Understanding Pathways to Successful Aging: How Social and Behavioral Factors Affect Health at Older Ages: Workshop in Brief (2015)

Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

WORKSHOP IN BRIEF Board on Behavioral, Cognitive, and Sensory Sciences September 2015 Understanding Pathways to Successful Aging: How Social and Behavioral Factors Affect Health at Older Ages— Workshop in Brief Research has identified many behavioral, social, and biological factors that are associated with healthy aging. Less well under- stood are possible causal relationships between such factors and healthy aging outcomes or the mechanisms through which these factors may influence the aging process. Improved understanding of these relationships is needed to support the design of interventions to promote healthy outcomes at midlife and older ages. To advance understanding about antecedents to healthy aging, the Board on Behavioral, Cognitive, and Sensory Sciences held a workshop, “Understanding Pathways to Successful Aging: How Social and Behavioral Factors Affect Health at Older Ages,” on June 11-12, 2015. Supported by the Division of Behavioral and Social Research at the National Institute on Aging (NIA) and chaired by Susan Fiske, Eugene Higgins professor of psychology and public affairs at Princeton University, the workshop provided an opportunity for a diverse group of researchers to explore research strategies and ways to build on what is known about influences on healthy aging. Workshop participants were asked to: • review what is known about three exemplar factors which research has demonstrated are associated with healthy aging: optimism, marital satisfaction, and educational attainment; • define a set of objective criteria for delineating causal pathways and the causal relationships that underlie the associations; and • identify the strongest approaches for investigating how these antecedents are linked to favorable outcomes and the most promising targets for interventions. This document summarizes the presentations and discussion; more information about the project and the workshop can be found at www.nas.edu/SuccessfulAging. SETTING THE STAGE Lis Nielsen of NIA’s Division of Behavioral and Social Research set the workshop context by describing the division’s goals and the range of research tools used in the work on aging supported by the division. The division’s portfolio integrates diverse data collected by researchers in the population and behavioral and psychological sciences. Overall, the division takes a “life-course perspective” to aging, she noted. Its goal is to support research on factors that promote healthy aging trajectories, as well as opportunities to intervene throughout the life-cycle to promote healthy aging—not just to focus on the last stages of life. David Reiss, also of the Division of Behavioral and Social Research, elaborated on the research issues that prompted the workshop, noting that there is a long time gap between many antecedents and the life stages at which their influence may be significant. This time gap is a key reason that it has been difficult to establish causal links and to identify the specific mediators that alter the pathways for individuals. The workshop, he noted, was designed to explore research that addresses this challenge in creative ways, such as: detailed observations taken over a short period to fill in the picture of how mechanisms might operate; true exper- iments, in which subsamples of larger study groups are examined in greater detail; and randomized microtrials, studies that use the methods of larger controlled trials to investigate the effects of very specific interventions. 

THREE ANTECEDENTS TO HEALTHY AGING of which, in turn, influence cardiovascular disease. Positive The workshop planning committee asked three experts to psychological functioning may reduce stress or mitigate its discuss the state of evidence regarding causal relationships effects, and it may also promote processes that are primarily between optimism, marital satisfaction, and educational restorative and mitigate processes that are primarily deterio- attainment and outcomes at older ages. The presenters also rative in relation to health. summarized what is known about the specific behavioral, There are several challenges in establishing causality, she social, and biological mechanisms through which these explained. There may be variables that are not measured that antecedents might influence aging, as well as the research affect both optimism and cardiac health. If these confound- challenges that remain for both delineating causal pathways ing variables are not accounted for, researchers might errone- and identifying targets for intervention. ously conclude that there are associations between optimism and cardiac health. Ideally, one would rule out all such con- Optimism founders, but in practice this is not possible. Another issue is Laura Kubzansky, professor of behavioral and social science that if self-reports are used to measure both optimism and at Harvard University, emphasized the importance of con- health-related factors (e.g., physical activity), the findings sidering not only deficits associated with aging, but also the could be biased if optimistic people are more likely than those influence of assets and positive factors. Positive psychologi- who are not optimistic to emphasize the positive in their cal functioning is not simply the absence of “ill-being,” she health reports. noted, and it includes such commonly researched constructs The direction of causality may also be difficult to establish as a sense of purpose in life, a sense of autonomy and self- because many effects are bidirectional. That is, not only does efficacy, personal growth, self-acceptance, vitality, and opti- being optimistic promote cardiac health, good health likely mism. causes people to feel more optimistic. She noted that the evidence for a relationship between opti- Kubzansky discussed ways to address the problem of estab- mism and health is strongest for heart disease: Figure 1 shows lishing causality. Using longitudinal data can help to estab- the mechanisms that may explain the association. Positive lish whether optimism precedes changes in health or health psychological functioning is involved in the development behaviors, and statistical approaches can be used to adjust and progression of disease, as well as in the maintenance of for potentially confounding variables. For example, individ- health. A variety of positive psychological attributes can influ- ual studies that use such approaches have demonstrated that ence both health behaviors and biological processes, both optimism is associated with a decreased risk of developing coronary heart disease and other adverse health outcomes. In general, Kubzansky Health Effects of Positive observed, it is clear that optimism is asso- Psychological Functioning ciated with longevity, cardiovascular and pulmonary health, and some immune markers. Positive Psychological – A number of psychological, behavioral, Functioning Stress and biological mechanisms that might – explain how optimism influences health + – + have been studied, Kubzansky noted. Some lie on a continuum from restor- ative to deteriorative (e.g., from a healthy Restorative Health Biological Deteriorative to an unhealthy diet, range of levels of Processes Behaviors Function Processes HDL cholesterol), while others are either primarily restorative (e.g., meditation) or primarily deteriorative (e.g., smoking). – + With regard to psychological mecha- Cardiovascular Disease nisms, optimism promotes two psycho- logical processes—self-regulatory capac- ity and effective interpersonal skills—that may influence problem solving and the Figure 1 Model of the relationship between positive psychological functioning and cardiovascular capacity to set goals and priorities. These disease. Adapted from Boehm, J.K., and Kubzansky, L.D. (2012). The heart’s content: The association processes may, in turn, influence the between positive psychological well-being and cardiovascular health. Psychological Bulletin, 138(4), 655-691. Published by the American Psychological Association. Reprinted with permission. likelihood of engaging in healthy behav- 2 

ior, another key pathway that links optimism with health. Behavioral data has helped researchers understand how There is research linking optimism with both behaviors that marital quality may influence health, explain differences affect health (e.g., cigarette smoking, physical activity, sleep among people, and better predict outcomes, and those habits) and biological processes that are linked to cardiac data provide more objective measures than data from health and longevity (e.g., blood pressure and levels of self-reports. Kiecolt-Glaser noted that negative behaviors, lipids and antioxidants). It may be important to consider such as psychological abuse, hostility, and withdrawal these mechanisms jointly. “People look at behavior and from interaction, are closely tied to damaging physiologi- biology as separate, but they are bidirectional,” Kubzansky cal changes. For example, chemical changes in the blood observed. that occur during distressing interactions persist for many hours after the interaction has ended. Hostility and other Optimism can be modified by the social environment, she negative behaviors can retard wound healing and contrib- noted, and research has also identified some strategies ute to depression, emotional stress responses, and other that may enhance optimism, including cognitive behavior detrimental health outcomes for the individual receiving the therapy. What has not yet been established, however, is negative behavior. Poor health behaviors, in turn, promote whether increasing optimism—or, more generally, improv- inflammation, which is associated with numerous diseases. ing psychological functioning—leads to improvements in Research also shows that marital behaviors are “amazingly physical health. “We don’t know whether optimism is truly consistent,” she said: it is very difficult for interventions such a causal factor for health,” Kubzansky explained, or by what as counseling to have a lasting impact on a bad marriage. precise mechanisms it may promote health. It is also not known whether increasing optimism at any particular period Kiecolt-Glaser ended by noting two areas in which further may lead to healthier aging or whether there are periods in research would be useful: an exploration of cross-cultural people’s life-cycles during which prevention or intervention and cohort differences among married people and the inte- may be particularly effective. Experimental approaches have gration of data from qualitative observational data on long- been limited so far, but they have yielded insights that could term marriages with longitudinal evidence. be the basis for future work. Taking a life-course perspective is another promising avenue for research. Educational Attainment Jennifer Manly, associate professor of neuropsychology at Marital Satisfaction the College of Physicians and Surgeons of Columbia Uni- Janice Kiecolt-Glaser, director of the Institute for Behavioral versity, explained that the sources of evidence about the Medicine Research at Ohio State University, noted that mar- connections between education and health are varied. riage is the central relationship for most adults. Married Some randomized trials have established the benefits of people have lower rates of cancer, heart attack, and surgery, early childhood education—with data from Project STAR, and they live longer than unmarried people. However, the Perry Preschool, the Abecedarian studies, and studies being married is not necessarily protective of health because of Head Start—but there are few of them, and they have a troubled marriage is itself a prime source of stress, and it not been adequately followed up, she noted. Quasi-ex- may also limit a partner’s ability to seek support in other periments—using data associated with events such as the relationships. Kiecolt-Glaser emphasized that both the pos- opening of many colleges to women and changes in com- itive and negative effects of marriage are cumulative over pulsory school laws, as well as data on children whose birth- time. People’s social networks decrease with age, and older days fall near school entry cutoff dates—can also be used adults are more vulnerable than are younger adults to trou- to explore the effects of differences in educational attain- bled marital relationships. ment on health outcomes. However, these studies generally show small effect sizes, she noted. Another approach, twin Studies have shown an association between marital quality studies, has generally shown lower correlations between and cardiovascular health; this influence is stronger for years of schooling and later outcomes, a result that has women than for men. Marital discord is also associated puzzled observers. with depression, an association that is clearly bidirectional, Kiecolt-Glaser said. There is evidence that the effects of a Longitudinal observational research, such as in birth cohort negative marital experience on people’s self-reported state studies, has been used to control for potentially confound- of health increase as they age. With age, she added, highly ing variables; this work has pointed to possible mechanisms coordinated mechanisms that support health at the molecu- and mediators of the effects of education on health at older lar, cellular, and organ levels are altered. The effects include ages. Manly also noted well-supported findings related to mitochondrial and DNA damage; changes in metabolism; cognitive decline: for example, education is associated with functioning of the endocrine, immune, and cardiovascular slowed rates of cognitive loss or reduced risk of late-life systems; and frailty and increased susceptibility to chronic dementia. However, the effects of education appear to be diseases. strongest for diseases that are preventable, such as those 3 

related to smoking, though there are challenges in establish- State University of New York at Stony Brook, began by noting ing causal relationships for these outcomes, Manly cautioned. that identifying a predictor of a health outcome, which is dif- Unobserved variables may confound the results, and the ficult on its own, is not enough. In order to know whether effects of childhood health and of level of educational attain- intervening would actually be helpful, one needs to know to ment may be bidirectional. Another issue is survival bias. what extent the predictor might change either on its own or Because mortality is higher at all ages among people with low in response to an intervention and when and how it might be levels of educational attainment, those who live to be studied susceptible to change. as older adults tend to be “hardier” than the larger group of Most experimental studies establish how particular mecha- survivors with high levels of educational attainment. Even if nisms can work, but long-term follow-up is needed to demon- hardiness is not initially related to educational attainment, strate causality. For example, he explained, the research on the selective survival rates could bias estimates of the effect stress has clearly shown its acute influence on cardiovascular of education on survival. This bias may be present not only in reactivity and other health factors. However, in order to estab- mortality rates, but also for other dichotomous health events. lish that stress causes poor outcomes, it will be necessary to Research has identified possible mechanisms by which edu- measure exposure to chronic stress, possible associated medi- cational attainment may influence health and aging. Mate- ators and moderators, and health outcomes of interest over rial factors—such as income, occupational and environmen- time. Many of the psychosocial factors that have been identi- tal hazards, and access to medical and social services—are fied as harmful or beneficial are likely to affect multiple organ all associated with education level and have implications for systems. In longitudinal studies, cumulative exposure to such health. Manly noted that biological changes associated with harmful or beneficial factors can be measured repeatedly education also appear to be protective: enriched environ- over time with a focus on the specific outcomes of interest. ments lead to the development of more complex neural cir- Schwartz commented that although it is difficult to get the cuitry, denser and more complex synaptic connections, and temporal ordering of data correct, or to know that you have increased neuroplasticity. Education is also associated with the causal direction right, studying the relationship between a reduced “cumulative allostatic load,” which is the long-term change in X and a change in Y is a better approach than exam- physiological wear and tear associated with chronic stress. ining cross-sectional differences for increasing understanding These associations suggest that educational attainment is of causality. really a marker for all sorts of other factors,” Manly noted. It is Maria Glymour, associate professor of epidemiology and difficult to trace the interactions among a wide range of psy- biostatistics at the University of California at San Francisco, chosocial and behavioral traits across the life course—includ- observed that all causal inferences, even those based on ran- ing social capital and proximity to educated people, problem domized controlled trials, rest on some unverifiable assump- solving, healthy behaviors, conscientiousness, cognitive tions. Causal inferences are strongest if they are supported engagement—and education and health. by evidence from multiple studies that are based on different Disentangling these factors will be difficult, Manly added. assumptions and use different research designs and types of One challenge is measuring the quality of schooling. Another evidence. is identifying the critical points in the life-cycle when educa- Glymour suggested that several analytic methods have been tion may play a particularly important role for health. Also underused in the search for supportable causal inferences. needed is research on which mediators are most malleable; Analysis of instrumental variables—factors that are not neces- the role of gender, race, and cultural differences; and whether sarily intentionally randomized in the original design but are the health effects of education are specific to particular health correlated in some way with the antecedents being studied outcomes. and may provide “pseudo-randomization”—is one powerful complement to other designs. These variables provide a way CLARIFYING PATHWAYS, ENHANCING CAUSAL of estimating causal effects when the relationship between the ANALYSES, AND DELINEATING MECHANISMS antecedent and the outcome may be bidirectional and when Each of the antecedents discussed by Kubzansky, Kiecolt- controlled experiments are not practical. Policy changes, such Glaser, and Manly point to consistent methodological chal- as extending the length of the school year or desegregating lenges, noted Arun Karlamangla, geriatrician and clinical public schools, provide opportunities for the use of natural epidemiologist at the University of California, Los Angeles. instrumental variables, she noted. Studying such changes can Karlamangla moderated a session focused on three strategies provide immediate information about the possible effects of for addressing these problems: longitudinal studies, molec- other interventions. ular and quantitative genetic approaches, and experimental Figure 2 illustrates this point, showing that a policy change, approaches. such as changing the age for compulsory school entry, will affect educational outcomes, which, in turn will affect the Longitudinal Studies health of adults as they age. It also shows that parental educa- Joseph Schwartz, professor of psychiatry and sociology at the tion and general health endowment both affect how individ- 4 

uals respond to that education, apart from any effects of the Molecular and Quantitative Genetic Approaches policy change. Matt McGue, professor of psychology and behavioral genet- The standard toolkit to support causal inferences from obser- ics at the University of Minnesota, described ways that co-twin vational data centers around adequate control for factors that studies can contribute to understanding of aging outcomes, may have influenced both the antecedent and the outcome, using the example of underage alcohol use. Given pairs of but recent innovations have improved methods to incorpo- twins who have identical genetic attributes and have grown rate such control with complex time-varying exposures and up in the same family environment, with the same exposure to confounding structures. For example, a marginal structural risk, he explained, one might expect the same outcome. If, for model, she explained, can adjust for confounding variables example, one such twin tried alcohol at an early age and the that may both mediate the effects of past exposure and con- other did not, and the early-using twin later became alcoholic found the effects of future exposure. They are thus important while the other did not, this outcome would support a hypoth- for the study of exposures that vary across time—behaviors esis about the effects of early experimentation with alcohol. or conditions that may change in the course of the life-cycle, This method has helped to answer questions about the genetic such as smoking or socioeconomic status. Other tools that influence on life-style factors that appear to influence healthy hold promise to address various challenges in causal inference aging, including smoking, drinking, diet, physical activity, include meta-analysis, the use of latent variable models, and intellectual activity, and social behaviors. McGue noted that substudies embedded in larger ones to test the quality of mea- these studies are very useful in correcting for unmeasured surement strategies. confounding variables, but they also have drawbacks. Obser- New computational tools have brought new possibilities for vational findings from twin studies can be difficult to general- integrating different sorts of data, Glymour noted. There are ize, and twin studies can have limited statistical power, which many reasons to question the use of observational designs as may compound measurement errors. At the biological level, the basis for causal inference, but they do “much better than monozygotic twins are genetically identical, but they may not we had any reason to expect” at identifying determinants of necessarily be identical in other factors that could confound a health, at least in the few areas for which we have formal com- study of the antecedents to healthy aging. Thus, this method parisons of observational and trial-based evidence, she said. may produce biased estimates if the twins studied are not perfectly matched on confounding variables. Yet twin studies The magnitude and consistency of health disparities across can also strengthen inferences based on observation: to gain social groups is compelling, even though research has not yet the most benefit, it is important to use large study samples, fully explained the functioning of the mediators, and available report the correlation in exposure, pay close attention to the interventions seem to have limited effects. What is needed in potential for measurement error, and identify any factors that designing interventions, she said, are more precise definitions may contribute to a difference in exposure for pairs of twins. of exposures and development of clear guidelines for when and with whom interventions should be used, that is, closer Dalton Conley, professor of sociology at New York Univer- ties between the observational evidence and the intervention sity, discussed his research on whether genetic differences as design. Integration of evidence from observational and inter- measured by polygenic risk scores could account for varia- vention research will be very helpful, she concluded. tion in later life outcomes after exposure to the same experi- ence. Using sociodemographic and genotype data for several cohorts of men who were exposed to the draft lottery (and possible con- scription into military service during Parental the Vietnam War), he examined possi- Education ble causal links between conscription and long-term outcomes on earnings, IV: Policy Late Adult labor market participation, and health Education Health and mortality. He examined such Change factors as years of schooling attained, average number of hours worked per Health week, number of cigarettes smoked Endowment per day, and blood pressure and linked them to genetic data. The use of this natural experiment—the random Figure 2 Example of policy change as an instrumental variable. From Glymour, M.M. (2015). Strategies assignment of draft numbers to men for Clarifying Pathways, Enhancing Causal Analyses and Delineating Mechanisms: Exploiting Longitudinal Data Sources. Presentation at the June 11-12, 2015, Workshop on Understanding Pathways to Successful of particular ages—allowed explora- Aging: How Social and Behavioral Factors Affect Health at Older Ages, National Academies of Sciences, tion of whether the outcomes differed Engineering, and Medicine, Washington, DC. Reprinted with permission. by genotype. Conley said that this 5 

research provided evidence that compulsory Viet- nam-era military service affected men’s attach- ment to the labor force and smoking behavior at older ages in ways that varied by genotype, an association known as a gene-by-environment interaction. Experimental Approaches Eric Loucks, assistant professor of epidemiol- ogy at Brown University, noted that while many research questions about whether given factors cause disease remain challenging, the research focus has shifted to the identification of effective interventions. He discussed an iterative experi- mental approach for developing interventions, illustrated with the stage model for intervention from the National Institute of Health, presented in Figure 3. The process begins with a mecha- nism identified through basic research as being associated with a particular outcome (Stage 0). An intervention is designed (Stage I) based on a hypothesis—such as that social integration will provide supports that promote healthy behav- iors, which in turn may have measurable bio- Figure 3: NIH stage model of intervention. From Onken, L.S., Carroll, K.M., Shoham, V., logical effects in older people, such as reduced Cuthbert, B.N., and Riddle, M. (2014). Reenvisioning clinical science: Unifying the disci- blood pressure, cholesterol, and obesity, which pline to improve the public health. Clinical Psychological Science, 2(1), 22–34. Published then may reduce the likelihood of coronary heart by the American Psychological Association. Reprinted with permission. disease. across contexts, he suggested. Research approaches that con- Efficacy research (Stages II and III) is then conducted through found the two are not particularly useful because “knowing research clinics to ascertain which aspects of the intervention something about one of these tells you absolutely nothing are efficacious. The intervention design may be modified at about the other.” this stage and then tested further in community clinics to collect larger-scale data about effectiveness and the optimal For example, he noted, one might ask whether people who means of applying the intervention in practice. The next steps report more frequent marital conflict are also those people would be randomized studies of effectiveness (Stage IV) and who have higher stress levels—a “between” question. One then a staged implementation and dissemination process might also ask whether an individual who experiences marital (Stage V) that is designed to ensure that the intervention can conflict also experiences increased stress and, if so, for how be effectively applied on a large scale. As the arrows in the long—and which came first. Research on these two questions figure show, the process may proceed in a nonlinear fashion. may not support the same conclusions, and may lead to dif- For example, findings from efficacy research conducted in ferent approaches to interventions. community clinics might spur refinement of the intervention Smyth suggested that intensive longitudinal data, which can (a move from Stage III to Stage I). The dotted arrows denote be collected in many different ways, can provide important pathways that warrant careful consideration of alternatives information about ways that individuals’ responses may alter before proceeding. over time and contexts. Data that can be collected for this Joshua Smyth, professor of behavioral health and medicine purpose include self-reports; biomarkers; GPS and geocod- at Pennsylvania State University, focused on the “thin-slice” ing data; physical activity and sleep logs; and digital data approach to understanding the role of time, process, and including that from sensors, cameras, texts, e-mails, and the context in mechanisms that may influence aging, and using like. Data that shed light on within-person variability and that knowledge to design, tailor, and adapt interventions. person-specific triggers can be used to design personalized This approach, he noted, can “help unpack hints” gleaned (or precision) interventions that can be delivered at the right from large-scale studies. Understanding variability among time and in the right place, and have the intervention content different individuals in their responses to a construct such as or components most needed at that moment for that indi- stress is completely different from understanding variation in vidual. the way particular individuals respond to stress over time and 6 

RESEARCH DESIGNS idea was to add a measure of marital quality or satisfaction to Workshop participants were asked to meet in separate groups, a large-scale national study, to capture both spouses’ expe- each focused on one of the three antecedents discussed, opti- riences. The challenge of finding an accurate measure that mism, marital satisfaction, and educational attainment. Their could easily be included is difficult, Kiecolt-Glaser noted, but task was to build on the presentations by suggesting new this kind of data would be invaluable. Finally, the group sug- research designs. gested, twin studies could be useful in untangling the bidi- rectional relationship between marital satisfaction and health Maria Glymour summarized the discussion of the group that because they provide a way to avoid many confounding vari- had focused on educational attainment. They identified two ables. primary design challenges: linking early-life interventions to late-life outcomes and trusting causal inferences based on Joshua Smyth summarized the group that focused on opti- observational data. It will never be possible to have experi- mism. They suggested that a good next step would be to mental data on all the factors that may be important from a use expert consensus panels to identify the “meta-construct policy perspective, she noted. Drawing on features from some of positive psychological resources,” to set the stage for sys- of the most useful past studies, such as PROJECT TALENT and tematic investigation of existing datasets. This process could ProjectSTAR,1 this discussion group highlighted promising identify places to enhance existing measures and identify new ideas, including: ones for use in longitudinal studies of health outcomes that are already under way. For example, improved understanding • focusing interventions on middle-aged individuals who of the meta-construct and improved measures would allow are not yet ill but for whom health outcomes are begin- for more detailed study of biological processes and genetic ning to be evident; information. • using a factorial design to examine multiple mecha- On the experimental front, Smyth explained, the group nisms, such as the mental health effects of mindfulness focused on supplementing traditional designs with thin-slice programs or the effects of literacy on cognitive skills and approaches that could improve internal validity and better engagement; support understanding of mechanistic and causal pathways. • embedding new analyses in existing studies, such as They suggested using hybrid measurement designs that can the New England Family Study, so that existing infor- establish links across the life span with good reliability. For mation on sample populations that have been studied example, researchers might use detailed data episodically as children can be used to study their responses to later collected over a 3- to 6-month period (such as ecological interventions; and momentary assessments [EMAs] or daily diaries) to identify proximal mechanisms and then use follow-up at longer inter- • using quasi-experimental methods, such as exploiting vals to test changes over time and longer-term outcomes natural policy changes, to bridge the gap between ob- (e.g., using measurement burst designs). servational and experimental evidence. Janice Kiecolt-Glaser reported for the workshop discussion SYNTHESIS group that focused on marital satisfaction. They suggested Robert Levenson, professor of psychology at the University using data that provides repeated measures of a population of California at Berkeley, synthesized the presentations and to explore how marital status may affect long-term health discussions, identifying recurring points and issues for further outcomes. By examining such exogenous events as job loss, research. incarceration, and military deployment, it might be possible to trace the effects of such stressors on family dynamics and Antecedent-consequent studies, which identify correlations health. but do not identify mediators, have been the norm, he noted, but funders are increasingly focused on the mechanisms by The group also proposed collecting real-world observational which the antecedents produce the outcomes. The goal is to data to explore a thin-slice hypothesis. For example, research- find the “active ingredients” that are responsible for behavior ers could collect observational data on patients admitted to change, he explained. the hospital for stroke or congestive heart failure to explore the nature of the marital relationship (such as the degree of Many newer studies involve the use of interventions to test support provided to the patient) and then follow the couple hypotheses about mechanisms and how they might be for one year to collect data on such outcomes as which altered. Increasingly sophisticated designs are also allowing patients were readmitted to the hospital, changes in the researchers to trace the moderators that influence the medi- marital relationship, and the health of the spouses. Another ators—that is, to identify the relative effects of such factors as race and gender on the mechanisms to better understand dif- 1 ferences in people’s responses. For information about PROJECT TALENT, see http://www.projecttal- ent.org/ [June 2015]. Project STAR was a study of the effect of class size on The challenge of moving “both upstream and down” in student performance; see http://www.princeton.edu/futureofchildren/ publications/docs/05_02_08.pdf [June 2015]. studying causal relationships remains difficult, Levenson 7 

said. The relationships among the processes and character- another promising approach, provide a close-up look at istics that are associated with aging outcomes are “clearly behaviors in context, but the observed behaviors may not be bidirectional”: for every statement—such as “optimism pro- representative of behaviors in other contexts or at different motes good health”—the converse is also surely true. Another times. In addition, thin-slice studies may not capture critical challenge is identifying what constructs should be measured, time periods when intervention could be most effective. New and how. There are many possible measures of psychologi- statistical methods are needed to address these challenges, cal well-being, for example, and researchers should to distin- Levenson noted. guish among inherent traits and states that change over time Levenson closed with a cautionary note. Researchers could and can be altered. They also have to consider whether such learn from the experience of psychotherapy researchers, who tools as self-reports provide adequate measures and how often “throw everything” at a serious problem because the accurately moderators such as race and socioeconomic status need is urgent. This “kitchen sink” approach may be effective can be measured. These characteristics may seem simple in producing behavior change, but it provides little evidence on the surface, but are deceptively complex: for example, of what active ingredients actually helped the patient and when considering race, it is not only often difficult to estab- how such a complex treatment might compare to a simpler lish who should be included in a particular group, but also treatment. Psychotherapy research also points to the impor- what aspects of group membership are salient to the research tance of testing the effectiveness of interventions in the real question (e.g., self-identified ethnicity, exposure to cultural world. He noted that interventions that work well in univer- traditions, or familial ethnicity or race). sity settings with carefully selected populations and carefully Longitudinal studies hold promise for establishing how trained doctors or other clinicians often do not work as well antecedents influence outcomes, but they, too, have chal- outside those controlled settings. Interventions that are long lenges, Levenson noted. For example, when studying a late- lasting “are difficult” he concluded, and “every intervention life disorder such as dementia, it is difficult to decide how has a shelf life.” Promoting health—addressing the range of early in the life-cycle to begin data collection, as well as how factors that may be harmful—throughout the life-cycle may often to collect data and for how long. Thin-slice studies, require different interventions at different points. WORKSHOP PLANNING COMMITTEE SUSAN T. FISKE (Chair), Princeton University; JASON BOARDMAN, University of Colorado Boulder; MARIA GLYMOUR, Uni- versity of California, San Francisco; ARUN KARLAMANGA, University of California, Los Angeles; JANICE KIECOLT-GLASER, The Ohio State University College of Medicine; TINA WINTERS, Study Director; BARBARA WANCHISEN, Director, Board on Behavioral, Coginitive, and Sensory Sciences. Disclaimer: This Workshop in Brief has been prepared by Alexandra Beatty as a factual summary of what occurred at the meeting. The statements made are those of the author or individual meeting participants and do not necessarily represent the views of all meeting participants, the planning committee, the Board on Behavioral, Cognitive, and Sensory Sciences, or the National Academies of Science, Engineeering, and Medicine. The planning committee was responsible only for organizing the workshop, identifying topics, and choosing speakers. Reviewers: The summary was reviewed in draft form by Jason D. Boardman, Institute of Behavioral Science and Department of Sociology, University of Colorado; James S. Jackson, Research Center for Group Dynamics, Institute for Social Research, University of Michigan; Laura D. Kuzansky, Department of Social and Behavioral Sciences, Harvard School of Public Health; and Joshua M. Smyth, Biobehavioral Health and Medicine and Social Science Research Institute, Pennsylvania State University to ensure that it meets institutional standards for quality and objectivity. The review comments and draft manuscript remain confidential to protect the integrity of the process. Sponsor: The workshop was sponsored by the Division of Behavioral and Social Research of the National Institute on Aging, U.S. Department of Health and Human Services. Copyright 2015 by the National Academy of Sciences. All rights reserved. 8