The present longitudinal study was set out to investigate the hypotheses that two memory markers for AD recently recommended by consensus  differently predict dementia throughout its continuum. Based on previous evidence we predicted that the VSTMBT would be able to discriminate between older adults who are asymptomatic and those who are in the very early stages of cognitive decline more effectively than the FCSRT (H1). However, the FCSRT would discriminate between older adults in the prodromal stages (MCI) who later convert versus those who do not convert to dementia more accurately than the VSTMBT (H2). Our results supported both hypotheses and have some implications for our understanding of neuropsychological assessment to track the transition from normal to pathological ageing and to monitor progression through the prodromal stages towards conversion to dementia.
Before discussing these implications, it is worth considering some observations drawn from the background neuropsychological assessment. HOA and eMCI participants differed on a number of neuropsychological tasks, yet eMCI participants were not seeking professional help. As patients with MCI progressed along the disease continuum (i.e., MCI non-Converter and Converter), discrepancies in the neuropsychological scores decreased. Hence, standard neuropsychological tests used in our study appear to be effective for detecting impairments but less so for differentiating risk phenotypes. These shortcomings of off-the-shelf neuropsychological tests have been acknowledged previously [8, 60–63] and called for new tests to better phenotype dementia and detect risk profiles [8, 60]. Notwithstanding such limitations, the ability of some neuropsychological tests used in our assessment battery to detect very early cognitive impairments, particularly of memory, is also worth highlighting.
The HVLT revealed significant memory differences along the disease continuum, particularly between groups informing the very early stages. Lonie et al.  had previously demonstrated that the Delayed Recall component of the HVLT can discriminate between MCI converters and non-converters over a 4-year follow up period as accurately as the Visuospatial Paired Associates (PAL) Task from CANTAB . Regarding the experimental tasks, the FCSRT and VSTMT showed differential abilities to predict group membership along the disease continuum. Opposite patterns of sensitivity were observed at the extreme ends of the continuum here explored (preclinical: VSTMBT > FCSRT; advanced prodromal: FCSRT > VSTMBT), and varying levels of complementary throughout its intermediate stages. These findings lend support to the two hypotheses investigated in this study and suggest that these recently recommended tests [9, 13, 31, 66] shall form part of new memory toolkits to assess and monitor AD.
To investigate the individual and complementary values of the two experimental tasks in informing about the transition from normal to pathological ageing we focused on data from HOA, eMCI and MCI non-Converter. We predicted that the VSTMB should outperform the FCSRT in the earlier stages because it would be able to detect gradually increasing levels of impairments whereas the function assessed by the FCSRT would still be preserved. Didic et al.  proposed a hypothesis that is very much in line with this prediction (see Figure 1). The authors argued that in the early preclinical stages of AD when the hippocampus is still unimpaired, context-free memory tests, which seemingly rely on structures of the Anterior Medial Temporal Lobe network (i.e., entorhinal, perirhinal cortices), would stand a better chance to identify impairments than context-rich memory task, which are reliant on the integrity of hippocampus. Not much work has been done to test this hypothesis, but studies are now piling up confirming its potential validity [32, 35, 38, 67]. Tests that are sensitive to the very early preclinical stages of AD would likely reach floor levels when patients enter the symptomatic stages as such a function would have been declining for years before the disease onset (see [68, 69] for early discussions of this issue). At this point, such tests would become little informative about group membership whether such groups involve converters or non-converters MCI. This study provides the first empirical evidence supporting such a model. Tests sensitive to the symptomatic stages of the continuum (e.g., FCSRT) would take over the predicting role at this stage (MCI, [70–72]).
In the current study, we demonstrated that both tests hold excellent abilities to discriminate between HOA, eMCI and MCI non-Converter (HOA > eMCI/MCI non-Converter). The VSTMB outperformed the FCSRT only in the preclinical stages (HOA > eMCI). As the disease progressed, (i.e., eMCI/MCI non-Converter) performance on the VSTMBT became less differentiated between groups, whereas that on the FCSRT continued to effectively discriminate between them. These findings, although encouraging, raise a number of concerns for promising neuropsychological assessment aimed at the preclinical stages of AD. Logie et al.  suggested that a good memory marker for AD should avoid very low performance levels when the symptoms become severe. Regarding the VSTMBT, which relies on the Change Detection Paradigm, chance levels are set at 50%. This is a constraint of the method. To overcome it, Parra et al.  suggested strategies such as titrating the task difficulty (i.e., memory load). For instance, the authors suggested that for studies aimed at the preclinical stages of AD, a VSMTBT using set size 3 would increase changes to detect impairments (see also [26, 27, 30]). However, for the symptomatic of dementia stages, set size 2 would be preferred as 3 would pose significant challenges (see [14, 26, 27]. In the present study, we chose to use one set size (i.e., 3) for the sake of comparability of findings along the disease continuum. Furthermore, Parra et al.  acknowledged that titrating, at least relying on current procedures, would be an unfeasible task to be implemented in clinical settings.
The literature supporting the validity of the FCSRT to predict dementia in longitudinal cohorts of MCI patients has grown significantly over the last few fears (e.g. [70–73]). This is the first report on the use of the VSTMBT in such longitudinal cohorts. These outcomes fit the notion that, at the prodromal stages of the disease memory tests sensitive to such a stage would stand a better chance to achieve reliable predictive outcomes relative to tests that assess functions sensitive to the very early and still silent stages of the disease (see Figure 1). Some considerations regarding these findings are warranted. First, they support the notion that the neuropsychological assessment of AD, in its new conceptualization (i.e., a continuum of clinical and pathological stages), ought to abandon the one-size-fits-all approach. Assessment protocols aimed at investigating AD related disorders (i.e., detection, prediction) need to consider this evidence. Belleville et al.  acknowledged that a cognitive toolkit intended to identify AD at the pre-dementia stage would need tasks that are early indicators and others that might suggest imminent progression. Second, further research is required to unveil not only “which” memory function is most sensitive to AD  but also the “when” in the disease continuum such memory phenotypes become apparent.
The fact that the VSTMBT detects AD related changes early (see [26, 27, 30]) and then performance drops to near or chance levels (see ) has pros and cons. The positive aspect of this is that we have for long needed tests that can detect the very early stages of the disease process, preferably, when people are unaware of or are very little concerned about any cognitive or functional impairment. We have learned that at this stage, the VSTMBT is taxing the early growth of amyloid in at risk individuals even before tau deposits or neurodegeneration become apparent [32, 35, 38]. Such a test would be an ideal tool for clinical trials aiming at dementia prevention as they could enhance recruitment strategies by selecting who will likely meet inclusion criteria (e.g., Aβ+).
Of note, the individual and combined predictive value of both tests to discriminate between stages closer to dementia (MCI non-Converter and MCI Converter) was rather low. That was predicted for the VSTMBT but not for the FCSRT, which performed very well in the early and middle stages of the continuum. This is an interesting finding which seems to suggest that as the VSTMBT predictive abilities decrease in the prodromal stages of the disease, the FCSRT might face similar limitations if used to predict stages closer to dementia. Therefore, the quest for the abovementioned toolkit ought to be mapped along the disease continuum including tools sensitive to the very early stages through the advanced dementia stages where cognitive assessment can still provide evidence to inform clinical practice and in so doing, ensuring patients will receive the best possible care until the end of life.
One final aspect concerns our control participants. Most participants who were allocated to the eMCI group entered the study as self-referred healthy volunteers (see Figure 2). Relative to those who met criteria for HOA, eMCI participants displayed significant differences on various standard neuropsychological assessments. This is striking, as these individuals, at the time of the study, had not sought help and a few were only mildly concerned about their cognitive abilities. There is consensus that in the new context of AD research and clinical practice (i.e., following the biological definition of AD), deciding who is a control individual is proving as challenging as deciding who is in the early stages of the disease . There are two issues worth considering here. First, the source of these control volunteers and second, awareness of and stigmas against early symptoms of dementia. Volunteers entering as controls were recruited from the Psychology Volunteer Panel at the University of Edinburgh o were relatives of patients with dementia. In the case of the former source, there is awareness about the impact that such selective samples could have on the interpretation of data . Older adults involved in such panels (1) regularly support research and (2) are often highly educated, thus representing a rather biased sub-sample of the relevant population. Importantly, they frequently undergo cognitive testing, which grants them additional cognitive reserves and resilience . Therefore, it is not entirely surprising that these older adults overlook or underestimate the level of decline in cognitive abilities here identified. Although volunteering has been considered a protective action against cognitive decline , managers of volunteer panels need to be aware of these risks. In the case of the latter source of recruitment (i.e., relatives of patients with dementia), there is evidence that the burden posed by the patients’ level of cognitive and behavioural problems cause caregiver stress, which in turn leads to impaired cognitive functioning [77, 78]. Therefore, volunteer panels and dyads of dementia patients, two common sources of recruitment in ageing and dementia studies, will need revised approaches if we are going to progress in the new dementia research context with more confidence and reliability. The second issue, awareness of and stigmas against early symptoms of dementia, is also relevant  and suggests that more work is needed to continue raising awareness about the fact that ageing is not a disease  and seeking help early is the best approach to mitigate the dramatic impact that departures from its normal trajectory will carry.
There are some limitations that need to be considered when interpreting the findings here reported. The first one is the rather small sample size. However, as shown by our inferential statistics, effect sizes were rather large for the hypotheses tested. Moreover, both experimental tests used in this study have demonstrated to hold informative value to identify individual patients and not just during group comparisons. For instance the VSTMBT test had shown sensitivity and specificity value of over 77% in completely asymptomatic individuals  and of 100% for patients with dementia (see Della Sala et al.  who reported an Area Under the Curve of 96% for the FCSRT). Nevertheless, efforts will be needed to expand such samples within diseases stages and along the continuum, and such efforts are already ongoing .
Another limitation is the nature of the control participants who entered this study. This is not a representative sample. Even if unpaired cognitively, it is still possible that some of these older adults were already accumulating disease pathology (see ). Together with the report by Parra et al.  this evidence suggests that some of those who entered our HOA group may still be classified as not healthy controls if the approach recommended by Bos et al.  is followed. This limitation is shared by many studies in the field and urgent strategies will be necessary to address this important caveat.
One final limitation of this study is that we did not have biomarkers evidence to assess the biological status of our MCI patients and hence we choose to adhere to the definition of the Alzheimer’s clinical syndrome as recently recommended [1, 5, 6].