Monday, Aug 4: 10:30 AM - 12:20 PM
4048
Contributed Posters
Music City Center
Room: CC-Hall B
Main Sponsor
Biometrics Section
Presentations
Evaluating the calibration of prognostic biomarkers is crucial for assessing how well predicted risks match observed event rates. A common approach bins predicted risks and compares the observed event rates in each bin against the predicted rates. However, small sample sizes can yield wide confidence intervals (CIs), which can obscure whether observed deviations reflect actual miscalibration or random variation.
A simulation framework was developed to determine the necessary sample size for reliable estimation of bin-specific event rates. We leverage assumptions on the marginal distribution of biomarker risk groups (RGs), the conditional bin distribution per RG, and event rates within RGs. Using a Dirichlet‐multinomial process, individuals are assigned to a RG, then randomly allocated to bins conditional on the RG assignment. Using an exponential survival model, event times are generated to calculate the bin‐level survival estimates and CIs at a fixed time point.
The method provides practical guidance for choosing sample size to ensure robust calibration assessments using an adaptable approach towards different bin schemes and biomarker assumptions.
Keywords
Calibration
Sample Size
Biomarker
Survival Analysis
Simulation
Confidence Interval
Abstracts
Analytical treatment interruption (ATI) trials involve interruption of antiretroviral therapy (ART) in order to evaluate the efficacy of novel treatments for HIV. In these trials, the primary outcome of interest is often the time to viral rebound: the time from ART interruption until the participant's viral load crosses a predetermined threshold. Due to the discrete nature of clinical visit schedules, this time to viral rebound is interval-censored, though many ATI studies implicitly right-impute the failure time. Furthermore, measurement error and the non-monotonicity of the HIV viral load trajectory may lead the first threshold-crossing event to be missed entirely by the intermittent viral load measurements. We conduct a simulation study to evaluate the performance of the Cox proportional hazards model and log-rank test on misclassified interval-censored data, both with and without right imputation of the rebound time. We also investigate how aspects of the ATI trial design may mediate the impact of these data characteristics on the Cox model and log-rank test performance. Finally, we make recommendations based on our results for best practices for ATI trials.
Keywords
HIV/AIDS
clinical trials
interval censoring
misclassification
Cox proportional hazards model
log-rank test
Abstracts
In time series modeling, it is common to assume that innovations follow a normal distribution. However, this assumption does not always hold in real-world scenarios. Environmental datasets, in particular, often contain extreme values that violate normality. Through a comprehensive simulation study, we demonstrate that traditional AR(q) models can produce inaccurate results when innovations deviate from normality, especially when they exhibit skewness. Our findings highlight that outliers can distort estimates, introduce bias, and compromise the generalizability of results.
Keywords
Autoregressive Models
Robustness
Outliers
Skew distributions
Abstracts
Co-Author
Evrim Oral, LSUHSC School of Public Health
First Author
Mohamed Mohamed, Louisiana State University Health Science Center
Presenting Author
Mohamed Mohamed, Louisiana State University Health Science Center
Longitudinal data and panel data are both obtained by repeatedly measuring a certain response variable over time for multiple subjects, and analytical methods for these data have been developed separately in several disciplines. In recent years, methodological integration is occurring. Here, we focus on dynamic models in which the previous response value, that is the lagged dependent variable, appears on the right-hand side of the equation. We compare our proposed autoregressive linear mixed effects models (Funatogawa et al., 2007; Funatogawa and Funatogawa, 2019) with similar dynamic models used in several fields. Our proposed model is an extension of the linear mixed effects model by combining autoregression with it, and has been developed with the main aim of expressing changes in responses over time. We have also provided the state-space representation and the relationships with nonlinear mixed effects models. On the other hand, in panel data analysis, which is used in observational studies, stable unobservable individual characteristics and the lagged dependent variable are often used for adjustments purposes. In this study, we focus on likelihood-based methods.
Keywords
Autoregressive
Dynamic
Longitudinal
Panel Data
Abstracts
Platform trials are multi-arm designs that simultaneously evaluate multiple treatments for a single disease under a shared protocol, benefiting from control data borrowing to improve statistical efficiency. Longitudinal outcomes can provide more precise estimates and increase statistical power for platform analysis. However, relatively few studies have addressed borrowing approaches for longitudinal outcomes in platform trials. In pain and depression studies, outcome trajectories are often nonlinear. To address these issues, we extend Bayesian hierarchical borrowing methods (BHM) to longitudinal endpoints, incorporating nonlinear features within a causal inference framework. We investigate the performance and benefits of various pooling methods: simple pooling, eligible pooling, BHM, and incorporating flexible and adaptable patient-level weights in BHM, in comparison to no borrowing methods. The BHM framework introduces hierarchical structures to balance the extent of borrowing based on data similarity across regimens, optimizing inference while maintaining Type I error control. Simulation results will be presented to compare different borrowing approaches.
Keywords
Bayesian Hierarchical Borrowing
platform trials
Longitudinal
non-linear
Abstracts
Numerous studies have shown that microbial metabolites, which represent the products of bacteria in the human gut, play a key role in shaping cancer risk and response to treatment. However, metabolite data typically contain a large proportion of missing values which are often recorded as zeros. These missing values may result from either low abundance or technical challenges in data processing. Moreover, given the compositionality of microbiome data, where the observed abundances can only be interpreted on a relative scale, standard variable selection methods are not applicable. In this project, we propose a novel Bayesian method to address challenges in both metabolite and microbiome data. Key features of our proposed model include adopting a z-prior to address the compositional characteristics of microbiome data and modeling the two different mechanisms of missing metabolite data. We demonstrate on simulated data that our proposed model can impute the unobserved true metabolite values and correctly select the relevant microbiome predictors. We illustrate our method on real data from a study on the interplay between the microbiome and metabolome in colorectal cancer.
Keywords
Bayesian variable selection
Compositional covariates
Metabolome outcome
Microbiome data analysis
Missing value imputation
Abstracts
Co-Author
Christine Peterson, University of Texas MD Anderson Cancer Center
First Author
Kai Jiang, The University of Texas Health Science Center as Houston
Presenting Author
Kai Jiang, The University of Texas Health Science Center as Houston
In the frequentist framework, Jiang et al. (2016) established the asymptotic properties of the restricted maximum likelihood (REML) estimator under misspecified linear mixed models (LMMs), demonstrating the consistency of the REML estimator for heritability. Our study extends these results to the Bayesian paradigm by considering a non-informative prior on the error variance. We derive the Bayesian marginal maximum likelihood estimator (MMLE) for the signal-to-noise ratio (SNR) and analyze its concentration properties.
Our analysis establishes that the Bayesian MMLE exhibits asymptotic consistency properties analogous to those of the REML estimator. Furthermore, we derive non-asymptotic convergence rates for the Bayesian MMLE, elucidating its behavior under model misspecification, particularly in high-dimensional settings. These results have direct implications for variable selection, uncertainty quantification in hierarchical models, and signal detection in complex data structures.
Keywords
Bayesian estimation
Restricted Maximum Likelihood Estimator (REML)
Model Misspecification
Signal-to-Noise Ratio (SNR)
Marginal Maximum Likelihood Estimator
Asymptotic Consistency
Abstracts
Being able to cluster data with incomplete records is vital in many disciplines. Here, we develop a model-based clustering approach for clustering multivariate discrete data with missing entries using a mixture of multivariate Poisson lognormal distributions. A multivariate Poisson lognormal distribution is a hierarchical Poisson distribution that can account for over-dispersion and can model the correlation between the variables. To illustrate the effectiveness of this method, we have designed a variety of simulation studies to show the robustness of this new method under different percentages of incomplete records and patterns of missing data. Additionally, the approach is used to demonstrate clustering partial records from a proteomics dataset.
Keywords
Clustering
Missing Data
Discrete Data
Multivariate Poisson Log Normal Distribution
Abstracts
Several factors make clustering functional data challenging, including the infinite dimensional space to which observations belong and the lack of a defined probability density function for functional random variables. Despite extensive literature describing clustering methods for functional data, clustering of error-prone functional data remains poorly explored. We propose a two-stage approach: first, clustered mixed-effects models are applied to adjust for measurement-error bias; second, cluster analysis is applied to measurement error–adjusted curves. Readily available methods (e.g., K-means, mclust) can be used to perform the cluster analysis. We use simulations to examine how complex heteroscedastic measurement error affects clustering, considering variations in sample sizes, error magnitudes, and correlation structures. Our results show that ignoring measurement error in functional data reduces the accuracy of identifying true latent clusters. When applied to a school-based study of energy expenditure among elementary school–aged children in Texas, our methods achieved enhanced clustering of energy expenditure.
Keywords
clustering
functional data
measurement error
physical activity
wearable device
Abstracts
The discovery of the urinary microbiome has ignited investigations into the mechanisms of disease and potential interventions for improved bladder health. Characterizing a patient's "baseline" is key to properly evaluate change due to exposures and treatments such as probiotics, antibiotics, and surgical intervention. In our previously published prospective observational study, we found that urological surgery altered the urinary microbiome, with differences in recovery to baseline in premenopausal versus postmenopausal women. This study is a secondary analysis of these data, capitalizing on additional samples to describe assay variability, evaluate stability across multiple available baseline samples (screening, pre-operative), and estimate the minimal detectable change in key microbiome features (diversity indices and prevalence/relative abundance of specific microbes). These data can inform those conducting longitudinal clinical studies in this field, where for convenience a urine sample at a single timepoint is most often collected to establish a baseline. Power analyses and sampling design should account for expected variability of the dynamic ecosystem in the bladder.
Keywords
microbiome
clinical research
Abstracts
Analyzing the effect of obesity on chronic disease risk is challenging due to endogeneity, measurement error, and complex dependencies between obesity, physical activity, and health outcomes. Standard statistical methods, such as generalized linear model, often fail to adequately address these issues, leading to biased estimates. To overcome these limitations, we develop a bivariate semi-parametric recursive copula model that flexibly accounts for non-linear relationships and intricate dependency structures. We evaluate the finite sample properties of our approach through simulation studies and apply it to NHANES 2011–2014, incorporating device-measured physical activity to enhance estimation accuracy. Results confirm the robustness of our method and reinforce the causal association between obesity and chronic disease risk. This study highlights the importance of advanced statistical techniques for improving average treatment effect (ATE) in epidemiological research.
Keywords
Obesity
cardiovascular disease
semi-parametric recursive copula model
endogeneity
physical activity
diabetes
Abstracts
Understanding treatment effects through biological pathways is an essential objective in biomedical investigation. Causal mediation analysis (CMA) provides a useful framework for such inquiries. However, the natural direct effect (NDE) and natural indirect effect (NIE) may depend on specific patient characteristics. To account for such heterogeneity, we include covariate-treatment and mediator-treatment interactions in the outcome model. We relax the strict hierarchical constraint by including interactions without requiring the corresponding main effects. NDE and NIE are then calculated for given values of the covariates. To maintain model parsimony in the presence of high dimensional covariates, we apply generalized LASSO regularization to select key covariate-treatment interactions. Simulation studies show that the method has good performance in selecting the interactions. The method can properly stratify individuals and achieve unbiased estimates for the NDE and NIE. The method represents a step forward in understanding the heterogeneity in the mediation pathway of the treatment within personalized medicine. Data from a real clinical study were used to illustrate the method.
Keywords
Causal Mediation Analysis
Heterogeneous Treatment Effects
Generalized LASSO
High-Dimensional Covariates
Natural Direct and Indirect Effects
Personalized Medicine
Abstracts
Inferring heterogeneity of treatment effect is a popular secondary aim of clinical trials. While there are several methods available to estimate conditional average treatment effects (CATEs) in clinical trials, they are often applied in settings with lower sample sizes than were included in corresponding seminal methodological work, making the validity of inference in these settings unclear. To provide practical guidance, we conducted a simulation study to evaluate the performance of different estimators for the CATE, including ordinary least squares (OLS) and causal forests, in a variety of settings. We evaluated 95% confidence interval coverage, bias, and variance under linear and non-linear data generating mechanisms (DGM) in the presence of 0-40 nuisance covariates and 0-16 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 poor performance under nonlinear DGMs. We conclude with recommendations.
Keywords
heterogeneous treatment effects
causal forests
machine learning
simulation
causal inference
Abstracts
In the past couple of years, the Clinical Research outlook has evolved, including rising costs for developing new treatments. This has led the industry to look for new innovations to help alleviate these costs, without the risk of losing scientific reliability. A few that I will be looking at, but not limited to, will include real-world data / real-world evidence and natural history studies. These discussions will include the benefits, pitfalls and unique considerations associated with each of these, as well as the stance that regulatory authorities currently have towards them. Each of these new innovations include their own different statistical methodology considerations to take into account. I will also look at some real examples where such sources were used in the industry thus far (the good, the bad and the ugly). Lastly, I will discuss what this could imply for the future and the practicing statistician.
Keywords
Real-World Data / Real-World Evidence
Purpose and utilization of natural history studies (i.e. external controls, and retrospective and prospective studies).
Sensitivity/covariate analysis between external data and recruited controls.
Doubly debiased machine learning to estimate average treatment effect (ATE)
The stratification score estimated probability of the outcome given potential confounders matching of external controls to mimic internal controls.
Abstracts
First Author
Ian Lees, MMS Holdings, Inc
Presenting Author
Ian Lees, MMS Holdings, Inc
Micro-randomized trials (MRTs) are often used in mHealth studies to assess app-based interventions. Participants are randomized to receive treatment at a series of decision points, traditionally using the same rule across individuals. Several recent MRTs utilize Thompson Sampling (TS), a reinforcement learning algorithm, to build individualized treatment strategies that optimize delivery with respect to a reward. Treatment may interact with several contextual features, but estimation of models in this setting can be unreliable. This is especially difficult with a binary reward where complete separation often occurs, even with a large sample and few features. We present an approach to balance algorithmic flexibility and computational cost in the context of a binary reward that (1) uses partial pooling and weakly informative priors that apply more shrinkage to higher-order interactions and (2) considers the amount of information available in the data when defining a model. Our approach is useful in MRTs where the TS algorithm must be automated. We demonstrate the empirical utility of our method in a digital twin of an ongoing MRT study, LowSalt4Life, compared to logical alternatives.
Keywords
Mobile health
Micro-randomized trials
Clinical trials
Reinforcement learning
Individualized treatment
Abstracts
Future precision medicine requires accurate assessment on the explainable variability in treatment effects, known as heterogeneous treatment effects (HTE), to guide the optimal clinical decision at individual level. Measuring HTE by the ratio of survival probabilities under structural failure time model, we develop a martingale R-learner to estimate HTE. Our martingale R-learner incorporates flexible estimators for 1) marginal survival or cumulative hazards for association between outcome and confounders, and 2) time-varying propensity score in risk sets, which enables leveraging advances in machine learning. To reduce the impact of estimation bias in these two nuisance models on HTE, we proposed a Neyman orthogonal score based on an orthogonal decomposition of conditional model martingale residuals into residuals of propensity score and marginal model martingale. The resulting martingale R-learner attains the quasi-oracle property, i.e. estimation error of nuisance models have no impact on HTE if their estimators are consistent at o(n^(-1/4)) rate. Numerical experiments in various settings demonstrated valid empirical performance consistent with theoretical properties.
Keywords
heterogeneous treatment effect
causal inference
survival analysis
orthogonal score
Abstracts
Co-Author(s)
Jue Hou
Ronghui Xu, University of California-San Diego
First Author
Yuchen Qi, UC San Diego, Department of Family Medicine & Public Health
Presenting Author
Yuchen Qi, UC San Diego, Department of Family Medicine & Public Health
Random-effects meta-analyses with only a few studies often face challenges in accurately estimating between-study heterogeneity, leading to biased effect estimates and confidence intervals with poor coverage. This issue is especially the case when dealing with rare diseases. To address this problem for normally distributed outcomes, two new approaches have been proposed to provide confidence limits of the global mean: one based on fiducial inference, and the other involving two modifications of the signed log-likelihood ratio test statistic in order to have improved performance with small numbers of studies. The performance of the proposed methods was evaluated numerically and compared with the Hartung-Knapp-Sidik-Jonkman (HKSJ) approach and its modification for handling small numbers of studies. Simulation results indicated that the proposed methods achieved coverage probabilities closer to the nominal level and produced shorter confidence intervals compared to those based on existing methods. Two real data examples are used to illustrate the application of the proposed methods.
Keywords
confidence interval
fiducial inference
modified LRT statistic
small sample asymptotics
rare diseases
Abstracts
This project proposes a propensity score (PS)-based stratified win ratio method to address challenges of small patient populations in clinical trials, especially for rare or pediatric diseases, by incorporating external control data. Our approach enhances traditional win ratio analysis by leveraging PS stratification to account for heterogeneity between the current and external studies. Additionally, down-weighting based on the overlapping coefficient of PS distributions of current treatment and external control groups further mitigates the patient bias due to heterogeneity. Our simulations showed significant improvements in statistical power for detecting treatment effects within the composite endpoint, over non-borrowing and pooling methods, with utilizing Mantel-Haenszel (MH)-type weights achieving the highest power. The proposed methods are also applied to an Amyotrophic Lateral Sclerosis (ALS) study incorporating the external control arm from a prior ALS trial. The proposed PS-based stratified win ratio method thus provides a rigorous framework for borrowing external data and analyzing composite endpoints with limited patient availability.
Keywords
Placebo borrowing
Win ratio
Composite endpoint analysis
Propensity score stratification
Abstracts
Parkinson's Disease (PD) is a devastating neurodegenerative disorder that affects millions of people around the globe. Many researchers are continuously working to understand PD and develop treatments to improve the condition of PD patients that affects their day-to-day lives. In the last decades, the treatment of Deep Brain Stimulation (DBS) has given promising results for motor symptoms by improving the quality of daily living of PD patients. In the methodology of the present study, we have utilized sophisticated statistical approaches such as Nonparametric, Semiparametric, and robust Parametric survival analysis to extract useful and important information about the long-term survival outcomes of the patients who underwent DBS for PD. Finally, we were able to conclude that the probabilistic behavior of the survival time of female patients is statistically different from that of male patients. Furthermore, we have identified that the probabilistic behavior of the survival times of Female patients is characterized by the 3-parameter Lognormal distribution while that of Male patients is characterized by the 3-parameter Weibull distribution.
Keywords
Survival Analysis
COX-PH
Deep Brain Stimulation
Movement Disorder
Parkinson’s Disease
Parametric, Non-Parametric and Semi-Parametric Survival Analysis
Abstracts
We introduce a novel Kernel Regression estimator, Kernel Regression with Tree-Exploring Aggregations (KR TEXAS), that learns a distance metric while allowing feature aggregation along a predefined tree structure. This approach is particularly relevant for microbiome analysis, where data is often collected at multiple taxonomic levels, and determining the appropriate level of aggregation is non-trivial. Unlike traditional aggregation methods that rely on uniform taxonomic levels, KR TEXAS leverages an L1-penalized distance metric to selectively aggregate features based on their importance, leading to biologically interpretable results. Our method extends prior work on metric learning and nonparametric regression, incorporating structured feature aggregation to improve predictive accuracy and interpretability. We demonstrate the utility of KR TEXAS through both simulations and real microbiome datasets, highlighting its advantages in capturing functional relationships that may be missed by conventional aggregation techniques.
Keywords
Microbiome
Compositional Data
Kernel Regression
Metric Learning
Abstracts