Single cohort:
Study II:
This is the first study to apply the PredictAD tool (DSI and DSF) in FTD. There was almost the same population as in Study I, with the main differences of a few cases added to the AD group and the combination of SMCI and PMCI into one single MCI group.
The highest accuracy was reached when comparing FTD with controls (0.84), followed by FTD compared with MCI (0.79) and AD (0.69). The lower accuracy in FTD vs. AD may be explained by the shared findings in some of the parameters included in DSI. When one focusses on each individual parameter, it was apparent that MRI was the most relevant feature when FTD were compared to MCI and AD, however in controls vs. FTD, the most relevant feature was the MMSE.
MMSE: In the controls vs. FTD comparison the most relevant feature was the MMSE.
Instead MMSE was not useful differentiating AD from FTD. This may be due to two reasons: the non-specificity of MMSE for helping to make a specific dementia diagnosis (Kertesz et al. 2003), and the recruitment of FTD patients already in a more advanced stage of the disease, where there may be an overlap between symptoms (memory impairment, behavioural changes, disinhibition) between bvFTD and AD, while other symptoms that could be differentiating such as apathy (Chow et al. 2009), are not assessed in MMSE.
The MMSE score includes questions concerning attention, orientation, language and memory. At least the two first parameters are usually impaired in bvFTD (these patients may have language difficulties and amnestic episodes, but these usually appear later as the disease progresses and they are not always present), and this can help in differentiating between healthy controls and bvFTD.
MCI patients could be differentiated from bvFTD with using MMSE, probably because of the typical amnestic-presentation of the MCI group.
DSI reflects the profile of cognition The scores for each individual item in MMSE, could provide support in differentiating AD from FTD e.g. a low score in recall questions leans more to AD while a low score in attention questions is more indicative of FTD.
Nevertheless, one should remember that MMSE is a screening tool. In the follow-up procedure, MMSE could help to differentiate AD from bvFTD, as both diseases exhibit different rates of decline (Tan et al. 2013).
Imaging methods: VBM was the most relevant technique in all the comparisons, although HV is highly relevant if one wishes to differentiate FTD from controls and MCI. This highlights the fact that the hippocampus is a region also affected in FTD (van de Pol et al.
2006), thus when comparing FTD with MCI and in particular AD one needs to study more regions than the hippocampus alone.
CSF and APOE: CSF and APOE were valuable when comparing FTD with MCI and AD.
CSF samples or APOE data were not available for controls. Interestingly, only P-Tau achieved statistically significant differences. This supports the belief that P-Tau can differentiate AD from other dementia diseases (Schoonenboom et al. 2012), although Aβ42 and T-Tau could help to make a MCI or AD diagnosis at an earlier stage. More studies with more incipient MCI cases and FTD and AD cases recruited at an earlier stage are needed.
Furthermore, there is no specific CSF profile for FTD (Chow, Alobaidy 2013).
In addition to the combination of P-Tau, T-Tau and Aβ42, the so-called CSF profile or AD profile, defined as ൌଶସାଵǤଵ଼୲ୟ୳ஒସଶ was also included. This profile was less relevant than the CSF values in AD vs. FTD, and had markedly lower relevance in the MCI vs. FTD comparison. Thus it is recommended to use the normal CSF values and to combine them with values from DSI; it is not recommended to use the so-called CSF profile.
The APOE genotype was particularly relevant in AD vs. FTD. In the AD group, 69% were carriers of one or two APOE ε4 alleles, whereas in FTD there were only 20% patients carrying one APOE ε4 allele and there were no ε4/ ε4 carriers. The proportion of AD ε4 carriers is in line with the published literature, and the ε4 frequency in the FTD group was comparable with the previous literature (Engelborghs et al. 2003, Lovati et al. 2010). The APOE genotype is not usually investigated in clinical practice, as the DSM-IV criteria for AD do not include this parameter. However, it could be advisable to include it since it is a major risk factor for AD and this has been recommended in recent guidelines (Dubois et al.
2007). However, it should be stressed that the presence of APOE ε4 is not exclusively a
property of AD, this allele may be present in FTD and other diseases (Engelborghs et al.
2003, Lovati et al. 2010).
Limitations and future directions: The main limitation of this study is the single inclusion of MMSE as a neuropsychological test, which was non-specific for AD or FTD. Study III tried to tackle this issue.
Study III:
This study was planned in order to amend the limitations of study II, in particular with respect to the presence of MMSE as the only neuropsychological test (which is a screening test and thus does not differentiate AD from FTD) (Kertesz et al. 2003), and to compare the importance of the manual volumetric analysis and SPECT, which is used in clinical practice mostly in uncertain cases because it has proven utility in differentiating FTD from AD (McNeill et al. 2007).
The highest accuracy was reached in differentiating controls from AD (0.99) and FTD (0.97). In addition, AD could be differentiated from FTD also with a high accuracy (0.86).
When one focusses into each individual category, then it seemed that clinical symptoms record and neuropsychological tests were the most relevant groups in differentiating AD from FTD.
Symptoms: among each category, the clinical category, in particular the symptoms record, was the most relevant group. This is as expected, as in the majority of AD and FTD cases the symptom profile differs, i.e. AD is characterized by amnesia and FTD by frontal symptoms (e.g. behavioural changes, alterations in conduct). However the simplicity of the recording of these two symptoms may mask differences in the pattern or profile of memory (Bertoux et al. 2013) and behavioral conduct (Bathgate et al. 2001); these are important in differentiating AD from FTD, and there is always a need to perform specific neuropsychological tests.
Although apathy and disinhibition are usually more frequent in FTD than in AD (Bathgate et al. 2001, Leger, Banks 2013), only the presence of disinhibition was found to be relevant in this study in helping to differentiate both diseases. In our initial population there were also patients diagnosed as PNFA, SD or PSP, but these were excluded from the final study due to the low number of cases.
Dysphasia, confusion and psychosis were relevant only for differentiating FTD from controls, while paranoia only differentiated controls from AD. Depression, apathy, sleeping disorders and tremor helped to differentiate controls from FTD and AD, however they were not relevant when comparing both dementia diseases. Previous studies have shown that depression is a common finding in AD and FTD (Leger, Banks 2013).
Clinical scales: HIS was not relevant in AD vs. FTD. As stated in the review of literature, HIS only identifies the vascular component; therefore its utility may be reduced in these cases where AD displays a clear manifestation of a vascular disorder, or for differentiating AD from VaD (Hachinski et al. 2012, Knopman et al. 2001). Webster total score differentiated AD from FTD, as extrapyramidal and other parkinsonian signs may be associated with FTD (Espay, Litvan 2011) and are less frequent in AD (Duker et al. 2012).
The Hamilton depression scale is relevant to a minor extent in differentiating AD from FTD. The reason why this scale is relevant, whereas the isolated symptom of depression is not helpful, may be that Hamilton depression scale includes some questions concerning loss of insight, agitation and compulsive behavior, which point more to FTD than to AD.
Neuropsychological tests: with respect to the neuropsychological tests, memory tests were clearly the most relevant group when comparing FTD with controls and AD, and separating controls from AD. The next most useful tests were those assessing executive-functions. The overall inclusion of all the neuropsychological tests was more relevant than the single use of MMSE when differentiating controls from patients with AD and patients with FTD, and in distinguishing AD from FTD.
Genetics: two genetic factors were included: APOE genotype (APOE ε4 alleles) and if there were other family members with dementia. APOE genotype was relevant in differentiating AD from controls and FTD, as the APOE ε4 allele is clearly more prevalent in the AD population (74%) than in controls (18%) or FTD (16%). In FTD, the proportion of family members with the disease has been described as being as high as 50% (See et al. 2010), while in AD it is significantly lower, although having a family member with AD is also a risk factor for developing AD (van Duijn et al. 1991). This explains why this is a relevant parameter in the FTD vs. AD comparison.
Imaging methods: MRI is particularly useful in differentiating the healthy state from both AD and FTD, while SPECT is more relevant in comparing between AD and FTD.
With respect to the ROIs to be incorporated into MRI, the hippocampus was the most relevant area in AD in the comparison with controls and FTD. In the FTD vs. controls comparison, both the hippocampus and the frontal lobe were particularly relevant. This implies that atrophy is present in both frontal lobes and hippocampus in FTD, but the frontal atrophy differences between AD and FTD are less marked than those in the hippocampus. This is not in line with the previous literature; there is one study which compared AD and FTLD proven autopsy cases, indicating that the MTL was a common area of GM reduction in both diseases and thus it did not help in discriminating between them, whereas atrophy in frontal cortices did differentiate AD from FTLD (Rabinovici et al.
2007).
With respect to asymmetries, one can observe that there were striking differences present in the temporal lobe, where only the left temporal lobe was relevant in the controls vs. FTD and the right temporal lobe in AD vs. FTD. One study reported that the left temporal lobe was smaller in AD than it was in FTD, while the right temporal lobe was smaller in FTD (Fukui, Kertesz 2000). This present study revealed that the left temporal lobe and the right frontal lobe were the regions that best discriminated AD from FTD (Fukui, Kertesz 2000).
Here the right hippocampus and the hippocampus were more relevant when comparing FTD with both AD and controls. This is in line with the results using HV in study I.
However asymmetries in the hippocampus for AD and bvFTD are not clearly defined in the literature. One meta-analysis found only minor differences among left and right hippocampal atrophy, although a left-less-than-right atrophy has been described (Shi et al.
2009).
With respect to ROIs in SPECT, the frontal cortex displayed some relevance in the left side when comparing between controls and FTD, otherwise it was not relevant. The parietal cortex and the left amygdala-hippocampus were the most relevant areas for differentiating AD from FTD. Previous studies also revealed that the parietal lobe could discriminate AD from FTD (McNeill et al. 2007).
CSF: Aβ42 is relevant in distinguishing AD from FTD, although to a much lower extent than Tau. This is in line with previous studies and also found in study II. The reason for this may be the age and stage of the disease of the patients recruited. The Aβ42 level is
considered to be the first marker which detects subjects with MCI and AD (Jack et al. 2010), and this occurs when there are few neurodegenerative changes causing the atrophy in the brain and as a result there is no marked elevation of Tau levels in CSF. AD patients, with a mean age of 70, may have already higher levels of Tau and P-Tau which can help to differentiate them from FTD. Studies on FTD cases have described discrepant profiles for Tau in CSF (Schoonenboom et al. 2012, Grossman et al. 2005).
Limitations and future directions: this study evaluated manual MRI volumetry and SPECT.
Although volumetric MRI with manual outlining of the hippocampus has been used for the diagnosis of AD, one can predict that it will be replaced by automatic methods that allow the quantification of the atrophy in several regions more accurately and more quickly. The MRI methods from study I were not included, instead the data evaluated included a manual outlining of the ROIs (frontal lobe, temporal lobe and hippocampus) and SPECT (frontal region, temporal region, parietal cortex, occipital area, basal ganglia and amygdala-hippocampus).
SPECT has been more widely used during the last decades because it is cheaper and more widely available than FDG-PET or PIB-PET. However nowadays FDG-PET is starting to substitute SPECT in many centres.
In future studies it would be interesting to examine the importance of more biomarkers and tests (e.g. FDG-PET, FBI) and their combinations in bvFTD not only at baseline but in the follow-up, and to attempt to determine whether their predictive value varies over time, as has been proposed (Krudop et al. 2013).
Conclusions: study III showed that a wider battery of neuropsychological tests and a detailed symptoms record was more useful than using MMSE alone, and SPECT could be useful in the differentiation of AD from FTD. Volumetric MRI is useful when comparing the healthy state with both AD and FTD. There were no automatic analyses performed (e.g.
HV, TBM and VBM) because this package of data was already closed. It would be interesting to compare the manual volumetric study of the main ROIs (e.g. frontal lobe, temporal lobe, parietal lobe) with their performance when they are determined by automatic methods.
Multiple-cohorts:
Study IV:
This study aimed to investigate the value of DSI at predicting the conversion from MCI to AD in four different cohorts.
In the prediction of how MCI subjects progressed to AD MRI features alone gave good accuracies (0.67-0.81) for the four cohorts studied. The accuracy was slightly improved if one incorporated MMSE, APOE, CSF and neuropsychological tests.
MRI: TBM was the most accurate method in all the cohorts except in AddNeuroMed, where VBM displayed the highest accuracy. Wolz et al., used two classifiers including several imaging methods from ADNI cohort, where TBM and HV accuracies were somewhat similar in differentiating SMCI from PMCI, and TBM was more accurate than HV in the controls vs. AD comparison (Wolz et al. 2011). In both methods, the accuracies in differentiating SMCI from AD were similar to the accuracies in all the cohorts included in this study (Wolz et al. 2011), except for the situation with AddNeuroMed, where both TBM and VBM achieved higher accuracies. The amygdala was the most accurate ROI again with
the exception of the ADNI cohort, where the most useful ROI was the right lateral ventricle (temporal horn).
Neuropsychological tests: In ADNI, the neuropsychological tests were the most accurate category, while in the other cohorts they were less accurate. This may be explained by the number of sets available in each cohort. From the ADNI the ADAS-cog test was included, where word recall, orientation and delayed word recall were the most relevant features, and only word recognition test was less accurate, while in the other three cohorts, other tests were included whose accuracies were very low and therefore they reduced the overall accuracy of the neuropsychological test category. Word list recall, Buschke test and word list recognition were the most accurate subtests in DESCRIPA, Kuopio-MCI and AddNeuroMed respectively.
MMSE: In Kuopio-MCI and AddNeuroMed, MMSE was not sufficiently relevant to be useful. The highest accuracy was reached in ADNI, although it was still not very high.
These accuracies compared to the neuropsychological tests accuracies, confirm that MMSE is a screening test for detecting a cognitive impairment but it is not specific for predicting conversion to AD.
CSF: In the DESCRIPA and Kuopio-MCI cohorts, CSF was the most accurate biomarker while in ADNI it was less accurate. This may be explained by the different analytical procedure followed, as in Kuopio-MCI and DESCRIPA ELISA was used while ADNI utilized xMAP technology.
APOE: With respect to the genetic biomarkers, only APOE genotyping was available. Its accuracy was higher in AddNeuroMed as compared with the other cohorts.
Recently a study was conducted using ADNI data (Munoz-Ruiz et al. 2013), in an attempt to predict the conversion from MCI to AD. The combination of neuropsychological tests, CSF, APOE, MRI and PET achieved the highest accuracy, although MRI and neuropsychological tests together were almost as accurate. Of all the parameters, neuropsychological tests and semi-automated HV achieved the highest accuracy. These results are partly in line with the results in the present study, where neuropsychological tests were the most accurate feature, although HV was not particularly accurate, and TBM was the most accurate imaging method.
Limitations and future directions: MCI is recognized as a heterogeneous entity. Data derived from four large cohorts, each one of which had its own criteria for selecting MCI patients, slightly different methods and follow-up times. It would be preferable to standardize the selection criteria in order to obtain more homogenized cohorts and to harmonize the methods.
Conclusions: DSI can compare and combine data coming from different populations. The prediction accuracy resulting from the combined cohort is close to each individual cohort accuracy. This is feasible if the different datasets are relatively similar.