Meta-Analysis of Amyloid and Cognition: Supplement 1

Meta-Analysis of Amyloid and Cognition: Supplement -- 1

Meta-Analysis of Amyloid-Cognition Relations in Cognitively Normal Older Adults

Trey Hedden1,2, Hwamee Oh4, Alayna P. Younger1,3, and Tanu A. Patel4

1Athinoula A. Martinos Ctr. for Biomed. Imaging, Dept. of Radiology, Massachusetts Gen. Hosp., Charlestown, MA; 2Dept. of Radiology, 3Dept. of Psychiatry, Massachusetts Gen. Hosp., Harvard Medical School, Boston, MA; 4Helen Wills Neuroscience Institute, Univ. of California, Berkeley, Berkeley, CA

Supplementary Materials

e-Methods

Methodological Assumptions. Our analysis incorporates two primary simplifying assumptions. First, that amyloid pathology can be detected via histopathological, CSF or plasma assays, or imaging biomarkers and all of these methods indicate the presence of potentially toxic forms of amyloid. To assess this assumption, we compared the effects of each broad class of measurement on cognition. Second, that individual tests of cognitive domains provide similar assessments of that domain. Because of the wide variability in tests, this assumption is relatively difficult to test. We examined episodic memory (including both verbal and spatial assessments), executive function (a broad category including tasks involving directed attention, inhibition, task-switching, and working memory), working memory (as a subdomain of executive function; this was the only subdomain reported in a sufficient number of studies to allow separate examination), processing speed, visuospatial function, semantic memory (including vocabulary and language tests), and global cognition.

Search Strategy: Selection criteria and the search strategy were defined in advance and were not modified during the search process. Keyword searches were performed on the PubMed database using the joint keyword search: (amyloid) AND ((“Pittsburgh Compound B” OR PIB OR florbetapir OR AV-45 OR florbetaben OR flutemetamol OR PET) OR (CSF OR plasma OR Abeta40 or Abeta42) OR (neuropathology)) AND (normal OR nondemented OR aging OR older OR Alzheimer’s OR dementia OR “cognitive impairment” OR MCI) AND (cognitive OR cognition OR memory OR executive OR speed OR visuospatial OR semantic). Filters were applied to limit searches to human studies in the English language and to include all dates through November 19, 2012. Database searching identified 1275 records (309 of which were reviews), with an additional 30 records from other sources (reviews, citations, and author contact), for a total of 1305 records to be screened after removal of duplicates. Of these, 1188 were excluded on the basis of title or abstract or as reviews, leaving 117 screened via full-text. Fifty-three of these were excluded because no specific cognitive domains were reported, a separate group of cognitively normal individuals could not be identified or separately assessed, because insufficient information was not provided or not available from the authors to compute effect sizes, or the selection criteria were not otherwise met. This left 64 studies meeting selection criteria to be included in the meta-analysis.

Amyloid Assessment. Histopathological methods included modified Bielschowsky, Hedreena, Hirano and Gallyas silver stains, thioflavine S & fluorescence microscopy, luxol fast blue, Nissle/hematoxylin and eosin staining methods. Also included were immunohistochemical procedures with anti-amyloid-beta antibodies (A; monoclonal antibody 10-D-5 and 6-E-10). Histopathological studies for which relations between cognition and joint amyloid plaque and neurofibrillary tangle assessments, but not a separable amyloid measure, were available were not included e1, e2. CSF and plasma assay methods included measurement of A40 and A42 monomers and the ratio of A40/A42 via enzyme-linked immunosorbant assay or xMAP Luminex technology (Luminex Corporation). PET amyloid imaging agents included were Pittsburgh Compound B (PIB), florbetapir, and florbetaben. No studies meeting the selection criteria were found for flutemetamol. Studies using FDDNP e3-e6 as an imaging agent were not included because this agent appears to have different properties from other amyloid imaging agents and labels both amyloid and tau pathology. Estimates of amyloid binding in PET studies were used from global cortex, large aggregate cortical regions, or precuneus and posterior cingulate regions (if no global or aggregate measure was reported).

Neuropsychological Assessment: Individual tasks classified as belonging to the memory domain were: Alzheimer’s Disease Assessment Scale — Cognitive Behaviour Section: delayed word recall, Associate Learning, Benton Visual Retention Test: recall, California Verbal Learning Test, East Boston Story, Face-Name Associative Memory Exam, Free and Cued Selective Reminding Test, Fuld Object-Memory Evaluation, International Shopping List Test, Logical Memory, Memory Capacity Test, One Card Learning (CogState), Paired Associate Learning (CogState), Rey Auditory-Verbal Learning Test, Rey-Osterrieth Complex Figure: recall, Selective Reminding Test, Visual Reproduction, and the Visual Association Task: recall. For the executive function domain, tasks were: Delis-Kaplan Executive Function System, Hayling task, Identification (Cogstate), Letter Fluency or Phonemic Fluency, Mental Control, Stroop: color-dot or color-word, Trail-Making Test B or B-A, and all tasks categorized in the subdomain of working memory. For the working memory subdomain, tasks were: Digit Span Forward & Backward, Letter-Number Sequencing, Listening Span,1-Back Task (CogState), and the N-Back Task. For the processing speed domain, tasks were: A Quick Test of Cognitive Speed, Conceptual Comparisons, Crossing-off, Detection (CogState), Digit Symbol, Number Comparison, Number Matching, Pattern Matching, Perceptual Comparisons, Sensorimotor Speed, Stroop: color and word subtests, Symbol Digit Modalities Test and Trail-Making Test A. For the visuospatial function domain, tasks were: Benton Visual Form Discrimination Test, Benton Visual Retention Test: copy, Rey-Osterrieth Complex Figure: copy, Block Design, Constructional Praxis and the Card Rotations Test. For the semantic memory domain, tasks were: Boston Naming Test, Category Fluency, Information, Similarities, the Visual Association Task: picture naming, and vocabulary scales. For global function, tasks were: Alzheimer’s Disease Assessment Scale—Cognitive Behaviour Section: total score, Activities of Daily Living, Blessed Dementia Scale – Cognitive Portion, Clinical Dementia Rating – Sum of Boxes, Mini-Mental State Examination, Modified Mini-Mental State, Short Blessed Test, and the Short Portable Mental Status Questionnaire. Composite scores were assigned to domains by evaluating the contribution of tasks to the composite score.

e-Results

Homogeneity Analysis and Evaluation for Outliers. Figure 1 (main text) displays funnel plots showing the distribution of effect sizes as a function of study weighting. When examining independent cohorts using PIB only or when examining independent cohorts using all amyloid assessment methods, no cognitive domain exhibited significant inhomogeneity (all ps > .11). Although episodic memory did not exhibit significant inhomogeneity, one study stands out as a large outlier for this domain: Fuld et al., 198739; because of its small size, if this study is excluded, no significant changes in the results are observed. Although some domains did exhibit studies with effect sizes outside the boundaries of the funnel plots, we kept all of these cases in the analyses because none of these domains exhibited significant inhomogeneity and this approach maximizes the available data; if any of these outlier studies are excluded, the substantive results are unchanged.

One potentially important source of inhomogeneity may be differences among the wide array of test variables included in each cognitive domain. For example, tasks classified as measuring executive function encompass multiple component processes including directed attention, inhibition, task switching, and working memory. Similarly, semantic memory encompasses multiple types of language tasks, including picture naming and category fluency. As noted above, none of the domains exhibited significant inhomogeneity in the independent cohorts analyses, suggesting that task variation was not a major source of variability in the reported effect sizes for any domain.

Effect of Assessment Method, Study Design, and Control Variables. Using the sample of studies with independent cohorts for all amyloid assessments, we examined the effects of amyloid assessment method (CSF/plasma, histopathology, or PET imaging), study design (cross-sectional or longitudinal), and whether control variables were included or not by conducting separate weighted one-way ANOVAs for each cognitive domain. For assessment method, no significant effects were observed. There were no CSF/plasma studies for working memory or visuospatial function, hence only histopathology and PET imaging studies were compared for these domains. Although it did not meet the significance threshold, there was a trend for working memory (Q(1,9) = 3.35, p = .07) such that higher effect sizes were observed with histopathological measures (z(r) = .15, SE = .05) than for PET imaging measures (z(r) = .02, SE = .04). When PIB was directly contrasted against all other assessment methods (including other PET ligands), no domain reached the FDR-corrected significance threshold. However, this comparison showed a trend for executive function (p = .08), working memory (p = .02), and speed (p = .03). Note that these results are in the context of non-significant effect sizes for both working memory and speed. Nonetheless, these results may provide an indication of greater variability across assessment methods in executive function, its subdomains, and speed than in other cognitive domains. However, we note that all but one of the studies using other assessment methods and showing a large positive effect size (r > .2) for any of these domains have sample sizes of fewer than 100 subjects, so it will be important to examine these differences across assessment methods in subsequent larger studies.

For study design, no significant effects were observed. There were no longitudinal studies in the independent cohorts analysis for working memory, and only one longitudinal study each for executive function and processing speed; hence no statistics could be computed for these domains. Because of the low number of longitudinal datasets available, even modest effect size differences due to study design are unlikely to be detected.

There was little consistency across studies in the control variables entered, although age was the most common control variable. We coded whether age (regardless of whether it was accompanied by other control variables) was entered into the analysis of the relation between amyloid burden and cognition. Whether control variables were included in the analysis or not was not significant for any domain.

Results in All Studies. As exploratory analyses to provide the possible range of effect sizes if more power were available, we examined the data from all available studies, not limited to those with independent cohorts. Studies could contribute both a cross-sectional and longitudinal dataset to this analysis, under the assumption that the two measurements capture different information. Removing this assumption and allowing each study to contribute only one dataset does not materially alter any result. The reported subject count (Table e-2) is adjusted so that such studies are only represented once. When all studies, representing a maximum of 7018 subjects, are included, all cognitive domains except visuospatial function had a significant mean effect size (Table e-2). Notably, the observed effect sizes are very consistent with or somewhat inflated relative to the independent cohorts analysis reported in the main text; the extent of these relative changes across analyses can be visualized in the shift from the solid to the dashed lines in Figure 1 in the main text for each domain. However, one-way ANOVAs examining the mean effect sizes in studies selected as independent cohorts versus studies not so selected found no significant effects of selection (all ps > .11), indicating that the non-independence of samples was unlikely to be a major contributor to the meta-analytic findings. Studies selected as independent cohorts tended to have the largest, and therefore most heavily weighted, samples in the meta-analysis. Nonetheless, results from the non-independent cohorts analysis should be interpreted with caution.

Meta-Analysis of Amyloid and Cognition: Supplement -- 1

Table e-1. Study Characteristics and Effect Sizes for All Studies Meeting Selection Criteria

Studies are grouped by cohort, with each cohort given a unique abbreviation. Studies listed more than once contributed multiple datasets to the analysis. Effect sizes are Fisher-z transform of r. The weight w for each study is n – 3. Bold values were selected to represent the cohort in the independent cohorts analysis. Italicized studies indicate that effect sizes were computed in part from additional unpublished information provided by the authors. aEwers et al., 2011 e10 used imputed PIB values estimated from CSF A42. bPike et al., 2007 e12 was included in the AIBL cohort because the authors indicated that most subjects in this study were later enrolled into AIBL (C. Rowe, personal communication). cEffect sizes for Rodrigue et al., 2012 e36 were computed using only the subsample of adults aged 60+ so as to be most comparable to the other studies (K. Rodrigue, personal communication). dJack et al., 2008 e41 was included in the MCSA cohort because the authors indicated these subjects were enrolled in the MCSA (C. Jack, personal communication). eEffect sizes for Mielke et al., 2012 e45 were computed using the 1.5 PIB threshold. fOnly 47 subjects had available data for the memory measure in Gomperts et al., 2012 e50 (S. Gomperts, personal communication). fCross-sectional data for Riley et al., 2011 e59 had a smaller sample size than the longitudinal data due to greater missing-ness at baseline. hEffect size for Gu et al., 2012 e63 was computed by comparing the highest and lowest A tertiles. EM = episodic memory, EF = executive function, WM = working memory, PS = processing speed, VS = visuospatial function, SM = semantic memory, GF = global function.

Table e-2. Effect Size Statistics for Exploratory Analysis of Non-independent Cohorts.

N = number of analytic datasets, n = number of subjects across datasets; r = weighted mean effect size (calculated with inverse Fisher-z transform); SD = weighted standard deviation. *pFDR < .05

Meta-Analysis of Amyloid and Cognition: Supplement -- 1

e-References

e1.Galvin JE, Powlishta KK, Wilkins K, et al. Predictors of preclinical Alzheimer disease and dementia: a clinicopathologic study. Arch Neurol 2005;62:758-765.

e2.Wilson RS, Segawa E, Hizel LP, Boyle PA, Bennett DA. Terminal dedifferentiation of cognitive abilities. Neurology 2012;78:1116-1122.

e3.Braskie MN, Klunder AD, Hayashi KM, et al. Plaque and tangle imaging and cognition in normal aging and Alzheimer's disease. Neurobiol Aging 2010;31:1669-1678.

e4.Ercoli LM, Siddarth P, Kepe V, et al. Differential FDDNP PET patterns in nondemented middle-aged and older adults. Am J Geriatr Psychiatry 2009;17:397-406.

e5.Ossenkoppele R, Tolboom N, Foster-Dingley JC, et al. Longitudinal imaging of Alzheimer pathology using [(11)C]PIB, [ (18)F]FDDNP and [ (18)F]FDG PET. Eur J Nucl Med Mol Imaging 2012;39:990-1000.

e6.Small GW, Siddarth P, Kepe V, et al. Prediction of cognitive decline by positron emission tomography of brain amyloid and tau. Arch Neurol 2012;69:215-222.

List of references meeting criteria for inclusion in the meta-analysis (listed Table e-1):

e7.Balasubramanian AB, Kawas CH, Peltz CB, Brookmeyer R, Corrada MM. Alzheimer disease pathology and longitudinal cognitive performance in the oldest-old with no dementia. Neurology 2012;79:915-921.

e8.Mormino EC, Kluth JT, Madison CM, et al. Episodic memory loss is related to hippocampal-mediated β-amyloid deposition in elderly subjects. Brain 2008;132:1310-1323.

e9.Schott JM, Bartlett JW, Fox NC, Barnes J, Alzheimer's Disease Neuroimaging Initiative Investigators. Increased brain atrophy rates in cognitively normal older adults with low cerebrospinal fluid Aβ1-42. Arch Neurol 2010;68:825-834.

e10.Ewers M, Insel P, Jagust WJ, et al. CSF Biomarker and PIB-PET-Derived Beta-Amyloid Signature Predicts Metabolic, Gray Matter, and Cognitive Changes in Nondemented Subjects. Cereb Cortex 2012;22:1993-2004.

e11.Vemuri P, Weigan SD, Przybelski SA, et al. Cognitive reserve and Alzheimer’s disease biomarkers are independent determinants of cognition. Brain 2011;134:1479-1492.

e12.Pike KE, Savage G, Villemagne VL, et al. Beta-amyloid imaging and memory in non-demented individuals: evidence for preclinical Alzheimer's disease. Brain 2007;130:2837–2844.

e13.Rowe CC, Ng S, Ackermann U, et al. Imaging beta-amyloid burden in aging and dementia. Neurology 2007;68:1718-1725.