. NIA-AA Research Framework: Toward a biological definition of Alzheimer's disease. Alzheimers Dement. 2018 Apr;14(4):535-562. PubMed.


Please login to recommend the paper.


  1. I support the proposal by Jack and colleagues to move the diagnosis of Alzheimer's disease to an etiological basis, rather than to continue to rely on syndromic diagnoses. I have advocated for this shift for many years, and am pleased that the evolution and growing acceptance of biomarkers for AD can help to accomplish this goal.

    At Washington University, our approach for our research cohort is to provide a clinical determination of the underlying etiology (or etiologies) of the cognitive impairment and use that etiology as an independent variable for correlation with the research biomarker findings. Hence, biomarkers do not enter into our research clinical diagnostic algorithm. 

    In our non-research clinic, i.e. faculty practice, where we provide patient evaluations and management, increasingly we use biomarkers (amyloid PET when it was supported by the IDEAS study, but now that IDEAS has ended we primarily use the ADMark CSF assay provided by Athena Diagnostics) to provide support (or not) for our clinical diagnoses in difficult cases (e.g., when the cognitive impairment is minimal, or there is an atypical presentation, such as early onset dementia or prominent unusual features).​

    We diagnose two stages of Alzheimer disease: asymptomatic (preclinical AD, diagnosed by biomarkers—this is only in research participants) and symptomatic, encompassing MCI due to AD and AD dementia. In our research, our investigators are examining the A/T/N system, particularly as the distinction between tau and neurodegeneration may be important. What seems less useful to me in the A/T/N paper is the call for a new system of syndromal staging. The Clinical Dementia Rating, which designates cognitive normality (CDR 0) and then the various levels of severity of symptomatic AD (CDR 0.5, 1, 2, and 3, designating very mild, mild, moderate, and severe impairment), has been well-established and widely adopted and, to my mind, already serves to indicate the stages of symptomatic AD. I don’t think that another, overlapping staging system is a good idea. 

    View all comments by John Morris
  2. In 2011, a task force for NIA and the Alzheimer’s Association conceptualized Alzheimer’s disease as a progressive sequence of pathophysiological changes which correspond roughly to preclinical (“cognitively unimpaired”), MCI, and dementia stages. The 2018 framework builds upon that conceptualization in thoughtful and important ways. As the authors note, the biomarker-based AT(N) criteria, as well as the cognitive stages with which different biomarker categories are associated, promise to advance the scientific study of AD and the development of new treatments, particularly in the preclinical and early clinical stages of the disease.

    The framework also notes limitations in categorical biomarker measurements of underlying pathology. It reminds us that the framework should not restrict alternative approaches to the study of the disease (e.g., using continuous biomarker measurements), and appropriately notes that additional work is needed before the framework should be considered in the clinical setting. As the authors note, there is a growing need and new opportunities to find less-expensive and -invasive biomarkers (including blood tests), such that this approach could be applied in the largest number of people and galvanize research and care. There is also a need to find biomarkers for comorbid pathologies which, along with the AD biomarkers, may help to inform a person’s diagnosis, prognosis, and treatment.

    It is possible to envision a time in which a person’s disease state will be characterized by a person’s AT(N) biomarkers, biomarkers of other neurodegenerative or cerebrovascular pathologies, genetic background, cognitive stage, and associated features. That kind of information would help to advance our understanding of AD and help to tailor the right treatment to the right person at the right preclinical or clinical stage of the disease.

    It has taken considerable time, patience, and persistence to articulate the relevant issues, respond to feedback from a large number of stakeholders (some of us more vocal than others), and advance the conceptualization of AD in this important way. I’m grateful to NIA, the Alzheimer’s Association, and the contributors for their important efforts.

    View all comments by Eric M. Reiman
  3. This consensus paper is a major step forward for the development of an accurate definition of Alzheimer’s disease. We are soon approaching a stage where at least one of the dementia disorders can be defined in vivo based on evidence of the presence of key pathologies, and not based mainly on symptomatology. This is a natural development, considering that many other diseases are defined using biological parameters and are not solely based on the symptoms they cause; e.g., colon cancer is colon cancer when certain histological criteria are met and the definition does not require that the tumor has yet caused any overt symptoms.

    I think the current proposal is well-balanced and thoughtful. I believe it is important that the definition of AD requires in vivo evidence of both Aβ plaques and pathologic tau deposits. Further, the possibility to not only classify the biomarkers into abnormal or normal, but also into clearly normal (0), intermediate range (1), and clearly abnormal (2) is most welcome, and this system might be even more refined in the future. Needless to say, a PET scan showing a very subtle increase in the retention of a certain tau PET ligand in the entorhinal cortex does not give the same information as a scan exhibiting widespread and high uptake throughout the brain. Also, several recent longitudinal studies show that declining CSF Aβ42 levels or increasing Aβ amyloid PET SUVRs are associated with AD-like progression even when the values are still within the normal range. Finally, I think it is an important development that other key pathologies, like cerebrovascular disease, can be incorporated in the model in the future.

    However, I personally think that we should be somewhat cautious when using CSF T-tau and P-tau as markers of neurodegeneration and tau pathology, respectively. We do still not fully understand which molecular and cellular events these markers actually represent when elevated in the CSF, even though they are clearly increased in most AD patients many years before symptom onset. For example, there is evidence from experimental studies that certain tau species are released by neurons when activated and they are not only released when cells are degenerating. This is evident when considering that T-tau and P-tau are present in the CSF even during childhood in the absence of overt neurodegeneration or tau pathology; actually CSF P-tau levels are very high in neonatal infants. As the authors write, it is probable that certain synaptic markers in CSF or maybe PET ligands reflecting synaptic density might be better markers of neurodegeneration.

    View all comments by Oskar Hansson
  4. This new framework is the natural evolution of the 2011 NIA-AA and earlier IWG guidelines. The authors are thoughtful and innovative in their attempt to unify the Alzheimer’s biomarker and clinical continuum. They acknowledge that a number of important challenges remain.

    Relying on a binary cutoff to determine biomarker status will not fully capture the important contribution that the severity of biomarker change plays in determining disease stage, clinical course, and response to treatment. Continuous measures would be ideal, if available, but at least three levels of biomarker status should be captured: negative, intermediate, and positive.

    Equating CSF p-tau with tau PET is problematic, as the timeline for the evolution of abnormal tau values can vary significantly between these measures in some forms of AD, as shown in the DIAN and Colombian cohorts. A measure of cerebrovascular disease is not yet included. Useful MRI measures of cerebrovascular disease are already available. The criteria for denoting (N+) for atrophy on MRI are unclear. New biomarkers, such as synuclein, TDP-43, and inflammatory markers etc., can be added as they become available.

    Separating the biomarker and clinical syndromes is imaginative, and a number of new terms are proposed. The five syndromal levels make sense but that is followed by a six-point staging syndrome for the AD continuum that may be confusing for some people, at least at first. Neurobehavioral symptoms are now included in the clinical phenotype of AD, which is an advance.

    Genetic factors are not currently included in the framework. For example, a designation for a CU-dominant mutation carrier without evidence of AD pathological change could be a good candidate for a treatment that prevents the accumulation of fibrillar amyloid.

    Overall, this framework represents an important advance for our field and I look forward to working with my colleagues to implement and test it.

    View all comments by Stephen Salloway
  5. Jack et al. have proposed a “research framework” for thinking about Alzheimer’s disease. It seeks to define Alzheimer’s disease by biomarkers rather than clinical findings. The authors admit that this approach has some inherent problems and thus describe it as a “research framework,” but they fail to appreciate some of the broader, unintended consequences of their effort. 

    Because they call their effort an “update” to 2011 NIA-AA diagnostic criteria for Alzheimer’s disease, it has the potential to undermine that existing consensus while the promise of biomarkers in clinical diagnosis is still unrealized. This proposal moves research even further away from clinical realities, so that research findings will be less relevant. The reality is that pathological biomarkers are not yet reimbursed or widely available. Some of the measures are susceptible to technical variability (Mattsson et al., 2013) and not all have been validated compared to autopsy diagnosis.

    The clinical implications of research proposals cannot be ignored. Without clinical access to biomarkers, the proposal actually reinforces the old, but still frequently stated opinions that “you can only diagnose Alzheimer’s at autopsy”—which of course hardly ever occurs—and “Alzheimer’s is a diagnosis of exclusion.” Today, most patients with Alzheimer’s disease are not diagnosed and are less likely to be after this proposal. This is not the intention of the authors, but this misunderstanding of the implications of the framework will find a receptive audience nevertheless.

    Part of their argument for biomarker-dependent diagnosis is that the presence of multi-domain amnestic dementia is inadequate. No one doubts that dementia syndrome is not a disease, but current clinical criteria for Alzheimer’s disease dementia don’t end there. Conflating the presence of pathology as disease gives solace to those who don’t value the painstaking efforts of clinicians to determine the specific causes of cognitive impairment using all available data and individualizing care. A disease has a known natural history and prognosis that can be used to guide management. This critical knowledge base is lacking with a definition based upon pathology alone. Redefining disease as pathology identified with biomarkers is appealing in its simplicity, but it risks “desktop medicine” that fails to consider clinical context and ignores the heterogeneity of disease expression and patient response (Karlawish, 2010). 

    There is also the unintended problem of greater confusion about the meaning of Alzheimer’s disease among clinicians, patients, families, and policymakers. Great efforts to increase public awareness of Alzheimer’s disease are at risk when current perception is at odds with expert opinion. The most common question I’m asked by both clinicians and patients is, “What is the difference between dementia and Alzheimer’s?” I believe that confusion between common and research terminology contributes to lack of consensus about the value of diagnostic testing and insurer decisions not to reimburse amyloid PET imaging and biomarker quality MRI. I have begun to be more careful to distinguish when I mean Alzheimer’s disease dementia, Alzheimer’s disease pathology, and mild cognitive impairment due to Alzheimer’s disease pathology. I fear that the framework dismisses these subtleties rather than clarifying them.  

    Finally, the inflexibility of the proposed framework that requires several types of biomarkers for Alzheimer’s disease may actually inhibit rather than enhance research. What is to be done when research participants can only be partially characterized with biomarkers? What happens when more than one disease pathology could be present? Is it progress for a research framework not to ignore this complexity and differences in clinical presentation, variable psychiatric and motor complications, or disease course? The framework we choose for observational cohort studies carries the risk of limiting the questions we ask. It is reasonable and important to characterize research participants as fully as possible. It is also reasonable to distinguish neurodegenerative, amyloid, and plaque biomarkers, but I believe the disadvantages outweigh the advantages of defining disease independent from clinical symptoms.


    . Desktop medicine. JAMA. 2010 Nov 10;304(18):2061-2. PubMed.

    . CSF biomarker variability in the Alzheimer's Association quality control program. Alzheimers Dement. 2013 May;9(3):251-61. PubMed.

    View all comments by Norman Foster
  6. The NIA-AA 2018 framework for Alzheimer’s disease changed scope and now has the ambition to provide a research-only framework, with the explicit disclaimer that it should not be seen as criteria for clinical use. The authors did a great job in reviewing and synthesizing the current state of the field.

    A novelty of the 2018 framework compared to the NIA-AA 2011 criteria is that instead of using separate criteria for three different clinical stages, AD is now conceptualized as a biological construct, independent of cognitive status. This makes NIA-AA 2018 aligned with the International Working group (IWG) criteria and is a major improvement (Visser et al., 2012). 

    Another change is that the 2018 AD classification scheme separates tau markers (T) from neuronal injury markers (N) (parentheses around N are intentional to indicate that markers are not specific for AD). This separation implies that these markers are distinct entities, but this is not the case given the correlations reported between T and (N) markers. In particular the distinction between phosphorylated tau (p-tau, a T marker) and total tau (t-tau, an (N) marker) is problematic given their high correlation in AD. The argument for distinguishing between these markers is that t-tau, unlike p-tau, is increased in several other neurodegenerative disorders. However, in amyloid-positive individuals it is conceivable that amyloid-related processes cause increases in both p-tau and t-tau. Another example is the association of tau-PET retention (T) with brain atrophy ((N) and with glucose hypometabolism (N) (Ossenkoppele et al., 2016; Schöll  et al., 2017; Gordon et al., 2018). Practically, the correlation between T and (N) markers means that, depending on the choice of biomarkers, the frequency of T–(N)+ and T+(N)– categories can become very low.

    While the new classification scheme has removed tau markers from the (N) category, this category still lumps together markers that reflect different processes. This has the disadvantage that, even though such processes may correlate, studies with an AT(N) classification with biomarker set X may give different results for prevalence or outcome than an AT(N) classification with biomarker set Y (Vos et al., 2016; Bertens et al., 2017). As each biomarker reflects unique aspects of brain dysfunction there is no solution other than describing pathologies separately.

    The previous NIA-AA criteria and IWG criteria staged individuals based on clinical markers (C). The 2018 framework makes it possible to use (N) biomarkers as well. Still, while (C) is set apart in a table, (N) is included in the AT(N) system, without a clear indication on how these markers can be used for staging.

    The plan is to add other neuropathological processes to the framework that are directly, indirectly, or not linked to AD, when biomarkers for these processes become available. Possible new categories mentioned are vascular pathology (V), astrocytosis (A), microgliosis (M), synucleinopathy (S), and TDP-43 pathology (T). In the same vein, one could imagine to add categories for endothelial dysfunction (E), reactive oxidative species (R), dystrophic neurites (D), axonal loss (A), and myelin loss (M), which would then give an ATNVAMSTERDAM framework. While there is a value in the assessment of these other pathologies for monitoring, staging, or prognostic purposes, including a wide range of such markers in an AD framework may overshoot the mark as eventually A and T markers may be sufficient to select individuals for AD specific treatments.

    The NIA-AA 2018 framework gives a comprehensive overview of the pathophysiology and diagnosis of AD. It will be useful for providing structure in future research, while minding the caveats mentioned. As the understanding of AD pathophysiology is still incomplete, it is likely that novel discoveries will reshape the conceptualization of AD. With the exclusive focus on research in the NIA-AA 2018 framework, IWG and NIA-AA 2011 criteria remain a reference for the use of AD biomarkers in clinical practice.


    . Comparison of International Working Group criteria and National Institute on Aging-Alzheimer's Association criteria for Alzheimer's disease. Alzheimers Dement. 2012 Nov;8(6):560-3. PubMed.

    . Tau PET patterns mirror clinical and neuroanatomical variability in Alzheimer's disease. Brain. 2016 May;139(Pt 5):1551-67. Epub 2016 Mar 8 PubMed.

    . NIA-AA staging of preclinical Alzheimer disease: discordance and concordance of CSF and imaging biomarkers. Neurobiol Aging. 2016 Aug;44:1-8. Epub 2016 Apr 4 PubMed.

    . Distinct 18F-AV-1451 tau PET retention patterns in early- and late-onset Alzheimer's disease. Brain. 2017 Sep 1;140(9):2286-2294. PubMed.

    . The effect of diagnostic criteria on outcome measures in preclinical and prodromal Alzheimer's disease: Implications for trial design. Alzheimers Dement (N Y). 2017 Nov;3(4):513-523. Epub 2017 Sep 21 PubMed.

    . Cross-sectional and longitudinal atrophy is preferentially associated with tau rather than amyloid β; positron emission tomography pathology. Alzheimer's & Dementia: Diagnosis, Assessment & Disease Monitoring, 2018.

    View all comments by Betty Tijms
  7. This biomarker-based classification of AD should in my opinion not be promoted as an NIA-Alzheimer’s Association research policy recommendation, it should be considered a research proposal. I raise the concern that the proposed research scheme, published as the 2018 National Institute on Aging-Alzheimer’s Association (NIA-AA) Research Framework, is premature, even with the disclaimer that it is intended for use in observational and interventional research, and not routine clinical care. Adopting the suggestion of classifying as AD spectrum normal subjects who are biomarker-positive, while valuable for lobbying funders, is not good for the general public. Such a strategy raises the specter of increasing the current AD epidemic threefold and thus increasing the numbers of worried elders who do not have access to treatment. Currently, 15 percent of subjects over 65 years old are considered to be affected by AD-spectrum diseases where biomarker studies estimate that 35–45 percent in this age range are affected.

    The expected high prevalence of individuals with positive AD biomarkers who do not develop cognitive change begs for new ideas about dementia-triggering factors. What remains essential is continued focus on mechanisms conferring risk to brain and behavior. This need can be serviced by developing novel ways of exploring biomarker biology as the AD-related proteins in the brain and CSF are dynamically responding to energetics, diurnal influences, pressure dynamics, injury, clearance and degradation kinetics, etc. This need does not require an a priori predictive (staging) structure.

    Specifically, the proposed formulation, which is Aβ-centric, defines Aβ– and tau+ biomarker results as not AD. The evidence for this policy recommendation is from cross-sectional CSF studies of other dementias. Missing are longitudinal investigations and consequently, the recommendations do not even consider the possibility of an Aβ–/ tau+ pattern evolving with the passage of time into Aβ+/tau+. The modeling of Aβ+ as central to the pathological AD diagnosis and to its dementia and the presumed temporal primacy of Aβ+ in the AT(N) model is an operational restatement of the AD amyloid hypothesis that at considerable cost continues to be extensively studied. Simply put, we need to be more open-minded about the biomarker models and more critical of the available evidence before issuing research policy recommendations.

    The CSF and the PET characterizations of Aβ positivity are quite different. Our recent studies show that both high and low CSF Aβ42 levels in normal elderly are associated with a poor outcome (de Leon et al., 2018). Moreover, the trajectory of Aβ42 is not linear in aging or in the early stages of AD. The data show that both elevations and reductions in Aβ42 are prominent with normal aging and both changes are associated with elevated CSF tau levels. This is not the case for Aβ38 or Aβ40, which, while not diagnostic, demonstrate positive linear relationships to tau. Presumably the difference lies (in part) in the greater brain aggregation potential of the Aβ42 thus lowering the CSF levels. Why the Aβ38 and Aβ40 show good correlation with tau levels remains unclear and raises important research questions regarding the impaired clearance of brain-derived analytes to the CSF (Tarasoff-Conway et al., 2015; de Leon et al., 2017). Importantly, however, among CSF analytes it is the CSF tau elevation that is most consistently related to the risk for future cognitive decline. Longitudinal observations with PET tracking both Aβ and tau lesions are not yet available.

    Overall, we offer that earlier and more frequent biomarker observations are needed to demonstrate the trajectories of CSF and brain Aβ and tau levels and their relationship to cognition and postmortem evaluation. We are not yet ready to adopt a universal biomarker staging model. 


    . The nonlinear relationship between cerebrospinal fluid Aβ42 and tau in preclinical Alzheimer's disease. PLoS One. 2018;13(2):e0191240. Epub 2018 Feb 7 PubMed.

    . Clearance systems in the brain-implications for Alzheimer disease. Nat Rev Neurol. 2015 Aug;11(8):457-70. Epub 2015 Jul 21 PubMed.

    . Cerebrospinal Fluid Clearance in Alzheimer Disease Measured with Dynamic PET. J Nucl Med. 2017 Sep;58(9):1471-1476. Epub 2017 Mar 16 PubMed.

    View all comments by Mony de Leon
  8. I agree with the cautions by others that care must be taken not to let the new framework influence clinical practice until it is validated for that purpose. I’m sure that the authors will be the first to agree, but enforcing that might be difficult.

    The proposed system will classify most clinical probable DLB cases within the Alzheimer’s biological continuum, even though their clinical picture is typical of DLB and not AD. This suggests to me that if we are to move away from a clinical classification to a biological framework, it needs from the outset to be broadly based and provide information about more biological variables of interest, not just AT(N).

    Evidence of Lewy body (LB) pathology should be sought in all individuals, since in addition to those cases we currently call DLB, ADNI shows us that LB are also frequently present at autopsy in cases with ATN pathology and clinical diagnoses of AD. The significance of this isn’t yet clear, but we know from DLB autopsy studies that mixed pathology cases (diffuse LB / tau +) progress more rapidly. This group, previously called the LB variant of AD, represents a large and unrecognized population who will continue to be recruited into research studies alongside non-LB AD cases using ATN alone.

    It is true that we don’t have yet a robust direct measure of α-synuclein, but we do have some good surrogates. The recent DLB Consensus report (Oct 2017 webinar) suggested that the best indicative markers for LB disease in a person with dementia are reduced dopamine transporter uptake in basal ganglia demonstrated by SPECT or PET; abnormal (low uptake) 123iodine-MIBG myocardial scintigraphy; and polysomnographic confirmation of REM sleep without atonia. While it is true that the FDG PET cingulate island sign is often present in Lewy body disease, there are less data about this yet, and it is therefore recommended as a supportive marker.

    How any of these markers will perform in presymptomatic individuals is completely unknown.

    View all comments by Ian McKeith
  9. The new NIA-AA Framework for Alzheimer’s disease marks a big step forward in how we are looking at this disorder and will lead to better diagnostic principles. In addition, to base the diagnosis on pathobiology instead of clinical staging will help to improve clinical research as well as drug trials.

    The Food and Drug Administration has recently taken a similar decision, in which the importance of biomarker results in clinical trials has been upgraded. As most trials for Alzheimer’s disease nowadays are aimed at earlier disease stages, the cognitive functions of the participants are often close to normal, which makes it more difficult to demonstrate an improvement in cognition. With the new guidelines, however, an effect on biomarkers—such as CSF- and PET-based measures of Aβ and tau pathology—could be valid and more realistic trial readouts.

    The NIA-AA Framework system is flexible and possible to be further developed along with the continuous increase in knowledge. One can speculate, and hope, that biomarkers for vascular dysfunction, inflammation and concomitant pathologies—such as aggregated α-synuclein and TDP43—will be developed in the future.

    The understanding of AD pathophysiology has improved greatly over the years, but is still far from perfect. Thus, the new NIA-AA Framework must be regarded as a provisional, but still an important, acknowledgement of the need of a refined amyloid cascade hypothesis.

    Some words of caution:

    1. By following the current biomarker-based criteria too strictly, one runs the risk of misdiagnosing some patients. As one example, 5 to 10 percent of AD brains, including those from the Arctic mutation family, seem to display only or mostly diffuse plaques, which are likely to be negative on amyloid PET (Schöll et al., 2012Ingelsson et al., 2003). 

    2. Certain biomarkers are still not targeting the pathology with convincing precision. For example, tau PET needs to be further developed before it can be adopted in a reliable way.

    Taken together, the new NIA-AA Framework constitutes a significant step forward toward a diagnostic concept that mirrors the pathogenesis for Alzheimer’s disease and will most likely be very useful for research and clinical trials.


    . Low PiB PET retention in presence of pathologic CSF biomarkers in Arctic APP mutation carriers. Neurology. 2012 Jul 17;79(3):229-36. PubMed.

    . Lack of correlation between biochemical and neuropathological amyloid measures in the Alzheimer brain. Ana Aslan International Academy of Aging 2003; pp 191-8.

    View all comments by Martin Ingelsson
  10. In Finland, we are widely using biomarkers of AD in clinical work, in particular MRI and CSF biomarkers. The new framework emphasizing the biomarkers irrespective of clinical symptoms in detecting AD early in research settings is very welcome. These are very useful for screening participants for drug trials. 

    The major challenge, however, is how to get in contact with possible participants in the very early stages, when there are not clear symptoms in everyday life yet.

    View all comments by Hilkka Soininen
  11. We agree with previous commentaries emphasizing the limitations of the proposed framework for subject classification in research studies. Our main concern is the lack of reproducibility and consistency of subject classification when applying different biomarkers for the definition of cerebral amyloidosis, tau, and neurodegeneration. This is not a trivial issue, since it may hamper the generalization of the results obtained in a particular cohort to the rest of the scientific community.

    We just published the first article assessing the consistency and reproducibility of the A/T/N classification system in the AD continuum (Illán-Gala et al., 2018). Our results emphasize the limitations of interchanging biomarkers from different modalities (biomarkers in cerebrospinal fluid, nuclear medicine and magnetic resonance) for the classification of patients along the AD continuum. 

    We used data from the Alzheimer's Disease Neuroimaging Initiative to demonstrate that the A/T/N system showed important inconsistencies for subject classification when using different biomarker combinations. The observed inconsistencies in the subject classification were derived from insufficient agreement between biomarkers along the AD continuum. In our work we also highlight the existence of dynamic correlations between biomarkers along the AD continuum and the potential improvement of the agreement of CSF t-Tau and p-Tau with threshold modification.

    We believe that a balance between precision (i.e., number of pathophysiological categories) and reproducibility must be found to ensure generalizability of the results. Far from resolving the observed limitations of the A/T/N system, the consideration of additional categories (as anticipated by the authors) would likely increase the observed inconsistencies in subject classification. 


    . Challenges associated with biomarker-based classification systems for Alzheimer's disease. Alzheimer's & Dementia, April 19, 2018

    View all comments by Juan Fortea
  12. We applaud this new research framework, as it reflects the improved molecular understanding of AD. Therapies targeting disease modification, and potentially disease interception, will have to be based on the molecular biology, and such experimental therapies are in development. But even more than other approaches, these therapies will need a good definition of the individual biology for study participant selection and/or stratification. This can be achieved with these new research criteria.

    There is precedent that the stringent definition of MCI by Petersen et al. has enabled biomarker research leading to a better understanding of the continuum of the disease. Disease-modifying therapies are directed against the molecular pathology and need criteria that support understanding of these factors when including volunteers in trials.

    The plethora of available biomarkers reflects the need for biomarkers in AD research. Some of these biomarkers are not well understood yet, e.g., the relationship of MRI volumetric loss versus CSF tau increase versus CSF p-tau increase. While research on these questions continues, successful drug development will need clear categorization and alignment. The new framework provides very clear guidance, while still providing flexibility to add new biomarkers or change interpretation of current biomarkers when new data arise. As the field recognizes the continuity of the disease spectrum, a more flexible categorization, as suggested here, will be ideal and avoid arbitrary categorization in the continuum.

    We believe that this diagnostic framework may be an important step to advance disease understanding for the research field. We understand that this very careful description of research criteria will not replace or sideline clinical diagnosis or decision-making. At specialized centers, where these markers are readily available, this framework may inform physicians, but for the day-to-day care of patients with dementia it may not have an immediate impact. This may change in the future, when disease-modifying treatments finally become available.

    We expect these criteria to be used for patient selection, and more importantly initially, for patient stratification, until we understand which therapy is best directed to which part of the disease spectrum. From our drug discovery perspective, this new diagnostic research framework is particularly relevant as the field is expanding into treatment of presymptomatic disease, a step that will be important to preserve quality of life for a significant part of the population—unthinkable 20 years ago.

    View all comments by Johannes Streffer
  13. We read with the greatest interest the new NIA-AA research framework setting the base for a biological definition of Alzheimer’s disease. Whatever the opinion that might be formed on this framework, one has to acknowledge the quality of the text, as well as the cautiousness of the authors. Indeed, they exercise such prudence that it is hard to criticize the framework without paraphrasing them. However, as clinicians and researchers, we would like to put emphasis on several issues.

    Just how reliable are biomarkers?
    The AT(N) system relies entirely on biomarkers to diagnose AD, a situation that existed already for presymptomatic AD diagnosis in the 2011 NIA-AA and IWG-2 criteria. This conceptual shift is probably visionary, but we believe that it comes too soon considering the still-insufficient diagnostic accuracy of available biomarkers. First, as stated by the authors, tau PET is still under development. Second, the specificity of CSF biomarkers for AD has recently been questioned in a recent Cochrane DTA review (Ritchie et al., 2017). Low CSF Aβ42 is commonly found in normal-pressure hydrocephalus (Graff-Radford et al., 2014); a recent clinicopathological report showed that a subset of tauopathies can present with elevated levels of CSF P-tau (Irwin et al., 2017) or even with a full AD CSF profile (Le Guennec et al., 2016). Conversely, CSF tau and P-tau drop over time (Sutphen et al., 2018). Although it is not clear in what proportion of patients tau biomarkers may eventually reach normal levels, the data should raise concerns on possible false negative results at the dementia stage. Additionally, tau and Aβ42 might not be independent measures. In AD, neuronal secretion of truncated tau species correlates with Aβ load, which may explain why there is usually no increase in CSF tau in other tauopathies (Sato et al., 2018). While we only start to understand the links between Aβ and tau, using such biomarkers as gold standards for AD diagnosis might be premature.

    Hence, imaging and biofluid markers will continue to be a source of ambiguity for many patients. Placing them as the alpha and omega of AD diagnosis and ignoring the clinical context that may influence the results may prove unwise. Our position may evolve with the biomarkers’ accuracy and availability (amyloid PET may be superior to CSF Aβ and tau PET might be to CSF P-tau). In that regard, amyloid PET is still not available in routine care in some countries.

    Should clinical symptoms be thrown out with the bathwater?
    In that context where caution should still be exerted in the analysis of biomarkers, we find it hard to admit that the new criteria ignore the value of clinical symptoms not only to strengthen the diagnosis, but also to link biomarker results with cognitive impairment.

    Every semantic effort in the framework has been made so that the new nomenclature never implies a causal link between the AT(N) profile and the cognitive stage, acknowledging that cognitive impairment might be due to causes other than A, T, and even N. However, in clinical routine as in interventional research, the question of the link between the biological diagnosis and the symptoms remains central. The risk is to conclude that intervention is ineffective in AD (i.e., A+T+N+), when progression is due to a (concomitant) Lewy, vascular, or TDP-43 pathology.

    Even if they were at least in part intended for research, both the 2011 guidelines and the IWG-2 criteria for symptomatic AD offered operational criteria that could be used in clinical practice in selected cases, such as in patients with focal syndromes willing to engage in a thorough diagnostic procedure. Previous criteria acknowledged that clinical assessment has strong limitations, but did not discard the contribution of Alzheimer’s clinical syndromes for AD diagnosis. They probably lacked sensitivity but their specificity will remain essential both for clinical care and research purposes.

    In conclusion, we feel that the AT(N) framework is a step forward to simplify and harmonize the nomenclature. We welcome this contribution and shall certainly use it in observational and interventional research. However, until significant progress improves the reliability of CSF AD biomarkers, and until biomarkers of other causes of neurocognitive impairment become available, we shall continue to think hard and consider false positive results and comorbidities before blindly associating A+T+N+ with AD.  

    Florence Pasquier and Vincent Deramecourt of Univ. Lille, Inserm, CHU Lille, also contributed to this comment.


    . CSF tau and the CSF tau/ABeta ratio for the diagnosis of Alzheimer's disease dementia and other dementias in people with mild cognitive impairment (MCI). Cochrane Database Syst Rev. 2017 Mar 22;3:CD010803. PubMed.

    . Alzheimer CSF biomarkers may be misleading in normal-pressure hydrocephalus. Neurology. 2014 Oct 21;83(17):1573-5. PubMed.

    . Ante mortem cerebrospinal fluid tau levels correlate with postmortem tau pathology in frontotemporal lobar degeneration. Ann Neurol. 2017 Aug;82(2):247-258. Epub 2017 Aug 19 PubMed.

    . 17q21.31 duplication causes prominent tau-related dementia with increased MAPT expression. Mol Psychiatry. 2016 Dec 13; PubMed.

    . Longitudinal decreases in multiple cerebrospinal fluid biomarkers of neuronal injury in symptomatic late onset Alzheimer's disease. Alzheimers Dement. 2018 Jul;14(7):869-879. Epub 2018 Mar 23 PubMed.

    . Tau Kinetics in Neurons and the Human Central Nervous System. Neuron. 2018 May 16;98(4):861-864. PubMed.

    View all comments by Luc Buee

Make a Comment

To make a comment you must login or register.