Abstract
Objective: To test the hypothesis that MRI-based measurements of hippocampal volume are related to the risk of future conversion to Alzheimer’s disease (AD) in older patients with a mild cognitive impairment (MCI).
Background: Patients who develop AD pass through a transitional state, which can be characterized as MCI. In some patients, however, MCI is a more benign condition, which may not progress to AD or may do so slowly.Patients:— Eighty consecutive patients who met criteria for the diagnosis of MCI were recruited from the Mayo Clinic Alzheimer’s Disease Center/Alzheimer’s Disease Patient Registry.
Methods: At entry into the study, each patient received an MRI examination of the head, from which the volumes of both hippocampi were measured. Patients were followed longitudinally with approximately annual clinical/cognitive assessments. The primary endpoint was the crossover of individual MCI patients to the clinical diagnosis of AD during longitudinal clinical follow-up.
Results: During the period of longitudinal observation, which averaged 32.6 months, 27 of the 80 MCI patients became demented. Hippocampal atrophy at baseline was associated with crossover from MCI to AD (relative risk [RR], 0.69, p = 0.015). When hippocampal volume was entered into bivariate models—using age, postmenopausal estrogen replacement, standard neuropsychological tests, apolipoprotein E (APOE) genotype, history of ischemic heart disease, and hypertension—the RRs were not substantially different from that found univariately, and the associations between hippocampal volume and crossover remained significant.
Conclusion: In older patients with MCI, hippocampal atrophy determined by premorbid MRI-based volume measurements is predictive of subsequent conversion to AD.
For patients who develop AD, the transition from a normal cognitive state to clinically recognizable AD occurs gradually over years.1 It is presumed that the pathologic substrate of the cognitive decline that characterizes AD follows a similar slowly progressive course, with gradual accumulation of degenerative pathology of AD ongoing for years, perhaps even decades, before manifestation of unequivocal clinical symptoms. Memory impairment is usually the initial manifestation of dementia in AD. That the transition from normal cognition to AD is gradual, however, presents clinicians with a common and difficult diagnostic problem: Does evidence of a mild memory impairment in an older individual represent the earliest manifestation of AD or more benign forgetfulness that may not progress to dementia? Clinical criteria for the classification of patients with a mild cognitive impairment (MCI) have been established.2-4 The rate at which MCI patients convert to AD is substantially greater than that of the general older population,2,5-8 and MCI patients are the subject of several more recent treatment trials.
Structural and functional imaging findings are diagnostic markers of AD.9 Most imaging studies, however, have been cross-sectional in nature and have been designed to demonstrate differences between older controls and patients who were already demented. Prior studies addressing prediction of future dementia have been done with relatively few patients with familial AD,10 with the oldest old,11 or using subjective image assessment.12 We addressed this issue by conducting imaging studies and then longitudinally following-up individuals who were at increased risk of AD because of the diagnosis of MCI. The imaging measurement evaluated was MRI-based volume measurements of the hippocampi. We chose this measurement because 1) medial temporal lobe limbic structures, particularly the hippocampus, play a central role in memory function and are the site of the earliest neurofibrillary pathology in AD13,14; 2) memory impairment is the hallmark of early AD15; and 3) MRI detects subtle medial temporal lobe damage in AD.11,16-19 In this study, we tested the hypothesis that premorbid MRI-based hippocampal volume measurements in patients with MCI were related to the risk of subsequent conversion to AD. We also determined whether the predictive power of hippocampal volume measurements was independent of other potential predictor variables—age, apolipoprotein E (APOE)genotype, estrogen replacement, performance on selected measures of cognitive performance, hypertension, and ischemic heart disease.
Patients and methods.
Recruitment and evaluation of subjects.
Eighty consecutive MCI patients were recruited from the Mayo Clinic AD Center and AD Patient Registry (ADC/ADPR), which are prospective, longitudinal studies of aging and dementia.20 Informed consent was obtained for participation in the studies, which were approved by the Mayo Institutional Review Board. Study participants were assigned to diagnostic group categories during (ADC/ADPR) consensus committee meetings consisting of a geriatrician, neurologists, neuropsychologists, psychometrists, and nurses who had seen the patient. Relevant diagnostic categories were those of MCI and AD. Criteria for the diagnosis of MCI were the following2,3: 1) memory complaint documented by the patient or collateral source; 2) normal general cognitive function, as determined by measurements of general intellectual function and screening instruments; 3) normal activities of daily living, as documented by history and record of independent living21; 4) dementia ruled out by Diagnostic and Statistical Manual of Mental Disorders, 3rd ed., revised (DSM-III-R) criteria,22 and met no National Institute of Neurological and Communicative Disorders and Stroke–Alzheimer’s Disease and Related Disorders Association (NINCDS-ADRDA) criteria for AD1; 5) objective memory impairment, defined by performance at 1.5 standard deviations below age and education-matched controls on indices of memory function23; 6) age 60 through 89 years; and 7) Clinical Dementia Rating score of 0.5.24
Ascertainment of endpoint.
The primary endpoint or dependent variable in this study was the crossover of individual MCI patients to the AD category during longitudinal follow-up. Patients were enrolled throughout the study period. All study patients underwent clinical/neuropsychological reevaluations at approximately 12-month intervals. Twenty-six patients had a single serial follow-up assessment, 13 had two follow-up assessments, and 41 had three or more follow-up assessments. MCI patients who remained unchanged cognitively were characterized as stable, and the mean follow-up time for these patients was 33.5 ± 17.9 months. Patients who became demented, all of whom received the diagnosis of probable AD at that time, were designated as crossover patients; the mean follow-up time (from enrollment to crossover) for these patients was 30.8 ± 17.3 months. The diagnosis of dementia was made according to DSM-III-R criteria.22 The diagnosis of probable AD was made according to NINCDS-ADRDA criteria.1
An MRI examination of the brain was performed within 4 months of the initial clinical assessment in all patients. These MRI studies were used in the diagnostic process only to exclude treatable causes of cognitive impairment. The hippocampal volume data were unknown to the consensus committee throughout the study.
MRI methods.
All imaging studies were conducted at 1.5 T (Signa, General Electric Medical Systems, Milwaukee, WI), using a standardized imaging protocol.25 Measurements of intracranial volume were derived from a T1-weighted sagittal sequence with 5-mm contiguous sections. Volume measurements of the hippocampi were derived from a T1-weighted three-dimensional (3D) volumetric spoiled gradient recalled echo sequence, with 124 contiguous partitions, 1.6-mm slice thickness, a 22-cm × 16.5-cm field of view, 192 views, and 45° flip angle.
All image processing steps in every patient were performed by the same research associate who was blinded to all clinical information (age, sex, clinical course) to insure that the volumetric data were unbiased. Validation studies have shown the intra-rater test-retest coefficient of variation of hippocampal volumetric measurements to be 1.9% with this method.26 The 3D image data set of each patient was realigned into an orientation perpendicular to the principal axis of the left hippocampal formation. The imaging data were then interpolated in-plane to the equivalent of a 512 × 512 matrix and magnified 2×. The voxel size of the fully processed image data was 0.316 mm3. The borders of the hippocampi were manually traced on the work station screen for each image slice sequentially from posterior to anterior. Typically, 40 to 50 imaging slices were measured for each hippocampus. In-plane hippocampal anatomic boundaries were defined to include the CA1 to CA4 sectors of the hippocampus proper, the dentate gyrus, and the subiculum (figure 1).25 The posterior boundary of the hippocampus was determined by the oblique coronal anatomic section on which the crura of the fornices were defined in full profile.
Figure 1. Neuroanatomic boundaries. The column of images on the left are cropped oblique coronal MR images through the temporal lobes of a 75-year-old woman. The upper image is through the body of the hippocampus and lower image is through the head of the hippocampus. This mild cognitive impairment (MCI) patient remained stable over 49 months of clinical follow-up. At baseline her hippocampal W score was 0.21. On the right are matched imaging sections of a 70-year-old woman, who was initially categorized as having MCI, but became demented after 43.5 months of follow-up. Her hippocampal W score was −2.48 at entry into the study. The hippocampi of the patient who became demented (right) are visibly atrophic relative to the stable patient (left) despite the fact that the crossover patient was 5 years younger. The anatomic outlines of the left hippocampus are indicated.
Intracranial volume was determined by tracing the margin of the inner table of the skull on contiguous images from the sagittal sequence.
Apolipoprotein E genotyping.
DNA was extracted from peripheral leukocytes and amplified by PCR.27 PCR products were digested with HhaI, and the fragments were separated by electrophoresis on an 8% polyacrylamide nondenaturing gel. The gel was then treated with ethidium bromide for 30 minutes, and DNA fragments were viewed by ultraviolet illumination.
Assessment of clinical variables.
The presence or absence of hypertension and ischemic cardiac disease was assessed by review of medical records. Patients were recorded as being positive for hypertension if hypertension or its treatment was identified at any point in time in the medical record. Patients were considered to have ischemic heart disease if any of the following were identified: angina pectorus, myocardial infarction, coronary bypass surgery, or coronary angioplasty. The time of menopause and the presence or absence of estrogen replacement therapy were also extracted from the medical records.
Statistical methods.
The hippocampal volume measurements of each patient were normalized for interpatient variation in head size by dividing hippocampal volume by the total intracranial volume of that particular patient. We previously determined age- and sex-specific normal percentiles for normalized hippocampal volume in a group of 126 cognitively normal older controls using the MRI volumetric method described.25 Age- and sex-specific normal percentiles for each of the 80 MCI patients were determined using this normal-value database. Each percentile was then converted to a W score. The W score is the value from a standard normal distribution corresponding to the observed percentile. For example, for a standard normal distribution, the 50, 5, and 2.5 percentiles are given by 0, −1.645, and −1.96, respectively. Thus, a patient with a hippocampal volume (adjusted for age and sex) at the fifth percentile in the normal value database would receive a W score of −1.645. Similarly, a patient at the 50th percentile would receive a W score of 0. When this method of assigning W scores is applied to the normal older control patient database, the resulting W scores precisely follow a standard normal distribution. W scores in other study populations, including our MCI cohort, can then be compared directly to this standard distribution, providing a framework for comparing hippocampal volume measurements among individual patients, appropriately corrected for age, sex, and head size.
In addition to hippocampal volume, other predictor variables for crossover to AD that were evaluated included age, APOE genotype, Mini-Mental State Examination (MMSE),28 Dementia Rating Scale (DRS),29 Wechsler Memory Scale–Revised–Logical Memory II Subtest–Paragraph Retention score (WMS-R-LMRII),30 Auditory Verbal Learning Test–Percent Delayed Retention score (AVLT),31 the total free-recall and delayed-recall indices from the Free and Cued Selective Reminding Test (FCSRT),32 and the Controlled Oral Work Association Test total final score (COWAT). Estrogen replacement, hypertension, and ischemic heart disease were also modeled as potential predictor variables.33 In the APOE ε4 risk analysis, patients were stratified into those with genotypes known to increase the risk of AD (3/4, 4/4) and those who were ε4 noncarriers (2/3, 3/3).34 Six patients who were ε2/4 were excluded from the APOE risk analysis because the association between AD and ε2/4 is unclear.
Although a direct comparison of the W scores of patients who did and those who did not crossover seems natural, the length of follow-up varied among patients. Direct comparisons between patients who crossed over versus those who did not were therefore analytically inappropriate. To accommodate variable follow-up periods, life-table methods were used to evaluate patient characteristics relating to crossover rather than discriminate function analyses or logistic regression. Each predictor variable was evaluated univariately. Because of sample size limitations, extensive multivariate modeling was not feasible. The possibility of confounding between hippocampal W score and other variables was assessed by fitting bivariate models evaluating hippocampal W score with each of the other predictor variables individually. APOE status, estrogen replacement, hypertension, and ischemic heart disease were entered as dichotomous variables in all analyses. All tests were two-sided. Estimates of relative risk (RR) were obtained using usual, semiparametric Cox regression testing methods. A nonparametric version of Cox regression testing was used for hypothesis testing for quantitative variables. For these same reasons, confidence intervals are not reported for estimates of risk.
Tests of hypotheses for quantitative risk factors using Cox regression testing are sensitive to departures from normality; in this case, more accurate probability statements are obtained using logit rank tests.35 Because some of the data were skewed, with the degree of skew varying among successive risk sets, the logit rank test was used both univariately and bivariately in hypothesis testing.35 The logit rank tests were implemented computationally by treating the logit rank scores for each of the quantitative variables in each risk set as time-dependent covariates in a Cox regression model.
Kaplan-Meier survival curves showing the probability of crossover for patients stratified by hippocampal W scores into three groups (W ≥ 0, 0 > W > −2.5, W ≥ −2.5) were used for display purposes only, to illustrate the association between hippocampal volume and crossover to AD. The W score W ≥ 0 was selected a priori. This is a natural cut point, indicating values in patients that are equal or greater than the mean value among controls. The cutoff point W ≤ −2.5 was selected post-hoc to optimally display the gradient of the RR of crossover associated with hippocampal atrophy. A W score of −2.5 corresponds to approximately the 1 percentile (more precisely, the 0.6 percentile) among controls. These W score cut points were used for display purposes only (figure 2, table 1). All statistical analyses (tables 2 and 3⇓) were performed using hippocampal W score as a continuous variable.
Figure 2. Hippocampal W score and crossover. Kaplan-Meier curves of patients whose hippocampal W score at baseline was ≥0 (n = 13), 0 > W > −2.5 (n = 54), and ≤−2.5 (n = 13).
Characterization of patients with mild cognitive impairment
Risk of crossover: Univariate analyses
Risk of crossover: Bivariate analyses
Results.
Of the 80 patients who were classified as having MCI at entry into the study, 3 died during the follow-up period, 2 of these after they had converted to AD. The mean hippocampal W score of these 80 MCI patients was −1.24 ± 1.24, corresponding to the 11 percentile of volumes among controls after correction for age, sex, and head size. Of the entire group of 80 MCI patients, 13 had hippocampal W scores ≥ 0 at entry into the study, indicating hippocampal volumes that were at or above the mean value expected for age- and sex-matched controls. Thirteen had hippocampal W scores ≤ −2.5. Fifty-four had W scores between 0 and −2.5. Patients in the three W-score groups were similar on most demographic, clinical, and cognitive testing variables (see table 1). During the period of observation, 27 of the 80 MCI patients converted to AD. Of the 13 MCI patients with hippocampal W scores ≥ 0 at baseline, 2 converted to AD; 19 of 54 with W scores between 0 and −2.5 crossed over; and 6 of 13 with W scores ≤ −2.5 crossed over (see figure 2).
Only hippocampal volume, DRS, FCSRT free recall, and age were statistically significant predictor variables in univariate analyses of the risk of crossover (see table 2). The interpretation of hippocampal W-score result is that for each one-unit increase in the hippocampal W score (i.e., less atrophy), the RR for crossover declined by 31%. The risk of crossover declined with advancing decade of age. Carriers of the APOE ε4 allele were 49% more likely to cross over than noncarriers. Patients with better scores on the cognitive tests were less likely to cross over. The one exception was that of the COWAT, with a RR of 1.01 and a nonsignificant p value. The association between estrogen replacement and crossover was not significant (p = 0.864). Patients with a history of hypertension were more likely to cross over than those without, and patients with a history of cardiac ischemic disease were less likely to cross over than those without such a history, although neither of these associations were significant.
Separate bivariate analyses were performed with hippocampal volume, together with each of the other predictor variables (see table 3). Hippocampal volume was significant in all models. The RR ratios of hippocampal volume in all bivariate models were similar to those observed univariately, suggesting the independence of hippocampal volume as a risk factor for crossover relative to each of the other predictor variables evaluated.
Discussion.
Based on life-table analysis, we estimate that 9% of MCI patients with hippocampal W scores ≥ 0 at baseline will convert to AD within 3 years, compared with 26% of those with hippocampal W scores between 0 and −2.5 and 50% of those with W scores ≤ −2.5. There is considerable controversy whether all MCI patients will eventually progress to AD or whether MCI represents a relatively stable condition in some. Our results indicate only that the rate of conversion is greater in MCI patients with smaller hippocampi and do not address the lifetime risk of conversion.
Old age is an established risk factor for AD, and in a cross-sectional prevalence study, older age would be expected to be associated with a greater prevalence of AD.36 This is not the case in a study such as ours because the rate at which individuals with MCI progress to AD does not necessarily accelerate with age. The 80 MCI patients in this cohort shared a similar cognitive and demographic profile at study entry (see table 1). It is possible that because AD is a clinically heterogeneous disorder, patients with incipient AD who were younger may have had a more rapidly progressing form than those who were older.37
This work focused on the prediction of crossover using hippocampal volume measurements. The intent was not to conduct an exhaustive assessment of possible neuropsychological testing instruments as predictors of crossover but to test the hypothesis that the predictive power of imaging studies is independent of other potential predictor variables, such as standardized neuropsychological tests.2,38 The MMSE and DRS are measures of general cognitive function, whereas the AVLT, WMS-R-LMII, and FCSRT indices are tests of memory.28,32 The COWAT is a test of verbal fluency, measuring attention and language skills.39 In bivariate analyses paired with hippocampal W score, the associations with risk of crossover were statistically significant for the DRS and FCSRT free recall.
Other studies have found an association between postmenopausal estrogen replacement and decreased risk of developing AD.40,41 No significant association between crossover and estrogen replacement was observed in the women of this MCI group, however. The nonsignificant p value (p = 0.067) observed for hippocampal W score as a predictor of crossover when paired with estrogen bivariately (see table 3) may simply be the result of reduced statistical power when men were excluded from the analysis.
Polymorphisms of the APOE gene are a significant risk factor for developing late-onset AD.34 The ε4 allele confers both an increased risk of developing AD and also lowers the mean age at onset in a dose-dependent fashion, whereas the ε2 allele is protective. A trend was present in our data indicating that the ε3/4 or 4/4 genotypes conferred a 49% increased RR (see table 2) of crossover relative to an MCI patient with the ε2/3 or 3/3 genotypes. APOE genotype does not influence the rate of clinical progression in patients with established AD.42 We suspect, however, that if the follow-up period were extended to increase the number of crossover events, APOE ε4 would emerge as a significant risk factor for crossover. In an earlier study analyzing the risk of crossover as a function of several known risk factors (including age, family history, a variety of cognitive testing instruments, and APOE genotype), APOE ε4 emerged as the most powerful predictor variable2; however, imaging variables were not considered in that study.
Unlike genetic markers, which are present at birth, imaging studies can only identify progression of the disease itself. This is true for both structural imaging measures and functional measures such as PET, because imaging studies become abnormal only when the disease process itself has produced deviation from normal cerebral function or anatomy.43,44 Ideally, the imaging findings should represent markers of incipient disease. Our data suggest that MRI-based volume measurements of the hippocampi fulfill this criteria.
Our results generally agree with those of several studies. Fox et al.10 studied 7 patients in their 40s and 50s who were members of a family with an amyloid precursor protein 717 Val-Gla pedigree. Hippocampal volume declined more rapidly in those who declined cognitively than in those who remained stable. de Leon et al.12 used visual assessment of the size of the perihippocampal CSF spaces on CT and found individuals with atrophy were more likely to progress to dementia. As part of a study of the oldest-old, Kaye et al.11 found that the temporal lobes but not hippocampi were smaller in those who declined cognitively than in those who remained stable. The unique aspect of our study was the combined use of all the following features: 1) a rigorous quantitative MRI-based measurement; 2) a fairly large longitudinal cohort (n = 80); 3) MCI patients, who are at risk for typical sporadic AD; and 4) patients whose age was fairly typical for onset of AD rather than at the extremes of the age spectrum.
There are several limitations to this study that cannot be fully addressed at this time. The criterion used to determine the endpoint was the clinical diagnosis of AD. At our center, the clinical diagnosis of probable AD is 81% accurate compared with the pathologic diagnosis of definite AD. Thus, absolute determination of the endpoint in this study can only be ascertained in the future at autopsy. Despite the small sample size, a significant relation between premorbid hippocampal volume and crossover to AD was demonstrated, which illustrates both the strength of the association and the clinical potential of the technique. Moreover, this association was not altered when evaluated in the context of a series of bivariate analyses, which included many known risk factors for AD. Additional studies with larger sample sizes are needed, however, to more accurately delineate the rate at which risk of crossover increases with increasing hippocampal atrophy and the possibility that the relation may be nonlinear and the role of interactions with other predictor variables.
Acknowledgments
Supported by NIH-NIA-AG 11378, AG08031, AG06786, NINDS-NS29059, The DANA Foundation, and The Alzheimer’s Association.
- Received October 13, 1998.
- Accepted in final form January 27, 1999.
References
Disputes & Debates: Rapid online correspondence
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.