11/12/2025: 3:00 PM - 4:30 PM EST
Speed
Presentations
A Sequential, Multiple Assignment, Randomized Trial (SMART) is a type of clinical trial design that allows for the development, estimation and comparison of dynamic treatment regimens or tailored sequences of treatments. Typical methods for analyzing outcomes from a SMART utilize weighted and replicated regression. When dealing with count outcomes, especially in settings of substance use, zero-inflation can be a common issue.
One common approach to handling zero-inflated outcomes for standard, longitudinal data is a two-part hurdle model (HM) which addresses sampling or random zero outcomes. Sampling zeros occur when all individuals have zeroes in their outcome but are still at risk of an event. HMs have yet to be applied to SMART data.
We develop a two-part HM to address zero-inflated outcomes in the SMART setting with longitudinal, count outcomes. In this approach, one part of the model is a logistic regression with a binary zero / nonzero indicator outcome, and the second part is a truncated Poisson model for the nonzero count outcomes. We were motivated by and apply our method to a recently completed SMART, the SafERteens M-Coach trial which investigated the effects of brief interventions and text messaging on young adults considering the outcome of alcohol consumption.
Presenting Author
Hanna Venera, University of Michigan
First Author
Hanna Venera, University of Michigan
CoAuthor(s)
Kelley Kidwell, University of Michigan
Bingkai Wang, Department of Biostatistics, School of Public Health, University of Michigan
Multiple Sclerosis (MS) is a chronic inflammatory and demyelinating disease that affects the central nervous system (CNS) and is known to be more prevalent in women than in men. MS activity and progression are often monitored through clinical evaluations, such as the Expanded Disability Status Scale (EDSS) and the Multiple Sclerosis Functional Composite (MSFC), which includes Timed 25-Foot Walk, 9-Hole Peg Test, and PASAT scores, and magnetic resonance imaging (MRI), which can identify new and/or contrast-enhancing lesions.
The timing of diagnosis and treatment may influence the course of the disease and the severity of both MRI findings and clinical outcomes. Recent studies have shown that disease-modifying therapies are more effective when initiated early in the disease course. In patients who have only had one attack of neurological symptoms consistent with MS, the initiation of DMT right after the attack has been found to delay its conversion to Clinically Definite Multiple Sclerosis (CDMS). A disconnect between the clinical symptoms and MRI findings can occur, highlighting the limitations of relying solely on MRIs to monitor the disease.
Motivated by the heterogeneity and multifactoriality of MS, this study uses hierarchical mixed-effects models applied to a large clinical trial cohort from CombiRx to assess the relationship between the timing of MS diagnosis and disease trajectory based on multiple functional outcomes used to assess the MSFC scores. Additionally, we evaluate the association between the timing of MS diagnosis and the evolution of MRI-based markers of disease activity, based on joint modeling of the components of the MSFC Z4 composite score. Finally, we explore the impact of diagnosis timing on overall disease burden, including relapse rate and persistent functional decline related to disease progression.
Presenting Author
Christilene Tumsiah, The Ohio State University
First Author
Christilene Tumsiah, The Ohio State University
CoAuthor(s)
Fernanda Schumacher, Ohio State University
Yinan Zhang, The Ohio State University Wexner Medical Center
Exploratory biomarkers are crucial in clinical trials as they help identify potential therapeutic targets, understand disease mechanisms, and predict patient responses. However, their collection and analysis can be expensive and can impose a significant burden on patients, requiring sometimes frequent and invasive sampling procedures. Here, we introduce a biomarker optimization workflow that is aimed at enhancing the efficiency and effectiveness of exploratory biomarker planning by incorporating statistical power analyses and cost optimization models. The workflow leverages both analytical and simulation-based power calculations to identify opportunities for sampling a reduced number of subjects or reduced timepoints for each exploratory biomarker, while maintaining statistical power. An additional innovation is a user-friendly RShiny app for implementing the optimization workflow, facilitating real-time collaboration between clinical and statistical teams. Unlike general power analysis packages that offer a wide range of methods, this app focuses solely on the statistical tests commonly performed on exploratory biomarkers, including but not limited to association, pharmacodynamic, and survival analyses. This ensures a smooth user experience and maintains consistent assumptions across trials. In addition to informing biomarker collection based on statistical power, our methods can incorporate surrogate based optimization to account for complex cost functions, to account for dynamic pricing, site specific overhead and other pricing factors. In simulated late-stage oncology clinical trial plans, we demonstrate that this workflow can reduce costs by a large margin (mean = 15%, SD= 7%), while maintaining statistical merit (power= 0.8) across diverse assay types. Overall, we demonstrate a collaborative and data-driven approach to biomarker planning that enhances the efficiency of clinical trials, while preserving statistical integrity of exploratory biomarkers.
Presenting Author
Aditi Basu Bal, Bristol Myers Squibb
First Author
Coryandar Gilvary
CoAuthor(s)
Sanhita Sengupta, Bristol Myers Squibb
John Schwarz, Bristol Myers Squibb
Aditi Basu Bal, Bristol Myers Squibb
Mariann Micsinai-Balan, Bristol Myers Squibb
In interventional clinical trials, rescue medication refers to an alternative treatment administered, in addition to or as a replacement for the study intervention, to trial participants for reasons such as alleviating the worsening of symptoms. While it may benefit the patient, rescue medication may distort the true effect of the intervention being studied. It may then lead to study results that may not align with the original research question. The International Council of Harmonisation (ICH) comprising of representatives from regulatory agencies and pharmaceutical companies provides guidelines on how to handle intercurrent events of which rescue medication is a special case. ICH recommends five strategies. This talk will focus on one of these strategies, namely the hypothetical strategy, and discuss some methods to handle use of rescue. Of particular interest are methods based on the balanced estimand, inverse probability weights, and Loh g-estimation. We illustrate the concepts and methods applied to data from a completed trial on myasthenia gravis, a rare neurological disease. We also conducted simulation studies based on different scenarios to evaluate the performance of these methods and determine which is best in estimating the treatment effect in the presence of rescue medication regarding bias, variance, and mean squared error in the case of small to moderate sample sizes.
Presenting Author
Ernestina Boateng
First Author
Ernestina Boateng
CoAuthor
Inmaculada Aban, University of Alabama at Birmingham
School closures in 2020-2021 negatively impacted learning trajectories for young children. To facilitate academic recovery, one large, urban school district allocated additional resources in the post-closure years for direct school-based mental health services. We describe an evaluation of the impact of receiving such services on a student's early literacy trajectory. This presentation will also emphasize the common challenges faced in causal inference and the process for selecting a rigorous statistical analysis plan. Analytical challenges arise since the probability of receiving the intervention is substantially higher among students with lower literacy and slower trajectories with limited information for balancing. The selected method leverages a student's own pre-intervention trajectory to address the counterfactual question of whether the initiation and accumulation of services translated to better gains in literacy. Administrative data included 691,742 early literacy assessments completed by students in grades K-2 in three consecutive years. A linear mixed effects model was fit including student grade, an indicator of whether an initial service already occurred, cumulative services to date, and assessment time. Detailed trajectories were estimated using interactions and random effects accounted for nesting and repeated measures. Growth from beginning to middle and end of year was slower among students receiving services. Higher total number of prior services also translated to lower beginning of year scores. However, as services accumulated within a given year, the number of services received was positively correlated with the rate of increase in literacy. This study builds a flexible model to disentangle the impact of mental health service accumulation from the increased propensity for receiving services among students with lower reading achievement. The results display a promising pattern of recovery through the use of school-based mental health services.
Presenting Author
Naomi Wilcox, UCLA Jane and Terry Semel Institute for Neuroscience and Human Behavior
First Author
Naomi Wilcox, UCLA Jane and Terry Semel Institute for Neuroscience and Human Behavior
CoAuthor(s)
Hilary J. Aralis, Public Health Sciences, School of Medicine, University of California, Davis
Patricia Tan, UCLA Jane and Terry Semel Institute for Neuroscience & Human Behavior, VA Greater LA Health System
Alyssa R. Palmer, UCLA Jane and Terry Semel Institute for Neuroscience & Human Behavior
Sheryl H. Kataoka Endo, UCLA Jane and Terry Semel Institute for Neuroscience & Human Behavior
Alison Wood, UCLA Jane and Terry Semel Institute for Neuroscience & Human Behavior, VA Greater LA Health System
Roya Ijadi-Maghsoodi, UCLA Jane and Terry Semel Institute for Neuroscience & Human Behavior, VA Greater LA Health System
Mobile health interventions have become increasingly prominent in recent years as tools for supporting health behavior change. Recent studies have focused on improving medication adherence and self-care among individuals managing chronic conditions such as diabetes and hypertension. A key component of these mobile health interventions is designing them in a way that encourages engagement. However, because differential engagement with the intervention can translate to differential benefit, the sustained impact of the intervention faces challenges due to declining engagement patterns.
Understanding and quantifying the role of engagement is therefore critical for evaluating the effectiveness of mobile health interventions on outcomes of interest. Prior methodological development has leveraged causal inference tools, notably modifications to instrumental variable approaches via sensitivity analysis, to characterize how engagement drives improvements in outcomes when the exclusion restriction cannot reasonably be assumed to hold. These contributions, however, have been limited to settings with two treatment arms and a continuous measure of engagement. There is an opportunity to extend these methods more broadly to enhance their applicability to additional real-world interventions.
This work advances the methodological foundation for characterizing engagement by expanding upon prior research to consider multiple treatment groups and alternative distributions for the engagement variable. In this presentation, we focus on the proper interpretation and methodology of these approaches and demonstrate their utility through applications to recent studies of mobile health interventions.
Presenting Author
Alexis Fleming
First Author
Alexis Fleming
CoAuthor
Andrew Spieker, Vanderbilt University Medical Center
The concentration curve and closely related concentration index are methods that can be used to visualize and quantify the distribution of a health outcome versus another variable, such as measures of socioeconomic status or access to health-related resources. However, when fitting these models to real-world data, the data are often error-prone. One area where this issue arises is when investigating food access and individual dietary inflammation scores (DIS), as access to healthy foods is a neighborhood-level determinant of diet quality that is subject to multiple sources of error. To approach this issue, we conduct simulation studies to assess the impact of various sources of error in measuring access to healthy foods, such as errors that are dependent on other neighborhood-level factors like rural vs. non-rural geographies or the DIS outcome. We then compare these simulated results to the theoretical values of the concentration index under each of these conditions. Building on current measurement error literature, we seek to demonstrate how error-prone measures may bias a measure like the concentration index to direct future work in addressing this issue.
Presenting Author
Cassandra Hung, Wake Forest University
First Author
Cassandra Hung, Wake Forest University
CoAuthor
Sarah Lotspeich, Wake Forest University
Background: Autistic older adults experience higher rates of nearly all psychiatric conditions compared to non-autistic peers, yet psychiatric hospitalizations in this group remain underexamined. Most research on psychiatric service use among autistic individuals focuses on pediatric populations, leaving critical gaps in understanding how aging and autism interact to shape psychiatric service needs.
Objectives: We examined whether autistic older adults differ from population controls (PCs) on odds of psychiatric hospitalization, length of stay (LOS), and odds of psychiatric readmission. Secondarily, we assessed whether co-occurring intellectual disability (ID) modifies these associations.
Methods: We conducted a retrospective cohort study using national Medicare claims data, a federal health insurance program for adults aged ≥65 years. The analytic sample included 7,801 autistic older adults and 7,801 propensity score matched PCs. Multivariable logistic regression models estimated the odds of psychiatric hospitalization, adjusting for demographic and clinical characteristics. Among those with a psychiatric hospitalization, a negative binomial regression model assessed differences in LOS, and logistic regression models estimated the odds of 30-,90- and 180-day readmissions. All models were replicated after stratifying autistic beneficiaries by ID status.
Results: Autistic older adults did not differ from PCs in odds of psychiatric hospitalization. However, they had significantly longer LOS (IRR=1.23, 95% CI: 1.12–1.36) and higher odds of 90-day (OR=1.50, 95% CI: 1.12–2.00) and 180-day (OR=1.35, 95% CI: 1.03–1.76) readmission. Stratified analyses showed autistic older adults with ID had lower hospitalization odds than PCs, while those without ID had higher odds and longer stays.
Conclusions: Autistic older adults, particularly those without ID, face elevated risks for prolonged and recurrent psychiatric hospitalizations, highlighting unmet needs in inpatient care and post-discharge support.
Presenting Author
Alison Deitsch
First Author
Alison Deitsch
CoAuthor(s)
Brittany Hand, The Ohio State University
Melica Nikahd
Understanding the relationship between neighborhood food environments and obesity outcomes relies heavily on accurately measuring food access. However, proximity-based metrics frequently introduce measurement error and misclassification, leading to bias in regression estimates. This study evaluates five two-phase validation sampling designs to correct bias due to error-prone food access measures when modeling obesity prevalence using Poisson regression.
Using census tract-level data from the Piedmont Triad region of North Carolina, we defined true food access based on driving distances and compared it to an error-prone proxy based on underestimated straight-line distances. We implemented various validation sampling strategies, including simple random sampling, case-control sampling , balanced case-control sampling, extreme tail sampling, and residual sampling based on the "naive" model (using error-prone data). For each design, a validation subset of 48 tracts was selected to have true food access measures; all other tracts had only the error-prone versions. We then analyzed tract-level obesity prevalence using the partially validated data and maximum likelihood estimation, jointly modeling the outcome and error processes. Our findings provide practical guidance for designing efficient validation studies in food environment research and other public health applications, demonstrating that strategically selected validation subsets and MLE correction can substantially enhance inference even under limited validation resources.
Presenting Author
Yizhi Zhang
First Author
Yizhi Zhang
CoAuthor
Sarah Lotspeich, Wake Forest University
Inferring heterogeneity of treatment effect is a popular secondary aim of clinical trials. Recently, many trial analyses have moved from traditional subgroup analyses to more modern assessments of heterogeneity using machine learning. While there are several such methods available to estimate conditional average treatment effects (CATEs) in clinical trials, these methods are often applied in trial settings that have lower sample sizes than were considered in the simulations of corresponding seminal methodological work, making the validity of inference in these settings unclear. To provide guidance to practitioners, we conducted a simulation study to evaluate the performance of different regression and machine learning estimators for the CATE, including ordinary least squares (OLS), Bayesian Additive Regression Trees (BART), and causal forests with both default settings and cross-validation based hyperparameter tuning, in a variety of settings across a range of sample sizes.
We evaluated 95% confidence interval (CI) coverage, bias, and variance under linear and non-linear data generating mechanisms (DGM) in the presence of 0 to 40 nuisance covariates and 0 to 16 effect modifying covariates. We found that while tree-based ensembles like causal forests can be quite flexible to linear or nonlinear settings, they can have meaningfully impaired coverage in many settings at sample sizes which constitute most trial applications. As expected, OLS has superior performance under linear DGMs but has poor performance under nonlinear DGMs. We conclude with recommendations for practitioners.
Presenting Author
Lisa Levoir, Vanderbilt University
First Author
Lisa Levoir, Vanderbilt University
CoAuthor(s)
Andrew Spieker, Vanderbilt University Medical Center
Bryan Blette, Vanderbilt University Medical Center
During the COVID-19 pandemic physical activity and being vaccinated were shown to provide protection from serious/critical cases of COVID-19. In the wake of the pandemic, there emerged between 12%-20% of US adults who had COVID-19 and manifested symptoms for COVID-19 three months following their illness but lacked an active infection. Hence these adults have been diagnosed with Long COVID'(LC). The purpose of this study was to examine the associations of physical activity, vaccination status, and selected chronic conditions among US adults identified as having LC using CDC's 2023 Behavioral Risk Factor Surveillance System (BRFSS) with the hypotheses that those adults meeting the aerobic physical activity guidelines (PAG) and those having at least one COVID-19 vaccination (VAX) would have lesser odds of reporting LC. Methods: We examined the association of LC among the 46.4% of adults 18 years and older who had tested positive for COVID-19 (n = 201,248) and a subset these adults who reported having LC (n = 27,074, 13.6%). Both univariate and logistic regression analyses were conducted using SPSS (v29) for complex samples. A series of logistic regression analyses controlling for age, sex, overweight/obesity, type 2 diabetes, race/ethnicity, and educational attainment comparing the outcome variable of LC with the exposure variables of 1) not meeting the PAG and 2) having at < 3 VAX vs. 4 or more VAX were conducted. Results: Adults (n = 9,809; 15.9%) who did not achieve the PAG were at greater odds of reporting LC (OR= 1.19 95% CI = 1.06, 1.33) compared with those meeting the PAG (n = 13,449; 12.2%). Respondents reporting 3 or less VAX with LC (n = 5,092; 13.8%) were at greater odds of reporting LC (OR=1.42, 95% CI = 1.24, 1.49) compared with those reporting 4 or more VAX with LC (n = 1894; 10.04%). Discussion: The present findings support the hypothesis that adults who did not achieve the PAG recommendations manifest greater risk for LC following a case of COVID-19. In addition, those adults who have had 3 or less Vax compared to those with those receiving 4 or more VAX also independently demonstrate greater risk for LC. A suggested mechanism from these findings is the potential synergism between the immune protective effects of sufficient physical activity and having 4 or more COVID-19 vaccinations.
Presenting Author
Gloria Oppong
Harmful Algal Blooms (HABS) can produce toxins that pose serious health risks to humans, pets, and livestock. Data from Utah Lake over the past years have largely been collected in a preferential manner, meaning samples are primarily gathered when HABS are already suspected to be present. This introduces bias, as it fails to capture the full range of bloom conditions, including periods of absence or low risk. A further challenge is distinguishing between presence-only data and systematic presence/absence data, both of which are available and require different modeling strategies. To address these issues, we are developing a Bayesian hierarchical framework that incorporates seasonal patterns and environmental covariates to estimate latent bloom behavior. This ongoing work aims to improve inference on HAB dynamics and support more reliable risk assessment.
Presenting Author
Camilla McKinnon, BYU
First Author
Camilla McKinnon, BYU
Hearing loss affects over 1.5 billion people globally, representing approximately 20% of the world's population. Currently, 430 million individuals have disabling hearing loss, a number expected to rise to 700 million by 2050. Children with hearing loss often receive lower-quality education compared to their peers, and adults frequently experience higher unemployment rates or occupy lower-level jobs. According to the World Health Organization, unaddressed hearing loss results in an annual global economic burden of approximately US\$ 980 billion, encompassing healthcare costs, educational support, lost productivity, and broader societal impacts. Cochlear hair cell loss significantly impairs the conversion of sound into neural signals. To address this critical issue, we propose a novel, automated deep learning approach utilizing Graph Neural Networks (GNNs) to detect missing cochlear hair cells. In our approach, individual hair cells are represented as nodes within a graph, with edges capturing spatial and morphological relationships. This allows GNNs to effectively learn and identify complex patterns associated with hair cell loss and degeneration, providing improved accuracy and efficiency compared to traditional manual morphometric analyses. Our methodology offers a scalable and robust framework for advancing hearing-loss diagnostics and research.
Presenting Author
Ariana Mondiri, Creighton University
First Author
Ariana Mondiri, Creighton University
CoAuthor(s)
Alison Kleffner, Creighton University
Cole Krudwig, Creighton University
Adya Dhuler, Creighton University
Steven Fernandes, Creighton University
Generative artificial intelligence (AI) models have risen in popularity and increased their capabilities since the introduction of ChatGPT by OpenAI in late 2022. AI chatbots can be trained on a variety of materials provided by the user for specific interactions in any number of settings. The goal of this research is to develop and assess an AI-based chatbot designed to support pharmacy students in building essential patient counseling skills within the three-year Dispensing and Patient Care (DPC) course series. The pilot text-based chatbot will initially be created for third year (P3) students in PHA 489 Dispensing and Patient Care III to simulate a realistic patient interaction and provide objective feedback based on the course patient counseling rubric to help students improve communication skills and patient counseling proficiency. Data collection will include chatbot session transcripts, rubric scores, and self-assessment surveys, and will be compared across all pharmacy pathways (Omaha, Distance, and Phoenix). Results will inform the potential expansion of chatbot-based practice opportunities across all levels of the DPC course series. Expected outcomes include increased student confidence in patient counseling, greater readiness for fourth year experiential rotations, and potentially a broader adoption of AI tools in the pharmacy program.
Presenting Author
Danielle Carroll, Creighton University
First Author
Danielle Carroll, Creighton University
CoAuthor(s)
Sara Avila, Creighton University
Cole Krudwig, Creighton University
Jessica Cumber, Creighton University
Kevin Fuji, Creighton University
Halie Erwin, Creighton University
Steven Fernandes, Creighton University
Creighton University prides itself on pre-professional preparation, specifically advising, academic support, and co-curricular programming to help undergraduates achieve their dream of medical school. One of the most intimidating aspects of the medical school application, particularly for students who have focused on the hard sciences, is the personal statement: a brief, high-stakes explanation of why one wants to be a doctor. This research requests funding for the development of a custom chatbot for pre-med students that gives them real-time advice and feedback on the efficacy of the rhetorical elements of their personal statements, based on a rubric co-created by Creighton faculty and students. By reducing the intimidation of starting and revising a draft, the chatbot can encourage students to iteratively improve their statements, making the process more manageable. This feedback can enhance students' confidence and preparedness for medical school admission by making high-quality writing assistance accessible anytime. This research also makes a timely intervention into the larger integration of AI into teaching and learning at Creighton University. This project will test the feasibility of using specially trained AI for student feedback, create a model for the training and development of such a system others can use, and evaluate the effectiveness of AI systems for similar use cases.
Presenting Author
Izzie Nielsen, Creighton University
First Author
Izzie Nielsen, Creighton University
CoAuthor(s)
Faith Kurtyka, Creighton University
McHendry, George (Guy) McHendry, Creighton University
Sara Avila, Creighton University
Cole Krudwig, Creighton University
Sean Dore, Creighton University
Steven Fernandes, Creighton University