The combination of DSI and DSF has created a tool that integrates data coming from multiple sources, displaying it in an easy and understandable manner that could help the clinician to come to a confident diagnosis.
This thesis obtained results gathered by using DSI. The DSI integrates the data and helps the clinician to recognize the importance of each biomarker, by comparing the marker originating from a particular patient to healthy and diseased populations which have been already registered. The goal of DSI is to try to resolve the question of how one can best combine the plethora of data in a logical manner to obtain a precise diagnosis of AD.
Particularly, it helps to assess the results, since it takes into account which method or test would be more relevant for achieving a certain diagnosis.
What to include in DSI
Another major issue is which parameter should be included in DSI. DSI is a statistical tool for collecting, analyzing data and finally displaying it in a comprehensible graphical manner via the DSF. This means it could include almost any parameter available in an appropiate format. For example it could be used for many other diseases. However, it can be predicted that clinicians would prefer to use a tool that would help them in their daily work. The DSI and DSF are designed for this task since they provide an unbiased perspective of the patient.
It is important to consider the variation due to gender, age and education (Koikkalainen et al. 2012). These studies did not include age, gender or education as parameters in the DSI although their importance is noted in the literature. Most of the AD cases are women (Alzheimer's Association 2012) and while there are no differences in terms of age in FTD (Rosso et al. 2003a); most of FTD cases are younger than AD cases (Ratnavalli et al. 2002). It is known that a lower level of education is associated with a higher risk of developing AD (Anttila et al. 2002).
The problem is that not all the cases follow the same pattern concerning age and gender, giving those parameters weightings on the DSI would probably bias the results, and could well blur the overall outcome. If a parameter can be considered by the clinician without using the tool and that parameter does not necessarily follow a pattern (e.g. AD cases are usually women, but that condition is not helpful in making a diagnosis, it is simply a risk factor; the same is the case for age or the level of education), then it is better not to include it in DSI. On the other hand, the volume of the hippocampus or the amount of amyloid present in the CSF are biomarkers, not risk factors, and in these cases DSI can help to determine their importance for coming to a diagnosis.
Another question which could be asked is whether it is advisable to include medications, other risk factors and symptoms on DSI.
Study III had access to a larger battery of data, which included knowledge about the medications used by AD and FTD cases, and RFs present in the same group of patients.
What are the reasons for not incorporating medications or RFs into the tool? Medications do represent an enormous list of variables, and knowing that a patient is taking some particular medication could help us to deduce whether the symptoms are iatrogenic and thus the reason why they are present in that patient, but again this could blur other results.
Thus it may well be better to consider this as additional information and interpret it on an individual basis (e.g. a patient taking NMDA inhibitor and not displaying any clear benefit, more likely could be suffering from FTD than AD).
However, in terms of lifestyle and vascular RFs, it could be really advisable to include these parameters into DSI when predicting AD. During the last decade some studies have demonstrated the benefits of controlling both vascular and other risk factors (lack of exercise, no diet), and therefore it would be very advantageous if it could be possible to integrate these kinds of factors postulated to influence the development of AD. A different question would be how should this kind of risk factor data be collected, and not simply incorporated as a binary (yes/no) scale. Instead, it would be preferable to register actual values of midlife of heart rate, sleeping patterns, gait, combined with more precise questionnaires on diet and exercise.
In selecting the ROIs used in study I and II and IV the Hammer atlas was utilized (Heckemann et al. 2006). This has been used previously in other studies; here it was decided to gather data from a total of 83 regions from this atlas, and then the ROIs with the highest accuracy were selected.
This thesis has compared bvFTD with AD, MCI and controls. However, bvFTD only represents about half of the cases within the FTLD spectrum. BvFTD was chosen because it may be difficult to differentiate this particular syndrome from AD e.g. a patient over 65 years of age presenting with amnestic memory, an AD pattern on imaging or even APOE ε4 allele. In future studies, features from other dementia diseases such as other forms of the FTLD spectrum and VaD should be investigated. It is important not only to make a differential diagnosis between the major dementia diseases, but also to identify MCI cases and to be able to predict their probable conversion to dementia. It is also important to make a differential diagnosis in the mixed cases which tend to be the rule in very old patients.
It is concluded that conceivably DSI can be a supportive tool for profiling patients and it can help the clinician to make as accurate a diagnosis as possible.
The combination of several biomarkers has been recently conducted by applying many different procedures. There are many classifiers such as support vector machines or linear discriminant analysis (Wolz et al. 2011, Mattila et al. 2011). However within the dementia literature, only one multivariate data analysis has been done that attempted to combine different features and display the findings in a graphical manner in a similar way to the DSF: the orthogonal projections to latent structures (OPLS) (Westman et al. 2011). The use of MRI data predicted conversion from MCI to AD in the AddNeuroMed cohort over a 1 year period with an accuracy of 86% (Aguilar et al. 2013). In that study, the combination of CSF and MRI parameters at baseline, the accuracy for discriminating between AD and controls, MCI and controls and for predicting future conversion from MCI to AD, was higher than the accuracy obtained by using either MRI or CSF separately (Westman, Muehlboeck & Simmons 2012).
Diagnosis using PredictAD tool by the clinician:
There are two clinical studies describing the use of DSI. Both studies utilized data from ADNI, and in both cases it helped the clinician in the task of classifying subjects with MCI into six possible stages, ranging from non-clear AD to clear AD (non-clear AD, probable non-AD, subtle non-AD, subtle AD, probable AD and clear AD). Simonsen et al., (Simonsen et al. 2012) reported that the classification accuracy with the tool (70%) was higher than without its use (62.6%). Liu et al., (Liu et al. 2013) the accuracy reached by the PredictAD itself and the clinician using the tool was very similar (72% vs. 71%).
DSI in clinical practice:
Finally, one question to remains to be discussed: when should DSI best be used in clinical practice? It has recognized utility comparing population datasets for research purposes, and it is also known how it performs in the examination of a single case in determining whether he/she is more likely to be AD or FTD. Nevertheless the studies in this thesis do not investigate when this tool could become part of the daily clinical routine. Since this issue will be relevant in the near future, its possibilities will be discussed in this final section.
It can be a long road from the time when a patient presents in the physician’s office with a memory complaint or behavioural problem (or other cognitive function) until there is an initial diagnosis; the tests or methods which can be applied to ascertain that diagnosis and then to monitor the patient in follow-up, are depicted in Figure 16. This is a general overview, not all of the steps are taken nor are all the tests or methods applied in every single patient.
One very useful stepwise approach provided for the diagnosis and assesses of AD in the primary care was depicted in the guideline by Galvin and Sadowsky (Galvin, Sadowsky &
NINCDS-ADRDA 2012). The figure illustrates one possible line to follow, as is done in Kuopio region, Finland.
With respect to the screening test, one could recommend the well-known MMSE test, or CERAD such as is the case in Finland; CERAD includes MMSE as well as some other tests.
CERAD contains sections more focused on the study of memory and also behavioural components, therefore it can initially point towards to a more amnestic or behavioural presentation. Additionally the performance of the patient while undertaking daily life activities is conducted with ACDS-ADL. These tests along with a detailed interview of the patient and the caregiver, lead to the initial evaluation, i.e. deciding if the patient needs to be refered for further examinations. A detailed interview provides the highest amount of information, in particular when the first symptoms appeared and the age of the patient, which usually orientates more to either AD or FTD, although one should remember that both diseases can be present in every range of age, and both diseases can start with amnesia or behavioural symptomatology.
Once the patient is referred to the hospital, a new evaluation is done and it is then decided if the diagnosis would benefit from the input of some other tests and/or imaging. If the tests have been conducted 3 months earlier than the present time, they will be required to be done again in the hospital because it is essential to have access to up-to-date data.
This initial evaluation is followed by a first visit to the memory clinic and subsequent visits if needed. AD patients usually are directed to the memory clinic, while patients with
a behavioral profile may initially be referred to a psychiatrist. If there is a suspicion of FTD, the FBI battery is performed.
Finally, one has gathered a detailed clinical story of the patient, along with several neuropsychological tests, MRI (normal protocol; or memory protocol, focusing on temporal lobe atrophy in the hippocampus with Scheltens scale, general and focal atrophy, sulcus widening, ventricular enlargement, white matter hyperintensities in Fazekas scale, MBs.
Visual analysis is done although quantitative techniques are starting to become available), and possibly CSF biomarkers and SPECT or FDG-PET.
Although there have been used different versions of the PredictAD tool, Figure 13 shows a screenshot of the latest version of PredictAD (Windows application).
The MRI images go through a pipeline that estimates the results for HV, TBM and VBM (Figures 14 and 15).
96 Figure 13: Screenshot of patient overview in PredictAD tool, patient overview window. The PredictAD tool on a computer, selecting one single case, the clinician has access to the demographic and identifying information of the patient, a central image showing detailing information to allow selection (e.g. MMSE individual results, CERAD individual results, etc.), at the bottom the clinician can examine the entry timeline specifying all the times that the data was included in the tool and when it was actually obtained from the patient, and on the right side we can find the fingerprint of the patient and the patient’s position according to the distributions from the two population groups selected.
Figure 14. TBM analysis results visualised for one AD case (left) and for one healthy control (right). Shades of red and blue indicate areas which expand and shrink, respectively, in a way typical to AD. The AD case has clear expansion in the ventricles and shrinking in the MTL, not visible in the healthy control. Courtesy of Dr. Jyrki Lötjönen, VTT.
Figure 15. VBM analysis results visualised for one AD case (left) and for one healthy control (right). Shades of red and blue indicate areas which have higher or less GM concentration, respectively, in a way typical to AD. The AD case has clear less GM concentration in the MTL, not visible in the healthy control. Courtesy of Dr. Jyrki Lötjönen, VTT.
These three automatic methods are applied from a basic T1 image, and since it uses a sophisticated and elaborated protocol, this is not a time consuming procedure to conduct these studies, and could provide additional information for the clinician.
Data has originated from multiple sources, which different levels of importance, thus combining the information using the PredictAD tool could be useful. The question then is should it become part of the routine in the first visit and follow-up? Or should it be reserved for the situation when there is a plethora of biomarkers and tests? Perhaps it should be only utilized just in complicated and difficult cases?
To start, there is a basic problem: the availability of data. Not simply if these tests are done or not, or if the procedures for collecting data and samples are homogeneus, but the possibility of collating all of them together in comparable software format in a reasonable time. Even today most of the information is collected on paper, in some cases that is the only way it can be done (e.g. drawing tests). ACDS-ADL, GDS, CDR and FBI tests are not added to any software, sometimes the score is mentioned in the clinical anamnesis but often is not. With imaging, there is a different problem: images are in different formats, and one would need a direct transfer of these images to PredictAD, in order that they can be utilized. Finally, other tests are rarely conducted, such as taking a CSF sample in order to analyse biomarkers or alternatively conducting genetic profiling.
Nevertheless now ideally one has access to all the appropriate data and next one must decide whether or not to use PredictAD. There are two occasions when the PredictAD tool could be wisely used: the first time the patient comes to the memory clinic, if the clinician is unexperienced or not familiar with dementia diseases, this tool could be supportive or help to point to a certain diagnosis. However, in pure dementia cases, the clinician might not need any supportive tool, nonetheless in rare forms of dementia, where the different biomarkers do not clarify the state of the patient, the PredictAD tool might be helpful to the clinician. An initial diagnosis could be done, to allow initiation of therapy and counselling.
The second occasion is that in the follow-up, one could attempt to use PredictAD. In the future, more tests or biomarkers will be added with the intention of confirming the initial diagnosis. Furthermore, sometimes diagnoses need to be updated or modified. It is important that with ageing there tends to be an overlap of comorbidities, which may complicate the diagnosis e.g. the appearance of depression, extrapyramidal signs, cardiovascular diseases. PredictAD may help the clinician to elucidate this overlap and derive a diagnosis from it.
Finally, it is worth mentioning that while it is recommended to utilize the PredictAD for research purposes, it can only be viewed as a supportive tool in the clinic. A positive biomarker identified in the PredictAD tool just as any other individual biomarker (e.g.
medial temporal lobe atrophy) is not in itself sufficient to diagnose a disease, it is simply a risk for having a disease. In other words, the tool provides information which always has to be interpreted.
99 Figure 16. Assistance chain for the diagnosis and follow-up of a patient suspected of dementia
6.5 FUTURE STUDIES
Future projects will no doubt attempt to find new biomarkers and to combine novel biomarkers in order to come to an accurate diagnosis as early as possible in AD and FTD. In fact the work done in this thesis will be developed further in three projects: PredictAD pilot project, PredictND and VPH-DARE@IT.
Nowadays this PredictAD tool is being used in a pilot clinical project, the PredictAD pilot project, in two centers in Finland: Kuopio University Hospital and Turku University Hospital, to determine whether it helps clinicians to make a differential diagnosis of patients with dementia.
The VPH-DARE@IT will validate previously known methods or develop new biomarkers, attempting to combine the mechanistic and phenomenological models of the ageing brain and how these are influenced by environmental factors. This is a consortium involving 21 European centres and it started in 2013. There are new methods being assessed e.g. magnetic resonance elastography, RSfMRI, ASL, a model for integrating risk factors and DTI. All these biomarkers presumably could be integrated into the PredictAD tool.
PredictND which started in 2014 is intended to use biomarkers and tests from daily clinical use and develop PredictAD tool further also for other neurodegenerative diseases.
Finally, DSI could be used not only for diagnosing a patient at a specific moment from certain data, but to monitor the progression of the disease and determine whether the patient status has changed with time, in a longitudinal study (Runtti et al. 2014).
7 Conclusions
1. HV, TBM and VBM provide accurate results in these research studies when comparing the healthy state with disease and for predicting the conversion to AD. They may also help in differentiating between AD and FTD (study I).
2. DSI collates data from several tests and biomarkers, and can be supportive in the profiling of a patient with a certain dementia disease, i.e. whether it is MCI, FTD or AD.
DSF can help to profile a particular patient by displaying the findings in an easy-to-interpret picture format (studies II-IV).
3. Imaging is the most relevant feature in differentiating FTD from MCI and AD in comparison with MMSE, CSF and APOE, while MMSE is the most useful test distinguishing a healthy state from FTD (study II).
4. Clinical symptoms and neuropsychological tests are the most important studies in differentiating a healthy state from dementia and in distinguishing AD from FTD. MRI and particularly SPECT, APOE genotyping and CSF can be useful in distinguishing patients with AD from those with FTD (study III).
5. SPECT differentiates FTD from both controls and AD while manual hippocampal volumetry may be particularly helpful in the differential diagnosis between a healthy state and AD. It is recommended that a broad battery of neuropsychological tests is conducted rather than the single use of MMSE in the differentiation of a healthy state from dementia and AD from FTD (study III)
6. MRI features alone achieve good accuracies in predicting the progression from MCI to AD, this is only slightly improved by the addition of MMSE, APOE, CSF and neuropsychological tests (study IV)
7. The combination of data coming from multiple-center studies and their comparison is feasible using DSI. The accuracy of the combined cohort is close to the average of the individual cohorts. It is feasible to use different cohorts as training sets for the DSI, as
7. The combination of data coming from multiple-center studies and their comparison is feasible using DSI. The accuracy of the combined cohort is close to the average of the individual cohorts. It is feasible to use different cohorts as training sets for the DSI, as