Application of Digital Cognitive Biomarkers for Alzheimer's Disease: Identifying Cognitive Process Changes and Impending Cognitive Decline.

BACKGROUND: Recent Alzheimer's disease (AD) trials have faced significant challenges to enroll pre-symptomatic or early stage AD subjects with biomarker positivity, minimal or no cognitive impairment, and likelihood to decline cognitively during a short trial period. Our previous study showed that digital cognitive biomarkers (DCB), generated by a hierarchical Bayesian cognitive process (HBCP) model, were able to distinguish groups of cognitively normal individuals with impending cognitive decline from those without. We generated DCBs using only baseline Auditory Verbal Learning Test's wordlist memory (WLM) item response data from the Mayo Clinic Alzheimer's Disease Patient Registry. OBJECTIVES: To replicate our previous findings, using baseline ADAS-Cog WLM item response data from the Alzheimer's Disease Neuroimaging Initiative, and compare DCBs to traditional approaches for scoring word-list memory tests. DESIGN: Classified decliner subjects (n = 61) as those who developed amnestic MCI or AD dementia within 3 years of normal baseline assessment and non-decliner (n = 442) as those who did not. MEASURES: Evaluated the relative value of DCBs compared to traditional measures, using three analytic approaches to group differences: 1) logistic regression of summary scores per ADAS-Cog WLM task; 2) Bayesian modeling of summary scores; and 3) HBCP modeling to generate DCBs from item-level responses. RESULTS: The HBCP model produced posterior distributions of group differences, of which Bayes factor assessment identified three DCBs with notable group differences: Immediate Retrieval from Durable Storage, (BFds = 11.8, strong evidence); One-Shot Learning, (BFds = 4.5, moderate evidence); and Partial Learning (BFds = 2.9, weak evidence). In contrast, logistic regression of summary scores did not significantly discriminate between groups, and the Bayes factor assessment of modeled summary scores provided moderate evidence that the groups were equivalent (BFsd = 3.4, 3.1, 2.9, and 1.4, respectively). CONCLUSIONS: This study demonstrated DCBs' ability to distinguish , at baseline, between impending cognitive decline and non-decline groups where individuals in both groups were classified as cognitively normal. This validated findings from our previous study, demonstrating DCBs' advantages over traditional approaches. This study warrants further refinement of the HBCP DCBs to predict impending cognitive decline in individuals and other factors associated with AD, such as physical biomarker load.

Novel Therapy of Hyperhomocysteinemia in Mild Cognitive Impairment, Alzheimer's Disease, and Other Dementing Disorders.

OBJECTIVES: Studies have produced conflicting results assessing hyperhomocysteinemia (HYH) treatment with B vitamins in patients with normal cognition, Alzheimer's disease and related disorders (ADRD). This study examined the effect of HYH management with L-methylfolate (LMF), methylcobalamin (MeCbl; B12), and N-acetyl-cysteine (CFLN: Cerefolin®/Cerefolin-NAC®) on cognitive decline. DESIGN: Prospective, case-control study of subjects followed longitudinally. SETTING: Outpatient clinic for cognitive disorders. PARTICIPANTS: 116 ADRD patients (34 with HYH, 82 with No-HYH) met inclusion and exclusion criteria to participate. No study participant took B vitamins. INTERVENTION: HYH patients received CFLN, and No-HYH patients did not. MEASUREMENTS: Cognitive outcome measures included MCI Screen (memory), CERAD Drawings (constructional praxis), Ishihara Number Naming (object recognition), Trails A and B (executive function), and F-A-S test (verbal fluency). Dependent or predictor measures included demographics, functional severity, CFLN and no CFLN treatment duration, ADRD diagnosis, memantine and cholinesterase inhibitor treatment. Linear mixed effects models with covariate adjustment were used to evaluate rate of change on cognitive outcomes. RESULTS: The duration of CFLN treatment, compared to an equivalent duration without CFLN treatment, significantly slowed decline in learning and memory, constructional praxis, and visual-spatial executive function (Trails B). CFLN treatment slowed cognitive decline significantly more for patients with milder baseline severity. CFLN treatment effect increased as baseline functional severity decreased. The analytical model showed that treatment duration must exceed some minimum period of at least one year to slow the rate of cognitive decline. CONCLUSION: After covariate adjustment, HYH+CFLN significantly slowed cognitive decline compared to No-HYH+No-CFLN. Longer CFLN treatment duration, milder baseline severity, and magnitude of homocysteine reduction from baseline were all significant predictors. There are a number of factors that could account for disagreement with other clinical trials of B vitamin treatment of HYH. Moreover, CFLN is chemically distinct from commonly used B vitamins as both LMF and MeCbl are the fully reduced and bioactive functional forms; CLFN also contains the glutathione precursor, N-acetyl-cysteine. The findings of other B vitamin trials of HYH can, therefore, only partly account for treatment effects of CFLN. These findings warrant further evaluation with a randomized, placebo-controlled trial.

The development of structure and function in the postnatal human cerebral cortex from birth to 72 months: changes in thickness of layers II and III co-relate to the onset of new age-specific behaviors.

HYPOTHESIS: The thickness of a cortical layer is a composite measure of neuronal, axonal, dendritic, synaptic, and glial numbers and sizes that may relate to thefunction of a cortical area. METHODS: 35 age-specific behaviors with defined cortical localization whose onset lies between birth and 72 months were selected. Each behavior's function localized to one or more of 12 cytoarchitectonic areas (Brodmann areas 4, with homuncular subdivisions for leg, trunk, face, and hand, plus 17, 18, 19, 20, 21, 24, 36, and 37). Data on cortical thickness for each layer of 41 cytoarchitectonic areas (including the 12 above) of the postnatal human cerebral cortex from birth of 72 months were analyzed for general patterns of change. For the 12 cortical areas functionally related to theage-specific behaviors, we searched for layer thickness changes that co-related to when the behaviors began. RESULTS: Without exception, all layers of the 41 cortical areas of the postnatal human cerebral cortex studied develop through a series of repeated thinning and thickening in a wave-like fashion. With regard to the co-relation of behavioral onset and changes in cortical layer thickness, from birth to 15 months, only layer II has agreater than expected frequency of being the layer with the greatest relative change in thickness (relative to its previous value). From 15 to 72 months, only layer IlI has a greater than expected frequency of being the layer with the greatest absolute change in thickness (81% involved a change in its direction of growth (thinning <--> thickening)). The co-occurrence of directional growth change and having the greatest layer thickness change were only statistically significant for layer III when an age-specific behavior began and was not seen for the 41 cortical areas overall (p = 0.014). CONCLUSIONS: Cortical laminar development exhibits aprocess that is mathematically consistent with a random walk with drift and with boundaries so that uncontrolled proliferation and pruning are prevented. The directional changes in layer growth could be controlled by feedback coupled with growth promoting and growth inhibiting factors. Layer II, with its function of establishing local corticocortical connections, appears to be most important in establishing age-specific behaviors of infants from birth to 15 months. Such a process tends to produce relatively simpler behaviors. LayerIII, with its function of establishing longer distance corticocortical connections, appears to be most important in establishing age-specific behaviors of children from 15 to 72 months. This process tends to produce richer, more cross-modal behaviors than those mediated primarily by local corticocortical interactions.

Acceptance of rules generated by machine learning among medical experts.

OBJECTIVES: The aim was to evaluate the potential for monotonicity constraints to bias machine learning systems to learn rules that were both accurate and meaningful. METHODS: Two data sets, taken from problems as diverse as screening for dementia and assessing the risk of mental retardation, were collected and a rule learning system, with and without monotonicity constraints, was run on each. The rules were shown to experts, who were asked how willing they would be to use such rules in practice. The accuracy of the rules was also evaluated. RESULTS: Rules learned with monotonicity constraints were at least as accurate as rules learned without such constraints. Experts were, on average, more willing to use the rules learned with the monotonicity constraints. CONCLUSIONS: The analysis of medical databases has the potential of improving patient outcomes and/or lowering the cost of health care delivery. Various techniques, from statistics, pattern recognition, machine learning, and neural networks, have been proposed to "mine" this data by uncovering patterns that may be used to guide decision making. This study suggests cognitive factors make learned models coherent and, therefore, credible to experts. One factor that influences the acceptance of learned models is consistency with existing medical knowledge.

Cortical atrophy in Alzheimer's disease unmasks electrically silent sulci and lowers EEG dipolarity.

Alzheimer's disease (AD) patients show lower dipolarity (goodness-of-fit) for dipole localizations of alpha or other dominant electroencephalography (EEG) frequency components in the occipital cortex. In the present study, we performed computer simulations to discover which of distributions of dipole activity lower dipolarity in a manner similar to that seen in severe AD. Dipolarity was estimated from simulations of various electric dipole generator configurations within the occipital cortex under conditions of widened cortical sulci (a severely demented AD case) or no sulcal widening (a normal subject). The cortical and scalp surfaces, derived from the subjects' MRI's, were assumed to be uniformly electrically conducting. Randomly placed, nonoverlapping lesions ranging from 1 to 4 mm2 per lesion were used in both the normal and AD models to simulate the electrical effect of neuropathological AD lesions. In both models, dipolarity decreased as total lesion size increased. However, the AD model showed lower dipolarity than the normal model for both individual lesion sizes and for larger total lesion sizes. The larger decline in dipolarity in the AD model appears to be due to sulcal widening which unmasks the effect of lesions buried within sulci. These simulations identify a possible mechanism explaining why sulcally-located neuropathological changes plus progressive cortical atrophy in AD brains (and presumably other cortical disorders producing atrophy) alter EEG patterns and dipolarity differently from normal cortex damaged by similar lesions.

Two-Stage Machine Learning model for guideline development.

We present a Two-Stage Machine Learning (ML) model as a data mining method to develop practice guidelines and apply it to the problem of dementia staging. Dementia staging in clinical settings is at present complex and highly subjective because of the ambiguities and the complicated nature of existing guidelines. Our model abstracts the two-stage process used by physicians to arrive at the global Clinical Dementia Rating Scale (CDRS) score. The model incorporates learning intermediate concepts (CDRS category scores) in the first stage that then become the feature space for the second stage (global CDRS score). The sample consisted of 678 patients evaluated in the Alzheimer's Disease Research Center at the University of California, Irvine. The demographic variables, functional and cognitive test results used by physicians for the task of dementia severity staging were used as input to the machine learning algorithms. Decision tree learners and rule inducers (C4.5, Cart, C4.5 rules) were selected for our study as they give expressive models, and Naive Bayes was used as a baseline algorithm for comparison purposes. We first learned the six CDRS category scores (memory, orientation, judgement and problem solving, personal care, home and hobbies, and community affairs). These learned CDRS category scores were then used to learn the global CDRS scores. The Two-Stage ML model classified as well as or better than the published inter-rater agreements for both the category and global CDRS scoring by dementia experts. Furthermore, for the most critical distinction, normal versus very mildly impaired, the Two-Stage ML model was 28.1 and 6.6% more accurate than published performances by domain experts. Our study of the CDRS examined one of the largest, most diverse samples in the literature, suggesting that our findings are robust. The Two-Stage ML model also identified a CDRS category, Judgment and Problem Solving, which has low classification accuracy similar to published reports. Since this CDRS category appears to be mainly responsible for misclassification of the global CDRS score when it occurs, further attribute and algorithm research on the Judgment and Problem Solving CDRS score could improve its accuracy as well as that of the global CDRS score.

Approximate doubling of numbers of neurons in postnatal human cerebral cortex and in 35 specific cytoarchitectural areas from birth to 72 months.

From 1939 to 1967, J.L. Conel quantitatively studied the microscopic features of the developing human cerebral cortex and published the findings in eight volumes. We have constructed a database using his neuroanatomical measurements (neuronal packing density, myelinated large fiber density, large proximal dendrite density, somal breadth and height, and total cortical and cortical layer thickness) at the eight age periods (0 [term birth], 1, 3, 6, 15, 24, 48, and 72 postnatal months) he studied. In this report, we examine changes in neuron numbers over the eight age-points for 35 von Economo areas for which Conel gave appropriate data. From birth to 3 months postnatal age, total cortical neuron number increases 23-30%, then falls to within 3.5% of the birth value at 24 months, supporting our previous work showing that the observed decrease in the number of neurons per column of cortex under a 1-mm2 cortical surface from birth to 15 months is almost entirely due to cortical surface expansion. The present study also shows a 60-78% increase in total cortical neuron number above the birth value from postnatal ages 24 to 72 months. The generalization, to humans at least, of the finding of no postnatal neurogenesis in rhesus macaques, a species belonging to a superfamily that diverged from that of Homo sapiens more than 25 million years ago, is not warranted until explicitly proven for humans. The data of the present study support the existence of substantial postnatal neurogenesis in humans for the 35 cortical areas studied.

Approximating dipoles from human EEG activity: the effect of dipole source configuration on dipolarity using single dipole models.

Dipolarity is the goodness-of-fit of the observed potential distribution with one calculated using specific assumptions about the source of the electrical potential distribution. We used computer simulations to examine the effect of different distributions of sources on their resulting dipolarity values. Electric dipoles were placed in a head-shaped model with uniform conductivity using four different dipole configurations (randomly oriented dipoles, a uniform dipole disk layer, a dipole disk layer with uniformly distributed holes, or one with randomly oriented dipoles). The best-fitting single dipole for each configuration was calculated and the dipolarity was computed as the mean squared error of the electrical potential distributions generated by the actual dipole configuration and by the best-fitting single dipole. The simulations show that: 1) a smooth dipole layer with or without holes gives dipolarities above 99.5% even when extended over areas as large as 1256 mm2; 2) randomly oriented dipoles under a smooth dipole layer also give dipolarities above 99.5%; and 3) randomly oriented and distributed dipoles, even if contained in a small portion of the total area, give dipolarities below 93.0%. These simulations show that inhomogeneity (holes) within a dipole disk layer per se do not lower dipolarity; rather, it is the random orientation and distribution of these dipoles which lowers dipolarity. Furthermore, dipolarity is not lowered by such randomly oriented and distributed dipoles when they are beneath a dipole disk layer.

Quantitative microscopic anatomy, illustrated by its potential role in furthering understanding of the processes of structuring the developing human cerebral cortex.

In this study, we searched for patterns in selected, quantitative microscopic features of the developing postnatal human cortex for 35 cytoarchitectural areas at eight age points from birth to 72 months. These data come from the largest extant survey of the microscopic features of the developing postnatal human cerebral cortex (JL Conel, 1939-67). In contrast to Jacobson's proposal that cortical surface area increases in proportion to brain weight, Conel's data show that surface area increases as brain weight2/3, with the maximal rate of increase in both brain weight and cortical surface area occurring from 1 to 3 months. We computed the numbers of cortical neurons per cortical layer under 1 mm2 of cortical surface (neurons/layer per mm2 column) and divided these values by the total neuron number/mm2 column. For all areas, these data show plateaux in most of the layers for periods of months to years, often preceded and followed by changes in their neuron number in a sinusoidal fashion. The age points of the maxima and minima of such laminar values differ across the 35 cortical areas, indicating that their sinusoids are phase-shifted with respect to each other. Ranking the six layers in each area at each age point by their neuron number/layer per mm2 column shows that, by 72 months, the first areas to receive thalamic auditory or visual input (primary sensory and unimodal association cortices) have the most neurons in layer 4 and in either layer 3 or 6. In contrast, by 72 months, other areas have the most neurons in layers 3 and 6, with the primary motor cortex reaching this ranking earlier than any other area. For temporal and parietal association areas, layers 2 (short cortico-cortical connections) and 4 (thalamo-cortical connections) have the most neurons from birth to 6 months, whereas layers 3 (long cortico-cortical connections) and 6 (cortico-thalamic connections) have the most neurons by 72 months. The quantitative, statistically non-random patterns demonstrated by our analyses suggest that hierarchical correlations between such structural changes and age-specific behavioral acquisitions exist during the first 72 months of postnatal development.

Evidence for a postnatal doubling of neuron number in the developing human cerebral cortex between 15 months and 6 years.

The generalization of the finding of no postnatal neurogenesis in non-human primates to humans may be incorrect because: (1) rhesus macaques belong to a superfamily that diverged more than 25 million years ago from the superfamily including the genus Homo; (2) the pulse thymidine labeling method, which demonstrates DNA synthesis rather than mitosis per se, is less reliable than some have assumed. This study examines changes in the number of neurons in a column underneath a cortical surface area of 1 mm2, extending through all cortical layers (mm2-column) for 35 gyri (representing about 73% of the human cerebral cortex) based on the data of J.L. Conel (1939 to 1967). We corrected these data, derived from his measures of cortical neuronal packing density, somal breadth and height, and cortical layer thickness at postnatal ages 0, 1, 3, 6, 15, 24, 48, and 72 months, for shrinkage and stereological errors. In all 35 gyri, neuron number/mm2-column: (1) initially declines (mu = 46% decline, sigma = 8%), 95% of which is due to surface area expansion (mean age of nadir value = 15.8 months); (2) then increases to age 72 months by 70% (mu = 1.7-fold increase, (mu rate = 1.1% per month). Because of a a concomitant 1.3-fold increase in cortical surface from 15 to 72 months, total cortical neuron number increases 2.2-fold. The close agreement between neuron number/mm2-column for Conel's age 72-month data to the corresponding values reported by others for adult human and primate cortex using more modern methods suggests the finding is not an artifact. Neuronal proliferative fate-determining factors provide at least four mechanisms for increasing cortical neuron number postnatally, with or without DNA synthesis.

Developmental patterns in the cytoarchitecture of the human cerebral cortex from birth to 6 years examined by correspondence analysis.

This paper uses correspondence analysis to examine the developmental patterns in the cytoarchitecture of the human cerebral cortex from birth to 72 months. The study is based on data collected by the late J. L. Conel, which consist of over 4 million individual measurements of six microscopic neuroanatomic features for each of six cortical layers in 46 cytoarchitecturally distinct regions. We analyze 1,727 profiles of development over eight age-points (term birth, 1, 3, 6, 15, 24, 48, and 72 postnatal months) resulting from the combinations of neuroanatomic feature, cortical layer, and brain cytoarchitectural region in the Conel data. The profiles for any given combination of feature and layer are found to be remarkably similar in all regions of the brain, and therefore the developmental patterns of different cytoarchitectural regions are not distinguishable from one another. Developmental change is most rapid at the earlier stages; of the total change in profile patterns observed, more than one-third occurs between birth and 6 months, about one-third occurs between 6 and 15 months, and less than one-third occurs between 15 and 72 months. The majority of the variance in developmental profiles is accounted for by the six microscopic, neuroanatomic features. Correspondence analysis shows that Conel's data are highly consistent and reliable.

Constructing the human cerebral cortex during infancy and childhood: types and numbers of cortical columns and numbers of neurons in such columns at different age-points.

This study examines JL Conel's data on neuron numbers in 35 human cortical areas for eight age points from 0 (birth) to 72 months, to analyze cortical columns, the presumed functional units of the cortex. For each cortical area at each age point, cortical surface divided by the square root of the area's neuron number gives cross-sectional areas with radii ranging from 180 microns at birth to 250 microns at 72 months. For the prefrontal cortex at birth and 48 months, these radii are approximately 2.10 and 1.19 times the longest radial basal dendrites, suggesting similar dimensions between these two measures of column radius. The logarithm of neuron number per cortical area and age point was examined in relation to the Weber-Fechner law governing the relationship between stimulus intensity and perception. A mechanism for this law consistent with the cortical model of Douglas et al. illustrates the importance of local circuit neurons. The cross-sectional areas of hexagonal columns for prefrontal cortex, using as radius, the longest radial extent of layer 5 pyramidal neuron basal dendrites, ranging from 0.013 mm2 at birth to 0.064 mm2 at 48 months, suggests that functional cortical columns increase cross-sectional area during development. These cross-sectional areas are 55-100-fold larger at birth, and 229-277-fold larger at 48 months, than those computed from somal width in prefrontal, layer 5 pyramidal neurons. Comparison of radial extent of pyramidal basal dendrites to their soma-to-soma distances shows that layer 3 pyramidal basal dendrites reach 1.5 and 4.0 other pyramidal neurons at 15 and 60 months, respectively, while layer 5, extra-large pyramidal basal dendrites reach 1.14 and 1.72 other such neurons at birth and 48 months, respectively. If such a relationship holds for other cortical areas, then the Conel data can be used to estimate basal dendrite extent, for which there currently is a paucity of data.

Simple models for estimating dementia severity using machine learning.

Estimating dementia severity using the Clinical Dementia Rating (CDR) Scale is a two-stage process that currently is costly and impractical in community settings, and at best has an interrater reliability of 80%. Because staging of dementia severity is economically and clinically important, we used Machine Learning (ML) algorithms with an Electronic Medical Record (EMR) to identify simpler models for estimating total CDR scores. Compared to a gold standard, which required 34 attributes to derive total CDR scores, ML algorithms identified models with as few as seven attributes. The classification accuracy varied with the algorithm used with naïve Bayes giving the highest. (76%) The mildly demented severity class was the only one with significantly reduced accuracy (59%). If one groups the severity classes into normal, very mild-to-mildly demented, and moderate-to-severely demented, then classification accuracies are clinically acceptable (85%). These simple models can be used in community settings where it is currently not possible to estimate dementia severity due to time and cost constraints.

A multinomial modeling analysis of memory deficits in Alzheimer's disease and vascular dementia.

Data from the immediate recall task of the Consortium to Establish a Registry for Alzheimer's Disease neuropsychological test battery were disaggregated into nine subject groups and analyzed with traditional statistics as well as with a general processing tree (GPT) model of free recall. The groups represented four levels of severity of Alzheimer's and vascular dementia, as well as a ninth group of healthy elderly controls. It was demonstrated that the patterns of success and failure of recall to individual items across successive trials contained much more information than the marginal trial-to-trial performance scores traditionally used in scoring the test. The GPT model analyzed recall performance in terms of three levels of item storage: unstored, intermediate, and long-term. Associated with the intermediate and long-term storage levels were respective retrieval parameters. Statistical methods enable one to estimate the parameters for each group, and the analyses revealed group differences in long-term storage that were not evident in a statistical analysis of the marginal trial-to-trial performance scores.

Reduced inferior olivary neuron number in early Down syndrome.

Counts of total neuron number per section and of neurons per microscopic field of inferior olivary principal nuclei were made on sections from 10 patients with Down syndrome (DS) aged 0.36 to 28 months and seven control (C) patients aged 1 to 29 months. After stereologic and appropriate shrinkage corrections of the count data, the ratios of values for DS/C were calculated. For mean principal olivary nucleus neuron number, DS/C = 0.64; for mean number of neurons per field, DS/C = 0.84; for mean volume of olivary neuronal band per section, DS/C = 0.79; and for mean volume of neuronal band per neuron, DS/C = 1.27. The data are in accord with other data suggesting that (1) numbers of cells in various cell populations, including various areas of the cerebrum, in DS approximate two-thirds normal (DS/C approximately 0.67); (2) for the volumes of such cell populations, DS/C = 0.82 normal; and (3) for volumes of individual cells, DS/C = 1.22 normal. The data of the present study suggest that the inferior olivary nuclei in DS are affected in the same way and to a similar degree as other brain areas, with the age distribution and histologic features of the specimens studied suggesting that the reduced olivary principal nucleus number in early Down syndrome results from reduced initial neuron production rather than postnatal neuron loss.

Ishihara test performance and dementia.

Ishihara trail-tracing (TT) and number-naming (NN) tests were administered to a clinical sample of 378 demented patients. Error counts on TT and NN tests were best fit by a negative binomial (overdispersed Poisson) distribution. TT, NN, and combined (TT + NN) error counts were regressed on patient characteristics (sex, age, and education), severity of cognitive impairment (Mini-Mental State Exam: MMSE), dementia stage (Clinical Dementia Rating: CDR), etiology, onset age, and symptom duration in a negative binomial generalized linear model. Patient characteristics, onset age, and symptom duration had no significant effects on any error count. The effects of MMSE, CDR, and etiology, on the other hand, were highly significant and appear to help discriminate vascular dementia from Alzheimer's disease. MMSE (which taps cognitive skills) correlated with both TT and NN errors. CDR (which taps both cognitive and functional skills) correlated only with TT errors and dementia etiology correlated only with NN errors. These distinct correlational patterns reflect differences between the TT and NN tasks (i.e., trail-tracing vs. number-naming) related to specific brain loci and associated functions. This aspect of the phenomenon suggests that Ishihara tests have useful clinical applications in dementia.

Full-information models for multiple psychometric tests: annualized rates of change in normal aging and dementia.

The rates of change for five widely used psychometric tests were analyzed to compare how much more variance reduction can be achieved using full-information methods relative to the single-equation methods previously used in dementia research. Nondemented controls and subjects with Alzheimer disease (AD), probable/ possible vascular dementia (VD), or mixed dementia (MD) were evaluated. A cohort design was followed, with follow-up of three demented groups and one normal control group; data were analyzed in a multiple-equation regression model estimated with full-information methods. The study was conducted at Alzheimer's Disease Research Center sites at the University of California, Irvine, and at the University of Southern California. In all, 226 patients and controls who had completed initial assessment and at least one annual reassessment were included in the study. Dependent variables were annualized rates of change in the Mini-Mental State Examination (MMSE), the Short-Blessed Dementia Rating Scale (DRS), the Consortium to Establish a Registry for Alzheimer's Disease drawings test (CD), the WAIS-R Block Design test (WRB), and the Boston Naming Test (BNT). Independent variables were dementia severity, diagnosis (AD, VD, MD, or control), sex, age, marital status, education, and age at onset. Full-information methods reduced the variance in the change scores by > or = 25% compared with previous studies. The model's prediction of a test's rate of change was almost entirely due to dementia stage and diagnosis. The effects of other explanatory variables (sex, marital status, age, and education) were weak and statistically insignificant. When the effects of other independent variables were controlled, AD and MD patients were found to decline at significantly faster rates than VD patients. Full-information methods, relative to single-equation methods, substantially reduce the variance of rates of change for multiple psychometric tests. They do so by simultaneously considering the correlated error terms in the regression for each dependent psychometric change score variable. The robustness of these results to minor variations in follow-up time suggests that annualization is a reasonably valid procedure for making change scores comparable. This study's results suggest that change scores in psychometric tests provide information that can be used to aid differential diagnosis. However, the large variances of change scores preclude many other uses. Finally, since standardization of psychometric change scores translates all tests to the same scale (0-100%), standardized change scores are easier to interpret. The analysis of standardized change scores deserves further investigation.

Acquisition and long-term retention of a fine motor skill in Alzheimer's disease.

This study investigated the acquisition and long-term retention of the rotary pursuit task under varying amounts of practice in 12 moderate-to-severely demented AD patients and 12 healthy older adults. Equal numbers of AD and control subjects were randomly assigned to either 40, 80, or 120 trials of training (40 trials/day) on the rotary pursuit task, followed by 15-trial retention tests 20 min, 2 days, 7 days, and 37 days following the end of practice. Performance improved significantly in both groups during the first 40 trials, while additional practice provided no ensuing positive or negative effects. Further, subjects in both groups showed minimal forgetting across the four retention tests. Therefore, the results demonstrated that AD patients can effectively learn and retain a motor skill for at least 1 month.

Neural-network-based classification of cognitively normal, demented, Alzheimer disease and vascular dementia from single photon emission with computed tomography image data from brain.

Single photon emission with computed tomography (SPECT) hexamethylphenylethyleneamineoxime technetium-99 images were analyzed by an optimal interpolative neural network (OINN) algorithm to determine whether the network could discriminate among clinically diagnosed groups of elderly normal, Alzheimer disease (AD), and vascular dementia (VD) subjects. After initial image preprocessing and registration, image features were obtained that were representative of the mean regional tissue uptake. These features were extracted from a given image by averaging the intensities over various regions defined by suitable masks. After training, the network classified independent trials of patients whose clinical diagnoses conformed to published criteria for probable AD or probable/possible VD. For the SPECT data used in the current tests, the OINN agreement was 80 and 86% for probable AD and probable/possible VD, respectively. These results suggest that artificial neural network methods offer potential in diagnoses from brain images and possibly in other areas of scientific research where complex patterns of data may have scientifically meaningful groupings that are not easily identifiable by the researcher.

Data analysis for dynamic contrast-enhanced MRI-based cerebral perfusion measurements: correcting for changing cortical CSF volumes.

The time evolution of the histogram (number of pixels versus signal intensity) is used to calculate delta R2 parameters from dynamic contrast-enhanced magnetic resonance (MR) imaging of the brain. This method partially corrects for partial volume effects and is an improvement over the approach using the signal intensity as a function of time when confounding factors such as changing cortical cerebrospinal fluid volumes are involved. The maximum value for delta R2 is found to correlate with relative cerebral blood flow as assessed by xenon inhalation and can be used to discriminate between vascular dementia and healthy volunteers. With this method, the normal range for delta R2 values is found to be the same for both young (19-40 years old) and elderly (65-85 years old) healthy volunteers.