An examination of antecedents of individual differences in the maintenance or decline of intellectual functioning would surely begin by inquiring into the impact of health on cognition. As shown here, however, it does not necessarily follow that the relationship between health and intellectual functioning is strictly unidirectional. Reviews of the literature (see Elias, Elias, & Elias, 1990; Siegler, 1988; Siegler et al., 2009) suggest that the relationship indeed may be reciprocal: a healthy body facilitating intellectual competence, and competent behavior facilitating the maintenance of physical health. In this chapter, I therefore consider models that may have bidirectional as well as reciprocal implications. I first report our efforts to assess health histories in a manner suitable for relating them to behavioral development. Second, I consider the diseases that seem to affect the maintenance of cognitive functions. Third, I describe our work on the study of health behaviors. Finally, I return to the role of intellectual functioning as a predictor of physical health, as a predictor of medication use, and as a possible early indicator of impending mortality.
One of the interesting aspects of our panel of study participants is the fact that all panel members, during their time of participation in our study, received all of their health care (with the rare exception of emergency procedures when away from home) from the health maintenance organization (HMO) that forms the base of our sampling frame. Virtually complete records are therefore available on the frequency and kinds of illnesses requiring medical care, as well as anecdotal records of treatment history. Subsequent to the third study cycle, we were finally able to obtain the necessary resources to take advantage of the existence of these records for our study participants (Hertzog, Schaie, & Gribbin, 1978; Schaie, 1973a). We have worked intensively with the medical and research staff of the HMO to develop procedures designed to quantify the health records of our panel in such a way that it is possible to index them in terms of both the age of the individual and the points in time when incidents of ill health occurred.
We soon discovered that, although most physicians generate voluminous medical histories, only a few have experience or interest in retrieving data from such histories, particularly in a form that lends itself to research. At the time we became interested in this problem, a number of formal ways existed to code medical data. We elected to use the American version of the international system sponsored by the World Health Organization (U.S. Public Health Service, 1968). Unfortunately, as others interested in abstracting from medical records (e.g., Hurtado & Greenlick, 1971) have found, it is quite difficult to relate descriptive disease categories to individual outcome parameters. To do so requires the specialized services of medical record librarians, who must detect and decipher the information required for such coding.
If there is interest in relating medical histories to outcomes of a nonmedical nature, moreover, concern must also be given to the issue of incidents and episodes of medical care in addition to simply recording diagnostic entries. Such an approach was first used by Solon and associates (Solon, Feeney, Jones, Rigg, & Sheps, 1967; Solon et al., 1969). For our purposes, we decided to code data for our panel members with respect to incidents of a given disease condition and to apply the episode or “spell of illness” approach.
If there is concern with the impact of disease on behavior, we must also deal with the relative significance of a particular diagnostic condition as it affects the life of the person experiencing that condition. That is, we would like to assign a “severity” weight to particular diagnostic entities. A special study was therefore conducted to obtain such weights (Parham, Gribbin, Hertzog, & Schaie, 1975).
In this study, the health records for 150 participants were coded for a 14-year period. Although the eighth edition of the International Classification of Diseases (ICDA-8; U.S. Public Health Service, 1968) contains over 8,000 disease classifications, only about 820 were actually encountered in our sample. By collapsing and overlapping categories, we further reduced this number to 448 of the most frequently occurring classifications. Then, 12 physicians Q-sorted these classifications on an 11-point severity scale ranging from benign to extremely severe. The physicians were asked to rate severity according to the long-range impact of the particular disease on the general health and well-being of the patient. The 448 categories were divided into four decks of 112 cards each, on which the disease name and a brief description were typed. Each physician Q-sorted two of these decks; a total of four physicians sorted each deck. Interrater reliabilities for the disease severity ratings ranged from .82 to .90. Average weights were then computed, and these severity weights were used in some of the analyses described in this chapter.
Fortunately, our collaborating HMO was able to shift to an automated patient record system shortly after our sixth data collection. This means that current and future analyses related to health and disease will be able to benefit from the use of computerized databases; our manual coding utilized ICDA Version 8, and the automated coding utilizes ICDA Version 9. Although no simple transformation programs have been available to us, it turns out that this is not a major problem for our analyses because our data aggregations utilize primarily the first two digits of the diagnostic codes, which remain similar across versions.
Before discussing our data, I first must consider how a disease model can be related to the study of cognitive aging. A model then is explicated that might explain the complex relationship between health breakdown and decline in cognitive functioning.
I begin by raising the question of whether there is a direct relationship between raw indices of disease diagnoses (or incidence of medical care) and behavioral outcomes (e.g., Wilkie & Eisdorfer, 1973). I follow here the lead of Aaron Antonovsky (1972), who has addressed this question in some detail. Antonovsky held that the social or behavioral scientist dealing with the consequences of physical illness should not be concerned with the particular physiological dimensions involved in a disease but should, rather, address the behavioral and social consequences embedded in the concept of “breakdown.”
More specifically, Antonovsky argued that there are basically four dimensions of breakdown that deserve attention. First, a disease may or may not be directly painful to the individual; second, it may or may not handicap individuals in the exercise of their faculties or performance of social roles; third, it can be characterized along the dimension of acuteness-chronicity with respect to its possible threat to life; and fourth, it is or is not recognized by society’s medical institutions as requiring care under the direction of such institutions.
A very similar system of classification can be suggested with respect to the impact a particular disease may have on cognitive functions, particularly if medical histories are reclassified in terms of the degree of breakdown presented therein rather than in terms of the specific disease represented by the history.
Although we must pursue the impact of specific diseases on behavior, we should also be conscious of the possibility that it may not be fruitful to insist on a direct connection between specific diseases and behavior. It would therefore be better to organize information on illness and the utilization of medical care in terms of more psychologically meaningful organizing principles, such as the concept of breakdown. Such an approach would reduce the conceptual dilemma of having to distinguish between the contribution of the actual disease process and the manner in which the individual responds to the disease condition. It is often the latter that may be the more direct mediator of the behavioral consequences, at least at the macrolevel most easily accessible to observation and analysis.
Given the methods of coding described in this chapter concerning the analysis of disease histories (as well as the modification of the methods for application to computerized data), we can first study the impact of specific diseases in relation to cognitive change. The cumulative impact of health trauma can also be charted in at least two ways. First, a cumulative index of physical health breakdown can be assigned, which is simply the summation of all incidents observed as weighted by their impact on the life of each study participant. Second, we can graph the average level of breakdown at each measurement point for which behavioral data are available and relate the slope of physical health states to the slope of observed cognitive change.
Throughout these analyses, it is just as necessary as it was with the cognitive data to be concerned with the effects of cohort differences in the utilization of medical care as well as with the tendency of previous episodes of ill health to elicit different patterns of subsequent ill health than would be true if the occurrence of health breakdown were the first. Fortunately, our system of data acquisition permits a reasonable modicum of controls.
The first two analyses bearing on the question of how disease might affect cognitive functioning involved relatively small data sets. The first focused on the relation of cardiovascular disease (CVD) and maintenance of intellectual functioning; the second attempted to include a somewhat broader spectrum of diseases.
In our first effort to chart the relationship between CVD and maintenance of cognitive functioning systematically (Hertzog et al., 1978), we studied the health records of 156 subjects tested in 1956 and 1963, of whom 86 also participated in the 1970 testing. For this study, subjects were classified as having the diagnoses of hypertension, atherosclerosis (constriction of arterial pathways by fatty deposits), hypertension and atherosclerosis, cerebrovascular disease, miscellaneous CVD, benign CVD, or as having no evidence of CVD. After exclusion of subjects who had only hypertension and those with benign CVD, the remaining subjects with CVD were then compared with those without CVD. It was first noted that there was an excess of persons with CVD among the dropouts at the 1970 test occasions, thus making CVD a major factor in experimental mortality (see chapter 8). The CVD dichotomy was then related by analyses of variance to the maintenance of intellectual performance over time.
Significant main effects favoring the subjects without CVD were found for Verbal Meaning, Inductive Reasoning, Number, and the composite indices, as well as Motor-Cognitive Flexibility. However, an increase in the additional risk over time was found only for Psychomotor Speed. Breaking down CVD into subgroups, greater risk for those affected over time occurred for those with atherosclerosis and cerebrovascular disease for the Space and Number tests as well as for the composite IQ and the Psychomotor Speed measure. However, those with hypertension without other manifestations of CVD actually improved over time.
The next set of analyses of the health data was conducted as part of a doctoral dissertation by Stone (1980). She examined a sample of 253 subjects for whom psychological and health data were available at three time points: 1963, 1970, and 1977. Disease codes were aggregated into 16 systemic categories, of which 11 were sufficiently well represented to warrant investigation. These were as follows:
1. Diseases of the blood and blood-forming organs
2. Diseases of the circulatory system
3. Endocrine, nutritional, and metabolic disorders
4. Diseases of the digestive system
5. Diseases of the genitourinary system
6. Infectious diseases
7. Diseases of the musculoskeletal system and of connective tissues
8. Diseases of the skin and subcutaneous tissues
9. Neoplasms
10. Diseases of the nervous system
11. Diseases of the respiratory system.
Structural models were examined to link illness variables to Primary Mental Abilities (PMA) performance. Because there was little covariation among disease entities, separate structural models were tested for those categories that were well represented for all three time periods. Three of the concatenated disease categories listed in this section appeared to be significant predictors of time-related change in intellectual functioning: circulatory disorders, neoplasms, and musculoskeletal disorders. However, when the sample was divided by age into halves (35 to 58 and 59 to 87 years at T3), the relationship held for both age groups only for circulatory disease, and the other two variables were significant only in the older group. Interestingly, there was an unexpected positive relationship between diagnosed neoplasms at T2 and intelligence at T3. It might be speculated that the more able persons were more likely to seek earlier diagnosis and had better opportunities for effective treatment, thus increasing their survival rate. Another possible problem with the findings was the failure to disaggregate diagnoses of neoplasms into malignant and benign types.
Our most complete analysis of the prevalence of disease and its impact on cognitive functioning was conducted as part of the dissertation research of Ann Gruber-Baldini (1991a) for a sample of 845 subjects for whom data were available through the completion of the fifth cycle (1984). These subjects entered and departed the Seattle Longitudinal Study (SLS) at various measurement points (1956, 1963, 1970, 1977, 1984), but all had at least two points of measurement. Analyses either organized data by age at testing or utilized the 1970–1977 period, for which most of the subjects had complete data.
One goal of this analysis was to utilize the data from the HMO medical records to assess patterns of disease occurrence, prevalence and incidence of diseases, comorbidity of chronic disease conditions, and the progression and complications of disease categories (Gruber-Baldini, 1991b). Findings from HMO records were comparable to rates in other studies using self-report methods. Diseases increased in prevalence, incidence (except for neoplasms), and comorbidity with age. Arthritis was the most prevalent condition, followed by vision problems, neoplasms, and hypertension. Differences were found in rates of disease occurrence for men and women and in rates across time periods. Women had higher rates of arthritis, benign CVD, benign neoplasms, essential hypertension, osteoporosis and hip fractures, and depression. Males had higher frequency of more serious conditions (in terms of risk of mortality), such as atherosclerosis, cerebrovascular disease, and malignant neoplasms. The number of physician contacts and hospital days was more frequent in the latest measurement period, whereas the average number of chronic conditions in the sample peaked during the period from 1964 to 1970.
Rates of chronic conditions varied by time of measurement as well as age. Overall, arthritis, benign CVD, neoplasms, and osteoporosis peaked from 1957 to 1963 (the first measurement period), but all others peaked from 1970 to 1977. Rates from 1977 to 1984 appear to be lower than for other periods for arthritis, vision problems, and benign CVD. This period had lower rates of conditions even when rates were examined only for the age range 60 to 67 years. However, subjects with data from 1977 to 1984 were members of a training study in the SLS and may have been a more select sample because they were able to undergo five 1-hour sessions of cognitive training. Also, differences because of cohort effects in risk factors associated with medical utilization were confounded with time of measurement.
Average age of disease onset occurred after 50 years for all chronic conditions. Rheumatoid arthritis, benign CVD, and nonmalignant neoplasms have earlier disease onset; osteoarthritis, atherosclerosis, cerebrovascular disease, and malignant neoplasms have average onset after age 60 years.
The impact of diseases on cognitive functioning was examined longitudinally by three sets of analyses (Gruber-Baldini, 1991a). The first included logistic regression and event history analyses predictors for the occurrence of and age at onset of significant cognitive decline. The second used latent growth curve models (LGMs; McArdle & Anderson, 1990; McArdle & Hamagami, 1991) to examine longitudinal patterns of PMA functioning from ages 53 to 60 years. The third examined a LISREL (linear structural relations) path model for the direct effects of diseases on cognitive level and change and for indirect effects on level and change through measures of inactive lifestyle (leisure activities and an obesity measure).
Prior research suggested that ratings of poor overall health predict lowered cognitive functioning. In the current study, measures of the number of chronic conditions, total number of physician visits, and number of hospital days were examined. The number of chronic conditions had negative influences on cognitive level for Verbal Meaning, Number, and Word Fluency and predicted greater decline on Verbal Meaning and Number. However, the age of experiencing significant decline occurred later for persons with more chronic conditions on Number and Word Fluency. A greater number of physician visits predicted an increased hazard of cognitive decline for Reasoning and a later age of onset of decline for number. Physician visits were positively correlated with arthritis episodes, and arthritis was predictive of lower cognitive level on a number of the PMA measures. Hospitalization had no significant impact on PMA functioning except that it was positively correlated with number of physician visits, number of chronic conditions, and number of episodes for some chronic diseases (especially arthritis), all of which were predictive of PMA performance.
Research on the relation of specific diseases and cognitive functioning has focused mostly on CVD, particularly on hypertension. This research has found mixed results with respect to the direction of influence of hypertension on cognition. Most studies in the literature suggested that more severe CVD (atherosclerosis, cerebrovascular disease, etc.) has a negative impact on cognitive functioning. However, much of the prior research was cross-sectional. Longitudinal studies in this area have often involved small samples, have included a limited number of testing occasions (i.e., fewer than three points), have failed to compare hypertension groups with groups with more severe CVD, and did not have information on cognitive functioning prior to disease onset.
In the analyses summarized here, multiple CVD groups were examined for the influence of the disease on cognitive functioning. Results suggest that atherosclerosis is associated with lower cognitive functioning and greater decline on Space and Number. LGM results suggest, however, that less decline occurred for Verbal Meaning in people with atherosclerosis; the decline occurred after age 60 years and resulted in little level difference at age 81 years in groups with and without atherosclerosis. Cerebrovascular disease was also negatively associated with cognitive level and increased the risk of and amount of cognitive decline (although the age of onset of significant decline was later than average for Spatial Orientation in the event history analyses).
The LGM results suggest that hypertensives with other CVD complications performed worse over time than uncomplicated hypertensives and normotensives. Noncomplicated hypertensives had higher performances and less decline than complicated hypertensives and normotensives. The total number of hypertension episodes predicted increased hazard of significant decline and overall level on Word Fluency, but significantly later decline onset for Spatial Orientation and Reasoning. Also, hypertension was the only significant disease predictor for people under age 60 years in the path models (again predicting lower performance and negative change over time for Word Fluency). Miscellaneous CVD predicted earlier decline on Word Fluency.
Results from logistic and event history models suggest that miscellaneous CVD predicted less decline and later onset on Spatial Orientation. However, miscellaneous CVD also indirectly predicted (through leisure activities involving phone calls, game playing, and daydreaming) decreased Spatial Orientation level and change over time. Persons with miscellaneous CVD also performed at lower levels on reasoning from ages 53 to 81 years, although their onset of decline was later than for people free of miscellaneous CVD. Benign CVD was associated with a lower rate of cognitive decline.
Thus, the more serious CVD conditions (atherosclerosis and cerebrovascular disease) have generally negative influences, and benign CVD has more positive influences on cognition. Miscellaneous CVD and hypertension appear to fall between serious and benign CVD, with uncomplicated hypertensives maintaining higher cognitive functioning. Studies are needed to confirm the differences between CVD groups on a different longitudinal sample.
Studies on the influence of diabetes have shown a negative impact of diabetes on functioning. However, these studies have also been cross-sectional and did not screen subjects for complications from other chronic diseases. We had available only a limited number of diabetics (n = 51) and thus were not able to examine the longitudinal pattern of functioning for this group by LGM analyses. However, other types of analysis showed that diabetes had an indirect positive effect on Inductive Reasoning level and longitudinal change (mediated via the measure of body mass index). It is conceivable again that long-term survivors like those included in our study who are functioning at high intellectual levels may be able to manage their chronic conditions more adequately. These findings contradict prior findings and need to be replicated.
Only a few prior studies have examined the influence of arthritis on cognitive functioning, despite the high prevalence of this disease among the aged. Results from the LGM analyses suggest that arthritics have lower functioning and greater decline on Verbal Meaning, Spatial Orientation, and Inductive Reasoning. Logistic regression results, however, found that arthritis presented a lower proportional hazard of decline for Spatial Orientation and later average onset of significant decline on Verbal Meaning, Spatial Orientation, and Reasoning. Also, LISREL path models showed that arthritis had a direct negative effect on Spatial Orientation level and change from 1970 to 1977, and an indirect negative effect (through diversity of leisure activities) on Number and Word Fluency levels and change (less decline). Dividing arthritics by age of occurrence, LGM results indicate that persons who developed arthritis after age 60 years had lower levels and experienced greater decline on Verbal Meaning. However, persons with arthritis before age 60 years had lower levels and greater decline on Inductive Reasoning. A mixed pattern resulted for Spatial Orientation, with arthritics before age 60 years experiencing greater decline, whereas the post-60 years group had lower overall levels of functioning by age 81 years. Future studies may need to examine longitudinally the separate effects of rheumatoid arthritis and osteoarthritis—subdiagnoses of arthritis that differ by age of diagnosis—on cognition to clarify the results of arthritis influences on cognitive level and cognitive change.
In prior research on neoplasms, Stone (1980) found positive effects of neoplasms on cognitive performance in the SLS but used a global measure of neoplasms (the entire ICDA category). The analyses described here employed more specific categories, considering differences between malignant and benign neoplasms and between skin (the most frequent neoplasms) and other neoplasms. Results suggest that the positive effects found by Stone might be caused by the large frequency of benign neoplasms in the neoplasms category. Benign neoplasms (not skin) were found to produce earlier onset of decline, but less overall decline. Malignant neoplasms and benign skin neoplasms had indirect (through activity factors) negative influences on performance. Results of the influence of neoplasms on cognition might be specific to combinations of type (malignant versus nonmalignant) and location (skin, bone, etc.). Small subsample sizes again limit detailed examination of these effects.
Other conditions found related to cognitive functioning included osteoporosis and hip fractures and sensory problems. Osteoporosis and hip fractures were predictive of earlier decline on Word Fluency. Hearing impairment was associated with an increased risk of experiencing Verbal Meaning decline but was associated with better performance and later decline on Space. Vision difficulties predicted later age at onset of decline for Verbal Meaning and Space.
Thus far, we have dealt with issues concerned with the relation of disease and cognition and with the cognitive and environmental influences that may affect morbidity. We now turn to the issue of the role of cognitive and sociodemographic risk factors in the prediction of mortality.
Lower cognitive performance in older adults has long been associated with earlier mortality (cf. Berg, 1996; Bosworth & Schaie, 1999; Maier & Smith, 1999). Although the evidence is fairly consistent, it has not been clear whether this association holds true across different abilities or whether the association is affected by gender, education, or age.
Three classes of theories have been offered for the association between cognitive decline and mortality. The first class includes the implication of early cognitive impairment, general systems decline, and disease (Swan, Carmelli, & LaRue, 1995). All of these might be detected early by cognitive tests because these tests assess the integrity and efficiency of the central nervous system (Muscovith & Winocur, 1992).
The second type of theory suggests that decline in cognitive function may have an indirect role in affecting general health. Intact memory, for example, is required to follow any treatment regimen. Also, decline in perceptual speed may reflect the primary aging of the central nervous system, which reduces adaptive capability in general, hence increasing susceptibility to a number of internal and external events that increase the probability of death (e.g., Swan et al., 1995).
A third class of theories would suggest that low cognitive test performance may be an indicator of general cognitive disturbance that can be attributed to diseases that affect the brain only indirectly. Such diseases would include cardiovascular problems as well as metabolic or toxic disturbances.
We have previously dealt with these issues in the context of the popular construct of terminal decline and the issue of participant attrition (e.g., Cooney, Schaie, & Willis, 1988). Given the 7-year interval of assessment, we cannot directly deal with short-term decline occurring shortly before death. Rather, we deal with the question of the extent that cognitive functions can predict mortality from cognitive and other variables available to us at the last assessment point prior to death.
The analyses reported in this section address three questions: First, we ask whether there is a general relationship between cognitive function and risk of mortality when controlled for gender, education, and age. We hypothesized that Psychomotor Speed would most likely show the highest relationship with mortality. Second, we ask whether the relationship between cognition and mortality operates differentially by age. Third, we consider whether magnitude of decline over the last two measurement points available to us indicates increased risk of mortality (cf. Bosworth, Schaie, & Willis, 1999a; Bosworth, Schaie, Willis, & Siegler, 1999).
These analyses include the 601 deceased SLS participants (342 males and 259 females) whose date of death could be established by December 1995. A control group of 609 survivors (296 males and 313 females) who were last assessed in 1991 was selected whose members were within 2 years of age and within 1 year of education of the decedents.
The following analyses included the basic PMA battery (Verbal Meaning, Spatial Orientation, Inductive Reasoning, Number, and Word Fluency), immediate and delayed Verbal Memory, perceptual speed (Finding A’s, Identical Pictures, and Number comparison), and the Psychomotor Speed measure from the Test of Behavioral Rigidity. Demographic information included participants’ age, gender, and years of education.
The SAS PHREG procedure was used to compute Cox’s semiparametric proportional hazard models assessing the effects of level of cognitive performance and cognitive decline on the probability of mortality. It is then possible to compute odds ratios that will represent probability of mortality given level and decline of cognitive performance.
Hierarchical models were estimated successively introducing variables hypothesized to confound the association between cognitive performance and mortality. Groups of independent variables were introduced in the following order: (a) with cognitive performance at last measurement categorized as quartiles as the only variable in the model, (b) with the addition of sociodemographic factors (age, gender, education), and (c) Psychomotor Speed. Psychomotor Speed was treated as a covariate for all other cognitive measures. However, Psychomotor Speed was included in the model as an independent predictor of mortality. Dummy codes represented each quartile of cognitive performance, and the reference point for each individual cognitive ability was performance in the top 25th percentile. The use of dummy codes followed the conventional practice in previous studies using self-rated health (Bernard et al., 1997; Idler & Kasl, 1991). Age also was categorized by quartiles, and the youngest quartile was used as the reference age category. Education was categorized by: less than 11 years of education, 12 to 15 years of education, and 16 years of education or more. The last category was used as the reference category for education.
Because we were operating in a Thurstonian framework, we were interested in ability-specific change. Hierarchical analyses were first conducted for level of cognitive performance for all participants. Then, decline in cognitive level, as assessed by difference scores, was assessed in a similar hierarchical fashion as the level models. Finally, analyses were conducted in a similar hierarchical fashion except that they were stratified by age; a median split was used to categorize the old-old (≤74 years of age) and young-old adults (≥73 years of age).
Examining the unadjusted association between cognition and mortality, participants in the lower 50th percentile on Spatial Ability, Delayed Recall, Immediate Recall, Psychomotor Speed, and Identical Pictures had a significantly higher risk of mortality than those in the top 25th percentile on these measures. Individuals who performed in the lower 25th percentile of Verbal Meaning and Numerical Ability had a significantly higher risk relative to their reference categories. Significant odds ratios ranged from 1.38 to 2.32 for the lowest percentile of performance (table 10.1). The odds ratios for many of the cognitive abilities decreased after adjusting for sociodemographic factors.
The excess risk of low levels of Verbal Meaning, Numerical Ability, Spatial Ability, and Psychomotor Speed ceased to be a significant predictor of mortality once demographic factors were adjusted. However, low levels of Verbal Memory (Immediate Recall, Delayed Recall) and a measure of perceptual speed (Identical Pictures) remained significantly related to a higher probability of mortality after adjusting for sociodemographic factors. However, once Psychomotor Speed was introduced into the model, only Identical Pictures remained significantly related to mortality.
The influence of cognitive decline on mortality/survival is shown in table 10.2. Greater magnitude of decline was related to increased risk of mortality. Participants who had the greatest magnitude of 7-year decline in Verbal Meaning, Spatial Ability, Inductive Reasoning, and Psychomotor Speed had a 63%, 113%, 50%, and 97% greater likelihood, respectively, for mortality relative to their reference category even after adjusting for sociodemographic factors and psychomotor speed.
We next stratified participants by median age. The oldest adults were 75 to 95 years of age or older, and the younger adults were 34 to 74 years of age. Level of performance on Identical Pictures remained a significant predictor of mortality for both age groups. However, the relationship appeared to be more substantial among older adults, with those individuals in the lowest 50th percentile having a significant risk of mortality. Among the younger sample, only those in the bottom quartile had a significant risk of mortality compared with their reference categories. Examining the odds ratios, younger adults who had significant declines in Spatial Ability and/or Inductive Reasoning had a greater risk of mortality than older adults. Older adults who had significant declines in Verbal Meaning and Psychomotor Speed had greater risks of mortality than the younger group (see table 10.3).
TABLE 10.1. Odds Ratios and Confidence Intervals for Level of Performance
Level of cognitive function was related to subsequent mortality, but the relationship of performance level and early mortality varied across abilities. Lower levels of crystallized abilities, spatial abilities, Verbal Memory, and perceptual speed as well as declines in Verbal Meaning, Spatial Ability, Inductive Reasoning, and Psychomotor Speed were significant predictors of mortality. Level of cognitive performance was related to mortality, but Psychomotor Speed mediated the relationship for many of these cognitive variables. Seven-year decline in cognitive performance was a better predictor of subsequent survival than level of performance at last measurement, even after adjusting for demographic factors and psychomotor speed. Overall, the pattern of results in this study indicates that not all cognitive abilities are equally affected by terminal change or directly associated with mortality after adjusting for age, gender, level of education, and Psychomotor Speed.
TABLE 10.2. Relative Risk Ratios and Confidence Intervals for 7-Year Change Before Death
The relationship of Psychomotor Speed with mortality was one of the most consistent findings in this study, and the findings regarding Psychomotor Speed are similar to those reported by Birren (1968) and Kleemeier (1962): A significant decline in Psychomotor Speed is a risk factor for subsequent mortality. This study also indicated that Psychomotor Speed may mediate some of the relationship observed between cognitive function and mortality. There is growing evidence that perceptual speed is an important contributor to the decline in cognitive performance that occurs in many tasks (Earles & Salthouse, 1995; Salthouse, 1993). Perceptual speed has been suggested as a general mechanism underlying age-related differences on many cognitive tasks, including reasoning and spatial cognition (Mayr & Kliegl, 1993; Salthouse, 1985).
TABLE 10.3. Odds Ratios and Confidence Intervals for Level of Performance by Age Group
Only a few studies have analyzed memory in relation to survival. Our finding that Verbal Memory is related to survival agrees with the results of Deeg, Hoffman, and van Zonneveld (1990), Small and Bäckman (1997), and the Gothenburg Longitudinal Study (Johansson & Berg, 1989).
Verbal Meaning is known to be less sensitive to age than most other cognitive abilities, and vocabulary performance may be sensitive to the disease processes primarily responsible for death in the present deceased participants (Bosworth & Schaie, 1999; Cooney et al., 1988).
There was also evidence that decline in Spatial Ability, a measure of visualization ability, was related to mortality. Few studies have included Spatial Ability and examined its relationship with mortality. Survivors had significantly higher mean levels of Spatial Ability than decedents but did not show significant decline over time (Bosworth & Schaie, 1999).
The association between cognitive performance and subsequent mortality in younger adults (≤74 years) differed from older adults. Declines in age-related cognitive functions (i.e., Spatial and Reasoning abilities) were more of a risk factor among younger adults, whereas Verbal Meaning (crystallized ability) and two types of perceptual speed represented greater risk factors among adults older than 75 years. A possible explanation for this association between cognitive function and mortality between the two age groups might be found in the different age-related trajectories of these variables. That is, adults tend to decline earlier on visualization and fluid abilities than on crystallized abilities. Nevertheless, this relationship among age, cognition, and mortality may explain why past research has reported that certain age groups may appear more likely to exhibit terminal change than others (Bosworth & Schaie, 1999; Cooney et al., 1988).
We have also investigated the reverse side of the coin, more specifically, whether we can demonstrate that our measures of cognitive abilities and cognitive style might be useful in predicting the occurrence and onset of physical disease (Maitland, 1993).
Our first analysis (Maitland & Schaie, 1991) involved a sample of 370 subjects (169 males, 201 females; mean age 66.5 years) who were studied between 1970 and 1984. Two logistic regression models were tested. In the first model, all subjects diagnosed as having CVD were contrasted with normal controls. Age, gender, education, Attitudinal Flexibility, and decline in Spatial Ability were identified as significant predictors of the CVD condition [χ2(13) = 80.96, p < .001]. When examined separately by gender, age and Attitudinal Flexibility were significant for the females [χ2(12) = 39.02, p < .001], whereas age, Attitudinal Flexibility, and education were significant predictors of CVD for males [X2(12) = 44.67, p < .001].
The second model included only those subjects who developed a CVD condition after the second data point. The prediction model for recent occurrence of conditions implicated age, Attitudinal Flexibility, decline in Psychomotor Speed, Spatial Ability decline, and decline in Reasoning Ability as significant [χ2(13) = 43.14, p < .001]. Separately by gender, age, decline of Reasoning Ability, and decline in Word Fluency were significant for males [χ2(12) = 29.13, p < .01], whereas Attitudinal Flexibility was the only significant predictor for females. Lifestyle predictors of change in activity level and smoking surprisingly did not contribute to the prediction of recent CVD in this particular sample.
The most interesting finding from this analysis is the establishment of an association between the cognitive style measure of Attitudinal Flexibility and the presence of CVD regardless of time of onset of the disease. Explanations for these findings would be that very rigid participants may already have dropped out of the study prior to the onset of CVD or that rigid individuals might be more attentive to their health behaviors and thus more likely to avoid disease outcomes.
In a second analysis, as part of a master’s thesis, Maitland (1993) extended these findings to the possible effects of behaviors on the occurrence of arthritic disease and applied survival analysis to the prediction of the occurrence of both cardiovascular and arthritic disease. In this sample, women were more likely than men to have arthritis, but measures of social status, obesity, and smoking were not significant predictors. Greater Attitudinal Flexibility and decline of Psychomotor Speed were associated with preexisting arthritis, whereas declines in Spatial and Reasoning abilities were predictive of the occurrence of arthritis.
Life table and survival analysis methods were used to predict age of onset of cardiovascular and arthritic disease. Declines in Verbal and Reasoning abilities predicted later occurrence of first diagnosis of CVD; high levels of Spatial Ability were associated with earlier onset. For the arthritic participants, greater declines in Verbal and Spatial abilities, as well as higher level of education and Psychomotor Speed base levels, were associated with later ages of disease onset; decline in Psychomotor Speed and low baseline function on Spatial Ability predicted earlier onset.
Another influence that might affect the onset, course, and outcome of illness is the broad spectrum of social support. In this study, we investigated the extent to which both actual social networks and perceived social environments relate to health outcomes and service utilization in the middle to older years. The study of the relationship between social support and disease at older ages is important because the elderly are at greatest risk for almost all morbidity and mortality outcomes.
In the following analyses, we address these issues by applying both variable-oriented and individual-oriented approaches to contribute further information on the relation between social environment and health (cf. Bosworth, 1994; Bosworth & Schaie, 1997).
The subsample used in the following analyses consists of 387 participants (173 men and 214 women) who participated in the sixth cycle (1991) data collection; the participants had a mean age of 58.3 years (range 36 to 82 years) for those whose medical records were abstracted. Their mean health care characteristics in 1991 were as follows: primary care visits, M = 3.6, SD = 1.6, range 2–10; hospital days, M = 1.7, F = 5.3, range 0–50; medication usage, M = 1.9, SD = 2.0, range 0–11; total annual health care cost, M = $3,203, SD = $2,270, range $465–$11,835.
Included in this analysis as predictors were structural social network measures from the Life Complexity Index and perceived social environment measures from the family environment scales (see chapters 3, 11, and 16). Health outcome variables included number of diagnoses, number of disease episodes, number of hospital visits over a 1-year period (1991), the number of medications used regularly for at least 1 month, and an estimate of total annual health expenditures for each study participant.
Structural equation modeling with LISREL VIII (Jöreskog & Sörbom, 1993) was used to evaluate the model shown in figure 10.1. As an alternate approach, cluster analysis using the agglomerative method with cosine similarities was used to cluster individuals on the constructs of perceived social environment and social networks (cf. Blashfield, 1976).
Confirmatory factor analysis of the model in figure 10.1 showed a quite satisfactory fit, [χ2 (199, N = 387) = 329.25, p < .001; GFI = .936; RMSEA = .046]. Increased social networks were negatively correlated with hospital visits, and being unmarried was negatively correlated with higher estimated total health care costs, outpatient costs, and number of primary care visits. Increasingly favorable perceptions of social environment were negatively correlated with lower estimated primary care costs. Increased age was positively correlated with all health care variables; lower socioeconomic status was positively correlated with increased number of disease episodes, medications used, and estimated health care utilization.
As shown in figure 10.1, females had more physician visits, used more medications, and had more primary care visits. The magnitude of perceived social relations was negatively related to number of hospital and primary care visits. Advanced age predicted greater health problems.
FIGURE 10.1. Structural model for the relation among structural support, demographic factors, and health care outcomes. SES, socioeconomic status; SS, social support.
Differences between married and unmarried adults were examined using multiple group structural equations modeling. The model was the same as for the total group except for the exclusion of marital status. Married women had more disease episodes, greater medication usage, and more outpatient care visits than married men. For married persons, age was negatively related to all health variables; structural social networks were negatively related total health care costs, outpatient costs, and primary care visits. For unmarried persons, age and the magnitude of perceived social environment were negatively related to disease episodes. The models separately by marital status are shown in figure 10.2.
FIGURE 10.2. Structural model for the relation among social support, demographic factors, and health care outcomes by marital status: (a) married; (b) unmarried. SES, socioeconomic status; SS, social support.
Four clusters were retained for each of the structural social network and perceived social network data. It was found that the cluster that had significantly lower levels of social interaction also had the highest estimated health care costs even though they had higher church attendance than the other clusters. The two clusters that had the most positive health outcomes were also the younger members of the sample.
With respect to the perceived social environment clustering, it was found that the cluster of study participants with the most health problems was characterized by low levels of education and low levels of intellectual-cultural and recreational orientations in their current families. These individuals could be described as socially isolated. The two clusters with the least health problems had high levels of family cohesion and expressiveness.
These analyses confirm that married persons tend to have fewer physical health problems than unmarried persons, and that higher socioeconomic status is also associated with better health. On the other hand, we account for relatively little variance in health outcomes from more extensive structural support systems. However, perceived support appears to be important in those individuals who are married as well as for those clusters of individuals who are experiencing more serious health problems. These findings underline the importance of developing typologies of subsets of individuals who exhibit different causal mechanisms for outcomes of interest.
We have been able to obtain access to the medication use of our study participants because most of our participants receive their prescription medications from the collaborating HMO. These data began to be automated in the 1980s, and a 4-month window of prescriptions was examined prior to the date of our test administrations. We have not yet been able to develop appropriate models of how we should correct our cognitive test findings for medication use. However, we have been able to begin initial studies of the impact of cognition on medication use. In this section, we report findings on the effects of cognition on understanding the purpose of one’s medication at different ages (cf. Bosworth, Schaie, & Willis, 1999b).
Although a majority of older adults understand the purpose of their medication, many individuals do not for various reasons. A study of a large sample residing in a rural community found that as many as 90% of all reported drug purposes were considered accurate (Semla, Lemke, Helling, Wallace, & Chrischilles, 1991). But other studies have found smaller percentages of individuals who understood the purpose of their medication (e.g., Caskie, Willis, Zanjani, & Schaie, 2006; O’Connell & Johnson, 1992).
Much of this research is problematic because of a lack of uniform description of what constitutes patient knowledge, and operational definitions of patient knowledge vary significantly in published research. Medication knowledge in this study is defined as a general understanding of the purpose of the medication.
Considerable attention has also focused on the relationship between cognitive function and compliance (Morrell, Park, & Poon, 1989; Park, Willis, Morrow, Diehl, & Gaines, 1994). A major cause of noncompliance may be related to cognitive function, particularly among older adults. Older adults may have difficulty understanding what they are to do with their medication or, if they do understand, remembering what to do with it.
Despite the attention on the relationship between cognitive function and compliance, few have examined the relationship between cognitive level of functioning and simply understanding the therapeutic purpose of one’s medication. A basic understanding of the therapeutic function of prescription medications is necessary, but not sufficient, to ensure compliance in medication use.
There remains a paucity of information regarding the correlates of the knowledge of therapeutic purpose of medications used among adults, and research is lacking on identifying predictors of misidentification of medication purpose across the life span. In addition, despite the information on the ramifications of medication noncompliance, there is a lack of information on understanding the purpose of one’s medication and its relation to health outcome and service utilization.
The SLS provides an opportunity to examine individuals’ therapeutic understanding of their medication for a large, sociodemographically diverse group of adults across the life span and to examine whether knowledge of one’s medication mediates the relationship between cognition and health status and service utilization.
This study was designed to explore four issues: First, we investigated medication usage patterns across the life span of adults who are members of an HMO. Second, we examined participants’ knowledge of the therapeutic function of their prescription medications. Third, we examined the relationship of level of cognitive abilities and sociodemographic variables with understanding the purpose of one’s medication. Fourth, we were interested in whether the understanding of the therapeutic purpose of one’s medication mediated the relationship between cognition and health outcome and service utilization.
Prescription medication use was examined in a sample of 816 males and 986 females (N = 1802) with a mean age of 59.7 years (range 22–94 years) at the time data were collected in 1991. The perceived therapeutic purpose (PTP) of CVD medication was examined in a subsample of 110 males and 82 females (n = 192) with a mean age of 68.99 years (range 35–92 years). The PTP of nonsteroidal anti-inflammatory medications (NSAIDs) was examined in a subsample of 60 males and 56 females (n = 116) with a mean age of 66.37 years (range 31–92 years).
All participants were asked to bring all medications taken regularly for at least 1 month to the session at which their cognitive function was assessed. The name of the medication, dosage level, subjects’ perceived purpose for the medication, and physicians’ instructions were recorded by the examiner. Each medication was assigned a drug code based on the American Hospital Formulary Service (1991) coding scheme. CVD and NSAID medication categories were examined because they are two of the most commonly prescribed types of drugs (Baum, Kennedy, Knapp, Juergens, & Faich, 1988).
The psychometric tests used in this study included three dimensions of mental abilities: Verbal Memory, Inductive Reasoning, and Verbal Comprehension. These were measured by the PMA Verbal Meaning and Reasoning tests and by the Immediate and Delayed Recall tests. Various demographic and personal information was extracted from the Life Complexity Inventory, including age, family income, and education (see chapter 3 for details).
In 1993, participants rated their health on a 6-point Likert scale (1 = very good to 6 = very poor). Individuals also reported the number of doctor visits and hospital stays for 1993. Number of medications in 1991 was also considered. Estimated total care costs for 1991 were based on a Chronic Disease Score (CDS). The CDSs were based on empirically derived weights based on age, gender, and pharmacy utilization of the HMO’s pharmacies. These weights were then used to calculate a predicted score for total care costs (Clark, Von Korff, Saunders, Baluch, & Simon, 1995).
Individual PTPs were classified according to whether they corresponded to the actual therapeutic purpose (ATP) of CVD and NSAID medications. The ATPs of medications listed by participants were often broad responses. If the PTP did not correspond with the ATP of the medication, the response was categorized as inaccurate. For example, if an individual used a CVD medication and reported the PTP as “hypertension,” the response was coded as correct. If an individual used CVD medication but reported it was used for the treatment of gout, the response was coded as incorrect. To avoid overrepresentation of individuals who may take two of the same types of medications, only the individual’s first reported medication was included in analyses and then used to determine the sociodemographic variables and cognitive abilities that may be risk factors for misunderstanding the purpose of one’s medication.
First, age, gender, income, and education were modeled. Then individual cognitive abilities (i.e., PMA Verbal Meaning, PMA Inductive Reasoning, Immediate Recall, and Delayed Recall) were added to assess the magnitude of increased prediction above the influence of sociodemographic factors. The final model tested assumed that accurately reporting the purpose of one’s medication is dependent first on sociodemographic variables (i.e., age, gender, income, and education), then on Verbal Ability, Inductive Reasoning, and finally Verbal Memory.
Multiple measures of Verbal Meaning, semantic memory, and Inductive Reasoning were treated as latent variables with three to four indicators. Age, income, education, and gender were entered as manifest variables; they were unique factors in and of themselves. The observed dependent variables were treated as latent variables to assess whether they were impacted by cognitive functioning, gender, education, income, age, or medication knowledge.
Participants were classified by frequency of drug use. Given the low frequency of multiple drug use, subjects who used seven or more drugs were combined. Medication usage was examined across ten 7-year age cohorts that corresponded to the data collection intervals of this study. Analyses of variance and covariance were used to assess the difference between cohorts and medication use as well as the relationship of medication use and demographic variables, controlling for age.
Lack of any current medication usage was reported by 41% of the sample (n = 731). The average number of medications reported by the remaining subjects was 1.68. There was a significant difference in number of drugs used by age cohort [F(9, 1756) = 23.21, p < .001]. Medication usage increased significantly for successively older age cohorts.
A significant main effect for gender when controlling for age was obtained for the number of drugs consumed [F(1, 1761) = 127.33, p < .001]. Females used significantly more drugs (M = 1.92; SD = 2.03) than males (M = 1.31; SD = 1.77). Participants were then split into three levels of education: high school and less, college level (12–16 years), and graduate level (more than 16 years of education). Subjects with different levels of education also differed in the number of drugs consumed when controlling for age and gender [F(4, 1759) = 68.93, p < .001]. Individuals with less than a high school education used more drugs (M = 1.92), followed by those with some college education (M = 1.67); least use was by those with some graduate education (M = 1.33). A significant difference was also found for income level when controlling for age [F(2, 1763) = 101.23, p < .001]. Those with incomes lower than $30,000 used more medications (M = 2.01; SD = 2.10) than those with higher incomes (M = 1.4; SD = 1.76).
Of those who reported the use of a CVD medication, 90% accurately reported its therapeutic purpose. There was a significant negative correlation between age and accurately understanding the purpose of one’s CVD medication (r = −.18, p < .01); increased age was associated with greater understanding. Those individuals who reported accurate responses were 7 years older on average (accurate M = 69.56 years; inaccurate M = 62.84 years) and had used the medication for approximately 21 months longer (accurate M = 62.81 months; inaccurate M = 42.06 months). Individuals who reported inaccurate responses used fewer medications on average (M = 2.56) than those who responded accurately (M = 3.44).
Verbal Ability, Inductive Reasoning, Immediate Recall, and Delayed Recall failed to be significant risk factors for misunderstanding the purpose of one’s medication. However, individuals who were younger and male and had lower levels of education and income were more likely to be inaccurate (Caskie, Willis, Zanjani, & Schaie, 2006).
The PTP for 84% of those who reported using NSAIDs was correct. A significant negative correlation between age and accurately understanding the purpose of one’s NSAIDs was observed (r = −.30, p < .001); increased age was associated with greater understanding. In addition, length of use of an NSAID was negatively correlated with knowing the purpose of the NSAID (r = −.25, p < .001). Those individuals who reported accurate responses used the medication for approximately 40 months longer (accurate M = 62.77 months; inaccurate M = 23.22 months) and were 11 years older on average (accurate M = 68.07 years; inaccurate M = 57.28 years). Individuals who reported inaccurate responses used fewer medications on average (M = 2.71) than those who responded accurately (M = 3.09).
Age and length of use were significant predictors of understanding the purpose of NSAIDs. Once sociodemographic factors were controlled, performance on Verbal Meaning and Inductive Reasoning did not add significant risk of inaccuracy. However, both Immediate and Delayed Recall accounted for differences in knowing the purpose of one’s NSAIDs. In fact, low performance on Delayed and Immediate Recall placed subjects at an increased risk of 24% and 19%, respectively, in misinterpreting their medication’s purposes.
This study found significant age differences in medication use, with older adults using more medications than younger adults. However, lower medication usage than observed in other studies was found (e.g., Lamy, Salzman, & Nevis-Olesen, 1992). The low levels of medication usage reported here may be because of three possible reasons: First, identification of low levels of medication usage could be an artifact of the method by which reports of drug usage were obtained. Although subjects may have been prescribed and may have purchased many other drugs, they were asked to report only those drugs that they were using regularly for at least 1 month prior to the survey. Second, low levels of medication use may be a function of the philosophy of practice of the HMO. HMOs emphasize ambulatory and preventive care. Members of this HMO may therefore represent an unusually healthy group who require less medication (Gold, Jolle, Kennedy, & Tucker, 1989). Third, because this population, on average, is relatively well educated and financially affluent (Schaie, 1996b), better health practices and lifestyles may have led to lower medication use.
Women were found to use more medications than men, as has also been shown in previous research. Women are more likely to encounter types of illness, problems, and conditions that are amenable to drug therapy (e.g., hypertension and menopause). Gender differences in the number of medications used may also result from differences in health behaviors. Women are more likely to seek medical attention then men; women accounted for about 60% of all office visits and for a majority of the visits in each age group except for the youngest (U.S. Department of Health and Human Services, 1992).
An examination of PTP indicated that individuals who reported using CVD and NSAID medications were generally correct in reporting the purpose of their medications. However, no evidence was found for the hypothesis that community-dwelling older adults have lower congruence between PTP and ATP of one’s medication, specifically CVD and NSAID medications, than do younger adults. In fact, they were more accurate than their younger peers.
Despite the increased number of medications and likely increased complexity of prescribed regimen, older adults understood the purpose of their medication better than younger adults. Similar findings have been obtained in the compliance literature; age had no adverse effects on compliance (e.g., Lorenc & Branthwaite, 1993).
We did find differences in the length of time individuals were using their medications. There was an approximately 40-month difference between accurate and nonaccurate NSAID users and a 21-month difference between accurate and nonaccurate CVD medication users. As hypothesized, the longer an individual has been taking a medication, the less likely the person is to forget the purpose of it.
There was evidence that a lack of understanding of the therapeutic purpose of one’s medication does have health consequences in this study. In the CVD model, medication knowledge was predictive of increased health care costs, decreased self-reported rating of health, and increased hospital visits. In contrast, in the NSAIDs model, a lack of medication knowledge was associated only with increased health care amount.
This chapter discusses the manner in which physical disease may affect the maintenance of function and how disease onset might be affected by behavioral antecedents and concomitants. A series of studies is described that relates the role of disease and intellectual functioning in the SLS. The first study implicated CVD as associated with earlier onset of decline of intellectual functioning. Decline in Psychomotor Speed was also identified as a risk factor for the occurrence of CVD. However, when hypertensives were disaggregated, they did not show any increased risk of cognitive decline.
The second study, concerned with the structural relationship between disease processes and maintenance of intellectual functioning, also implicated cardiovascular and musculoskeletal conditions as leading to excess risk for cognitive decline. Surprisingly, the diagnosis of neoplasms was negatively correlated with cognitive decline.
A more comprehensive analysis has extended these findings to the disease categories of diabetes, neoplasms, and arthritis as well as to a measure of overall health. Overall health status as measured by number of chronic disease diagnoses was found to have a negative effect on the abilities of Verbal Meaning, Number, and Word Fluency. The more serious CVD conditions (atherosclerosis and cerebrovascular disease) have generally negative influences, whereas benign CVD (including uncomplicated hypertension) has a slightly positive influence on cognition. Diabetes was correlated positively with intellectual functioning, perhaps reflecting the survival of those who can manage this condition more intelligently. The onset of arthritis was associated with timing of cognitive decline in Inductive Reasoning and Spatial abilities. When neoplasms were disaggregated into benign and malignant forms, the malignant cases were associated with cognitive decline.
Finally, we examined cognitive level at last assessment as well as 7-year change in performance as predictors of mortality. Lower levels of performance on crystallized abilities, Spatial Ability, Verbal Memory, and perceptual speed, as well as 7-year declines on Verbal Meaning, Spatial Orientation, Inductive Reasoning, and Psychomotor Speed, were associated with earlier mortality. However, decline in fluid abilities was a better predictor of mortality in the young-old; declines in crystallized abilities and Psychomotor Speed were better predictors of mortality in the old-old.