Impact of Health Conditions on Cognitive Change in Later Life: A Cross-Study Comparative Analysis
by
Catharine Sparks
B.S., Southern Oregon University, 2008
A Thesis Submitted in Partial Fulfillment of the Requirements for the Degree of
MASTER OF SCIENCE in the Department of Psychology
Catharine Sparks, 2011 University of Victoria
All rights reserved. This thesis may not be reproduced in whole or in part, by photocopy or other means, without the permission of the author.
Supervisory Committee
Impact of Health Conditions on Cognitive Change in Later Life: A Cross-Study Comparative Analysis
by
Catharine Sparks
B.S., Southern Oregon University, 2008
Supervisory Committee
Dr. Scott M. Hofer, (Department of Psychology)
________________________________________________________________________
Supervisor
Dr. Andrea M. Piccinin, (Department of Psychology)
________________________________________________________________________
Departmental Member
Dr. Stuart W. S. MacDonald, (Department of Psychology)
________________________________________________________________________
Abstract
Supervisory Committee
Dr. Scott M. Hofer, (Department of Psychology)
Supervisor
Dr. Andrea M. Piccinin, (Department of Psychology)
Departmental Member
Dr. Stuart W. S. MacDonald, (Department of Psychology)
Departmental Member
Relatively few studies have considered how changes in health are associated with changes in cognition in aging populations. Even fewer have investigated the similarities and differences of the health-cognition link evidenced across independent longitudinal studies of aging that differ in country and birth cohort. The main objective of the current research is to evaluate aging-related cognitive change in the context of physical health conditions and to compare patterns and synthesize results across several longitudinal studies of aging. This cross-study evaluation is based on data from three longitudinal studies of aging: 1) the OCTO-Twin Study, a longitudinal investigation of same-sex twin pairs drawn from the population-based Swedish Twin Registry (N = 702; 67% female; mean age is 83.5), 2) the Health and Retirement Study (HRS), a study of middle-aged and older adults in the U.S. (N = 21,364; 57% female; mean age is 65.8), and 3) the English Longitudinal Study of Aging (ELSA), a study of middle-aged and older adults in the U.K. (N = 11,397; 54% female; mean age is 65.3). Data were analyzed using latent growth curve (LGC) analysis to evaluate 1) the impact of diagnosed health conditions and 2) the additive impact of comorbidity on level and rate of change in distinct cognitive outcomes. Our findings indicate that particular health conditions significantly impact initial status and rate of change in cognition, but do so differently across longitudinal studies of aging. The argument is made that the inclusion of health in our predictive models is essential as we try to parse out the effects of pathological aging vs. normative age-related change in cognition. The results of this study show the importance of replication in longitudinal research and for contrasting patterns of effects across
independent studies in order to build a cumulative basis for further understanding of the dynamics among aging, health, and cognition in populations that differ in cohort, culture, and country.
Table of Contents Supervisory Committee ... ii Abstract ... iii Table of Contents ... iv List of Tables ... v Acknowledgments... vi Introduction ... 1 Method ... 10 Participants ... 10 Procedure ... 11 Measures ... 11 Statistical Analysis ... 17 Results ... 19 Hypertension ... 19 Cardiovascular Disease (CVD) ... 21 Diabetes... 23 Stroke ... 23 Cancer ... 25 Comorbidity Index ... 27 Discussion ... 30 References ... 35
List of Tables
Table 1: Sample size and participant characteristics for cognitive outcomes ... 12 Table 2a: Frequency distribution and percentages of condition counts (full samples) ... 15 Table 2b: Frequency distribution and percentages of condition counts (restricted samples) 15 Table 3: Frequency of bivariate comorbidity across health conditions ... 16 Table 4: Effect of ever having been diagnosed with hypertension on level and rate of change
in memory outcomes ... 20 Table 5: Effect of ever having been diagnosed with cardiovascular disease on level and rate
of change in memory outcomes... 22 Table 6: Effect of ever having been diagnosed with diabetes on level and rate of change in
memory outcomes ... 24 Table 7: Effect of ever having been diagnosed with a stoke on level and rate of change in
memory outcomes ... 26 Table 8: Effect of ever having been diagnosed with cancer on level and rate of change in
memory outcomes ... 26 Table 9: Effect of having more than one health condtion (comorbidity index) on level and
Acknowledgments
I would like to express my deepest gratitude to my supervisor, Dr. Scott Hofer, for his guidance, encouragement, and unconditional support. His dedication to excellence in research and his eternally optimistic approach to life helped turn an ordinary process into an extraordinary experience. I also thank my committee members, Drs. Andrea Piccinin and Stuart MacDonald, for their generous support and sagacious advice in both work and life. In addition, I thank Dr. Mikael Jansson for his enthusiastic and continued support. Finally, I send kind regards to my good friend, Dr. Daniel Bontempo, whose colourful blend of mentorship and friendship made it all just a little more interesting.
This work was facilitated by the Integrative Analysis of Longitudinal Studies of Aging (IALSA) research network which is funded by NIH/NIA (AG026453).
Introduction
The study of cognitive aging involves the description and explanation of aging-related changes in cognitive abilities. There is increasing evidence from longitudinal studies that pathological changes related to chronic and acute health conditions contribute to cognitive impairment and accelerated decline in functioning. However, many studies of cognitive aging, whether
longitudinal or cross-sectional in design, omit measures of health and terminal decline from the predictive model, leading to potentially inflated estimates of the effect of “age” on cognitive change. In this study, we identify the extent to which later life declines in cognitive function are driven by changes in health, considering both independent and interactive effects of particular health conditions. Indeed, if health conditions commonly associated with an elderly population (e.g., cardiovascular disease, cerebrovascular disease, cancer, diabetes, hypertension) are not considered as predictors of cognitive change, the potential for implementing a more preventive science of cognitive aging is diminished. Examination of the impact of these health conditions on cognitive functiong will help to elucidate mechanisms and related risk factors. In addition, an evaluation of the impact of changes in health (that may be amenable to interventions) on
cognitive and physical changes will help to differentiate changes in health from other underlying mechanisms of change that might be considered “age normative”.
Among the most prominent nonnormative factors associated with cognitive functioning is that of health. Health conditions are by definition considered to be nonnormative because not all individuals in a population will experience similar health events, although the risk for particular health conditions and comorbidity increases with population aging. Instead of assuming that aging-related declines are strongly associated with chronological age, we can model change as a function of the diagnosis of chronic disease processes (e.g., Alzheimer’s disease [AD]; e.g.,
Sliwinski, Hofer, & Hall, 2003). For example, while the risk of stroke or developing AD increases with age, the relationship between these events and age is nonnormative in that the majority of individuals will not experience these events, and the occurrence may not be well predicted by chronological age (e.g., Baltes & Nesselroade, 1979). From this perspective, including health and physiological markers as explanatory variables will provide an explanatory basis above that which is typically modeled by chronological age alone (e.g., MacDonald, DeCarlo, & Dixon, 2011).
An inherent complexity in studying the association between aging and health-related change has contributed to the slow development of integrated theories, particularly in the area of cognitive aging (Alwin & Hofer, 2011; Birren & Birren, 1990; Hofer & Piccinin, 2010). These simultaneously unfolding processes have long motivated researchers to conceptualize the distinction between maturational processes (i.e., primary aging) and the effects of disease (i.e., secondary aging) and mortality (tertiary aging; e.g., Busse & Pfeiffer, 1969). Certainly, our understanding of normative population aging processes are at least partially confounded because the increasing risk of various diseases in later life is associated with increasing chronological age. Therefore, the degree to which population aging is due primarily to normative (i.e., universal) processes or our understanding of the degree to which population change is due to abnormal or pathological processes remains largely unresolved (Blumenthal, 2003; Newman & Ferrucci, 2009; Rabbitt, Lunn, Pendleton, & Yardefagar, 2011). The challenge of distinguishing normative and nonnormative processes is that potentially confounding or explanatory variables related to both age and cognitive performance, such as health and health-related change, are often not consistently included in the predictive model.
Birren and Schroots (1984) offered an early conceptualization of aging and development as two parallel, but related processes. For example, during early childhood, the developmental process is quite evident while the aging process is less apparent, and vice versa in late life. Baltes (1987) developed a similar theory suggesting that development throughout the lifespan is a joint expression of gains (i.e., growth) and losses (i.e., decline). However, neither theory really addressed the obvious problem that there is no essential difference between development/gain and aging/loss, except that they are opposites; aging and development can both be considered age-related changes that occur throughout the lifespan. Researchers continue to imply that aging and age-related changes are relatively uniform processes of change, but this view does not address the variety of patterns seen among aging rates in time of onset, range or magnitude of impact, reversibility, and outcomes (Schroots, 1995).
A more integrative model of aging is the cascade model, which presents three interactive patterns of aging (Birren & Cunningham, 1985). As mentioned earlier, two of these patterns are primary aging, identified as the slowing of the speed at which information is processed, and secondary aging, identified as chronic or acute disease, resulting in loss of functioning. The third pattern, tertiary aging, also known as terminal decline or the terminal drop hypothesis (Bäckman & MacDonald, 2006; Kleemeier, 1962; Riegel & Riegel, 1972), suggests there is a period prior to death in which the rate of functional decline (most specifically related to cognitive
performance) increases exponentially. Terminal decline has been a topic of aging research for many years and is well documented in the aging literature (e.g., Johansson, Hofer, Allaire, Maldonado-Molina, Piccinin, Berg, et al., 2004; Laukka, MacDonald, & Bäckman, 2006, 2008; Piccinin, Muniz, Matthews, & Johansson, 2011; Thorvaldsson, Hofer, & Johansson, 2006).
The aging-health debate has generally taken three points of view: 1) aging as a disease process (e.g., Goodwin, 1991; Forbes & Hirdes, 1993), 2) aging as separate from disease or physiological deterioration (aging as intrinsic, progressive and universal vs. disease as extrinsic; Blumenthal, 1993), and 3) aging and disease exist on a continuum (at one end, aging is clearly separated from disease while the two blend at the other end – in the middle, fluid, continuous, and complex interactions between pathology and physiology exist; Newman & Ferrucci, 2009; Rowe, 1985). Within cognitive aging literature, the debate as to the relationship between normative (primary) aging and pathology (secondary aging) continues, with few researchers explicitly addressing the often parallel processes of intrinsic maturation and extrinsic effects of environment and disease (for exceptions, see Anstey, Stankov, & Lord, 1993; Spiro & Brady, 2008; 2011).
The Vascular hypothesis, a more recent theory integrating health and cognition in the
psychology of aging literature, is reflective of the continuum point of view and has been gaining momentum in recent years (e.g., Casserly & Topol, 2004; de la Torre, 2002, 2004; Kuller, 2006; O’Brien, 2003; Raz & Rodrigue, 2006; Spiro & Brady, 2008; 2011). Findings from these studies support the idea that vascular diseases (e.g., stroke, myocardial infarctions, atherosclerosis) affect the brain as well as the heart, thus affecting cognition. In a recent article, Spiro & Brady (2011) argue that various aspects of preventable and/or treatable diseases (e.g., incident, severity, treatment) are likely to account for a considerable portion of aging-related variation in cognitive performance within and between individuals. The critical take home message from this argument is that by preventing or treating the disease, as well as associated risk factors, we not only slow the progression of the disease, but we could potentially reduce the magnitude and rate of subsequent declines in cognition.
Studies of the impact of specific health conditions on cognitive outcomes can be found in the aging literature. Consistently, we are seeing associations of potentially preventable and/or treatable health conditions (e.g., hypertension, diabetes, cardiovascular disease) on cognitive outcomes (e.g., memory, processing speed, and reasoning). These associations are often demonstrated after controlling for the effects of age alone. We briefly review evidence for particular health conditions that have been demonstrated to have an impact on cognition in later life.
Hypertension. Knopman, Boland, Mosley, Howard, Liao, Szklo, et al. (2001) found
associations of hypertension on memory, speed, and fluency outcomes in middle and older age groups. Brady, Spiro, & Gaziano (2005) examined the influence of age and hypertensive status on cognition in nondemented older men (mean age = 67) whose hypertensive status was stable over a 3-year period, and who had no other medical comorbidities. Age was negatively
associated with performance on all but one cognitive test. Age interacted with hypertensive status on verbal fluency and word list immediate recall: older persons with uncontrolled hypertension exhibited significantly larger age differences on these tests, compared with individuals in the normotensive range. These findings suggest that uncontrolled hypertension produces specific cognitive deficits beyond those attributable to age alone.
Diabetes. Findings from population-based studies have consistently reported weak negative correlations between diabetes and cognitive status/change among the elderly (Arvanitakis, Bennett, Wilson, & Barnes, 2010; Arvanitakis, Wilson, Bienias, Evans, & Bennett, 2004; Christman, Vannorsdall, Pearlson, Mecocci, Williams, & Senin, 2010; Hassing, Johansson, Pedersen, Nilsson, Berg, & McClearn, 2003). Only a few studies have provided data on multiple cognitive assessments examining longitudinal changes related to type 2 diabetes. Gregg, Yaffe,
Cauley, Rolka, Blackwell, Narayan, & Cummings (2000) found that diabetes was associated with lower levels of cognitive function at baseline and greater decline in two cognitive tests. In a sample of nearly 1000 persons aged 59-71, Fontbonne, Berr, Ducimetiere, & Alperovitch, (2001) found no diabetes-related differences in cognitive functioning at baseline (sample was dementia free at baseline) but after a four-year follow-up, diabetics were more likely than non-diabetics to have experienced a pronounced cognitive decline in four of nine cognitive tests. Hassing, Hofer, Nilsson, Berg, Pedersen, McClearn, & Johansson (2004) reported significant effects of type 2 diabetes mellitus across a 6-year interval on a variety of cognitive outcomes in a population-based sample (OCTO-Twin) of 274 elderly participants. In general, these findings point to the conclusion that type 2 diabetes is associated with accelerated cognitive decline in old age.
Cardiovascular Disease (CVD). Evidence concluding that vascular risk factors are predictive of cognitive decline and impairment in late life is numerous and includes genetic factors as modifiers (e.g., APOE; Duran & Hanon, 2008), disease (e.g., hypertension, diabetes; Kloppenborg, van den Berg, & Biessels, 2008), and behavioural factors (e.g., obesity, diet; Hughes & Ganguli, 2009). More specifically, Verhaeghen, Borchelt, & Smith (2003)
investigated the relationship between CVD and cognition in a population age 70 and older (n = 206) and found CVD to be negatively associated with cognitive outcomes, with the greatest impact seen in measures of perceptual speed and memory. Importantly, findings from this study suggest that CVD may differentially impact distinct cognitive outcomes.
Cancer. The majority of studies linking cancer with cognitive performance are considering either 1) the effect of treatments (e.g., chemotherapy, radiation) or 2) the link between cognitive performance and mortality risk in cancer patients (for review, see Anstey, Mack, & von Sanden, 2006). A recent study evaluated the effect of chemotherapy treatment in breast cancer patients (n
= 60; mean age of 52) on processing speed and verbal ability domains at 3 follow-up occasions post-treatment (Ahles, Saykin, McDonald, Li, Furstenberg, Hanscom, et al., 2010). They found a significant negative effect of treatment at the first occasion, with older patients showing lower performance but improved with additional follow-up, indicating that the effect of chemotherapy on cognitive performance may be temporary. Further research supports genetic influences as potential modifiers of the impact of cancer treatments on cognitive outcomes (Small, Rawson, Walsh, Jim, Hughes, Iser, et al., 2011).
Stroke. Brady, Spiro, McGlinchey-Berroth, Milberg, & Gaziano (2001) examined the
relationship between overall stroke risk and cognitive decline in 235 healthy older men. Whereas increasing age was associated with decline in all cognitive functions examined, increasing stroke risk was associated only with decline in executive function measures. Furthermore, the relation between stroke risk and executive decline was nearly as large (80%) as the relation between age and executive decline. These results suggested that overall stroke risk in relatively healthy older men exerted specific effects on decline in executive function but not on memory or visuospatial functions, and that the magnitude of these effects rivals those of age effects (which, as described above, may represent to a large extent, other health-related declines; e.g., Kloppenborg, et al., 2008; Gregg, et al., 2000).
Comorbidity Indices. Many studies have investigated the association between cognitive decline and independent health conditions (e.g., Cherubini, Lowenthal, Paran, Mecocci,
Williams, & Senin, 2010; Hassing et al., 2003; Rafnsson, Deary, Smith, Whiteman, & Fowkes, 2007), often excluding those participants with comorbid conditions. However, research looking at comorbid conditions has indicated that having two distinct health conditions can significantly increase the effect of independent conditions on cognitive outcomes (e.g., Hassing et al., 2004).
The effects of comorbidity are often nonadditive (i.e., interactive) and there are multiple ways that comorbidity can be operationalized (Borson, Scanlan, Lessig, & DeMers, 2010; de Groot, Beckerman, Lankhorst, & Bouter, 2003). Hassing et al. (2004) found that individuals with comorbidity of diabetes and hypertension exhibit the greatest cognitive decline compared to individuals with diabetes or hypertension alone. As with diabetes, CVD and cerebrovascular events are all prominent risk factors for cognitive decline and dementia, and analyzing their possible effects involves some complexity. With age, comorbidity becomes the rule rather than the exception. For example, in the US, adults aged 60 and over were found to average 2.2 chronic diseases (Wolff, Starfield, & Anderson, 2002).
As is evidenced here, chronic and acute health conditions have been found to account for individual differences in cognitive impairment and decline, but these findings have not been consistent across cognitive outcomes or across studies of aging. The replication of research findings across longitudinal studies is essential to our understanding of aging-related processes and provides a foundation on which to build our knowledge of population and individual aging (e.g., Hofer & Piccinin, 2009; 2010). From this perspective, it is essential that findings obtained from models based on different population samples, different measurement intervals, and different variables are compared in terms of general patterns of effects (i.e., direction and
magnitude), providing evidence for cross-study, cross-national validation of research findings. A better understanding of the similarities and differences of results across independent studies will enable us to make the most of available data and possibly move us beyond current limitations and toward a more progressive understanding of normative cognitive change in later life.
In the present study, the association between specific health conditions and distinct cognitive outcomes is investigated. The aim is to evaluate the impact of each health condition, as well as
the comorbidity (i.e., count) of conditions, on initial level and rate of change in memory performance. We use distinct measures of memory to further evaluate consistency across measures of the same domain. In addition, we compare patterns of results across three cross-national longitudinal studies of aging.
Method
This study involves the analysis of three longitudinal studies of older adults in the U.S., England, and Sweden.
Participants
HRS. The Health and Retirement Study (HRS) is a nationally representative sample of middle-aged and older adults in the United States (Juster & Suzman, 1995). Initiated in 1992, the HRS continues to survey more than 22,000 Americans every two years. In this research, we use five waves of data collected between 1998 and 2006. The full sample is comprised of 21,364 adults (57% female) aged 50 – 99 (M = 65.8, SD = 11.1). An exclusionary criterion for this analysis was having scored less than 6 on the 9-point Telephone Interview for Cognitive Status (TICS) scale at any time during the years represented here (i.e., 1998-2006).
ELSA. The English Longitudinal Study of Ageing (ELSA) study was modeled after the HRS; therefore, the aims, design, and standard measures are similar for both studies. Individuals were recruited from three years of the Health Survey of England (1998-2001) to represent the middle-aged and older population of England. ELSA continues to survey these individuals every two years. More complete details of this study have been published elsewhere (e.g., Banks, Lessof, Nazroo, Rogers, Stafford, & Steptoe, 2010; Banks, Breeze, Lessof, & Nazroo, 2008). We use the first four waves of ELSA data (2002-2006) in this research. The full sample is comprised of 11,391 adults (54% female) aged 50 to 99 (M = 65.3, SD = 10.4). An exclusionary criterion for this analysis was having ever been coded as cognitively impaired during the years represented here (i.e., 2002-2006)
OCTO-Twin. The Origins of Variance in the Old-Old: Octogenarian Twins (OCTO-Twin) Study (McClearn, Johansson, Berg, Pedersen, Ahern, Petrill, et al., 1997) is a longitudinal
investigation of same-sex twin pairs drawn from the oldest cohort of the population-based Swedish Twin Registry (Cederlof & Lorich, 1978)
For all studies, participants were excluded if they had missing data on the outcome variable of interest at time 1. This criterion resulted in a slightly different sample size for each of the
cognitive outcomes within each study. Participant characteristics for each study and each outcome are summarized in Table 1.
that were 80 years old or older and both alive during the 2-year testing period that began in 1991. Understanding the relative contributions of genetic and contextual influences to aging-related variability in measures of cognition, physical health, relationships, personality, and mental health was an aim of the study. The sample at the first measurement occasion consisted of 351 intact twin pairs. Five waves of OCTO-Twin data are used in this research. The full sample is comprised of 702 older adults (67% female) aged 79.37 to 97.92 (M = 83.5, SD = 3.2). Exclusionary criteria for the current analyses included having received a physician-diagnosis of dementia at anytime during the study.
Procedure
For HRS and ELSA, participants were assessed via structured telephone interviews conducted by trained research staff. For OCTO-Twin, participants were assessed in their homes by
registered nurses. For all three studies, the participants were assessed at 2-year intervals. The OCTO-Twin data collection is now complete, while the HRS and ELSA collections are ongoing. Measures
The analyses for the current research focus on measures of memory functioning and self-reported diagnoses of major health conditions (hypertension, diabetes, CVD, stroke, and cancer). In addition to the variables of interest, baseline measures of age, sex, and education are included as covariates.
Table 1
Sample size and participant characteristics for cognitive outcomes.
Cognitive Outcome N Mean ± SD (range) Age at Baseline % Female health condition N with ≥ 1 OCTO-Twin
Digit Span Forward 444 83.3 ± 3.0 (79 - 98) 66% 301
Digit Span Backward 441 83.3 ± 3.0 (79 - 98) 66% 298
Prose Recall 407 83.1 ± 2.9 (79 - 98) 65% 277
Memory-in-Reality Recall 409 83.2 ± 2.8 (79 - 93) 66% 274 HRS
Immediate Word Recall 17166 65.5 ± 9.7 (50 - 98) 58% 13721 Delayed Word Recall 17166 65.5 ± 9.7 (50 - 98) 58% 13721
Subtract 7 16703 65.5 ± 9.7 (50 - 98) 58% 13345
ELSA
Immediate Word Recall 8170 64.0 ± 9.7 (50 - 99) 54% 4403
Delayed Word Recall 8165 64.0 ± 9.7 (50 - 99) 54% 4398
Cognition
HRS. Three tests representing both short-term and long-term memory were used. Immediate and Delayed Word Recall tests ask that the participant recall as many words as possible, in any order, immediately after being read a list of 20 words and again after a short time has passed. For the Subtract 7’s test, participants are asked the question: One hundred minus 7 equals what? The question is followed by four sequential questions asking, “And 7 from that?” until the respondent has subtracted 7 from the current number five times.
ELSA. Similar to HRS, we used the Immediate and Delayed Word Recall tests as well as the Prospective Memory Test. Prospective memory concerns memory for future actions. Early in the cognitive assessment, participants are informed of a task they will need to complete later (i.e., signing initials on a paper attached to a clipboard) and asked to remember as they will not be reminded again. A correct response requires that the participant performs this task at the appropriate time without being prompted (Huppert, Gardener, & McWilliams, 2006).
OCTO-Twin. Four tests representing both short-term and long-term memory involving processing of verbal and nonverbal material were used. Digit Span Forward and Digit Span Backward tests measure short-term memory for orally presented digits (Wechsler, 1991); subjects are asked to recall the digits in the order they are presented (i.e., Digit Span Forward) with a maximum possible score of 9 and to recall the digits in the reverse order (i.e., Digit Span Backward) with a maximum possible score of 8 (Johansson, Whitfield, Pedersen, Hofer, Ahern, & McClearn, 1999). Prose Recall is a Swedish language verbal memory test similar to the prose passages in the Wechsler Memory Test (Wechsler, 1991): subjects are asked to recall a brief humorous story (100 words) immediately after its presentation, with a maximum possible score of 16. The Memory-in-Reality Free Recall test involves a three-dimensional apartment model in
which subjects are initially asked to place 10 common household objects and then asked to recall the same 10 items 30 minutes later (Johansson et al, 1999).
Health Predictors
For all analyses, we included those individuals that were ever diagnosed (i.e., prior to and throughout the course of the study) with the specified health condition. The health conditions of interest for subsequent analyses are hypertension, diabetes, CVD (i.e., heart attack, coronary heart disease, angina, congestive heart failure, arteriosclerosis), stroke, and cancer. For HRS and ELSA, this is represented by self-reports of disease conditions garnered by a question: “Did a doctor ever tell you that you had…?” Health conditions in the OCTO-Twin Study were obtained through three sources: medical records (information from medical records was available for 99.1% of participants), medication usage, and self-reports. Using information from these sources, diagnoses were classified according to the International Classification of Diseases (ICD-10) (Nilsson, Johansson, Berg, Karlsson, & McClearn, 2002).
Tables 2a (full samples), 2b (restricted samples) and 3 summarize the total number of diagnoses for each condition within each study, the frequency of condition comorbidity, and the total comorbidity index (i.e., count of multiple health conditions). Hypertension was the most common health condition across the three studies. In the OCTO-Twin restricted sample, 39% had one of the five health conditions and approximately 27% had two or more conditions. In the HRS restricted sample, 33% had one of the five health conditions and approximately 47% had two or more conditions. In the ELSA restricted sample, 37% had one of the five health
Table 2a
Frequency distribution and percentages of condition counts (full samples).
OCTO-Twin HRS ELSA
Count of Conditions Freq. % Freq. % Freq. % 0 276 39.32 10978 35.53 5087 44.66 1 254 36.18 8661 28.03 4329 38.00 2 131 18.66 6809 22.04 1606 14.10 3 38 5.41 3350 10.84 324 2.84 4 3 0.43 970 3.14 44 0.39 5 0 0.00 126 0.41 1 0.01 Table 2b
Frequency distribution and percentages of condition counts (restricted samples).
OCTO-Twin HRS ELSA
Count of Conditions Freq. % Freq. % Freq. % 0 160 33.54 3804 20.00 4021 46.88 1 187 39.20 6323 33.25 3232 37.68 2 101 21.17 5303 27.89 1093 12.74 3 28 5.87 2710 14.25 206 2.40 4 1 0.21 772 4.06 26 0.30 5 0 0 104 0.55 0 0
Table 3
Frequency of bivariate comorbidity across health conditions.
OCTO-Twin High BP CVD Diabetes Stroke Cancer
High BP 267 51 45 48 30 CVD xx 107 25 22 5 Diabetes xx xx 98 12 11 Stroke xx xx xx 104 14 Cancer xx xx xx xx 66 HRS High BP 12101 4783 3596 1656 2370 CVD xx 6453 2039 1147 1448 Diabetes xx xx 4420 728 894 Stroke xx xx xx 2041 444 Cancer xx xx xx xx 3652 ELSA High BP 3465 339 520 260 285 CVD xx 630 111 73 58 Diabetes xx xx 929 90 92 Stroke xx xx xx 428 39 Cancer xx xx xx xx 688
Covariates
For all models used in this study, baseline age was included and was centered appropriately (i.e., at age 80 for OCTO-Twin; at age 65 for HRS and ELSA), education was included and
dichotomized (i.e., less than highschool equivalent = low education; highschool equivalent or greater = high education). This created reference groups of males (age 80 for OCTO-Twin; age 65 for HRS and ELSA) with low education.
Statistical Analysis
Latent Growth Curve Analysis. Longitudinal data were analyzed using latent growth curve (LGC) analysis, for the estimation and prediction of population average and individual
differences in trajectories of change and initial status. Briefly, this type of analysis summarizes individual-level data in terms of an intercept (i.e., the “true” initial level of performance), linear and possibly curvilinear slopes (i.e., constant and acceleration in rate of change; other functional forms are possible), and residual (error) parameters. In this application, we use LGC analyses to account for systematic variation in growth parameters that are attributable to specific health conditions, other covariates (i.e., age, sex, education), and their interactions. The LGC analyses were performed using Mplus Version 6 (Muthen & Muthen, 2010). The results presented here are based on ML estimation assuming that participant dropout and incomplete data are missing at random (i.e., that missingess is related to prior values of the outcome and covariates included in the model).
Assessing Model Fit. A nested series of unconditional (baseline) models (linear only, linear + fixed quadratic, and linear + fixed and random quadratic) were used to evaluate the best fitting functional form necessary to capture the sample average change. While we did see significant parameter estimates for some outcome variables, these estimates were not substantively
significant. For example, the Immediate Word Recall measure (HRS) had a chi-square of 847.216 with 3 degrees of freedom (statistically significant), but the variance estimate for the fixed + random quadratic model was ≤ 0.001. It is possible that the sample size for HRS was large enough to produce the significant random quadratic, but it leaves us interpreting estimates that are not substantively meaningful. Another complexity to consider when choosing to use quadratic functions is the collinearity between parameter estimates that are competing for variance as significant estimates can be absorbed by the inclusion of the quadratic. For these reasons, and for ease of comparing results across study outcomes that were better fit with linear only models, we used the linear-slope baseline model for these analyses.
Cluster identifiers were used to address the dependency among observations associated with twin status in the OCTO-Twin study and same-household status in HRS and ELSA. Full-information maximum likelihood estimation was used for analysis with incomplete data under the assumption that the data are missing at random.
The current study assesses the effects of age, sex, education, and each of the five health conditions, as well as the interactions of age and sex, age and education, sex and education, and the interactions of age, sex, and education with each of the health conditions, on each cognitive outcome. An additional model was used to assess the impact of having multiple conditions (i.e., comorbidity index) and the interactions between the condition count and the covariates of age, sex, and education.
Results
The results of the growth curve analysis for each of the health conditions are reported in Tables 4 – 9. All results reported here are from the final model that included the health condition of interest, age, sex, education, and all interaction terms. We report the conditional effects from the full models because when the interactions are included (whether significant or not) the coefficient for the predictor is the effect of that variable when other variables involved in the interaction are zero. Only the statistically significant effects of the condition and the interaction of the condition with sex, age, and education on the set of memory outcomes in each study are summarized below. In all models, effects of the conditions were adjusted for sex, age, and education as well as the interactions of sex and age, sex and education, and age and education. Hypertension
The parameter estimates for the effect of ever being diagnosed with hypertension on level and rate of change in memory outcomes are presented in Table 4.
OCTO-Twin. Hypertension is associated with a significant decline in memory function (as
measured by the Memory in Reality Recall test) in 80 year old men with low education by -0.279 per year more than for such individuals without hypertension. Initial level of functioning at the first occasion was not related to hypertension status. A significant interaction effect of 0.028 for hypertension and age on rate of change was observed for the Digit Span Backward test,
indicating that older hypertensives in this study decline slightly less than younger individuals. HRS. The main effect of hypertension is associated with between-person differences in memory function (across all three HRS measures) on initial level for 65 year old men with low education. A significant effect is also observed for the interaction of hypertension and sex on initial level of the Subtract 7 test.
Table 4
Effect of ever having been diagnosed with hypertension on level and rate of change in memory outcomes.
Intercept Linear Slope
Memory by Study M C CxA CxS CxE M C CxA CxS CxE
OCTO-Twin
Digit Span Forward 5.543* 0.153 0.013 -0.286 0.022 -0.052 -0.022 0.006 0.003 -0.013
Digit Span Back 3.343* 0.134 -0.050 -0.094 0.035 -0.031 -0.046 0.028* -0.029 -0.016
Prose Recall 10.115* -0.028 0.194 -0.351 0.265 -0.002 -0.199 0.009 0.153 -0.040
Mem-in-Reality 6.976* 0.106 -0.036 0.070 0.053 0.060 -0.279* 0.029 0.207 -0.022
HRS
Delayed Word Recall -0.341* -0.085* 0.003* -0.018 0.046 -0.027* -0.004 0.000 -0.004 0.001
Immediate Word Recall -0.389* -0.074* 0.004* -0.031 0.060* -0.030* -0.008 0.000 0.002 -0.001
Subtract 7s -0.234* -0.099* 0.004* -0.079* 0.093* -0.001 -0.007 0.000 0.004 0.003
ELSA
Delayed Word Recall 3.637* -0.030 0.002 -0.008 -0.038 0.011 -0.011 -0.002 -0.013 0.000
Immediate Word Recall 5.175* -0.022 0.002 -0.012 0.001 -0.011 0.002 0.000 -0.014 -0.023
Prospective Memory 3.335* -0.023 0.001 -0.036 0.009 0.028* 0.028 0.001 -0.029 0.000
Note. C = condition; CxA = condition by age interaction; CxS = condition by sex interaction; CxE = condition by education interaction; M = mean value for individuals who have not been diagnosed with hypertension. Age (A), sex (S), education (E), AxS, AxE, SxE terms where included in the model but are not listed here.
In other words, men with diagnosed hypertension performed more poorly at time 1 than non-diagnosed individuals across all memory measures; women non-diagnosed with hypertension perform -0.079 less well than men at time 1 for the Subtract 7 test. There is also a significant positive interaction of hypertension and age although this effect is substantively small. The interaction of education and hypertension shows a positive effect such that hypertensives with higher education will perform less poorly at time 1 than non-hypertensives with low education. We do not observe any significant effects on rate of change.
ELSA. No significant effects of hypertension are seen for the U.K. sample on either initial level or rate of change in any of the memory measures.
Cardiovascular Disease (CVD)
The parameter estimates for the effect of ever being diagnosed with cardiovascular disease on level and rate of change in memory outcomes are presented in Table 5.
OCTO-Twin. There is a significant main effect of CVD on the Digit Span Backward and Memory in Reality performance at time 1. In addition, both the Memory in Reality test and the Digit Span Backward test exhibit a significant interaction effect of age by CVD on initial level. While the main effects of CVD are negative for initial level, the interaction effects of age by CVD are positive, indicating that CVD-related differences in time 1 performance were smaller for older individuals.
HRS. The main effect of having CVD is associated with a -0.082 difference in memory performance (i.e., Delayed Word Recall) on initial level for a 65 year old man with low
education. There are also significant interaction effects of CVD with sex and age. The interaction effect of sex (-0.101) by CVD indicates that women diagnosed with CVD perform significantly more poorly at time 1 relative to men.
Table 5
Effect of ever having been diagnosed with cardiovascular disease on level and rate of change in memory outcomes.
Intercept Linear Slope
Memory by Study M C CxA CxS CxE M C CxA CxS CxE
OCTO-Twin
Digit Span Forward 5.654* -0.196 -0.044 0.217 0.011 -0.056 -0.026 0.013 0.004 0.017
Digit Span Back 3.539* -0.802* 0.089* 0.364 0.102 -0.063 0.093 -0.015 -0.054 0.012
Prose Recall 10.227* -0.488 0.045 0.553 0.268 -0.072 -0.283 -0.016 0.258 -0.011
Mem-in-Reality 7.267* -1.256* 0.228* 0.387 0.180 -0.041 -0.084 0.000 0.104 0.005
HRS
Delayed Word Recall -0.364* -0.082* -0.001 -0.039 0.065* -0.030* 0.000 0.000 0.001 -0.005
Immediate Word Recall -0.417* -0.051 -0.001 -0.037 0.052 -0.037* 0.003 0.000 0.001 -0.011*
Subtract 7s -0.301* 0.003 0.003* -0.101* 0.040 -0.005 -0.003 0.000 0.002 -0.001
ELSA
Delayed Word Recall 3.639* -0.211 0.006 0.084 0.039 0.008 -0.017 -0.001 0.012 -0.017
Immediate Word Recall 5.169* -0.039 0.004 -0.124 -0.260 -0.003 -0.056* -0.001 0.056 0.014
Prospective Memory 3.339* -0.156 0.010 0.204 0.137 0.040* 0.117 -0.002 -0.037 -0.061
Note. C = condition; CxA = condition by age interaction; CxS = condition by sex interaction; CxE = condition by education
interaction; M = mean value for individuals who have not been diagnosed with cardiovascular disease. Age (A), sex (S), education (E), AxS, AxE, SxE terms where included in the model but are not listed here.
ELSA. The only significant effect found in the U.K. population is a negative main effect of CVD on the rate of change for the Immediate Word Recall test, such that individuals with CVD decline -0.056 more per year relative to those not diagnosed with CVD.
Diabetes
The parameter estimates for the effect of ever being diagnosed with diabetes on level and rate of change in memory outcomes are presented in Table 6.
OCTO-Twin. We observed only the conditional interaction effect of diabetes and education. This interaction effect is associated with a difference of -0.114 in initial levels of performance on the Digit Span Backward test for diabetic individuals with higher education.
HRS. The main effect of diabetes is associated with between-person differences in memory function (across all three HRS measures) on initial level. Significant effects on initial level are also observed for the interactions of diabetes and age (Delayed and Immediate Word Recall tests) and diabetes and sex (across all three HRS measures). These results indicate that women diagnosed with diabetes perform more poorly at time 1 relative to men (with or without diabetes) and women without diabetes. Also, older individuals with diabetes perform better at time 1, although this effect is substantively small. A significant interaction effect of diabetes and education (-0.010) was observed on the rate of change in the Immediate Word Recall test.
ELSA. The main effect of diabetes is associated with between-person differences in memory function (Delayed and Immediate Word Recall tests) on initial level. We do not observe any significant effects on rate of change.
Table 6
Effect of ever having been diagnosed with diabetes on level and rate of change in memory outcomes.
Intercept Linear Slope
Memory by Study M C CxA CxS CxE M C CxA CxS CxE
OCTO-Twin
Digit Span Forward 5.661* -0.247 0.017 0.198 -0.106 -0.065 0.025 0.008 -0.048 0.025
Digit Span Back 3.399* -0.042 0.034 -0.260 -0.114* -0.043 -0.036 0.007 0.019 0.023
Prose Recall 10.085* 0.889 0.099 -1.505 -0.386 -0.111 -0.283 -0.047 0.150 0.023
Mem-in-Reality 7.072* -0.425 0.029 0.491 0.210 -0.068 0.045 0.039 -0.232 -0.028
HRS
Delayed Word Recall -0.372* -0.075* 0.005* -0.091* 0.014 -0.029* -0.002 0.000 -0.004 -0.005
Immediate Word Recall -0.406* -0.099* 0.004* -0.081* 0.021 -0.037* 0.006 0.000 -0.004 -0.010*
Subtract 7s -0.272* -0.087* 0.004 -0.131* 0.043 -0.004 -0.006 0.000 -0.001 0.004
ELSA
Delayed Word Recall 3.647* -0.196* 0.001 -0.056 -0.228 0.004 0.009 0.002 -0.006 0.015
Immediate Word Recall 5.189* -0.174* 0.010 0.027 -0.106 -0.011 0.014 -0.001 -0.044 -0.030
Prospective Memory 3.342* -0.106 -0.003 0.151 -0.026 0.045* -0.024 0.001 -0.030 0.027 Note. C = condition; CxA = condition by age interaction; CxS = condition by sex interaction; CxE = condition by education
interaction; M = mean value for individuals who have not been diagnosed with diabetes. Age (A), sex (S), education (E), AxS, AxE, SxE terms where included in the model but are not listed here.
Stroke
The parameter estimates for the effect of ever being diagnosed with a stroke on level and rate of change in memory outcomes are presented in Table 7.
OCTO-Twin. The only significant effect found in the Swedish sample is a negative interaction effect of stroke and education (-0.030) on the linear rate of change for Digit Span Forward, suggesting that individuals with higher education who have had a stroke will decline -0.030 more per year relative to individuals with lower education and to individuals with higher education who did not suffer a stroke. We do not observe any significant effects on initial level.
HRS. The main effect of stroke is associated with between-person differences in memory function (across all three HRS measures) on initial level. We also observe a significant negative main effect of stroke (across all three HRS measures) on the linear rate of change. In other words, individuals who report having ever received a stroke diagnosis perform more poorly at time 1 and decline at a faster rate relative to those never having a stroke diagnosis. We also observed a significant effect for the interaction of stroke and age on the linear rate of change, but this effect is substantively small.
ELSA. No significant effects of stroke are observed for the U.K. sample on either initial level or rate of change.
Cancer
The parameter estimates for the effect of ever being diagnosed with cancer on level and rate of change in memory outcomes are presented in Table 8.
OCTO-Twin. Only the interaction effect of cancer and age was observed in both initial level (-0.192; a negative effect) and in rate of change (0.060; a positive effect), such that older
Table 7
Effect of ever having been diagnosed with a stroke on level and rate of change in memory outcomes.
Intercept Linear Slope
Memory by Study M C CxA CxS CxE M C CxA CxS CxE
OCTO-Twin
Digit Span Forward 5.692* -0.334 -0.04 0.274 0.061 -0.058* -0.007 -0.001 0.035 -0.030*
Digit Span Back 3.532* -0.667 -0.011 0.658 0.053 -0.047 0.004 -0.018 0.064 -0.016
Prose Recall 10.550* -1.586 0.117 2.283* 0.020 -0.145 0.139 -0.114* -0.098 -0.094
Mem-in-Reality 7.112* -0.498 0.074 0.411 0.092 -0.044 -0.071 0.004 0.047 -0.023
HRS
Delayed Word Recall -0.377* -0.156* -0.001 0.018 -0.014 -0.029* -0.017* 0.001 -0.007 0.008
Immediate Word Recall -0.415* -0.177* -0.001 0.001 -0.006 -0.034* -0.016* 0.001* 0.005 0.005
Subtract 7s -0.277* -0.180* 0.004 -0.069 0.057 -0.004 -0.017* 0.000 0.008 0.001
ELSA
Delayed Word Recall 3.633* -0.245 0.005 -0.181 0.227 0.008 -0.026 -0.003 0.028 -0.008
Immediate Word Recall 5.181* -0.220 -0.002 0.034 0.190 -0.008 -0.020 -0.002 -0.031 -0.051
Prospective Memory 3.342* -0.225 -0.005 0.082 0.352 0.041* -0.004 0.002 -0.034 0.071 Note. C = condition; CxA = condition by age interaction; CxS = condition by sex interaction; CxE = condition by education
interaction; M = mean value for individuals who have not been diagnosed with a stroke. Age (A), sex (S), education (E), AxS, AxE, SxE terms where included in the model but are not listed here
*p < .05.
Table 8
Effect of ever having been diagnosed with cancer on level and rate of change in memory outcomes.
Intercept Linear Slope
Memory by Study M C CxA CxS CxE M C CxA CxS CxE
OCTO-Twin
Digit Span Forward 5.606* 0.153 -0.071 0.027 0.047 -0.073* 0.053 0.011 -0.066 0.007
Digit Span Back 3.321* 0.581 -0.192* -0.235 0.020 -0.037 -0.105 0.060* -0.053 0.010
Prose Recall 10.063* 0.961 0.034 -0.318 -0.106 -0.109 -0.069 0.039 0.061 0.006
Mem-in-Reality 6.949* 0.412 0.017 -0.122 -0.074 -0.094 0.195 -0.005 -0.149 -0.074
HRS
Delayed Word Recall -0.406* 0.044 0.002 -0.008 0.002 -0.030* -0.001 -0.001* -0.001 -0.001
Immediate Word Recall -0.447* 0.048 0.001 -0.021 0.009 -0.035* 0.000 0.000 -0.003 -0.003
Subtract 7s -0.307* 0.043 0.000 0.018 0.005 -0.007* 0.006 0.000 -0.004 -0.005
ELSA
Delayed Word Recall 3.600* 0.228* -0.003 -0.079 -0.329 0.008 -0.023 -0.002 -0.002 0.043
Immediate Word Recall 5.156* 0.127 -0.012 -0.062 -0.231 -0.011 0.005 0.001 0.002 -0.016
Prospective Memory 3.317* 0.108 -0.005 0.019 -0.286 0.043* -0.016 -0.001 -0.005 0.025 Note. C = condition; CxA = condition by age interaction; CxS = condition by sex interaction; CxE = condition by education
interaction; M = mean value for individuals who have not been diagnosed with cancer. Age (A), sex (S), education (E), AxS, AxE, SxE terms where included in the model but are not listed here.
time, older individuals with a cancer diagnosis decline less quickly relative to younger individuals with a cancer diagnosis.
HRS. The only significant effect found in the U.S. sample is the interaction effect of cancer and age on rate of change for the Delayed Word Recall test, although this effect is substantively small. We do not observe any significant effects on initial level.
ELSA. The only significant effect found in the U.K. sample is a positive main effect of cancer on the Delayed Word Recall test performance at time 1, such that 65 year old men with low education ever diagnosed with cancer score higher by 0.228 at time 1 relative to those never diagnosed with cancer. We do not observe any significant effects on rate of change.
Comorbidity Index
The parameter estimates for the effect of having more than one health condition (based on the comorbidity index) on level and rate of change in memory outcomes are presented in Table 9.
OCTO-Twin. No significant effects of comorbidity operationalized as an additive index are observed for the Swedish sample on either initial level or rate of change.
HRS. Significant negative effects of disease count are associated with between-person
differences in memory function (across all three HRS measures) on initial level. We also observe significant interaction effects for comorbidity and sex (across all three HRS measures) on initial level, such that with each additional diagnosed health condition at time 1, women perform more poorly relative to men with similar diagnosed health conditions. We observe a positive
interaction effect for comorbidity and education (Immediate Word Recall and Subtract 7’s tests) on initial level, such that individuals with more diagnosed health conditions and with higher education perform less poorly at time 1 relative to those with more diagnosed health conditions or higher education alone.
Table 9
Effect of having more than one health condtion (comorbidity index) on level and rate of change in memory outcomes.
Intercept Linear Slope
Memory by Study M C CxA CxS CxE M C CxA CxS CxE
OCTO-Twin
Digit Span Forward 5.730* -0.090 -0.016 0.060 0.016 -0.071 0.007 0.007 -0.014 -0.001
Digit Span Back 3.613* -0.198 -0.005 0.098 0.036 -0.038 -0.008 0.010 -0.015 -0.002
Prose Recall 10.316* -0.137 0.080 0.188 0.075 -0.005 -0.084 -0.022 0.110 -0.028
Mem-in-Reality 7.329* -0.292 0.050 0.206 0.096 0.040 -0.084 0.014 0.045 -0.013
HRS
Delayed Word Recall -0.317* -0.049* 0.001 -0.023* 0.022 -0.026* -0.002 0.000 -0.001 -0.001
Immediate Word Recall -0.361* -0.046* 0.001 -0.027* 0.026* -0.034* -0.001 0.000 0.000 -0.004
Subtract 7s -0.234* -0.039* 0.002* -0.054* 0.038* 0.000 -0.003 0.000 0.001 0.000
ELSA
Delayed Word Recall 3.671* -0.065 0.001 -0.013 -0.065 0.016 -0.010 -0.001 -0.003 0.006
Immediate Word Recall 5.206* -0.050 0.000 -0.010 -0.056 -0.003 -0.006 0.000 -0.008 -0.017
Prospective Memory 3.366* -0.049 0.000 0.041 -0.005 0.038* 0.004 0.000 -0.022 0.006 Note. C = comorbidity count; CxA = comorbidity count by age interaction; CxS = condition by sex interaction; CxE = condition by
education interaction; M = mean value for individuals who have no diagnosed health conditions. Age (A), sex (S), education (E), AxS, AxE, SxE terms where included in the model but are not listed here.
We do not observe any significant effects on rate of change.
ELSA. No significant effects of comorbidity are observed for the U.K. sample on either initial level or rate of change.
Discussion
We observed significant effects of diagnosed health conditions on initial level and rate of change in memory function in the OCTO-Twin study, but these effects were not consistent across health conditions or across memory measures. The Digit Span Backward and Memory in Reality tests demonstrated to be the most sensitive and showed between-person differences at time 1 in the impact of CVD (main effect and interaction effect of age), Diabetes (interaction effect with education), and Cancer (interaction effect of age), as well as significant rates of change in Hypertension (main effect and interaction effect of age) and Cancer (interaction effect of age).
Within the HRS, the observed significant effects were primarily associated with between-person differences in memory function at time 1. Hypertension, CVD, and Diabetes
demonstrated the greatest impact on initial performance levels. We also observed significant interaction effects of age, sex, and education across these same three health conditions. The impact of specific health conditions was less likely to be associated with rate of change, but we did observe significant effects for Diabetes and Stroke diagnoses. Having more than one health condition had a significant effect on initial performance levels, but not on rates of change.
We observed very few significant effects within the ELSA study. Having received a diagnosis of Hypertension or Stroke did not have an impact on memory function for the U. K. sample. Having more than one health condition was also observed to not have an impact. Significant main effects (no interaction effects were observed) were observed for CVD (Immediate Word Recall test) on rate of change, Diabetes (Immediate and Delayed Word Recall tests) on initial level, and Cancer (Delayed Word Recall test) on initial level.
In summary, significant main and interaction effects were demonstrated for each study, but these effects were not consistent across studies or across outcome measures of memory function. We found significant interaction effects of diagnosed heath conditions and age, sex, and
education for the HRS and OCTO-Twin studies, but we did not observe these effects within the ELSA study. We also found significant main effects of diagnosed health conditions primarily associated with between-person differences at time 1 for HRS and primarily on rate of change for OCTO-Twin. We observed little impact of the specific conditions within the ELSA study.
With increased age, there is increasing variability in the patterns of change (i.e., onset, rate of change, magnitude and direction) of most cognitive functions and processes. However, much of what we know about late life changes in cognition comes from cross-sectional designs that confound between-person age trends (Hofer, Flaherty, & Hoffman, 2006; Hofer & Sliwinski, 2001; Hofer, Sliwinski, & Flaherty, 2002) and can lead to spurious inferences about the interdependency among age-related functions.
In this study, we examined the impact of specific health conditions on level and change in memory function across three longitudinal studies of aging. As noted earlier, comparing results across different longitudinal studies can be challenging. A key issue is the comparability of results based on different measurement instruments (or the same instrument that may have been administered differently). This challenge is further compounded by differences in sample composition (e.g., birth cohort, culture, social system) and in attrition patterns. Dropout is likely to be related to the chronic conditions and comorbidity evaluated in this study, with those individuals more likely to exhibit greater change and to dropout from the study. The impact of this differential dropout related to health conditions may make it more difficult to find a
multiple independent data sets in ways that optimize the comparison of results. To this end, we chose measures that were as similar as possible and represented the same memory constructs (e.g., Delayed Word Recall for both HRS and ELSA; Memory in Reality Recall for OCTO-Twin). We also tried to create samples that were as similar as possible by excluding those with possible cognitive impairment (though this was done differently depending on available measures within each study) and restricting the age range. Results from these analyses
demonstrated that the effects of different health conditions were not consistent across cognitive outcomes or across studies.
Banks, Marmot, Oldfield, & Smith (2011) recently assessed the relative health status of older individuals in England and the United States using the HRS and ELSA data. Their study was cross-sectional and included a limited age range of 55 – 64. However, despite this difference in study design, the results from the current study are similar to this previous work in that we also observed a difference between the U.S. and U.K. samples in the frequency of diagnosed
conditions, suggesting that the U.K. sample may be more “healthy” relative to their U.S. counterparts. In addition, in considering the impact of diagnosed health conditions on cognitive outcomes, the current study found further differences, such that the U.K. sample exhibited little to no effect of health changes on cognition. The Swedish sample exhibited effects similar to that of the U.S. population. Restricting the U.S. and U.K. samples to match the Swedish sample in age (i.e., examining only those age 80 or older at time 1) did not alter these patterns.
It is possible that given the changing societal conditions across birth cohorts and countries, the impact of particular health conditions may lead to greater and earlier cognitive impairment (e.g., Mirowsky, 2011). This illustrates the importance of cross-study comparisons that differ in cohort and country, as we begin to see differences in level and rate of change of cognitive outcomes
across samples that differ in disease prevalence and other risk factors and health behaviours. This lends support to Spiro and Brady’s (2011) argument that researchers begin to view health as an explanatory domain in the study of cognitive aging. They argue that differences in health can account for differences in cognition both within and between individuals over time. As
demonstrated in the findings of this study, it should be further considered that specified diseases can affect cognitive functions differentially.
Health conditions provide a basis for understanding individual differences. While we did not include severity measures of particular conditions, the ever-diagnosed variable provided a first and important step in cross-study comparisons. An important next step might be to evaluate the differences between those diagnosed prior to Time 1, those diagnosed during the study, and those that are never diagnosed. This will help to elucidate the differences between individuals with long-term chronic conditions, those who develop health changes more abruptly, and those that remain disease free (i.e., those experiencing primary or normative aging). We also did not explore the interaction among particular health conditions as in the Hassing et al. (2004) paper finding an interactive effect of diabetes and hypertension. This will be an important extension to be evaluated in subsequent research as interactions among conditions may result in increased risk for rate of change in cognitive functioning.
Until we develop a more integrated science, and consistently apply more appropriate analytical designs for evaluating the complexities outlined here (for discussion of analytical approaches, see Sliwinski & Mogle, 2008), it will be challenging to tease apart the effects of disease from those of aging per se, but it is fundamental that we begin to more appropriately recognize age as a way to chart developmental changes (as suggested by Wohlwill, 1970; 1973) and not as a causal mechanism for physiological and cognitive changes late in life. It is critical that we make
a distinction between non-normative or pathological influences and normative aging that by definition affects all individuals.
As we continue to observe evidence that pathological changes related to chronic and acute health conditions contribute to cognitive impairment and accelerated decline in function, the impetus will be on disentangling normative aging from health-related change. The understanding that we are working towards, and that will lead us in the direction of a more preventive science, is identifying sources of change in cognition that may be amenable to prevention and/or
intervention efforts. Health conditions are often modifiable by a variety of risk factors (e.g., diet, exercise); such risk factors on cognition may be one and the same and work through the disease pathway.
References
Ahles, T. A., Saykin, A. J., McDonald, B. C., Li, Y., Furstenberg, C. T., Hanscom, B. S., . . . Kaufman, P. A. (2010). Longitudinal Assessment of Cognitive Changes Associated With Adjuvant Treatment for Breast Cancer: Impact of Age and Cognitive Reserve. Journal of Clinical Oncology, 28(29), 4434-4440.
Alwin, D. F., & Hofer, S. M. (2011). Health and Cognition in Aging Research. The Journals of Gerontology Series B: Psychological Sciences and Social Sciences, 66B(suppl 1), i9-i16. Anstey, K. J., Mack, H. A., & von Sanden, C. (2006). The relationship between cognition and
mortality in patients with stroke, coronary heart disease, or cancer. European Psychologist, 11(3), 182-195.
Anstey, K., Stankov, L., & Lord, S. (1993). Primary aging, secondary aging, and intelligence. Psychology and Aging, 8(4), 562-570.
Arvanitakis, Z., Wilson, R. S., Bienias, J. L., Evans, D. A., & Bennett, D. A. (2004). Diabetes Mellitus and Risk of Alzheimer Disease and Decline in Cognitive Function. Archives of Neurology, 61(5), 661-666.
Arvanitakis, Z., Bennett, D. A., Wilson, R. S., & Barnes, L. L. (2010). Diabetes and cognitive systems in older Black and White persons. Alzheimer Disease and Associated Disorders, 24(1), 37-42.
Bäckman, L., & MacDonald, S. W. S. (2006). Death and cognition: Viewing a 1962 concept through 2006 spectacles. European Psychologist, 11(3), 161-163.
Baltes, P. B. (1987). Theoretical propositions of life-span developmental psychology: On the dynamics between growth and decline. Developmental Psychology, 23(5), 611-626. Baltes, P.B., & Nesselroade, J.R. (1979). History and rationale of longitudinal research. In J.R.
Nesselroade & P.B. Baltes (Eds.), Longitudinal research in the study of behavior and development, (pp. 1-39). New York: Academic Press.
Banks, J. A., Breeze, E., Lessof, C., & Nazroo, J. (2008). Living in the 21st century: Older people in England: The 2006 English Longitudinal Study of Ageing (Wave 3). London: The Institute of Fiscal Studies.
Banks, J., Lessof, C., Nazroo, J., Rogers, N., Stafford, M., & Steptoe, A. (2010). Financial circumstances, health and well-being of the older population in England: The 2008 English Longitudinal Study of Ageing (Wave 4). London: Institute for Fiscal Studies. Banks, J., Marmot, M., Oldfield, Z., & Smith, J. P. (2006). Disease and disadvantage in the
United States and in England. JAMA: The Journal Of The American Medical Association, 295(17), 2037-2045.
Banks, J., Oldfield, Z., & Smith, J. P. (2011). Investigations in the Economics of Aging: Childhood Health and Differences in Late-Life Health Outcomes between England and the United States. In D. A. Wise (Eds.), NBER Book Chapter Investigations in the Economics of Aging, Stanford University.
Birren, J. E., & Birren, B. A. (1990). The concepts, models, and history of the psychology of aging. In J. E. Birren & K. W. Schaie (Eds.), Handbook of the psychology of aging (3rd ed) (pp. 3-20). San Diego, CA US: Academic Press.
Birren, J. E., & Cunningham, W. R. (1985). Research on the psychology of aging: Principles, concepts and theory. In J. E. Birren & K. W. Schaie (Eds.), Handbook of the psychology of aging (2nd ed.). (pp. 3-34). New York, NY US: Van Nostrand Reinhold Co.
Birren, J. E., & Schroots, J. F. (1984). Steps to an ontogenetic psychology. Academic Psychology Bulletin, 6(2), 177-190.
Blumenthal, H. T. (1993). Editorial: The aging-disease dichotomy is alive, but is it well? Journal of American Geriatrics Society, 41, 1272-1273.
Blumenthal, H. T. (2003). The aging-disease dichotomy: True or False? Journal of Gerontology: Medical Sciences, 58(2), 138-145.
Borson, S., Scanlan, J. M., Lessig, M., & DeMers, S. (2010). Comorbidity in aging and dementia: Scales differ, and the difference matters. The American Journal of Geriatric Psychiatry, 18(11), 999-1006.
Brady, C. B., Spiro, A., III, McGlinchey-Berroth, R., Milberg, W., & Gaziano, J. M. (2001). Stroke risk predicts verbal fluency decline in healthy older men: Evidence from the normative aging study. Journals of Gerontology: Series B: Psychological Sciences and Social Sciences, 56(6), P340-p346.
Brady, C. B., Spiro, A., III, & Gaziano, J. M. (2005). Effects of Age and Hypertension Status on Cognition: The Veterans Affairs Normative Aging Study. Neuropsychology, 19(6), 770-777.
Busse, E. W., & Pfeiffer, E. (1969). Behavior and adaptation in late life. Oxford England: Little, Brown.
Casserly, I., & Topol, E. (2004). Convergence of atherosclerosis and Alzheimer's disease: Inflammation, cholesterol, and misfolded proteins. The Lancet, 363(9415), 1139-1146.
Cederlöf, R. & Lorich, U. (1978). The Swedish Twin Registry. In W. E. Nance, G. Allen, & P. Parisi (Eds.), Twin research: Biology and epidemiology (pp. 189–195). New York: Alan R. Liss.
Cherubini, A., Lowenthal, D. T., Paran, E., Mecocci, P., Williams, L. S., & Senin, U. (2010). Hypertension and Cognitive Function in the Elderly. Dm Disease-a-Month, 56(3), 106-147.
Christman, A. L., Vannorsdall, T. D., Pearlson, G. D., Hill-Briggs, F., & Schretlen, D. J. (2010). Cranial volume, mild cognitive deficits, and functional limitations associated with diabetes in a community sample. Archives of Clinical Neuropsychology, 25(1), 49-59. de Groot V, Beckerman H, Lankhorst G, Bouter L. (2003). How to measure comorbidity: a critical review
of available methods. J Clin Epidemiology, 56, 221-229.
de la Torre, J. C. (2002). Alzheimer disease as a vascular disorder: Nosological evidence. Stroke, 33, 1152-1162.
de la Torre, J. C. (2004). Is Alzheimer’s disease a neurodegenerative or a vascular disorder? Data, dogma, and dialectics. Lancet Neurology, 3, 184-190.
de la Torre, J. C. (2010). Vascular risk factor detection and control may prevent Alzheimer's disease. Ageing Research Reviews, 9(3), 218-225.
Duron, E., & Hanon, O. (2008). Vascular risk factors, cognitive decline and dementia. Vascular
Health and Risk Management, 4, 363–381.
Fontbonne, A., Berr, C., Ducimetiere, P., & Alperovitch, A. (2001). Changes in cognitive abilities over a 4-year period are unfavorably affected in elderly diabetic subjects: results of the Epidemiology of Vascular Aging Study. Diabetes Care, 24(2), 366-370.
Forbes, W. F. & Hirdes, J. P. (1993). The relationship between aging and disease: Geriatric ideology and myths of senility. Journal of American geriatrics Society, 41, 1267-1271. Goodwin, J. S. (1991). Geriatric ideology: The myth of the myth of senility. Journal of American
geriatrics Society, 39, 627-631.
Gregg, E.W., Yaffe, K., Cauley, J.A., Rolka, D.B., Blackwell, T.L., Narayan, K.M., & Cummings, S.R. (2000). Is diabetes associated with cognitive impairment and cognitive decline among older women? Study of Osteoporotic Fractures Research Group. Archives of Internal Medicine, 160(2), 174-80.
Hassing, L. B., Johansson, B., Pedersen, N. L., Nilsson, S. E., Berg, S., & McClearn, G. (2003). Type 2 diabetes mellitus and cognitive performance in a population-based sample of the oldest old: Impact of comorbid dementia. Aging Neuropsychology and Cognition, 10(2), 99-107.
Hassing, L. B., Hofer, S. M., Nilsson, S. E., Berg, S., Pedersen, N. L., McClearn, G., &
Johansson, B. (2004). Comorbid type 2 diabetes mellitus and hypertension exacerbates cognitive decline: evidence from a longitudinal study. Age And Ageing, 33(4), 355-361.
