New criteria for Alzheimer’s disease: which, when and why?

This scientific commentary refers to ‘Prevalence and prognosis of Alzheimer’s disease at the mild cognitive impairment stage’ by Vos et al. (doi:10.1093/brain/awv029).

Until relatively recently, a clinical diagnosis of Alzheimer’s disease could only be made when an individual had acquired sufficient memory and other cognitive impairments to interfere with activities of daily living, and no more likely cause for their cognitive impairment was apparent. The prospect of targeted disease-modifying therapies predicted to have maximum effects when given early required diagnostic criteria that would both allow earlier disease detection, and be specific for Alzheimer pathology. The former led to the designation of ‘mild cognitive impairment’ (MCI) in individuals with isolated, subjective and objective memory impairment not sufficient to interfere with daily living, latterly revised to incorporate amnestic and non-amnestic forms, and single and multiple domain impairments (reviewed in Petersen, 2011). Individuals with MCI were shown to convert to dementia at 5–20% per year as opposed to the 1–2% incidence in age-matched general populations, paving the way for numerous therapeutic and observational trials at this disease stage. Particularly with the expansion to include multiple types of cognitive impairment, it became clear that MCI is not always a pre-dementia form of Alzheimer’s disease, but can be a precursor of any type of dementia, or in some cases even a transient and reversible state. Improving detection of Alzheimer pathology required the development and validation of new biomarkers. As well as established MRI-based measures of medial temporal lobe atrophy and FDG-PET measures of temporo-parietal hypometabolism, molecular biomarkers of Alzheimer pathology have more recently become available—CSF amyloid-β_1‐42 is reduced, and CSF total or phosphorylated-tau elevated in Alzheimer’s disease (Blennow et al., 2010); and PET scanning using a range of amyloid-specific ligands allows fibrillar amyloid deposition to be visualized and quantified in vivo (Herholz and Ebmeier, 2011). It was hypothesized and shown that conversion from MCI to Alzheimer’s dementia was higher in the presence of one or more of these biomarkers, as a result of which a diagnosis of MCI began to be combined with biomarkers to ‘enrich’ study populations for clinical trials.

To provide a more formal structure for pre-dementia diagnosis, a number of new diagnostic criteria have since been proposed. In 2007, the International Working Group (IWG-1) proposed research criteria for ‘prodromal Alzheimer’s disease’ that required the presence of episodic memory impairment and at least one abnormal biomarker, either of topography (medial temporal lobe atrophy or FDG hypometabolism), or molecular pathology (reduced CSF amyloid-β_1‐42, elevated tau, or amyloid PET deposition) (Dubois et al., 2007). Revised in 2014, the IWG-2 criteria now allow for prodromal Alzheimer’s disease to be diagnosed in the presence of cognitive impairment in domains other than memory, or in the presence of either increased amyloid PET deposition or the combination of lowered CSF amyloid-β_1‐42 and elevated CSF tau. Furthermore, IWG-2 criteria allow subdivision into ‘typical prodromal Alzheimer’s disease’ where there is memory impairment, and ‘atypical prodromal Alzheimer’s disease’ where there are non-memory cognitive deficits (Dubois et al., 2014). In 2011, alongside criteria for presymptomatic Alzheimer’s disease and dementia due to Alzheimer’s disease, the National Institute on Ageing/Alzheimer’s Association (NIA-AA) proposed new research/clinical criteria for ‘MCI due to Alzheimer’s disease’ (MCI-AD) (Albert et al., 2011). These criteria allow for cognitive impairment in any domain, and incorporate combinations of amyloid (CSF or PET) or ‘neuronal injury’ markers (medial temporal lobe atrophy, CSF tau, temporo-parietal FDG-PET hypometabolism). These markers are combined to allow several designations: (i) unlikely due to Alzheimer’s disease, i.e. MCI but normal/negative amyloid and neuronal markers; (ii) high-likelihood of MCI due to Alzheimer’s disease, i.e. MCI and both abnormal amyloid and neuronal markers; and (iii) intermediate likelihood of MCI due to Alzheimer’s disease where information from only one biomarker—either a neuronal injury marker or amyloid marker—is available, and that biomarker is abnormal. The designation ‘uninformative’ is applied if the biomarkers are unavailable, conflicting or indeterminate. In cases where amyloid markers are negative but measures of neuronal injury are positive, the term ‘MCI suspected non-Alzheimer’s pathology’ (MCI sNAP) has since been proposed (Petersen et al., 2013). Whilst each of these consensus-based criteria has been investigated to a greater or lesser extent in isolation, it is uncertain how these criteria compare in their ability to detect MCI due to underlying Alzheimer’s disease and to predict the subsequent development of Alzheimer’s dementia; how easily each can be operationalized for use in multicentre cohort studies; and what can be concluded from cases with discordant biomarker results.

In this issue of Brain, Vos et al. report results from a retrospective analysis of data from 13 different cohorts comprising a total of 1607 subjects with MCI (Vos et al., 2015). All subjects had clinical assessments, including various measures of memory and other cognitive deficits, and were classified as MCI; were seen for follow-up; and had at least one of CSF amyloid-β_1‐42, CSF tau, an MRI-based measure of medial temporal lobe atrophy, or measure of FDG temporo-parietal hypometabolism. CSF biomarkers were defined as positive/negative according to local cut-offs, and imaging markers using combinations of quantitative and qualitative measures. Applying IWG-1 criteria to the whole cohort, 53% fulfilled criteria for prodromal Alzheimer’s disease, of whom 50% converted to Alzheimer’s dementia over 3 years. Of the subset (48% of the cohort) with both amyloid and neuronal markers, 40% fulfilled IWG-2 criteria for prodromal Alzheimer’s disease—of whom 61% converted to Alzheimer’s dementia, independent of whether they had typical or atypical prodromal disease. Forty-six per cent of those with both neuronal and amyloid markers had NIA-AA high-likelihood of MCI due to Alzheimer’s disease, of whom 59% converted. Conversion rates were markedly lower in those not fulfilling IWG-1 and IWG-2 criteria, at 21% and 22%, respectively, and just 5% for those classified as MCI unlikely due to Alzheimer’s disease. Of those with amyloid and neuronal markers, 29% fulfilled criteria for MCI sNAP, of whom 24% subsequently converted to Alzheimer’s dementia.

These results demonstrate that even in a retrospective multicentre study using pragmatic approaches to classify individuals on the basis of available and often incomplete biomarker data, it is possible to distinguish groups of individuals with MCI at higher and lower risk of conversion to Alzheimer’s disease. While the three sets of criteria performed broadly similarly, these data suggest some relative advantages and disadvantages of each: where identifying individuals destined to convert to dementia is important, IWG-2 and NIA-AA criteria perform somewhat better, but fewer fulfil criteria which additionally require both amyloid and neuronal biomarkers to be acquired. Conversely, the proportion of individuals classified as prodromal Alzheimer’s disease is slightly higher using IWG-1 criteria, which require only the more widely available neuronal biomarkers to be used, but these criteria are slightly less accurate at predicting conversion to Alzheimer’s dementia. Vos et al. acknowledge many of the limitations of the study, particularly that they did not have amyloid PET data, which in practice may well be the favoured means of determining amyloid status for clinical trials. While combining data sets provides impressive numbers and increases power, without close standardization there will inevitably be considerable variability in how CSF assays are performed and interpreted, MRI is acquired and quantified, and how clinical criteria, e.g. those for MCI and dementia, are applied. It may therefore be that the criteria will prove more sensitive and specific when applied in a strict clinical trial setting.

While the field is likely to benefit considerably from the development of these new criteria, important questions remain. It is notable that a considerable proportion of individuals are classified as negative for IWG prodromal Alzheimer’s disease, or as having NIA-AA inconclusive level disease: aside from an estimated ∼20% rate of conversion to Alzheimer’s disease over 3 years, little is known about the cause of these individuals’ cognitive impairments or their longer term prognosis. Noting that the IWG criteria were explicitly designed for research and not clinical purposes, with the exception of the strong negative predictive value for developing dementia in individuals fulfilling NIA-AA low-likelihood criteria, considerable caution is required before using any of these criteria to guide individual patient management: even in individuals with neuronal and amyloid biomarkers both suggestive of underlying Alzheimer’s disease, more than 40% remained free of dementia after 3 years of follow-up. The intriguing finding that the unexpectedly and relatively high risk of conversion from MCI sNAP to Alzheimer’s dementia may in part relate to the choice of CSF amyloid-β_1‐42 cut-off point, with those having lower levels, i.e. higher amyloid load, having a higher risk of conversion, is a demonstration of how reliant all these criteria are on choice of biomarker cut-off point (Bartlett et al., 2012). Alternatively, it may reflect inaccuracies in the clinical diagnosis of Alzheimer’s dementia in the absence of post-mortem confirmation. At the risk of further complicating an already complex diagnostic process, it may be that refining the criteria to include more biologically plausible—and often substantial—biomarker ‘grey zones’ will provide more accurate prognostication, as may using more specific and sensitive neuropsychological test batteries. It will be important to know how these results compare with prospective studies using amyloid PET, and how they fare when strict, centralized and standardized measures of CSF processing and image analysis currently under development are used. Finally, it will be of considerable interest to see if incorporating emerging CSF or imaging biomarkers—and perhaps tau PET in particular (Villemagne et al., 2015)—can improve prediction of conversion from MCI to Alzheimer’s dementia on an individual patient basis.

Funding

J.M.S. acknowledges the support of Alzheimer’s Research UK, the NIHR Queen Square Dementia Biomedical Research Unit, UCL/H Biomedical Research Centre, and Leonard Wolfson Experimental Neurology Centre. R.C.P. receives support from the National Institute on Ageing of the National Institutes of Health and the Mayo Clinic and serves in a consulting capacity for Pfizer, Inc., Janssen Alzheimer Immunotherapy, Roche, Inc., Merck, Inc. and Genentech, Inc.

References