Cognitive Correlates of Human Brain Aging: A Quantitative Magnetic Resonance Imaging Investigation

Subjects were selected from participants in the Cardiovascular Health Study (CHS), an ongoing multicenter, population-based observational study of 5,888 adults 65 years and older, including 2,495 men and 3,393 women.^27–29 The major goal of the CHS is to identify risk factors related to the development and course of coronary heart disease and stroke in individuals living independently in the community. After providing informed consent, subjects undergo extensive clinical evaluation (home interview and physical examination) and laboratory testing (including brain MR imaging; see below) at baseline and at annual follow-up assessment. Additional details of the CHS have been published.²⁷

A detailed description of recruitment for the CHS subjects has been published.^28,29 For the present study, we identified from the CHS cohort a sample of 500 subjects recruited from two CHS sites (Pittsburgh, PA, and Hagerstown, MD) who gave written consent to participate in an ancillary investigation of cognitive functioning and aging (measures included the MMSE and a cognitive test battery; these data will be the subject of a future report).³⁰ All available subjects from these two sites in whom brain MR imaging was performed within 1 year of this cognitive testing were screened for inclusion in this study. We subsequently excluded from this cohort a total of 170 subjects for one or more of the following reasons: not right-handed (subjects were determined to be right-handed if they used their right hand to write, throw a ball, and brush their teeth),³¹ lifetime history of any psychiatric illness (including dementia) or of any illness or injury referable to the brain (per the CHS clinical evaluation described earlier), incomplete cognitive test data, incomplete MR imaging data (e.g., scan artifact, missing slices), or MR images with structural abnormalities (cortical infarct, n=5; hydrocephalus, n=1; tumor, n=1; markedly thickened calvaria, n=1).^32–34 For the present investigation, we excluded 10 additional subjects with scores less than 24 on the MMSE (administered per the ancillary investigation of cognitive functioning described above) to further ensure the absence of cognitive impairment. A cutoff score of 24 on the MMSE is a well-accepted measure of significant cognitive impairment consistent with dementia in the elderly.³⁵

The final sample (Table 2) consisted of 320 subjects, 198 women (62%) and 122 men (38%), ranging in age from 66 to 90 years. This sample was included in an earlier study of the relationship of education to brain size in normal aging.³⁶ The sample was similar to the CHS population as a whole with regard to age (CHS mean±SD=72.77±5.61 years), sex distribution (CHS=59% female), and years of education (CHS mean±SD=12.35±3.10 years). Years of education were determined by subjects' responses to a structured interview and defined as the number of years of full-time education normally required to reach the subject's reported final level (e.g., 10th grade=10 years; bachelor's degree=16 years). Subjects with a high school equivalency diploma (GED) were credited with 12 years. Of the 320 subjects, 226 (71%) were taking medications for one or more medical conditions (see Table 2). No subject was taking medication known to affect brain size (e.g., steroids). Additional subject characteristics are given in Table 2.

Cognitive function was assessed by using a one-and-a-half- to two-hour battery of standard neuropsychological tests. This assessment was completed at the CHS annual evaluation 2 and 3 years after CHS baseline (i.e., in 1991 and 1992). The battery included measures of attention and information processing speed, language, memory, and visuospatial ability. For purposes of the present study, the following six scores were derived from the test battery:

Trails B: Scores (time per connection) for part B of the Trail Making Test,³⁷ a measure of psychomotor speed and working memory. Time per connection was used as the measure of interest in order to distinguish among subjects who failed to complete the test in the allotted 300 seconds.

Fluency SP: Sum of scores from two verbal fluency tasks (number of words beginning with p and number of words beginning with s generated in 60 seconds), a measure of language and executive function.³⁸

Boston Naming: Total confrontation naming score (spontaneously correct plus correct following a stimulus cue) from the Boston Naming Test,³⁹ a language measure.

Verbal delayed memory: Comprises the delayed recall of the two short stories (Logical Memory) together with the weighted raw delayed recall score (Verbal Paired Associate Learning task) of the Wechsler Memory Scale–Revised,⁴⁰ a widely used standard test of explicit memory.

Visual delayed memory: Comprises the delayed recall score from the four Visual Reproduction Figures together with the weighted raw delayed recall score from the Visual Paired Associate Learning test, both of the Wechsler Memory Scale–Revised.⁴⁰

Block Design: Raw score on the Block Design subtest of the Wechsler Adult Intelligence Scale–Revised (WAIS-R),⁴¹ a measure of visuospatial/constructional ability.

As noted earlier, brain MR imaging was performed on all subjects as a part of their participation in the CHS. The standardized CHS brain MR imaging acquisition protocol has been previously described.⁴² MR imaging was performed on either a 1.5-tesla scanner (General Electric, Milwaukee, WI; n=244) or a 0.35-tesla scanner (Toshiba; n=76) at one of two CHS field centers (Pittsburgh and Hagerstown, respectively). Head position was oriented in the scanner and was stabilized during the scanning procedure by the use of Velcro straps and foam head supports. To establish slice orientation, the first scanning sequence consisted of a T₁-weighted sagittal series (TR repetition time=500 ms; TE echo time=20 ms; thickness=5 mm; gap=0 mm; matrix=128×256) centered at the midline to define the anterior commissure–posterior commissure (AC-PC) line. Then, a second series of proton density images (TR=3,000 ms, TE=30 ms, flow compensated) and T₂-weighted (TR=3,000 ms, TE=100 ms, flow compensated) images was obtained (thickness=5 mm; gap=0 mm; matrix=256×192; number of excitations, one-half (1 on the 0.35-T scanner), oriented parallel to the AC-PC line, and extending from the vertex to the skull base. A third series consisting of T₁-weighted axial images (TR=500 ms, TE=20 ms) was then obtained (thickness=5 mm; gap=0 mm; matrix=256×192; number of excitations=1), oriented parallel to the AC-PC line and extending from vertex to skull base. Images were stored on 9-track magnetic tape.

For the present study, the brain images were transferred from magnetic tape to read/write magneto-optical disks. Data were analyzed on a workstation (Power Mac 8100, Apple, Cupertino, CA) with high-resolution color graphic monitor. The measurements of regional brain size were made on the recalled T₁-weighted axial images by one of two trained technicians blinded to all subject characteristics. Window center settings were first standardized to ensure precision in boundary detection.⁴³ Structures were identified with the help of brain and MR imaging atlases^44,45 and then measured with a combination of computer-assisted edge detection and manual tracing, using graphic analysis software (MedVision, Imnet/Evergreen Technologies, Castine, ME). The area (in square centimeters) within the outline was calculated automatically; volume (in milliliters) was determined by multiplying the area by the slice thickness and summing over the multiple slices in which the structure appeared (described later).

The following regions were defined for volume measurement:⁴⁶

Intracranial volume (IV) was defined by the internal surface of the diploë⁴⁷ and measured in every slice between the vertex and the superior border of the midbrain; approximately 12 to 15 slices per subject were measured. Intracranial size could not be reliably measured inferior to this level because of the presence of structures such as the globes and sinuses. As such, this measure is an underestimate of the true total intracranial volume. There was no significant correlation between age and intracranial volume.

Cerebral hemisphere volume was measured in every slice between the vertex and the skull base (approximately 18 to 20 slices per subject). Ventricular volumes were excluded from this measurement.

Peripheral (sulcal) CSF volume was a calculated value derived by subtracting the cerebral hemisphere and ventricular volumes from the intracranial volume, for each slice on which intracranial volume was measured. As such, this measure is an underestimate of the true total peripheral CSF volume.

Lateral (Sylvian) fissure CSF volume provided an indirect estimate of atrophy of the temporal lobe, as well as of the frontal and parietal lobes. The lateral fissures were measured in each slice on which they were present, beginning at the level of the foramen of Monro. When the lateral fissure communicated freely with the peripheral CSF, the anterior boundary of the fissure was defined by a horizontal line connecting the anterior tip of the temporal lobe to the medial temporal region.

Lateral ventricle volume was measured in each slice on which lateral ventricles were present. We also measured the various subregions of the lateral ventricles, including the body, the frontal horns, the posterior horns, and the temporal horns.

Third ventricle volume was measured in each slice beginning at the level of the foramen of Monro and extending inferiorly to the superior border of the midbrain (approximately 3 to 4 slices per subject).

Extensive reliability studies of our measurement techniques have indicated that our measurements of these regions are highly reliable.⁴⁷ On the basis of a randomly selected sample of 10 brains from the current study,⁴⁶ intraclass correlation coefficients for interrater reliability of the two raters ranged from 0.85 (for small regions such as the third ventricle) to 0.99 (for large regions such as the cerebral hemisphere). Similarly, intraclass correlation coefficients for intrarater reliability ranged from 0.84 to 0.99.

Exploratory data analyses (boxplots, quantile-quantile plots, density plots, and resistant estimators of location and spread) were undertaken to review the distributions of the variables and identify distribution anomalies, outliers, extreme values, and floor or ceiling effects. These analyses revealed that all of the variables showed reasonably Gaussian distributions except for Trails B, which exhibited extreme right-skewness. A logarithmic transformation of Trails B yielded a nearly Gaussian distribution. All references to Trails B that follow are to the log-transformed version.

As previously reported,³⁶ age was highly significantly related to regional brain volumes. Adjusting for the effects of IV and sex, increased age was associated with smaller cerebral hemisphere volume (b=–2.46, P<0.003) and with larger volumes of the lateral fissures (b=0.13, P<0.0001), the lateral ventricles (b=0.98, P<0.0001), and the third ventricle (b=0.05, P<0.0001), where b is the unstandardized regression coefficient. In addition, significant correlations were observed among the volumes of most of the brain regions (Table 3).

Age was also significantly related to cognitive test performance. Adjusting for the effects of education and sex, increased age was associated with poorer performance on Trails B (b=0.03, P<0.0001), verbal delayed memory (b=–0.45, P<0.0001), visual delayed memory (b=–0.69, P<0.0001), and Block Design (b=–0.41, P<0.0001). In addition, a significant age-by-sex interaction was found for Boston Naming (P<0.0003), with males (b=–0.587) showing a greater age-related decrement in scores than females (b=–0.045). Significant correlations were also observed among the scores of most of the cognitive test measures (Table 4).

Initial analyses revealed no evidence for a differential effect of sex (male vs. female) or of cerebral hemisphere laterality (left vs. right) upon correlations between cognitive test performance and cerebral volumes. Consequently, all subsequent analyses were based on the total sample (males plus females) and used total (left plus right hemisphere) cerebral volumes.

The path analyses revealed significant brain–behavior relationships for two of the six cognitive test measures: Trails B and visual delayed memory (Table 5). Performance on adjusted Trails B was significantly related to the adjusted volumes of the cerebral hemispheres, the peripheral CSF, the lateral ventricles, and the third ventricle. Longer times (poorer performance) on adjusted Trails B were associated with smaller adjusted cerebral hemisphere volumes (r=–0.19, P<0.001, effect size [ES]=0.30), as well as with larger adjusted volumes of the peripheral CSF (r=0.16, P<0.006, ES=0.25) and the lateral ventricles (r=0.12, P<0.04, ES=0.19). Adjusted visual delayed memory was significantly related to the adjusted volumes of the lateral and third ventricles. Poorer recall on adjusted visual delayed memory was associated with larger adjusted volumes of the lateral ventricles (r=–0.15, P<0.007, ES=0.24) and the third ventricle (r=–0.15, P<0.01, ES=0.23). No relationships were observed between the four remaining cognitive test measures and any of the brain regions examined.

Our findings are subject to certain potential limitations. Although cross-sectional studies of age effects allow for relatively efficient and rapid acquisition of large amounts of data, they are subject to secular effects such as birth cohort.⁴⁶ This effect refers to the possibility that brain size, like cranial size, may exhibit systematic changes over successive birth cohorts in the general population. If such trends actually exist in the population at large and if they are not secondary to secular trends associated with correlates such as cranial size (note that in the present study intracranial volume was not correlated with age), then an assessment of the true effects of aging per se on brain volume will require longitudinal investigation.

A second issue relates to the health status of our subjects. Our sample represents a group that may be somewhat healthier than the entire elderly population because of selection criteria for the CHS and the current study.^28,29 As such, our findings may not be applicable to the entire population of seniors. There is also heterogeneity of health status within our subjects, in that 29% were free of major systemic illness and 71% had at least some mild physical disease, corresponding to the distinction between successful and usual aging.⁵² Such differences in health status could account for differences in brain aging, and indeed systemic disease such as hypertension has been found to be associated with changes in brain structure.^42,53,54 In our sample, however, the presence of hypertension/ischemic heart disease was not associated with brain size (Wilks' lambda=0.95, df=13,306, P=0.26). Finally, it is possible that at least some subjects in our sample may have experienced undetected cognitive decline despite our screening and exclusionary criteria. Still, our use of an MMSE cutoff score of 24 is widely accepted in the literature³⁵ and represents an appropriate strategy for selection of a sample that is representative of the community-resident “usual aging” population.

Third, there exists the possibility of a Type I error in our findings because many correlations were tested. Thus, our analyses must be considered exploratory, and our finding of a relationship requires confirmation by additional hypothesis-testing research.

Fourth, the tests used in this study constituted part of a larger test battery that was originally chosen to be maximally sensitive to aging rather than for its potential to provide information about the neuroanatomic basis of cognitive function. Many tests that are sensitive to aging, such as Trails B, are thought to be measures of functions (like information processing speed) that may be more diffusely represented in the brain. Therefore, the original purpose for the battery of tests used in this study may have led to the selection of instruments that are sensitive to global age-related changes but less suitable for the investigation of focal brain–behavior relationships. Consequently, tests such as Trails B may be particularly likely to correlate with the global anatomic variables reported in this study but less likely to be related to focal, lateralized brain changes.

Finally, the measurements of cerebral size in our study are subject to certain limitations.¹ First, because of limitations inherent in the CHS MR imaging acquisition protocol, our analyses of cerebral size were restricted to the axial plane; three-dimensional reconstruction was not possible without dramatic loss of resolution. Second, accurate delineation of regional boundaries can be affected by several sources of technical error, including improper window center settings, magnetic field inhomogeneity (resulting in spatial distortion of objects and object pixel nonuniformity), and differences in MR imaging technical variables.⁴³ The effects of these variables were minimized in this study by using a set of procedures that has been shown to optimize the accuracy of MR imaging size measurements.^43,47 Third, field strength differences between the two scanners could affect estimates of brain size.¹ Including scanner assignment in the path analysis did not, however, alter any of the correlations between cerebral volumes and neuropsychological test measures in our study.

The following cognitive correlates with age-specific changes in cerebral size emerged from the present study:

Attention and Psychomotor Speed. We found that the two hallmarks of brain aging, cerebral atrophy (as measured both by decreased cerebral hemisphere volume and increased peripheral CSF volume) and ventricular enlargement, were each associated with poorer performance on the Trail Making Test, a measure of attention, psychomotor speed, and working memory. Our results are consistent with those of Earnest et al.,⁵ who observed that ventricular enlargement on CT was associated with poorer performance on the Digit Symbol subtest of the WAIS-R, a measure of attention and psychomotor speed (Table 1). In contrast, most of the extant literature reports no relationship between measures of attention and psychomotor speed and either cortical atrophy^4,5,24–26 or ventricular enlargement,^5,25 including studies using the Trail Making Test (Table 1).^4,5,25,26 These negative findings are difficult to compare with the current study, however, because of differences in subject sample (age range, sex, exclusion criteria), brain imaging methods (e.g., CT vs. MR imaging) and measurement techniques (e.g., linear vs. volume measures), and approach to statistical analysis, including adjustments of cognitive test data for potential confounding variables. The negative findings from these studies^4,5,25,26 may also have been the result of low power due to relatively small sample sizes.

Memory. We found that lateral ventricle volume and third ventricular volume were each associated with poorer recall on visual delayed memory (Table 5). Our results stand in contrast to negative findings from the three previous studies that have examined relationships between ventricular size and visual memory (Table 1).^5,24,26 As discussed above, these negative findings are difficult to compare with the findings in the present study because of methodological differences, and they also suffer from low power associated with relatively small sample sizes. More recently, visual explicit memory (variously measured) has also been linked to the size of the visual cortex²⁰ and corpus callosum.²¹ Conflicting findings have been reported for the hippocampus (Table 1).^6,8,13,23

We found no relationship between verbal delayed memory and either age-specific cerebral atrophy (either cerebral hemisphere volume or peripheral CSF volume) or ventricular enlargement, a result that is in agreement with most of the extant literature (Table 1).^{5,7,10,20,24–26} In contrast, Carmelli et al.⁴ observed that the difference in brain volume between elderly male identical twins was associated with within-pair differences in memory performance on the California Verbal Learning Test (Table 1). Several other studies have examined relationships between verbal memory and discrete structures such as the medial temporal lobe, with conflicting findings reported for hippocampal size^{6–8,13,15,20,23} and negative results observed for the amygdala²³ and parahippocampus (Table 1).^13,20

Language and Visuospatial Function. We found no relationship between our two measures of language function (Fluency, Boston Naming) and either cerebral atrophy (cerebral hemisphere volume or peripheral CSF volume) or ventricular enlargement (Table 5). Others (Table 1) have likewise reported no relationship between CSF volume and various measures of language function,²⁵ including the Boston Naming Test.¹⁶ In contrast, Stafford et al.²⁴ found that total CSF volume on CT was related to a discriminant function of naming and abstraction that included the Boston Naming Test. O'Brien et al.¹⁵ recently reported that language function was not associated with age-related atrophy of the amygdala-hippocampal complex.

We also found no relationship between our measure of visuospatial construction (Block Design) and either cerebral atrophy (cerebral hemisphere volume or peripheral CSF volume) or ventricular enlargement (Table 5). All published studies of which we are aware have likewise reported no relationship between performance on the Block Design test and age-related changes in peripheral CSF volume^25,26 or ventricular size (Table 1).^5,25,26 Performance on other visuospatial tasks has likewise shown no relationship to age-related lateral ventricular enlargement.¹⁴ No previous study has examined relationships between Block Design performance and age-related changes in cerebral hemisphere volume per se, although the former was found to be unrelated to the size of corpus callosum, pons, or cerebellum (Table 1).²¹

In summary, our data suggest that global age-related changes in brain size, both cortical (e.g., cerebral atrophy and increased peripheral CSF volume) and subcortical (e.g., ventricular enlargement), are associated with age-related decrements in attention, psychomotor speed, and visual delayed memory. These results are consistent with other data suggesting that the neural substrates for these functions involve multiple systems distributed throughout the brain. Still other functions, such as verbal memory, language, and visuospatial functioning, were not related to age-linked changes in cerebral size in our study. Attempts by others to relate these functions to more discrete brain structures have met with inconsistent results (Table 1). These discrepant findings regarding relations between cognition and brain structure might mean that cognitive aging is truly a function of diffuse global brain changes, or that age-related cognitive decline is multifactorial, such that changes in several brain regions may have similar effects on cognitive test performance. Slowing on Trails B, for example, could be caused by reduced motor speed, impaired working memory, poor visual scanning, or several cognitive deficits. A finer-grained analysis of the relationships between the specific changes associated with aging and the morphology of more discrete brain structures is required to resolve this issue.