FINDING YOURSELF: PERSONALIZED MEDICINE, DATA SCIENCE, AND INTERACTIVE VISUALIZATIONS

Abstract Personalized medicine is care that is tailored to an individual patient. In contrast, randomized trials are designed to report evidence of benefits and harms “on average”. The average trial participant though, is often healthier than older patients seen in clinical practice. A variety of methods have been proposed which offer improvements to traditional (but questionable) practices of one-subgroup-at-a-time examinations of treatment effect heterogeneity, but translating evidence from these advancements has received less attention. Data Science initiatives have allowed broader data sharing, data harmonization and data synthesizing approaches, as well as an enormous maturing of interactive visualization techniques. Motivated by the SPRINT study and using advanced approaches of marginalized standardization and the Predictive Approaches to Treatment effect Heterogeneity (PATH) statement, we show how researchers, regulators, clinicians and patients can "find themselves" on the treatment effect continuum and be better informed of potential individualized evidence through interactive visualizations.


GERONTOLOGIC BIOSTATISTICS: MERGING WITH DATA SCIENCE AND TOWARD PERSONALIZED MEDICINE
Chair: Michelle Shardell Co-Chair: Terrence Murphy Discussant: Heather Allore The sub-discipline of gerontologic biostatistics (GBS) was introduced in 2010 to emphasize the special challenges encountered in the design and analysis of research studies of older persons. These challenges center on the multifactorial nature of human aging, characterized by the parallel and progressive deterioration of diverse organ and cellular systems that eventually results in death. Ten years after the introduction of GBS, which initially focused on important aspects of design and analysis that ensure their statistical validity, we update how GBS has been enriched by evolving practices. We present individual sessions on three seminal developments in the practice of GBS: integration of data science and multiple streams of data, including those automated and or multidisciplinary in nature; enhanced methods of addressing the heterogeneity of treatment effects from health-related interventions for older patients; and how interactive visualization can help specific patients locate themselves along the continuum of individualized treatment effects. We conclude our presentation with a session that reviews three prominent trends in the validation of the heterogeneity inherent to the assessment of health among older adults. Reflecting this era of big gerontological data, we discuss several established modeling approaches for validation, the proliferation of signal intensive behavior phenotypes, and the deep characterization of phenotypes through OMICS studies and multimodal approaches. All talks discuss pitfalls and areas of future development and draw from published studies. We are submitting as an interest group collaborative panel submission between two interest groups: Epidemiology of Aging and Measurement, Statistics and Design.

GERONTOLOGICAL BIOSTATISTICS AS THE FOUNDATION OF INTERDISCIPLINARY DATA SCIENCE IN AGING Thomas Travison, Harvard Medical School, Boston, Massachusetts, United States
Explosive growth in computing power has increased by orders of magnitude the complexity of data structures and the number of analyses that may be performed per unit time. Contemporary gerontology utilizes diverse data ranging from continuous longitudinal assessments (e.g. motion capture) to complex single-timepoint assessments (e.g. bioimages) to systems-level administrative descriptions of the healthcare delivery environment. Paradoxically, this abundance of resources presents a considerable challenge, as the availability of information threatens to overwhelm mechanistic models of aging supportive of intervention development. To inform precision medicine, gerontological biostatistics therefore embraces the opportunity of collaborating with allied quantitative disciplines to bolster the coherence, reproducibility, and generalizability of findings. This presentation will demonstrate the salient advantages of such interdisciplinary collaborations using the example of design and analysis of an intensely longitudinal study of wearable and environmental sensors, conducted with teams in exercise science and architectural design.

Qian-Li Xue, Johns Hopkins University, Baltimore, Maryland, United States
Adopting evidence-based medicine in clinical care of older patients is challenging because the "best" evidence available may not be directly applicable due to exclusion or underrepresentation of older adults in clinical trials. Interventions shown to be beneficial in trial populations therefore often exhibit heterogeneous treatment effects (HTEs) in older adults, particularly among the most vulnerable. This talk will review the concept, causes, and estimation of HTEs. We distinguish clinical heterogeneity from methodologic heterogeneity, defined respectively as the variation in biological mechanisms leading to similar aging phenotypes and variation in study design, measurement, and analysis. We use examples drawn from geriatric medicine to introduce novel study designs and data analytics used to study HTEs. This talk highlights the importance of moving beyond post-hoc subgroup analysis to an approach that integrates theories, observational and experimental data, and data science in the study of HTEs. Personalized medicine is care that is tailored to an individual patient. In contrast, randomized trials are designed to report evidence of benefits and harms "on average". The average trial participant though, is often healthier than older patients seen in clinical practice. A variety of methods have been proposed which offer improvements to traditional (but questionable) practices of one-subgroup-at-a-time examinations of treatment effect heterogeneity, but translating evidence from these advancements has received less attention. Data Science initiatives have allowed broader data sharing, data harmonization and data synthesizing approaches, as well as an enormous maturing of interactive visualization techniques. Motivated by the SPRINT study and using advanced approaches of marginalized standardization and the Predictive Approaches to Treatment effect Heterogeneity (PATH) statement, we show how researchers, regulators, clinicians and patients can "find themselves" on the treatment effect continuum and be better informed of potential individualized evidence through interactive visualizations.

SIGNAL DETECTION AND VALIDATION IN AN ERA OF BIG GERONTOLOGICAL DATA Karen Bandeen-Roche, Johns Hopkins Bloomberg School of Public Health, Baltimore, Maryland, United States
Older adult health assessment long has posed measurement challenges-multidimensionality of sentinel outcomes like functioning and frailty, for example. This presentation discusses three developments creating opportunities for gerontologic biostatistics (GBS) over the past 10 years. Firstly, modeling to internally validate measurements or to quantify systematic heterogeneity in assessing older adult health has become considerably more widespread. Confirmatory latent variable modeling, harmonization, and mixture models will be addressed. Secondly, signal intensive behavioral phenotypes are proliferating, e.g. accelerometry, sleep actigraphy, and ecological momentary assessment. Functional data analysis is described as a data analytic technique to extract signal capturing main behavioral features or most relevant for health outcomes. Thirdly, "deep" characterization is under hot pursuit-whether by single-or multi-'omics studies, or by multimodal phenotyping as is increasingly common in the study of cognition. Techniques to accomplish this replicably are discussed. Throughout, potential pitfalls and implications for gerontologic data science development are identified.

CHRONIC PHYSICAL HEALTH CONDITIONS, DEPRESSIVE SYMPTOMS, AND SELF-RATED HEALTH IN GRANDMOTHERS
Christina Henrich 1 , Carol Musil 1 , Jaclene Zausniewski 1 , Christopher Burant 2 , Elizabeth Tracy 1 , and Alexandra Jeanblanc 1 , 1. Case Western Reserve University, Cleveland,Ohio,United States,2. Case Western Reserve University,Parma,Ohio,United States Grandmothers often participate in caregiving to grandchildren, ranging from informal caregiving to having responsibility for raising grandchildren, but health problems may affect their ability to do so. To better understand potential health problems in grandmother caregivers, this secondary analysis examined 1) longitudinal relationships between chronic physical health conditions, depressive symptoms, and self-rated health with race, age, & marital status, and 2) causal ordering of depressive symptoms and self-rated health, controlling for baseline race, age, marital status, and chronic conditions of 328 Ohio grandmothers across 3 time points over 2 years. The study was guided using Lenz's Theory of Unpleasant Symptoms. Measures included chronic physical health conditions (modified Charlson), depressive symptoms (CES-D), self-rated health (SF-36), and demographic variables of race (white/ nonwhite), age, and marital status (married-partnered/not married). Analyses included correlations and structural equation modeling. Pearson product moment correlations indicated significant relationships (ranging from 0.26 to 0.56) between chronic conditions, depressive symptoms and self-rated health at all three time points. Next, a bivariate autoregressive model was tested using structural equation modeling. The data fit the model well. The goodness of fit results included: Chi square=55.39, TLI=0.94, CFI=0.97, and RMSEA=0.07. Results indicated consistency of depressive symptoms and self-rated health over time, and these were affected by chronic conditions; the causal order indicated that self-rated health impacts depressive symptoms. This study contributes new knowledge about the relationships and causal sequence of self-rated health and depressive symptoms. Implications on practice and research are discussed. Among older adults with multimorbidity (MM), diseaserelated stressors (e.g., pain) are associated with greater depressive symptoms. However, the contextual factors influencing this relationship remain understudied. We explored the moderating effects of interpersonal, sociocultural, and temporal factors as buffers of this relationship. Adults ≥ 62 years with MM (n=366) recruited through a national health volunteer registry and an online panel platform completed validated scales assessing diagnoses, disease-related stressors (pain intensity, subjective cognitive function, physical function, somatic symptoms), depressive symptoms. Potential moderators: age, expectations regarding aging, perceived social support, and difficulty affording medications (proxy for SES). Data were analyzed with structural equation modeling. Participants were 62-88 years old and living with many illnesses (M = 3.5; range: 2-9); 15% reported moderate-to-severe depressive symptoms. Among those with low social support, the effect of disease-related factors on depressive symptoms was greater (B = .70, SE = .06, p <.001) than for those with high social support (B = .46, SE = .06, p < .001)]. The negative effect of disease-related factors on depressive symptoms was stronger for those with worse expectations of aging (B = .68, SE = .07, p <.001), compared to those with more positive expectations (B = .47, SE = .06, p < .001). Age and difficulties affording medications were not significant moderators. Among older adults with MM, garnering social support and addressing low expectations for