Advancing complex longitudinal aging research: competing risks, composite outcome, measurement error

Focus and Timeliness: According to the United Nations report on World Population Ageing, the worldwide population of adults aged 60 years and older is growing faster than the overall population. This growth has catalyzed interest in biomedical research on aging simultaneously with new data modalities made possible through recent technological advancements. As a result, the availability of complex longitudinal data from older adults provides an unprecedented opportunity to advance research on aging to enhance healthspan (length of healthy life) and lifespan. However, longitudinal aging research is vulnerable to multiple threats to validity that lead to bias. Well established challenges of longitudinal aging research that may hamper biomedical breakthroughs include competing risks, measurement error, and informative irregular assessment. Addressing these challenges while concurrently handling additional complexities such as high dimensionality (e.g., via imaging data and biomarkers from high-throughput technologies), multivariate outcomes, and modern study design has become a hallmark of aging research. Thus, solutions require new analytical innovations that cross-fertilize advancements between existing statistical subspecialties and break down silos between these groups. Enhancing timeliness is the National Institute on Aging's Strategic Directions that explicitly emphasizes the importance of new analytical methods to capitalize on the wealth of longitudinal data on aging This topic is therefore particularly valuable for biostatisticians working in health sciences to link longitudinal aging research to other subspecialties (e.g., trial design, image analysis, -omics analysis, survival analysis). Therefore, a session on the nexus of longitudinal aging research with added analytical complexity will enhance crosstalk among biostatisticians working in many applications. Ultimately, the audience will learn both the opportunities and challenges arising from longitudinal data in aging research with the goal of enhancing healthspan. Content: In this session, we will present recent advances in longitudinal methods that address both ongoing technologically driven challenges in aging research. This collection of talks will probe topics that are salient to aging research, which will provide insight into the need for multiple statistical subspecialties to propel research on aging. First, Dr. Seonjoo Lee will present a novel unsupervised machine-learning method to identify correlates of brain morphology changes using high-dimensional brain imaging data and dementia biomarkers. This method extends canonical correlation analysis to longitudinal settings where variables are irregularly assessed at different time resolutions. Second, Dr. Fan Li will introduce strategies for designing cluster-randomized trials for win-ratio analysis using multivariate time-to-event endpoints. Motivated by a cluster-randomized trial evaluating a multifactorial intervention aimed at preventing fall injuries among older adults, a unified framework for power and sample size calculation for the win ratio will be presented. This framework builds upon the Finkelstein-Schoenfeld rank statistic and derives general expressions for the expectation and variance of win statistics in cluster-randomized trials. Third, Dr. Michelle Shardell will present new innovations that involve measured auxiliary variables and unmeasured latent variables to address informative observation times. These methods involve time-varying latent effects and a novel weighting approach and are applied to the US Medicare Minimum Data Set to quantify the role of time-varying cognition in physical recovery after hip fracture. Lastly, Dr. Chengjie Xiong will introduce new approaches to harmonize and bridge longitudinal error-prone biomarkers measured in multiple cohorts on different platforms to quantify disease-related rates of change. The approaches extend multivariate linear mixed models to examine how rates of change in cognition correlate with rates of change in biomarkers measured in cerebrospinal fluid. The proposed participants are all members of the American Statistical Association Statistics and Data Science in Aging Interest Group, an international network comprising almost 400 members that is focused on propelling analytic innovations in aging research through communication and collaboration across statistical subspecialties. Session participants represent a range of research experience from administrator-level senior faculty to mid-career investigators. Appeal: The target audience of this session includes applied statisticians engaged in aging research or who wish to engage in research on aging as well as theoretical statisticians interested in learning about open methodological problems in aging to motivate their theoretical research. Moreover, given that aging research crosses multiple scientific domains and involves multiple statistical subspecialties, statistical methods developed for aging research have potential for broad appeal among Joint Statistical Meetings attendees. Key Words: longitudinal data, unsupervised learning, cluster-randomized trials, informative observation times, composite outcomes; competing risks; multivariate outcomes; neuroimaging; high-dimensional data, dementia, biomarkers