Abstract
Big data analysis in the dementia field has provided a deeper understanding of underlying disease patterns, however, it has also led to a reproducibility crisis. To address this, we developed a comprehensive multi-time-point data preparation approach for analysing longitudinal data to reduce bias and variation across studies. Using a large open-source dataset, this approach was employed to analyse cardiometabolic risk factors associated with disease progression in Mild Cognitive Impairment (MCI) and dementia. Compared to baseline analysis, fewer significant risk factors were associated with progression using the multi-time-point approach. Sensitivity analysis revealed the findings to be robust with depression increasing Healthy-to-MCI/dementia risk, and hypertension reducing MCI-to-dementia risk. The same approach was then employed to identify drug classes associated with disease progression, while comparing whether different diagnostic measures (clinician diagnosis and Clinical Dementia Rating Sum of Boxes (CDRSOB) scores) affect the results. Analysis revealed that using CDRSOB as the diagnostic measure compared to clinician diagnosis, differentially affects the magnitude and significance of certain drug classes associated with disease progression.This thesis also analysed novel blood-based biomarkers for MCI and dementia. Peripheral levels of anandamide (AEA), 2-Arachidonoylglycerol (2-AG), N-palmitoylethanolamide (PEA), and N-oleoylethanolamide (OEA) were measured in controls, MCI, AD, and other dementia (OD) patients. Analysis revealed altered levels of plasma AEA, PEA, and OEA in MCI, AD, and OD groups. Finally, large-scale proteomic data analysis using machine learning identified 34 important protein markers from cardiovascular, neurology, immune response, and inflammatory panels, that predicted AD with a balanced accuracy of 88.75%
Overall, this thesis emphasizes the importance of efficient data preparation methods, potentially leading to improvements in prognosis and prescribing patterns among the ageing population. Additionally, it may lead to development of blood-based diagnostic tests for early dementia stages which will have widespread clinical benefits.
Thesis is embargoed until 30th November 2023
| Date of Award | Nov 2021 |
|---|---|
| Original language | English |
| Sponsors | INTERREG IVA administered by the SEUPB. |
| Supervisor | Magda Bucholc (Supervisor), Paula McClean (Supervisor) & Kongfatt Wong-Lin (Supervisor) |
Keywords
- cardiometabolic risk factors
- Dementia progression
- longitudinal data
- mild cognitive impairment
- multivariate logistic regression
- national Alzheimer's Coordinating Center data
- Endocannabinoids
- blood biomarkers
- random forest
- drug classes
- predictive model
- feature selection
- Alzheimer's disease
Cite this
- Standard