Brain volume preserved in healthy elderly through the eleventh decade

The relation between brain volume changes and the normal aging process has been the subject of many cross-sectional neuroimaging studies. These studies suggest age-related decreases in cerebral hemisphere volumes,^1-5 frontal lobe volumes,^2,3,6,7 temporal lobe volumes,^2,7,8 parietal-occipital region volumes,⁵ hippocampal size,^2,9-13 parahippocampal gyrus volumes,^5,12 corpus callosum size,^14,15 cerebellar volumes,⁵ subcortical nuclei size,^5,16 and volume of limbic structures.¹⁷ Additionally, increases in CSF amount and ventricular size^1,2,8,18-21 have been observed. Few of these studies have included healthy subjects over the age of 85.

There are limited longitudinal neuroimaging data available in healthy elderly and no studies of cortical regional volumes based on MRI quantification techniques. Thus, anticipated rates of brain volume change in healthy aging have not been established.

The current study uses quantitative volumetric MRI techniques to assess multiple cortical regions of interest (ROIs) in longitudinally studied, healthy elderly individuals from the Oregon Brain Aging Study (OBAS) whose cognition and health status have been monitored carefully over time. In addition to our volumetric analyses of multiple cortical regions, we investigated the rate of change with healthy aging by calculating rates of change, or slopes, for each individual, as well as subsequent mean group rates. It was hypothesized that, with normal aging, cortical regions demonstrate small but significant volume loss. However, we predicted that, unlike the aging brain in dementia patients, the rate of brain atrophy in individuals documented to be free of cognitive impairment over time does not change significantly in the last decades of life compared with younger elderly.

Methods. Subjects. Subjects consisted of 46 healthy volunteers who were recruited from retirement homes, senior citizens' organizations, and public relations activities. These subjects are participants in a longitudinal study of brain aging and cognition begun in 1989 (OBAS^12,22-24) at the Oregon Health Sciences University and Veterans Affairs Medical Center in Portland. Three age groups were formed: the oldest-old group (n = 20) included persons 85 to 95 years old, the middle-old group (n = 15) included persons 75 to 84 years old, and the young-old group (n = 11) included persons 65 to 74 years old. Subjects were required to be functionally independent and to score 12 or higher on the Instrumental Activities of Daily Living sections of the Older Americans Resources and Services Multidimensional Functional Assessment and Questionnaire.²⁵ Subjects had never sought evaluation for cognitive or behavioral impairment, and they considered themselves mentally fit for their age. All subjects had Clinical Dementia Rating Scale²⁶ scores of 0 and Mini-Mental State Examination (MMSE^27,28) scores ≥24. Subjects had annual neuropsychological and neurologic examinations. The tests that comprise each evaluation were selected to provide sensitive measures of age effects on cognition and early changes associated with AD.²² Of 37 oldest-old subjects initially entered into the study with three or more MRIs available for analysis, 44% had developed cognitive impairment consistent with early dementia and are not included here.

Special emphasis was given to excluding subjects with age-related diseases that might affect the brain. Exclusion criteria included a history of major medical illness, including diabetes, hypertension, angina, cardiac arrhythmia, chronic pulmonary disease, cancer, major affective disorder, transient ischemic attacks, stroke, head injury (i.e., loss of consciousness >5 minutes), or other neurologic disease. Subjects were also excluded if they had a history of cardiac or coronary artery surgery. There was no history of alcohol or other drug abuse. Subjects were not excluded if they took the following medications: vitamins, aspirin, thyroid hormone, and estrogen. Copies of medical records were reviewed to verify medical histories. Additional details of the recruitment procedures and inclusionary and exclusionary criteria, as well as extensive medical and cognitive data, are published elsewhere.^23,24 Each subject was scanned at time of entry into the study. The oldest-old subjects were scanned each subsequent year, and the young-old and middle-old subjects were scanned at least bianually. During 3 to 9 years of follow-up, the number of scans obtained per subject varied from three to nine.

Imaging and analysis procedure. MR images were taken with a 1.5-T magnet. The imaging protocol used for each study consisted of continuous-slice, multiecho, multiplanar image acquisition, with 4-mm-thick coronal slices and a 24-cm² field of view using a 256 × 256 acquisition matrix with two excitations. The brain was visualized using the following sequence: multiecho coronal sequence; repetition time, 3,000 msec; echo time, 30 and 80 msec. T1-weighted sagittal images centered in the midsagittal plane were used to orient the coronal plane. All brains were imaged with this protocol except for 14 scans acquired in the first year, which had a 2.5-mm gap. The coronal plane was determined as the plane oriented perpendicularly to a line drawn from the lowest point of the splenium to the lowest point of the genu of the corpus callosum on the midsagittal image. Analysis of the MR images was performed with computer-assisted techniques using a program called REGION, developed by us for use with any Macintosh (Apple Computer; Cupertino, CA) computer.¹²

The following ROIs were examined: supratentorial intracranial cavity volume (ICV); total brain volume (i.e., total supratentorial CNS tissue volume); frontal, temporal, and parietal-occipital cortices; basilar-subcortical region (including basal ganglia, thalamus, and internal capsule); CSF volumes; lateral ventricles; temporal horns; and subarachnoid spaces (SA). The left and right sides of the bilateral structures were analyzed separately and were subsequently combined for a total structure measurement. After all slices were analyzed, total pixel count for each ROI in the entire brain was determined by summing the corresponding pixel count for each slice. Pixel count was converted to volume by multiplying the ROI cross-sectional area by the slice thickness. To control for differences in mature adult brain size between men and women, total ICV was used as a covariate.

The ICV was defined as all non-bone pixels beginning with the first slice in which the frontal poles were present and ending at the occipital pole. At the base of the brain, brainstem structures were excluded from supratentorial structures by tracing manually those pixels to be excluded according to atlas-based rules. The total brain volume, or CNS volume, was determined by summing the volumes of the frontal, temporal, parietal-occipital, and basilar regions (defined later).

The temporal lobe volume was defined as all brain-assigned pixels-beginning at the temporal pole and extending posteriorly to the slice where the temporal horn of the lateral ventricle formed the trigone. On slices when the temporal lobe was contiguous with or connected to adjacent brain (e.g., temporal stem), the temporal lobe was isolated by tracing manually a line from the most inferior and medial point of the vertical Sylvian fissure across to the most superior point of the transverse or choroidal fissure.

The midsagittal image in which the central sulcus was best visualized was used to define the division between the frontal lobe and the parietal-occipital region.²⁹ The frontal lobe volume was defined as all brain-assigned pixels anterior to the central sulcus. The parietal-occipital region consisted of all pixels on those slices posterior to that point. The frontal lobe and parietal-occipital region were isolated from contiguous brain by drawing a line from the most superior and medial point of the body of the ventricle to the most superior point of the vertical Sylvian fissure. The basilar-subcortical region was defined as all brain-assigned pixels excluding the temporal and frontal lobes and the parietal-occipital region.

To ensure the least variable and most reliable measurements of hippocampal volume, we restricted our analysis to the body of the hippocampus. The body of the hippocampus was defined on the slice beginning at the level of the red nucleus and ending with the superior colliculus.^12,30 The parahippocampal gyrus region was the first gyrus identified after the collateral sulcus. The total ventricular CSF volume included the lateral ventricles and the third ventricle. Any CSF not included in the ventricular spaces was labeled SA.

The volumetric measures were determined with an automated technique called recursive segmentation, which is built into the REGION analysis program. Within each slice, tissue types of interest are sampled coincidentally on the spatially registered multiecho images by selecting a predetermined number of sample points within four tissue types: bone, brain (gray and white matter), CSF, and high-signal intensity areas within the white matter. This process has been described in more detail previously.¹²

The interrater reliability of the whole brain analysis technique (measured by intraclass correlation coefficient, or ICC) performed by four operators on the same set of five brain scans was 0.99 for the ICV, 0.98 for the CNS, 0.99 for the temporal lobe 0.91 for the frontal lobe, 0.93 for the parietal-occipital region, 0.98 for the basilar-subcortical region, 0.99 for the ventricular CSF, and 0.71 for the SA. The interrater reliability of the hippocampal measurement technique performed by three operators on the same set of five brain scans was 0.89 for the hippocampal body ICC and 0.81 for the parahippocampal body ICC.

Statistical analysis. Group characteristics at entry were compared with one-way analysis of variance (ANOVA). Multiple of the ROIs were analyzed, and baselines (i.e., intercepts) were calculated through a linear regression analysis. A regression line was calculated for each person, regressing the ROI volume against the subject's age at the time of the scan. Rates of change were calculated by a linear regression analysis to determine a slope for each individual subject. These subjects slopes were then combined to produce a mean group rate of change or slope.

The group slopes and ROI means were compared using ANOVA. Data were analyzed for volumes both with and without adjustment (i.e., covariance) for ICV (except for ICV analysis itself). All statistical results were essentially identical, and adjusted values are given in the Results section. Additionally, to take into account the potential effect of differing MRI protocols between year 1 and subsequent years, data were analyzed using MRI protocol as a covariate. The Newman-Keuls test of multiple comparisons was used in post hoc analyses (p < 0.05). Independent t-tests were used to evaluate whether the volumetric changes over time (i.e., slopes) were significantly different Pearson's partial correlations were used to assess the relation between the volumes of the ROI and age, as well as the rates of ROI change and age.

Results. Subject characteristics. The demographic characteristics of the subjects at entry (or first MRI) are described in table 1. As expected, the groups differed significantly in age. Additionally, reflecting lower mortality rates, the younger subjects had a significantly longer duration of clinical follow-up. There was a 1-point mean difference in MMSE scores of the oldest-old group, which was significantly lower than the other two groups, most likely due to the narrow variance of the MMSE in our healthy elderly subjects. MMSE change scores were calculated and no significant differences between the groups were observed. The groups did not differ with respect to sex, years of education, socioeconomic status,³¹ Cornell Depression Scale core,³² or duration of MRI follow-up period.

MRI ROI analysis. There were significant differences (p < 0.0001) between the young-old and oldest-old groups, as well as between the middle-old and oldest-old groups, in the total brain volume, left hemisphere, and right hemisphere (table 2). Also, the difference in the right hemisphere between the young-old and middle-old groups was significant at the p < 0.0001 level. The volumes of the basilar-subcortical region and the hippocampus were significantly different between the young-old and oldest-old groups, and between the middle-old and oldest-old groups at the p < 0.05 level. The differences between the young-old and oldest-old groups with regard to the temporal lobe and the SA were significant at the p < 0.05 level. Additionally, the middle-old and oldest-old groups differed significantly (p < 0.05) with regard to ICV. Lastly, in the young-old and middle-old groups, there were significant differences in the temporal lobe and the hippocampus (p < 0.05). For the significant differences in the brain regions reported earlier, the oldest-old subjects had significantly smaller volumes, followed by the middle-old and young-old groups. For the SA, the oldest-old subjects showed significantly larger volumes than the young-old subjects. There were no significant differences between the groups in the following ROI: CSF, lateral ventricles, temporal horns, frontal lobes, parietal-occipital region, and parahippocampal gyrus.

Correlations between age and ROI volumetric measurements were obtained. There were significant age-related correlations in the total brain (r = -0.57, p = 0.000), left hemisphere (r = -0.52, p = 0.000), right hemisphere (r = -0.64, p = 0.000), frontal lobes (r = -0.44, p = 0.002), temporal lobes (r = -0.47, p = 0.001), basilar-subcortical region (r = -0.51, p = 0.000), hippocampus (r = -0.57, p = 0.000), and parahippocampal gyrus (r = -0.33, p = 0.024). A significant age-related increase was observed in the temporal horns (r = 0.42, p = 0.003). No significant correlations with age were found for the ICV, CSF, lateral ventricles, SA, or parietal-occipital region.

Rates of change analysis. There were no gender differences in the mean rates of change in any of the regions except the temporal lobe (p < 0.01) and CSF (p < 0.05). When comparing the mean rates of change for each age group with zero (table 3), the total ventricular CSF and lateral ventricular volumes showed significant differences from zero in each of the groups. Within the young-old group, there were additional significant differences from zero with regard to the SA, hippocampus, and parahippocampal gyrus. The middle-old group demonstrated significantly differences from zero in the temporal horn, parietal-occipital, and basilar-subcortical regions, and the hippocampus. In the oldest-old group, the total CNS, parietal-occipital region hemisphere, and parahippocampal gyrus were significantly different from zero.

When comparing the mean rates of change, there were no significant differences between the groups with regard to any of the ROI rates of change. None of the correlations between age and rate of ROI change were significant, with the exception of the right hemisphere (r = 0.50, p = 0.000). The mean percents of the regional volume that each slope represents were calculated. Generally, these values were small (mean percent change, -0.80%), suggesting that the change over time is a minor percentage of the regional volumes. Discussion. In our cross-sectional analysis we demonstrated that in group of healthy elderly individuals, subdivided by age, there were significant age differences in the volumes of the total brain, the left hemisphere, and the right hemisphere. Furthermore, there were significant differences between groups in the volumes of the SA, temporal lobes, basilar-subcortical region, hippocampus, and ICV; however, these were smaller in magnitude. Additionally, when looking across a 40-year age spectrum, we found significant correlations between age and ROI volumetric measurements in the following regions: total brain, left hemisphere, right hemisphere, frontal lobes, temporal lobes, basilar-subcortical region, hippocampus, parahippocampal gyrus, and temporal horns. With regard to the rates of changes of the ROIs, we found no significant differences between the groups.

Thus, although the groups were not significantly different from each with regard to rates of change while under our observation, evidence for prior significant change over time within the groups was demonstrated. Therefore, our hypotheses were partially supported. In terms of our first hypothesis, we found that some cortical regions demonstrate significant volume loss with normal aging; however, this was not universal across all regions. Our second hypothesis was supported because we did not find the rate of brain volume change to differ significantly after age 65.

Similar findings have come from neuropatholgoic volumetric studies in which the brain has been shown to be free of pathology. Double et al.⁴ examined patients with AD and healthy elderly control subjects. They found a loss of brain volume with aging. However, they attributed this loss to white matter changes alone. These investigators discussed how age-related reductions in cortical and subcortical gray matter volumes are often found in MRI studies. However, they mention that such changes are typically only a reduction of 1% or less per year. Thus, although some volume changes are found, the rate at which the changes occur is likely to be very small. At the microscopic level, recent animal and human morphometric studies have suggested little neuron loss with health aging.^33-38

There do not appear to be previously published longitudinal studies of regional brain volumes specific to the normal aging process, as measured by MRI quantification techniques. Thus, this study serves as an initial work and, therefore, should be replicated. Shear et al.³⁹ conducted a longitudinal study that examined volumetric CT analysis of regional brain changes in normal aging and AD. They limited their study to analysis of CSF regions rather than cortical regions. Furthermore, they focused on the differences between the normal and AD groups and offered few data regarding the normal aging process. However, their control group's ventricular system change scores were similar to ours; that is, they demonstrated small yet positive slope values. They concluded that they were unable to support the hypothesis that frontal lobe volume is disproportionately affected by normal aging.

Similarly, there do not appear to be previously published studies that investigated the rate of change, via MRI quantification techniques, with regard to the brain in healthy aging process. Other studies have examined this phenomenon in alcoholics,⁴⁰ patients destined to develop dementia,¹² or using CT analysis³⁹ with varying results. Kaye et al.¹² compared nondemented, healthy elderly individuals with subjects who were initially healthy but experienced subsequent cognitive decline (i.e., preclinical dementia). They found hippocampal and parahippocampal atrophy rates to be similar in both groups. However, temporal lobe atrophy was present only in the preclinical dementia group. Although this study included a healthy elderly population, the number of subjects was limited (i.e., 18), and comparisons across several age groups was not done. Clearly, additional research investigating rates of change with regard to the healthy, aging brain is needed.

As is common in all longitudinal research, various challenges arose during our study. For example, changes in the research protocol could affect statistical analyses. Some have addressed this issue by using a statistical correction.^39,40 However, in our study we chose not to alter the raw data but rather to covary MRI protocol to address potential error, and we found no difference. Certainly, having different MRI protocols between subjects is not idea, and over- or underestimations could have occurred.

An additional methodologic problem common in all quantitative imaging studies, both cross-sectional and longitudinal, is head-positioning variability. This error is difficult to identify in a quantitative manner and may appear as a change in brain volume across time. Typical longitudinal methodologic problems are compounded when combined with standard MRI limitations. Error may result from the actual MR acquisition (e.g., protocol changes, head-positioning variability) or the subsequent image quantification.⁴¹ Simon et al.⁴¹ labeled this phenomenon the "error of serial studies" and concluded that it may exceed intraobserver error. They recommend acknowledgement of this potential source of error to guide the evolution of MRI quantitative analysis techniques. A new MRI quantification technique, the brain boundary shift integral (BBSI), has been proposed as an alternative measurement of brain atrophy in serial studies.^42-44 The BBSI is not as dependent on the accuracy of the segmentation and registration of serial scans. However, the method may be very sensitive to motion artifacts, especially in demented subjects. A study using the BBSI⁴² found a mean annual rate of brain atrophy of 0.24% in nine healthy elderly control subjects. Although the BBSI addresses some of the error problems of traditional MRI quantification techniques, limitations remain. For instance, because it is a measure of net volume loss, specific volume increases are not reflected in the total calculated atrophy. Furthermore, the BBSI has not been validated on regional cerebral volume changes. Thus, currently there is not an error-free MRI quantification technique, but there are empirically based options with diverse strengths and limitations.

Our results yielded a "cross-sectional finding" of certain structural atrophic changes, and a "longitudinal finding" of no change in the rate of brain atrophy after age 65. Taken together, these results may seem counterintuitive because the oldest-old subjects had to arrive at these reductions in some manner at some point in time. We have identified four potential explanations for this phenomenon. First, based on our sample size, there may have been too much variance to detect differences in the group rates of changes. This study, for the first time, establishes what were previously unknown rates of brain loss and allows us to calculate necessary sample sizes to show significant differences. Power analyses conducted showed that each group would require 70 subjects to have an 80% change of detecting a 0.5 SD change in the frontal lobes. However, currently, a longitudinal MRI study such as this with several hundred subjects would be extremely costly and labor intensive for ensuring a normal sampled throughout the study because the transition to dementia is very high among oldest-old subjects. Second, there could be cohort effects. Svennerholm et al.⁴⁵ state that the highest brain weights ever reported have been observed in the most recent studies due to "secular changes associated with high educational and nutritional levels." Their data support the hypothesis of a generalized increase in brain size over the past 90 years. Consistent with this observation, we observed significantly smaller ICVs in the oldest-old groups compares with the younger subjects. To adjust for these effects we were particularly careful about the health status across the age groups and covaried the ROI analyses for ICV. Nevertheless, we may have inadvertently selected oldest-old subjects who were least vulnerable to aging, whereas the young-old subjects, although now healthy, simply may never survive to be healthy, nondemented, oldest-old subjects. Third, the significant structural changes may have begun to occur before the age of 65. This is suggested in table 2, as the middle-old and oldest-old groups tend to be more similar to each other than to the young-old group. Although a similar study with younger healthy subjects has been completed,¹⁸ there do not appear to be any studies that encompass the entire adult age spectrum. Fourth, although the prevailing view has been that no new neurons are made in the adult brain, recent evidence⁴⁶ in adult monkeys suggests that the adult brain, at least in certain regions, is quite plastic, with the neurons being made throughout life. Thus, it is possible in some individuals that there are periods throughout life of volume increases, or at least stabilization, as well as decreases. However, with that is currently known about the adult human brain, the most plausible explanation for our finding of increases in some cortical regional volumes and decreases in some CSF spaces is measurement error. We will evaluate this further in the future by increasing the number of sampling point used to generate the slopes of change. Clearly, additional investigation is necessary to address such issues and advance the methodology.

We have demonstrated that in healthy elderly in whom the likely effects of mild dementia on brain atrophy have been minimized, there are only minor losses of brain volume with aging. These rates may be taken as benchmarks for assessing pathologic brain loss such as might be seen or monitored during a treatment trial.

Acknowledgment

The authors are grateful to A. Dame, S. Lehman, and D. Wasserman for their assistance on this project, and to Dr. R. Mueller for helpful discussions.