Abstract
Background: The use of volumetric MRI as a biomarker for assessing transitions to dementia presumes that more rapid brain loss marks the clinical transition from benign aging to mild cognitive impairment (MCI). The trajectory of this volume loss relative to the timing of the clinical transition to dementia has not been established.
Methods: The authors annually evaluated 79 healthy elderly subjects for up to 15 consecutive years with standardized clinical examinations and volumetric brain MRI assessments of ventricular volume. During the study period, 37 subjects developed MCI. A mixed effects model with a change point modeled the pattern of brain volume loss in healthy aging compared with subjects diagnosed with MCI.
Results: The brain loss trajectory of subjects developing MCI during follow-up differed from healthy aging in a two-phase process. First, the annual rate of expansion of ventricular volume decreased with age; however, the annual rates of expansion were greater in those who developed cognitive impairment during follow-up compared with those who did not. Further, subjects who developed MCI had an acceleration of ventricular volume expansion approximately 2.3 years prior to clinical diagnosis of MCI.
Conclusions: Ventricular expansion is faster in those developing mild cognitive impairment years prior to clinical symptoms, and eventually a more rapid expansion occurs approximately 24 months prior to the emergence of clinical symptoms. These differential rates of preclinical atrophy suggest that there are specific windows for optimal timing of introduction of dementia prevention therapies in the future.
GLOSSARY: AD = Alzheimer disease; BMI = body mass index; CDR = Clinical Dementia Rating Scale; MCI = mild cognitive impairment; MMSE = Mini-Mental State Examination.
Healthy brain aging is conventionally conceived of as a slow, selective loss of neuronal elements and associated brain parenchyma in contrast to Alzheimer disease (AD) where there is a more severe loss of neurons as well as an accumulation of characteristic lesions defining the disease.1 However, the contrast between normative aging and AD is not sharp, with most contemporary models encompassing a continuum of pathologic change. Characteristic AD neuropathology found at autopsy has been associated ante mortem with both rates of cognitive decline2 and rates of brain volume loss assessed with MRI3 in healthy aging and in those developing mild cognitive impairment (MCI) or dementia. Thus, brain volume loss may be used as an index of the degree to which one might predict the development of more benign aging change or MCI leading to dementia.4–7
Up until now, the use of MRI as a surrogate marker tracking preclinical transitions to dementia has been limited to a few studies of longitudinal samples with limited numbers of MRI assessments during the period prior to cognitive impairment. Accordingly, specification of rate or volume changes have applied simple two-point change scores or regression (“slope”) models covering only a few years prior to cognitive decline.4 Since the transition from normal to pathologic is difficult to discern as the clinical syndrome develops over several years with a prodrome or subclinical phase that may extend for an even longer period of time, it is not clear whether the recognized differences in simple rates of volume change are monotonic or whether there is a more complex trajectory of change unfolding during the long preclinical period. This question is particularly important for the application of volumetric MRI as a biomarker for assessing dementia prevention strategies that may be applied many years prior to clinical symptoms.
In this article, we address how the trajectories of brain aging differed between clinically normal subjects and those diagnosed with MCI by evaluating brain volume loss marked by ventricular volume change in a healthy aging cohort followed for up to 15 years. From this analysis, we are able to estimate the timing of disease associated increases in brain loss relative to both the aging and disease processes.
METHODS
Participants and clinical assessments.
Subjects were part of a longitudinal community-based cohort of healthy elderly, ages 65 to 100 from the Oregon Brain Aging Study.8–10 Participants were enrolled starting in 1989 at the NIA Layton Center for Aging and Alzheimer’s Disease at Oregon Health and Science University. At entry, all subjects were cognitively intact with a Clinical Dementia Rating Scale (CDR) score = 0 and a Mini-Mental State Examination (MMSE) score > 23. The subjects were community-dwelling, functionally independent adults, free of co-morbid conditions commonly associated with cognitive decline (e.g., stroke, heart disease, hypertension, cancer, diabetes, and neurologic disorders). Every 6 months subjects were assessed for medical history, functional limitations,11 MMSE,12 and CDR.13 Annually, the subjects received full physical, neurologic, neuropsychological, and MRI examinations.14 The assessments were performed until death. The annual attrition due to loss to follow-up is less than 1%.10 Additionally, blood samples were collected, DNA extracted, and APOE genotypes determined for each individual using standard methods.15 All subjects signed informed consent approved by the Oregon Health and Science University Institutional Review Board.
MCI was defined by at least two consecutive CDR scores ≥ 0.5 and an absence of functional impairment. The age at onset of cognitive impairment was then defined as the age of the subject at the first observation with a CDR ≥ 0.5.8,10 The majority of these subjects would be considered, by current criteria, to have amnestic MCI at the time of their designation of onset of cognitive impairment.16
MRI.
The 79 subjects in the Oregon Brain Aging Study cohort with a minimum of three volumetrically analyzed MRI scans were used for this analysis. Thirty-seven of the 79 subjects in this analysis developed cognitive impairment during follow-up. The analysis subsample is similar to the total cohort with two exceptions: 1) the conversion rate is higher than the total cohort rate of 33% and 2) the average age is older than the total population. These two differences reflect our balancing the number of clinically normal and impaired subjects when choosing the scans to analyze. Balancing the number of subjects with and without impairment improves the statistical power. Within each group the analyzed scans are essentially random.
The average age at baseline of the subjects in this analysis was 83.5 ± 7.4 years. On average, subjects had 5.8 ± 3.4 (mean ± SD) MRI scans and were followed for 9.5 ± 3.3 years. Other data and subject summary measures relevant to the analysis are given in tables 1 and 2.
Table 1 Description of the ventricular volume observations and length of follow-up on the subjects: mean (SD) and range, when applicable
Table 2 Summary statistics of subject characteristics
MRI scans were obtained within 2 weeks of enrollment and on annual visits using a 1.5 T scanner (GE Medical Systems, Milwaukee, WI). The MRI protocol was as follows: slice thickness of 4 mm (no gap), 24-cm field of view with a 256 × 256 matrix (0.86 mm × 0.86 mm pixel size), and 0.5 repetition per sequence. The brain was visualized in two planes using the following pulse sequences: 1) T1-weighted sagittal images centered in the midsagittal plane with the pituitary profile (including the infidibulum) and cerebellar vermis clearly delineated: repetition time = 600 milliseconds, echo time = 20 milliseconds, images; 2) T2-weighted coronal images perpendicular to the sagittal plane; multiecho sequence, repetition time = 2,800 milliseconds, echo time = 30/80 milliseconds. Ventricular, intracranial, and total brain volumes were measured. MRI volumes were determined by a standardized semiautomated recursive segmentation technique described previously with established reliability (intraclass correlation coefficients > 0.90 for all regions).7,17 Because of the longitudinal nature of this analysis, we note that the same MRI pulse sequence was used for the entire period. Software and hardware changes were investigated with bridging data and phantom controls. In addition, we studied ventricular volume because it is less sensitive to protocol changes over time and has been shown to change with progression across the spectrum of aging through AD.6,7,18 Finally, this population has continuous enrollment. Therefore, changes in the MRI protocol vary by age and time to onset of cognitive impairment. Of the 17 subjects with a MRI protocol change and a cognitive impairment diagnosis in follow-up, the range of time between protocol change and clinical diagnosis was 0 to 12 years with an average of 4.6 ± 4.5 years. This makes it unlikely that any change in the rate of change of ventricular volume expansion relative to clinical diagnosis would be due to a change in the MRI protocol.
Analysis.
Dr. Carlson performed the statistical analysis presented in this article. A longitudinal mixed effects model with a change point19 was used to estimate the pattern of ventricular volume change over time. An interaction term between age and clinical diagnosis was used to test whether the aging pattern for those diagnosed with MCI during follow-up differed from those not diagnosed with MCI during follow-up. A second part of the mixed effects model investigated whether the annual rate of expansion in ventricular volume changed at some point relative to clinical diagnosis. This was done by including a change point in the mixed effects model such that the rates of change in the mixed effects model are allowed to differ before and after the change point. The point of change in the coefficients is relative to the age of diagnosis with MCI, as opposed to age. However the timing of the point of change relative to diagnosis is common across all subjects. In the mixed effects model, ventricular volume values were the outcome of interest and were log (base e) transformed prior to analysis to address a skewed distribution and to address the larger variances in ventricular volume at older ages. The age of the ventricular volume measurement was centered at the population age of 87.6 years prior to fitting to reduce correlation between the quadratic terms. All analyses were adjusted for log (base e) of the baseline intracranial volume, gender, APOE-ε4 genotype, and body mass index (BMI).
In addition to estimating the coefficients in the mixed effects model, we needed to estimate the location of the change point. The location of the change point relative to clinical diagnosis was estimated by partial likelihood.20 Separate mixed effects models were fitted with the change point fixed at 0.1 time intervals from −6 years to +3 years from diagnosis. The model with the highest partial likelihood was the model used to summarize the results. A 95% CI of the location of the change point relative to diagnosis with cognitive impairment was calculated by a likelihood ratio approach. We tested whether there was a significant change in the rate of ventricular volume accumulation relative to clinical diagnosis by calculating a 95% CI around the parameter on the change point term in the mixed effects model. The significance of the other terms in the mixed effects model was determined using a Wald test statistic.21 The standard errors for the parameter estimates were calculated using the conditional variance as proposed previously.19 This SE calculation adjusts for the fact that numerous models were fitted to estimate the location of the change point. Significance was taken as p < 0.05.
RESULTS
Tables 1 and 2 show descriptive statistics of the demographics defining the normal and MCI subjects and the characteristics of the data used in the analysis. MCI and normal subjects did not differ in baseline characteristics or number of years or observations in follow-up. The figure shows the average ventricular volume curves for those remaining intact during follow-up and an example average ventricular volume curve for subjects with an MCI diagnosis at age 89.6, the population average age of diagnosis. Figure e-1 on the Neurology® Web site (www.neurology.org) shows the raw data in addition to the fitted curves.
Figure Expected ventricular volume trajectories
The figure exhibits the expected ventricular volume trajectories from the statistical model. The large dashed line is the average ventricular volume trajectory for those remaining cognitively intact during follow-up and the rates of change are more linear (p = 0.094) than those diagnoses with mild cognitive impairment (MCI) (p = 0.00079) represented by the solid black line; however, as an individual subject approaches clinical diagnosis, their curve diverges from the black line approximately 2.3 years prior to diagnosis. In this example the MCI curve is for a diagnosis at 89.6 years (the population average age of conversion in this study).
Ventricular volume aging trajectories.
On average, the annual percent increase in ventricular volume decreased with age (table 3, rows 1 and 2). The aging pattern marginally differed between those diagnosed with MCI during follow-up and those remaining cognitively intact during follow-up (p = 0.081; figure). The change in the annual rates of ventricular volume over time is stronger for those developing MCI during follow-up (p = 0.00079; table 3, row 2) compared with those remaining cognitively intact during follow-up (p = 0.094; table 3, row 1); however, the annual increases in ventricular volume appear consistently higher in those diagnosed with MCI during follow-up. In addition, in those developing cognitive impairment during follow-up, the annual rate of ventricular volume expansion accelerates 2.3 years prior to the age of diagnosis (95% CI: 0.3, 5.6 years.). The change in the annual rate of ventricular volume expansion relative to clinical diagnosis was such that there is a constant additional increase in ventricular volume expansion that starts approximately 2.3 years prior to clinical diagnosis with cognitive impairment. On average, the cognitively impaired subjects’ annual rate of change in ventricular volume increases an additional 2.3% (95% CI: 0.08%, 3.9%) starting approximately 2.3 years prior to diagnosis. Table 3 (row 3) shows how the annual rate of change is even larger compared with the cognitively intact when the subject is close to being clinically diagnosed with MCI.
Table 3 Example yearly changes in ventricular volume
The significance of the quadratic ventricular volume trajectory was verified by fitting separate models to the normal subjects and MCI subjects. The parameter estimates did not change and the p values remained similar. We additionally verified the results of the mixed effects model and the change point by fitting a model with time to MCI diagnosis as the time scale the subjects diagnosed with MCI at follow-up. The change point remained significant and was similar to the estimate in the presented analysis.
Other predictors of ventricular volume.
The presence of at least one APOE-ε4 allele and gender are associated with ventricular volume size after adjusting for baseline intracranial volume. The two factors interact (p = 0.022). For women, the presence of at least one APOE-ε4 allele is associated with a 48% (95%CI: 10%, 99%, p = 0.036) larger ventricular volume, while for men, the presence of at least one APOE-ε4 allele is not associated with ventricular volume (−10%, 95% CI: −33%, 23%, p > 0.99).
When fitting separate models to the subjects diagnosed with MCI and the normal subjects, the APOE-ε4–gender interaction with ventricular volume was attenuated, but marginal, in the subjects with MCI (p = 0.060); however, the APOE-ε4–gender interaction with ventricular volume was not found for the normal subjects (p = 0.41). Additionally, for the normal subjects, neither gender (p = 0.21) nor the presence of an APOE-ε4 allele (p = 0.58) was associated with ventricular volume trajectory in this model. BMI was not associated with ventricular volume (p = 0.49) and adjustment for BMI did not alter the results.
DISCUSSION
We have shown that, on average, changes in brain volume in subjects diagnosed with cognitive impairment differ from healthy subjects in a two-phase process. Years prior to diagnosis, the annual rate of ventricular volume expansion is faster in subjects destined to develop cognitive impairment compared with healthy subjects. In addition, at least 2 years prior to clinical diagnosis, the annual rate of ventricular volume expansion further accelerates. Our longitudinal model enabled us to compare the full aging trajectory of change in a normal aging population with a good number of subjects observed in the transition to cognitive impairment. By fitting these types of models in these populations we can begin to understand pathologic progression and address hypotheses that previously have been limited by the length of time and number of data points captured to describe change. Previous longitudinal MRI studies have shown that there are volumetric changes occurring years before clinical decline.6,7,22–26 The specific trajectory of change beyond a linear model has not been described. It is not clear what factors may influence this more complex trajectory.
The results in this article tie together previous findings that rates of change in ventricular volume are predictive of the development of MCI when measured in the 2 years prior to onset6 and that baseline levels of ventricular volume do not differ between those who develop and who do not develop MCI.5 In particular, we found that the ventricular volume trajectories are just beginning to diverge at young-old ages and it is the rates of change that are important, rather than the absolute levels. In addition, we quantify the timing of rate changes relative to clinical diagnosis and also add to the literature that rates of change may differ even earlier than the time courses studied thus far.
One may wonder if the findings would be similar when investigating other brain regions. For this analysis we chose to specifically focus on the longitudinal trajectory of ventricular volume as it has been consistently shown to be a sensitive, highly reliable measurement for longitudinal studies in aging and AD, and is anatomically readily comparable across multiple studies and methodologies in current practice.6,7,18 In the future, comparison of not only multiregional volume change, but also assessment of change in the same brain using both structural and functional imaging will be important. In this context it will be of great interest to see how our volumetric results map to trajectories of functional changes as assessed by other imaging modalities such as PET assessing glucose metabolism or amyloid accumulation.27
These results have some limitations. We are unable to characterize heterogeneity in the observed change point of the individual. There is likely heterogeneity in the change point as is shown by the wide CI on the location of the change point. Future areas of research could include obtaining subject level estimates of the change points, which would be useful for more precisely predicting, at the individual level, time to clinical diagnosis. However, this would require more annual data on each subject and a larger number of subjects.
In addition, the interpretation of our results is stronger at the population level, as opposed to the individual level, because of the number of observations on each subject and the number of subjects with observations both before and after the change point. However, we were still able to gain valuable information from all the subjects because subjects with only observations on one side of the change point contribute to the estimates of the population curve prior to or after a change point and contribute to the estimation of the parameters in the mixed effects model.19
The analysis does not rule out the possibility of an additional change point even longer before diagnosis. Given the current data, we can confidently conclude that there is a change point prior to the emergence of clinical symptoms, but the number of change points and precise timing warrant further investigation.
The fact that the rates of change differ years prior to diagnosis and even more importantly, the fact that the rate of change of ventricular volume accelerates in the few years prior to the clinical diagnosis of MCI emphasizes that MRI may be useful in detecting changes associated with AD prior to clinical measures and especially the need to begin preventative therapies as early as possible, ideally prior to the change point preceding the clinical diagnosis of cognitive decline.
ACKNOWLEDGMENT
The authors thank the Oregon Brain Aging Study volunteers and Oregon Layton Aging and Alzheimer’s Disease Center staff.
Footnotes
-
Supplemental data at www.neurology.org
Editorial, page 824
e-Pub ahead of print on November 28, 2007, at www.neurology.org.
Supported by National Center for Research Resources (grant UL1 RR024140 01), National Institute on Aging (grant AG08017), Department of Veterans Affairs.
Disclosure: The authors report no conflicts of interest
Received September 27, 2006. Accepted in final form July 5, 2007.
REFERENCES
Disputes & Debates: Rapid online correspondence
- Trajectories of brain loss in aging and the development of cognitive impairment
- Reply from the authors
REQUIREMENTS
If you are uploading a letter concerning an article:
You must have updated your disclosures within six months: http://submit.neurology.org
Your co-authors must send a completed Publishing Agreement Form to Neurology Staff (not necessary for the lead/corresponding author as the form below will suffice) before you upload your comment.
If you are responding to a comment that was written about an article you originally authored:
You (and co-authors) do not need to fill out forms or check disclosures as author forms are still valid
and apply to letter.
Submission specifications:
- Submissions must be < 200 words with < 5 references. Reference 1 must be the article on which you are commenting.
- Submissions should not have more than 5 authors. (Exception: original author replies can include all original authors of the article)
- Submit only on articles published within 6 months of issue date.
- Do not be redundant. Read any comments already posted on the article prior to submission.
- Submitted comments are subject to editing and editor review prior to posting.