Hippocampal volume as an index of Alzheimer neuropathology

Study population.

Participants in the Nun Study are members of the School Sisters of Notre Dame congregation and live in communities in the midwestern, eastern, and southern United States. The Nun Study, a longitudinal study of aging and AD, is described in more detail elsewhere.19-21⇓⇓ Cognitive and physical functions were assessed annually, and all participants agreed to brain donation at death. By early 1998, 256 sisters had died and received a neuropathologic evaluation. Postmortem MRI scans were initiated beginning in 1996, and the current sample is based on the first 56 brains that were scanned.

Assessment of dementia.

Cognitive function was evaluated annually using the Consortium to Establish a Registry for Alzheimer’s Disease (CERAD) neuropsychological battery, which assesses memory, concentration, language, visuospatial ability, and orientation to time and place.22 Assessments of social and daily function were made using performance-based testing.23,24⇓

Individuals considered to be demented in our study had each of the following conditions: 1) impairment in memory, indicated by a score of <4 on the Delayed Word Recall Test; 2) impairment in at least one other area of cognition, indicated by scores of <11 on the Verbal Fluency Test, <13 on the Boston Naming Test, or <8 on the Constructional Praxis Test; 3) impairment in basic or instrumental activities of daily living (defined by inability to use a telephone, handle money, or dress oneself); and 4) evidence of decline in ability from a previous level attributable to cognitive impairment. The cutoffs used to identify impairments in cognition were based on scores that were less than the fifth percentile for the normative data described by the CERAD group.25

Neuropathologic evaluation.

Gross and microscopic evaluations of the participants’ brains were performed by a neuropathologist who was blinded to the participants’ cognitive test scores. For histopathologic evaluation, multiple sections of neocortex, hippocampus, entorhinal cortex, amygdala, basal ganglia, brainstem, and cerebellum were stained with hematoxylin and eosin and the modified Bielschowsky stain. Senile plaques (both diffuse and neuritic types) and neurofibrillary tangles were counted in Bielschowsky-stained sections of the five most severely involved microscopic fields of the middle frontal gyrus (Brodmann area 9), inferior parietal lobule (areas 39/40), middle temporal gyrus (area 21), and CA1 region of the hippocampus. In addition, brains were staged for the severity of neurofibrillary pathology using Gallyas and Bielschowsky stains.2

To meet the study’s neuropathologic criteria for AD, participants were required to have 1) ≥16 senile plaques/mm² microscopic field in the frontal, temporal, or parietal lobe of the neocortex (sufficient to meet the Khachaturian criteria26); 2) neuritic plaques in the frontal, temporal, or parietal lobe; and 3) neurofibrillary tangles in the frontal, temporal, or parietal lobe.

Postmortem brain preparation and MRI acquisition.

The brain was removed from formalin, inspected for cuts or gross lesions, and immersed for 15 minutes in cold water. It was then placed in a watertight plastic cylinder containing dilute formalin and rotated to remove bubbles and trapped air from the ventricles. Paper packing was used to prevent movement of the brain within the container during imaging. This inner container was anchored in an outer water-filled loader aligned with fiducial markers. The loader was carefully centered in the head coil using the same fiducials.

MRI was performed on a 1.5 T scanner. Three-dimensional T1-weighted coronal images of the entire brain were acquired using a magnetization prepared rapid gradient echo [MP RAGE] sequence (repetition time [TR] = 11 milliseconds, echo time [TE] = 4 milliseconds, 180 × 2562 matrix, flip angle = 8°, true resolution = 1 × 1 × 2 mm). Additional axial slice series were acquired in a plane parallel to the anterior commissure–posterior commissure line (T1-weighted: TR = 400 milliseconds, TE = 18 milliseconds, true resolution = 0.5 × 0.5 × 3 mm; T2-weighted double echo: TR = 2,500 milliseconds, TE = 20 and 80 milliseconds, true resolution = 1 × 1 × 3 mm).

To reduce the influence of differences in fixation time, we attempted to keep this delay uniform in duration across brains. On average, MRI scanning was performed 6 weeks following death.

Imaging analyses.

Hippocampal volumes were calculated by an automated program called the Knowledge-Guided MRI Analyses Program (KGMAP). A more complete description of this program and its validation against manual tracing methods are given elsewhere.27,28⇓ KGMAP’s logic combines knowledge rules based on intensities with the spatial relationship of anatomic structures. The combination of these rules results in the localization and identification of regional structures of interest throughout the brain. Once identified, the volume of each region is calculated and regional analyses are performed.

Briefly, KGMAP analyses begin with the identification of the lateral ventricles, caudate nuclei, lenticular nuclei, and thalamus in the first image. The locations of these structures, once identified, are used to identify their possible presence in subsequent slices. KGMAP begins searching for the hippocampus on the first image. However, the hippocampus typically does not begin to appear until the image where the temporal horn and body of the lateral ventricle separate in the axial plane. This generally occurs within two axial images below the first image analyzed by KGMAP. As the analysis of KGMAP works downward from the first image into the temporal lobe of the brain, the gray matter comprising the hippocampus is initially identified as that tissue in the space separating the temporal horn and body of the lateral ventricle. In the axial plane, its superior border lies inferior to the remaining body of the lateral ventricle, while its lateral and inferior borders lie medial and superior to the temporal horn of the lateral ventricle. In the case of minimal atrophy, the CSF in this region can be convoluted by partial volume effects in the surrounding tissue. If a strong indication of CSF is not present, the lateral and inferior borders are identified by pixels indicating a strong presence of white matter. If these procedures do not completely isolate these borders, an algorithm-generated interpretative line, aided by pixels that were found, is used to close the bordering line. The use of hippocampal pixel locations from previous slices aids greatly in the total identification of this structure. Volumetric measures are calculated through a summation procedure that includes all images in which the hippocampus is identified.

Data analysis.

Because AD neuropathology was assessed using sections from the left hemisphere and our primary interest was the association of hippocampal atrophy and neuropathologic markers, we used left hippocampal volume to correlate with neuropathologic outcomes. When similar analyses were performed using right hippocampal volume and AD neuropathology from the left hemisphere, the findings were not significantly different. The sensitivity and specificity of left hippocampal volume as a predictor of Alzheimer neuropathology were obtained for every observed value and represented as receiver operating characteristic (ROC) curves.29 ROC curves are generated by calculating the sensitivity and specificity of every observed predictor value and plotting 1-specificity against sensitivity. The area under the ROC curve was used to assess the diagnostic accuracy of hippocampal volume in predicting fulfillment of AD neuropathologic criteria as well as differences in Braak staging. To assess whether age and education affected the diagnostic accuracy of hippocampal volume, all ROC curves were rerun using hippocampal volume adjusted for age and education. All analyses utilized SPSS 10.1 for Windows (SPSS, Chicago, IL). This program provides 95% CI for the area under the ROC curve and generates estimates of statistical significance of the departure from neutrality (area = 0.5). Finally, we compared different predictors of the same outcome (satisfaction of neuropathologic criteria for AD) using standard procedures.30 This approach reduces the standard error of between-area differences by taking into account correlations between two different variables measured on the same sample, resulting in an increased power of the test.

Although high sensitivity is desirable, the practical application of this technique to detect very early, asymptomatic cases of AD could lead to false identification of individuals as AD cases, with consequent ethical problems. Therefore, in specifying optimum cut points for separating individuals with specific levels of neuropathologic lesions, we adopted a minimum specificity of 80%. Optimum cut points were selected as those contributing to the highest sum of specificity and sensitivity in which the specificity was ≥80%.

Differences in means for age at death, interval between last exam and death, Mini-Mental State Examination (MMSE) score at the final exam before death, years of education, Braak stage, and left hippocampal volume were evaluated in four groups defined by satisfaction of neuropathologic criteria for AD and presence/absence of dementia using a one-way analysis of variance. Post hoc comparisons utilized Bonferroni adjustment for multiple comparisons.

Summary of cases.

The sample of 56 sisters analyzed in this study included 33 who fulfilled neuropathologic criteria for AD and 23 who did not. The relatively high percentage of sisters fulfilling neuropathologic criteria is likely the result of the high mean age at death31 and the fact that mortality is higher among individuals with this disease,32 which would increase their representation in the group that dies first. Table 1 shows characteristics of the four groups defined by neuropathology and dementia. Demented sisters had fewer years of education than nondemented sisters (p < 0.05), and Braak stage was higher in those meeting AD neuropathologic criteria (p < 0.001). The volume of the left hippocampus was smaller among demented sisters meeting AD neuropathologic criteria than in the other three groups (p < 0.01) and larger (p < 0.01) among nondemented sisters who did not meet neuropathologic criteria than in both groups who met these criteria. Finally, both groups of demented sisters had lower MMSE scores at the final evaluation prior to death in comparison with both groups of nondemented sisters (p < 0.01). Mean MMSE scores obtained by the two demented groups were not significantly different from each other, nor were mean scores obtained by the two nondemented groups significantly different from each other.

Predictive value of hippocampal volume in detecting neuropathologic AD.

Table 2 summarizes the findings of ROC analyses for the entire sample as well as several subgroups. The predictor is the inverse of left hippocampal volume (greater volumes correspond to lower values) and the predicted outcome is fulfillment of neuropathologic criteria for AD. For the entire sample, the area under the ROC curve is 0.91 (95% CI: 0.83 to 1.00; p < 0.001). Sensitivity at the optimum cut point was 0.82 and specificity 0.87. The positive predictive value of a hippocampal volume less than the cut point was 0.90 and negative predictive value 0.77. To determine the predictive value of hippocampal volume in individuals who did not exhibit dementia during life, we conducted similar ROC analyses for the 32 sisters who remained nondemented as well as two subgroups: 8 sisters who remained nondemented but were judged to have a significant memory impairment in the absence of other cognitive impairments (“memory impaired”) and 24 sisters who were intact with regard to memory and cognition at the final examination prior to death (“cognitively normal”). Memory impairment was determined by a score of <4 of a possible 10 on the CERAD Delayed Word Recall Test, whereas cut points for other cognitive impairments were the same as those described above in Assessment of dementia. Although lacking evidence of memory complaint, the nondemented group with a significant memory impairment would be similar to the group identified with clinical criteria for MCI.33 Finally, we conducted an ROC analysis for the 24 demented sisters.

Areas under the curve and their 95% CI, sensitivity, specificity, and positive (PPV) and negative (NPV) predictive values at the optimum cut points are shown in table 2 for these groups. In every group, hippocampal volume significantly predicted group membership. Areas under the respective ROC curves were similar to each other. All ROC analyses were rerun, adjusting hippocampal volume for age and education. In almost all cases, these adjustments had no effect on the ROC curve. The only difference seen was a slight, nonsignificant decrease in area under the curve from 0.923 to 0.917 for the analysis of the 32 nondemented sisters.

Hippocampal volume compared with delayed memory measure in nondemented sisters.

To determine whether hippocampal volume was a better indicator of AD neuropathology than a delayed memory measure among individuals who remained nondemented, we compared ROC curves for the inverse of left hippocampal volume with the number of words (up to 10) from the CERAD Word List Test not recalled after a delay (figure 1). Areas under the curves were 0.92 (95% CI: 0.83 to 1.02) and 0.78 (95% CI: 0.60 to 0.96) for hippocampal volume and delayed memory. The difference between areas was 0.14 (95% CI: −0.03 to 0.32; p = 0.11). Although the difference between areas was not significant, at the optimum cut points, the positive and negative predictive values for hippocampal volume (0.82 and 1.00) were larger than those for delayed memory (0.75 and 0.75).

• Download figure
• Open in new tab
• Download powerpoint

Figure 1. Receiver operating characteristic curves comparing prediction of fulfillment of AD neuropathologic criteria from the inverse of left hippocampal volume (solid line) and number of words not recalled in the final administration of the Consortium to Establish a Registry for Alzheimer’s Disease Delayed Word Recall Test before death (dashed line).

Hippocampal volume prediction of Braak staging in nondemented sisters.

To assess whether hippocampal volume was a sensitive and specific indicator of the Alzheimer degenerative process prior to spread of neurofibrillary tangles to the cortex, we investigated the value of hippocampal volume in predicting Braak stage in our nondemented subsample. Of the 32 nondemented sisters, 20 were rated Braak stage II or lower, whereas 12 were reported as having Braak stage III or VI. Figure 2A shows the ROC curve for this analysis, where the outcome is Braak stage III to VI vs Braak stage II or less. The area under this curve is 0.84 (95% CI: 0.70 to 0.99; p = 0.001). With use of the same definition for an optimum cut point, sensitivity was 0.83 and specificity 0.80. Positive and negative predictive values were 0.71 and 0.89.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2. Receiver operating characteristic curves showing prediction of Braak stages III and IV from Braak stages II and lower in 32 nondemented sisters (A) and prediction of Braak stage II from Braak stages I or lower among 18 sisters with Braak stages II or less who were cognitively normal (B).

We repeated these analyses restricting our subject group to the 18 individuals who maintained normal cognition and received a Braak stage of II or less. Figure 2B shows the ROC curve comparing the 12 individuals in stage II with the 6 in stage I or less. In this analysis, the area under the curve was 0.83 (95% CI: 0.63 to 1.04; p = 0.025), and sensitivity, specificity, positive predictive value, and negative predictive value at the optimum cut point were 0.58, 0.92, 0.88, and 0.50.