Performance of blood biomarkers in Alzheimer's disease diagnosis : Comparison with the CSF biomarkers currently in use
Silvestrini, Martina (2023)
2023
Master's Programme in Biomedical Sciences and Engineering
Lääketieteen ja terveysteknologian tiedekunta - Faculty of Medicine and Health Technology
This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Hyväksymispäivämäärä
2023-08-23
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:tuni-202308037442
https://urn.fi/URN:NBN:fi:tuni-202308037442
Tiivistelmä
The aim of this thesis is to analyze the potential use of blood biomarkers in the early diagnosis of Alzheimer’s disease and for monitoring its progression. In particular, the blood biomarkers have been studied in comparison with the existing solutions for Alzheimer’s disease diagnosis, the CSF test and PET scan, which are known to be accurate in the diagnosis (PET scan is the ground truth test for dementia diagnosis) but also have the disadvantage of requiring invasive procedures and being costly. Moreover, PET scans and CSF tests are performed only when there is an absolute need for them, or when the diagnosis is not clear, they are not screening tools and require specialized personnel and expensive machines.
In this context, data from the available blood biomarkers Aβ42, Aβ40, Aβ42/Aβ40, P-tau, Tau, Nfl, and GFAP, was taken from several assays stored in the ADNI database, and a preliminary statistical analysis was run on all the biomarkers to assess their capability to differentiate the different diagnostic groups. Correlation tests were conducted to identify bindings between the blood biomarkers and the currently in use hallmarks, the CSF biomarkers, and PET SUVR measurements. Lastly to study the potential of the blood biomarkers to serve as early diagnostic tools, models for predicting PET SUVR AB+ based on such blood biomarkers and other features were also done, which was the main work of this thesis.
As shown from the preliminary analysis in this thesis, Aβ42/Aβ40 seemingly better differentiates patients at an early stage of the disease compared to the other hallmarks, therefore was initially chosen as the main feature for the modeling of the prediction model for PET SUVR AB+. P-tau, on the other hand, showed high disease specificity, which made it also a good feature for the model, despite its low performance in the distinction between CN and MCI subjects. No other biomarker was included in the models. The blood biomarkers’ correlation with CSF and PET biomarkers was used as an initial exploration method to check for which ones the relationship with CSF and PET biomarkers could be described with a monotonic function, and the results exhibit good correlation between the CSF Aβ42 and the blood Aβ42/Aβ40, and between CSF and blood NfL, while P-tau’s concentration varied significantly in the two mediums and Tau also didn’t show significant correlation. PET SUVR showed little negative correlation with blood β42/Aβ40 and P-tau. The PET SUVR prediction models using Aβ42/Aβ40 and P-tau as main features have shown interesting potential, considering both their low invasiveness, good accuracy (AUC : 85%, SD : 5% for Aβ42/Aβ40 best models, AUC : 78%, SD : 2% for P-tau best models) and more approachable costs. Despite the comparison with similar CSF models showing lower accuracy of the blood biomarkers’ models, it was not significant enough to justify the high costs and invasiveness of the former rather than the latter in early diagnosis.
In this context, data from the available blood biomarkers Aβ42, Aβ40, Aβ42/Aβ40, P-tau, Tau, Nfl, and GFAP, was taken from several assays stored in the ADNI database, and a preliminary statistical analysis was run on all the biomarkers to assess their capability to differentiate the different diagnostic groups. Correlation tests were conducted to identify bindings between the blood biomarkers and the currently in use hallmarks, the CSF biomarkers, and PET SUVR measurements. Lastly to study the potential of the blood biomarkers to serve as early diagnostic tools, models for predicting PET SUVR AB+ based on such blood biomarkers and other features were also done, which was the main work of this thesis.
As shown from the preliminary analysis in this thesis, Aβ42/Aβ40 seemingly better differentiates patients at an early stage of the disease compared to the other hallmarks, therefore was initially chosen as the main feature for the modeling of the prediction model for PET SUVR AB+. P-tau, on the other hand, showed high disease specificity, which made it also a good feature for the model, despite its low performance in the distinction between CN and MCI subjects. No other biomarker was included in the models. The blood biomarkers’ correlation with CSF and PET biomarkers was used as an initial exploration method to check for which ones the relationship with CSF and PET biomarkers could be described with a monotonic function, and the results exhibit good correlation between the CSF Aβ42 and the blood Aβ42/Aβ40, and between CSF and blood NfL, while P-tau’s concentration varied significantly in the two mediums and Tau also didn’t show significant correlation. PET SUVR showed little negative correlation with blood β42/Aβ40 and P-tau. The PET SUVR prediction models using Aβ42/Aβ40 and P-tau as main features have shown interesting potential, considering both their low invasiveness, good accuracy (AUC : 85%, SD : 5% for Aβ42/Aβ40 best models, AUC : 78%, SD : 2% for P-tau best models) and more approachable costs. Despite the comparison with similar CSF models showing lower accuracy of the blood biomarkers’ models, it was not significant enough to justify the high costs and invasiveness of the former rather than the latter in early diagnosis.