Mounting evidence points to Aβ accumulation as a harbinger of Alzheimer’s disease, but can subtle slippage on memory tests also raise a red flag? Not according to a massive meta-analysis of cross-sectional studies. In the January issue of JAMA Psychiatry, a large collaboration of international research groups, led by Pieter Jelle Visser at Maastricht University in the Netherlands, reported that among cognitively normal people, or people with mild cognitive impairment (MCI), those with brain Aβ accumulation were more likely to perform poorly on memory tests. However, some Aβ-negative people also faltered on the tests, making the tests a poor screening measure for preclinical AD. Even so, the researchers used the data to estimate a 10–15 year lag between Aβ accumulation and the first signs of fuzzy memory, and this many years again between early memory loss and AD dementia, in keeping with previous data.

  • In large cross-sectional meta-analysis of cognitively normal and mildly impaired people, memory was likely to slip when Aβ accumulated.
  • By itself, low memory in cognitively normal people did not predict amyloid accumulation, and was useless for trial enrollment.
  • The data confirm a 10–15 year lag between Aβ positivity and low memory scores, and 10–15 years between poor memory and AD.

While multiple cross-sectional and longitudinal studies have revealed a strong correlation between Aβ accumulation and future development of AD, the relationship between Aβ and early memory problems in otherwise cognitively normal people has proven trickier to demonstrate. Cross-sectional studies have reached contrasting conclusions, with some finding a link but others coming up short (Sperling et al., 2013Petersen et al., 2016Rowe et al., 2010). However, a recent longitudinal study reported that having Aβ accumulation at baseline significantly boosted a person’s chance of showing signs of memory loss within a decade (Jun 2017 news). 

In an effort to settle the question using as much data as possible, first author Willemijn Jansen and colleagues looked to the Amyloid Biomarker Study, a meta-analysis of cross-sectional data on biomarkers, cognition, and other metrics from 53 studies. An initial report from the project predicted a 20–30 year lag between testing positive for elevated Aβ—as assessed by thresholds set for PET scans or CSF Aβ42 concentration in each study—and reaching symptom severity that meets the definition of AD dementia (May 2015 news). 

For the current study, the researchers investigated the connection between testing positive for Aβ accumulation in the brain and low memory scores, as well as the contributions of age, ApoE4 status, and educational attainment to memory performance. Verbal word learning tests were used to assess episodic memory, though the exact tests differed between individual studies used in the meta-analysis. The researchers used z-scores to normalize the data between studies, and categorized people in the lowest 10th percentile as having “low memory scores.”

Among 2,908 cognitively normal people included in the meta-analysis, low memory scores were more frequent in people with Aβ accumulation, but only in those 70 years or older. The correlation strengthened with age, and by 80 years, cognitively normal people with elevated Aβ were almost twice as likely to perform poorly on memory tests. A higher proportion of people with MCI had poor memory compared to cognitively normal people at any age, and Aβ also played a hand in this group. Among 4,133 people with MCI, Aβ positivity correlated with low memory scores, but the strength of this association decreased with advancing age.

Why would age have the opposite effect in cognitively normal versus cognitively impaired people? The researchers speculate that as cognitively normal people age, increasing amounts of Aβ and/or other comorbid pathologies, along with increased vulnerability to Aβ, bring on cognitive decline. In people with MCI, those who do not have brain amyloid are likely to suffer from a non-AD pathology as they age. As these non-AD pathologies ramp up, the incidence of low memory scores among Aβ-positive and -negative groups converges, the researchers proposed.

Might other factors, such as sex, ApoE4, and educational attainment, sway the relationship between Aβ and low memory scores? No, it turns out. The researchers found that associations between Aβ positivity and the incidence of low memory scores held steady regardless of any of these. However, each factor associated with low memory independently of Aβ—being male, having an ApoE4 allele, or low educational attainment—all increased the odds of poor memory scores by 8 percent.

In their previous study, the researchers reported that an age-related rise in the prevalence of AD dementia occurred 20–30 years after a similar rise in cases of elevated Aβ. In the current study, they placed low memory scores at roughly the halfway point along this disease trajectory, cropping up about 10–15 years after Aβ turned abnormal, but 10–15 years before AD dementia set in (see image below). 

Memory in Motion.

A 10–15 year lags exists between the prevalence of Aβ positivity (orange) and low memory scores (blue), and again between low memory scores and AD dementia (gray). [Courtesy of Jansen et al., JAMA Psychiatry 2018.]

Ultimately, the researchers found that while Aβ positivity correlates with low memory scores, the latter could not be used to predict the former: Low scores on memory tests in cognitively normal people or those with MCI did not add to the estimation of Aβ positivity above what could already be surmised by age and ApoE4 status. 

The data confirms consistent trends in the literature, commented co-author William Jagust of the University of California, Berkeley. “There is a relationship between Aβ and memory, but this relationship is relatively weak. For this reason, many smaller studies have conflicted about this relationship, although larger studies and meta-analyses have confirmed a small association,” he said. “The size of this association makes memory an unreliable screening method for the detection of amyloid,” he added.

Rachel Buckley of Massachusetts General Hospital in Boston also commended the study for its size. “The strength of this type of study is the ability to have the statistical power to detect relatively small effects, which is arguably the case with the cross-sectional relationship between amyloid and cognition,” she wrote to Alzforum. While Buckley found the lack of predictive power of low memory scores on amyloid positivity disheartening, she noted that the very nature of a meta-analysis limits its conclusions. For example, different memory tests were used across the studies, whereas performance on a single task might better predict Aβ. “More sophisticated data harmonization techniques will need to be applied in order to better align cognitive test performance across studies,” she added.—Jessica Shugart


  1. After having read the manuscript, I am struck (again!) by the powerful predictive effect of age and ApoE4 on the likelihood of high amyloid burden in clinically normal (CN) older adults. Time and again this has been shown, however, it is interesting to see another demonstration of the importance of these factors in such a large multicenter study.

    Another important component of this study that deserves a remark up front (although, Jansen and colleagues have previously published from these data in 2015), is the sheer size of the sample. Even with ~20 percent missing data across the cohorts, this study reports on ~2000 CN individuals with a memory score, which is quite staggering. The strength of this type of study is the ability to have the statistical power to detect relatively small effects, which is arguably the case with the cross-sectional relationship between amyloid and cognition, as accurately acknowledged by the authors. With this large dataset, it is clear that the MMSE screening tool does not perform as well as other memory tests to predict high amyloid burden, however, it is slightly disheartening to see that “low performance” on the combined aggregate memory tests did not survive forward selection to be included as a sensitive predictor of high amyloid. I agree with the authors, however, in that I don’t think that this throws out the idea of a cognitive test being a useful screening tool in clinical trials; the current analyses are attempting to determine the effect of “low memory” rather than the effect of a neuropsychological task per se. The rather large limitation of big data analysis is that cruder data gradients need to be applied in order to make conservative estimates about an effect across a range of different memory tests. Here, the authors have been somewhat limited in their ability to make a final statement about the efficacy of using a memory test as a screening tool, particularly those that are sensitive and challenging neuropsychological measures of memory. In this case, more sophisticated data harmonization techniques will need to be applied (i.e., item response theory or latent factor analysis, Gross et al., 2015) in order to better align cognitive test performance across studies.

    Further, and in the same vein, it will be nice to see big data analyses being reported with a continuous measure of amyloid rather than amyloid status alone. While harmonizing across CSF and PET measures is perhaps a future endeavor, more and more PET studies are being aggregated to allow for much larger analyses. Reprocessing of PET data, and transformations across different PET tracers, will allow for greater ease when analyzing and interpreting this data, which will be very exciting, and will give greater insights into gradients of amyloidosis and cognitive decline that are more likely to represent the insidious nature of the disease. 

    With regard to age, that low memory performance doubled the likelihood of high amyloid over the age of 80 is fascinating. One issue that is not an easy one to handle is related to survivor bias. At what point does the low memory/high amyloid group become so sparse with survivors that it becomes an entirely different group from the low memory/low amyloid group altogether? Clearly, the next stage in this project will be to look ahead to longitudinal and survival analyses—which I am looking forward to reading!


    . Prevalence of cerebral amyloid pathology in persons without dementia: a meta-analysis. JAMA. 2015 May 19;313(19):1924-38. PubMed.

    . Effects of education and race on cognitive decline: An integrative study of generalizability versus study-specific results. Psychol Aging. 2015 Dec;30(4):863-80. Epub 2015 Nov 2 PubMed.

Make a Comment

To make a comment you must login or register.


News Citations

  1. At Risk, or Already Alzheimer’s? Elevated Aβ Predicts Cognitive Decline
  2. Meta-Analyses Deliver Most Definitive Data Yet on Amyloid Prevalence

Paper Citations

  1. . Amyloid deposition detected with florbetapir F 18 ((18)F-AV-45) is related to lower episodic memory performance in clinically normal older individuals. Neurobiol Aging. 2013 Mar;34(3):822-31. PubMed.
  2. . Association of Elevated Amyloid Levels With Cognition and Biomarkers in Cognitively Normal People From the Community. JAMA Neurol. 2016 Jan;73(1):85-92. PubMed.
  3. . Amyloid imaging results from the Australian Imaging, Biomarkers and Lifestyle (AIBL) study of aging. Neurobiol Aging. 2010 Aug;31(8):1275-83. PubMed.

Further Reading


  1. . Transition rates between amyloid and neurodegeneration biomarker states and to dementia: a population-based, longitudinal cohort study. Lancet Neurol. 2016 Jan;15(1):56-64. Epub 2015 Nov 18 PubMed.
  2. . Cognitive and functional changes associated with Aβ pathology and the progression to mild cognitive impairment. Neurobiol Aging. 2016 Dec;48:172-181. Epub 2016 Aug 26 PubMed.

Primary Papers

  1. . Association of Cerebral Amyloid-β Aggregation With Cognitive Functioning in Persons Without Dementia. JAMA Psychiatry. 2018 Jan 1;75(1):84-95. PubMed.