectional measure of alcohol consumption and longitudinal electronic health record (EHR) data collected between 1981 and October 2023 from NIH’s All of Us Research Program participants. Setting All of Us is a large program to recruit and collect surveys, EHR, genomic, and wearable data from a wide array of Americans. The data presented here are from the All of Us Curated Data Repository version 8. Participants 393,596 All of Us participants with EHR data recruited across the United States. Exposure At least two GLP-1RA prescription records in the EHR. Main Outcomes and Measures Alcohol Use Disorders Identification Test (AUDIT-C) scores and responses to individual AUDIT-C questions. Results Among 15,447 participants with at least two recorded GLP-1RA prescriptions on separate days, 3650 had active GLP-1RA prescriptions, 5642 would have future GLP-1RA prescriptions (primary comparison group), and 544 had former GLP-1RA prescriptions. Those with active GLP-1RA prescriptions had statistically significant but modestly lower AUDIT-C scores on average compared with those with future prescriptions (incidence rate ratio [IRR] of 0.95; 95% CI, 0.91-0.99; P = 0.01). Participants with a former GLP-1RA prescription had lower AUDIT-C scores compared with those with future prescriptions, but this difference was not statistically significant. Results were similar using a propensity-score matched comparison group with a lower average AUDIT-C score for the current GLP-1RA group (IRR = 0.89; 95% CI, 0.85-0.93; P = <0.001) and no significant difference for the former prescription group. Analysis of individual AUDIT-C questions shows a significant association with GLP-1RA prescriptions and frequency of drinking but not drinks per occasion or binge drinking. Conclusions and Relevance This study’s findings indicate that GLP-1RAs may reduce alcohol consumption by decreasing use frequency. Experimental studies and randomized controlled trials are needed to test the mechanisms and potential efficacy of GLP-1RAs in people with AUD. KEY POINTS Question Is there an association between glucagon-like peptide-1 receptor agonist (GLP-1RA) prescriptions and alcohol consumption? Findings In this observational cohort study of 15 447 people with GLP-1RA prescriptions in NIH’s All of Us cohort, active GLP-1RA prescriptions were associated with significantly lower Alcohol Use Disorders Identification Test (AUDIT-C) scores compared with scores of those who would have GLP-1RAs in the future and of a matched comparison group with no GLP-1RAs. People with former GLP-1RAs did not have lower AUDIT-C scores than these comparison groups. Meaning Active GLP-1RA use may be effective in reducing alcohol consumption.

Year:

2026

Importance Glucagon-like peptide-1 receptor agonists (GLP-1RAs) are fast-growing treatments for type 2 diabetes, obesity, and sleep apnea and are under investigation as potential treatments for many other conditions. The National Institutes of Health’s (NIH’s) All of Us Research Program offers a robust observational data source for studying questions related to GLP-1RA use in real-world settings. Objective This article describes key characteristics of All of Us participants who have been prescribed GLP-1RAs. The goals are to present the utility of the All of Us data and describe the strengths and limitations of using this resource for future research on GLP-1RAs. Design Using the All of Us Controlled Tier Curated Data Repository version 8 (CDRv8), we provide a descriptive analysis of the cohort with GLP-1RA records using cross-sectional surveys, longitudinal electronic health record (EHR) data, and longitudinal Fitbit data. Setting The All of Us Research Program is a large, federally funded, longitudinal cohort study established in 2018 by NIH. Recruitment efforts are nationwide and target a range of populations to advance precision medicine for all. Participants Participants are U.S. residents, aged 18 or older at the time of study consent, who were enrolled between May 6, 2018, and October 1, 2023. Exposures The GLP-1RA cohort included participants with at least two GLP-1RA prescription records on different days at any time point based on their EHRs. Main Outcomes Frequencies and medians for a range of sociodemographic characteristics, health care utilization patterns, comorbid conditions, GLP-1RA prescription trends, laboratory and observation availability, and Fitbit data. Results The All of Us GLP-1RA cohort is large (n=15 477), with high data availability across a range of relevant data types. These participants are older and have more comorbid conditions than the entire CDRv8 population. Prescription trends indicate rapid uptake of GLP-1RA drugs since 2014. Conclusions and Relevance All of Us CDRv8 is a valuable resource for research on GLP-1RAs across a large, heterogeneous cohort of participants. The variety and availability of data offer many possibilities for future observational, real-world research to address unanswered questions about GLP-1RA use and replicate recent findings generated from other datasets. Key points Question What data are available and what are the patterns of GLP-1 receptor agonist (GLP-1RA) prescriptions among participants in the All of Us Research Program? Findings In this descriptive cohort study of 633 534 All of Us participants, 15 477 participants had at least two records of GLP-1RA prescriptions. These participants tended to be older, have more comorbid conditions, and have higher health care utilization than the All of Us population as a whole. Meaning The All of Us Research Program has a robust array of data to support observational studies of people receiving GLP-1RA prescriptions.

Year:

2026

Background Mortality is a critical variable in health care research, especially for evaluating medical product safety and effectiveness. However, inconsistencies in the availability and timeliness of death date and cause of death (CoD) information present significant challenges. Conventional sources such as the National Death Index and electronic health records often experience data lags, missing fields, or incomplete coverage, limiting their utility in time-sensitive or large-scale studies. With the growing use of social media, crowdfunding platforms, and web-based memorials, publicly available digital content has emerged as a potential supplementary source for mortality surveillance. Despite this potential, accurate tools for extracting mortality information from such unstructured data sources remain underdeveloped. Objective The aim of the study is to develop scalable approaches using natural language processing (NLP) and large language models (LLMs) for the extraction of mortality information from publicly available web-based data sources, including social media platforms, crowdfunding websites, and web-based obituaries, and to evaluate their performance across various sources. Methods Data were collected from public posts on X (formerly known as Twitter), GoFundMe campaigns, memorial websites (EverLoved and TributeArchive), and web-based obituaries from 2015 to 2022, focusing on US-based content relevant to mortality. We developed an NLP pipeline using transformer-based models to extract key mortality information such as decedent names, dates of birth, and dates of death. We then used a few-shot learning (FSL) approach with LLMs to identify primary and secondary CoDs. Model performance was assessed using precision, recall, F 1 -score, and accuracy metrics, with human-annotated labels serving as the reference standard for the transformer-based model and a human adjudicator blinded to the labeling source for the FSL model reference standard. Results The best-performing model obtained a microaveraged F 1 -score of 0.88 (95% CI 0.86‐0.90) in extracting mortality information. The FSL-LLM approach demonstrated high accuracy in identifying primary CoD across various web-based sources. For GoFundMe, the FSL-LLM achieved 95.9% accuracy for primary cause identification compared to 97.9% for human annotators. In obituaries, FSL-LLM accuracy was 96.5% for primary causes, while human accuracy was 99%. For memorial websites, FSL-LLM achieved 98% accuracy for primary causes, with human accuracy at 99.5%. Conclusions This study demonstrates the feasibility of using advanced NLP and LLM techniques to extract mortality data from publicly available web-based sources. These methods can significantly enhance the timeliness, completeness, and granularity of mortality surveillance, offering a valuable complement to traditional data systems. By enabling earlier detection of mortality signals and improving CoD classification across large populations, this approach may support more responsive public health monitoring and medical product safety assessments. Further work is needed to validate these findings in real-world health care settings and facilitate the integration of digital data sources into national public health surveillance systems.

Year:

2025

Introduction Despite the importance of clinical trial participation among cancer patients, few participate—and even fewer patients from ethnic and racial minoritized groups. It is unclear whether suggested approaches to increase accrual are successful. We conducted a scoping review to identify evidence‐based approaches to increase participation in cancer treatment clinical trials that demonstrated clear increases in accrual. Notably, more stringent than other published reviews, only those studies with comparison data to measure a difference in accrual rates were included. Methods We searched PubMed/MEDLINE, Embase, CINAHL, and Web of Science for English‐language articles published from January 1, 2012, to August 8, 2022. Studies were included if they were conducted in the United States, described single or multicomponent interventions, and provided data to measure accrual relative to baseline levels or that compared accrual rates with other interventions. Results Sixteen articles were included: six with interventions addressing patient barriers, two addressing provider barriers, seven describing institutional change, and one describing policy change. Key themes emerged, such as a focus on patient education, cultural competency, and building the capacity of clinics. Few studies provide comparative accrual data, making it difficult to identify with certainty any effective, evidence‐based approaches for increasing accrual. Some patient‐ and system‐level interventions studies showed modest increases in accrual primarily through pre‐post measurement. Conclusion Despite an extensive body of literature about the barriers that impede cancer treatment trial accrual, along with numerous recommendations for how to overcome these barriers, results reveal surprisingly little evidence published in the last 10 years on interventions that increase accrual relative to baseline levels or compared with other interventions. As clinical trials are a primary vehicle through which we improve cancer care, it is critical that evidence‐based approaches are used to inform all efforts to increase accrual. Strategies for increasing participation in cancer clinical trials must be developed and rigorously evaluated so that these strategies can be disseminated, participation in trials can increase and become more equitable, and trial results can become more generalizable.

Year:

2024

Year:

2024

Year:

2023

Collaborators (6)

Mehdi Farokhnia

Johns Hopkins University

UNITED STATES
View details

Lorenzo Leggio

National Institutes of Health

UNITED STATES
View details

Shirley Wang

Brigham and Women's Hospital and Harvard Medical School

UNITED STATES
View details

Mohammed Al-Garadi

Research Assistant Professor

Vanderbilt University Medical Center

UNITED STATES
View details

Michael Matheny

Associate Professor

Vanderbilt University Medical Center

UNITED STATES
View details

Philip Greenland

Professor

Northwestern University

UNITED STATES
View details