Periodic Reporting for period 1 - STRATA-ALZ (Optimizing stratification for trial design in Alzheimer’s disease)
Periodo di rendicontazione: 2023-04-01 al 2025-03-31
Sintesi del contesto e degli obiettivi generali del progetto
Background
Dementia is one of the greatest health challenges of our time. It causes major disability and dependence in older adults, and the number of people affected is growing rapidly. By 2050, an estimated 150 million people could be living with dementia worldwide. This has an enormous emotional and financial impact on families, healthcare systems, and society. The most common cause of dementia is Alzheimer’s disease (AD)—a brain disorder that begins quietly, with harmful changes starting many years before symptoms like memory loss appear. These early changes involve the buildup of specific proteins (amyloid and tau), which eventually damage brain cells and lead to cognitive decline.
We now know that a large portion of people over 50 already show early signs of these changes, even before they feel any symptoms. This means there is a huge opportunity to intervene earlier—before the damage becomes irreversible. But to succeed, we need better tools to identify and group the right people for the right treatments in clinical trials.
The Problem: Why dementia trials often fail
Despite major investments in research, most clinical trials for Alzheimer’s treatments have not been successful. One key reason is that these trials often include participants who are too far along in the disease, or whose disease is driven by different biological factors. Alzheimer’s is not a one-size-fits-all condition. People with the disease can differ widely in terms of symptoms, underlying biology, and other health problems like vascular disease. If these differences aren't accounted for, it becomes very hard to detect whether a treatment is truly working. In other words, poor patient selection and grouping ('stratification') can mask the effects of promising treatments. Recent trials that did show some success used smarter ways to group participants based on brain imaging. This shows that more personalized approaches to trial design can make a real difference—and that's exactly what this project aims to achieve.
The Project Goal: Tailoring trials to people’s unique disease patterns
This project aims to optimize how people are selected and evaluated in Alzheimer's clinical trials, especially in the earliest stages of the disease (before or just as symptoms appear). By better understanding and categorizing the different ways the disease progresses, we can:
1. Identify distinct disease "pathways"—for example, whether a person’s Alzheimer’s is mainly driven by protein buildup, blood vessel damage, or both.
2. Use this information to form more precise participant groups in trials.
3. Improve how treatment effects are measured by choosing the right cognitive tests and brain scans based on each disease subtype.
To do this, I will analyze data from over 2,900 participants in the BioFINDER studies—world-leading Swedish research studies that track brain health over time using advanced imaging, blood and spinal fluid tests, and cognitive assessments. As such, the project will map out different biological subtypes of early Alzheimer’s using imaging data (e.g. brain scans that show protein buildup or small vessel disease), explore new fluid biomarkers that reflect brain health and immune activity, using cutting-edge protein analysis tools, and match disease subtypes with specific patterns of cognitive decline, to improve how we measure treatment success in future trials.
Why it matters and what I aim to achieve:
This project comes at a crucial moment in the global effort to tackle dementia, a condition that is not only devastating for individuals and families but also poses a growing challenge for healthcare systems.
This project directly addresses that problem by bringing a more personalized, data-driven approach to how we design and run clinical trials. By better understanding the different ways Alzheimer’s can develop, and by matching participants to the right trials based on their specific disease pathway, we can significantly improve the chances of detecting true treatment effects. In turn, this means we can move faster and more confidently toward therapies that work.
Beyond its scientific innovation, the project also aligns with broader health and policy goals. It supports the push for earlier diagnosis, personalized medicine, and smarter, more efficient use of healthcare resources. The potential impact is significant: improving how we test treatments could accelerate the arrival of therapies that delay or even prevent dementia. This would not only improve quality of life for millions of people but also reduce the long-term societal and economic burden of the disease.
In short, by tailoring clinical trials to the biological reality of Alzheimer’s disease, this project aims to change the way we fight dementia—making the path to effective treatments clearer, faster, and more successful.
Dementia is one of the greatest health challenges of our time. It causes major disability and dependence in older adults, and the number of people affected is growing rapidly. By 2050, an estimated 150 million people could be living with dementia worldwide. This has an enormous emotional and financial impact on families, healthcare systems, and society. The most common cause of dementia is Alzheimer’s disease (AD)—a brain disorder that begins quietly, with harmful changes starting many years before symptoms like memory loss appear. These early changes involve the buildup of specific proteins (amyloid and tau), which eventually damage brain cells and lead to cognitive decline.
We now know that a large portion of people over 50 already show early signs of these changes, even before they feel any symptoms. This means there is a huge opportunity to intervene earlier—before the damage becomes irreversible. But to succeed, we need better tools to identify and group the right people for the right treatments in clinical trials.
The Problem: Why dementia trials often fail
Despite major investments in research, most clinical trials for Alzheimer’s treatments have not been successful. One key reason is that these trials often include participants who are too far along in the disease, or whose disease is driven by different biological factors. Alzheimer’s is not a one-size-fits-all condition. People with the disease can differ widely in terms of symptoms, underlying biology, and other health problems like vascular disease. If these differences aren't accounted for, it becomes very hard to detect whether a treatment is truly working. In other words, poor patient selection and grouping ('stratification') can mask the effects of promising treatments. Recent trials that did show some success used smarter ways to group participants based on brain imaging. This shows that more personalized approaches to trial design can make a real difference—and that's exactly what this project aims to achieve.
The Project Goal: Tailoring trials to people’s unique disease patterns
This project aims to optimize how people are selected and evaluated in Alzheimer's clinical trials, especially in the earliest stages of the disease (before or just as symptoms appear). By better understanding and categorizing the different ways the disease progresses, we can:
1. Identify distinct disease "pathways"—for example, whether a person’s Alzheimer’s is mainly driven by protein buildup, blood vessel damage, or both.
2. Use this information to form more precise participant groups in trials.
3. Improve how treatment effects are measured by choosing the right cognitive tests and brain scans based on each disease subtype.
To do this, I will analyze data from over 2,900 participants in the BioFINDER studies—world-leading Swedish research studies that track brain health over time using advanced imaging, blood and spinal fluid tests, and cognitive assessments. As such, the project will map out different biological subtypes of early Alzheimer’s using imaging data (e.g. brain scans that show protein buildup or small vessel disease), explore new fluid biomarkers that reflect brain health and immune activity, using cutting-edge protein analysis tools, and match disease subtypes with specific patterns of cognitive decline, to improve how we measure treatment success in future trials.
Why it matters and what I aim to achieve:
This project comes at a crucial moment in the global effort to tackle dementia, a condition that is not only devastating for individuals and families but also poses a growing challenge for healthcare systems.
This project directly addresses that problem by bringing a more personalized, data-driven approach to how we design and run clinical trials. By better understanding the different ways Alzheimer’s can develop, and by matching participants to the right trials based on their specific disease pathway, we can significantly improve the chances of detecting true treatment effects. In turn, this means we can move faster and more confidently toward therapies that work.
Beyond its scientific innovation, the project also aligns with broader health and policy goals. It supports the push for earlier diagnosis, personalized medicine, and smarter, more efficient use of healthcare resources. The potential impact is significant: improving how we test treatments could accelerate the arrival of therapies that delay or even prevent dementia. This would not only improve quality of life for millions of people but also reduce the long-term societal and economic burden of the disease.
In short, by tailoring clinical trials to the biological reality of Alzheimer’s disease, this project aims to change the way we fight dementia—making the path to effective treatments clearer, faster, and more successful.
Lavoro eseguito dall’inizio del progetto fino alla fine del periodo coperto dalla relazione e principali risultati finora ottenuti
This project used data that has already been collected from nearly 3,000 people who took part in two large research studies called BioFINDER-1 and BioFINDER-2. These participants include individuals who are at risk of developing Alzheimer’s disease, people with mild memory problems, and some who have reported early concerns about their cognition. All participants gave their consent to take part in these studies.
Because the data—including brain scans, fluid samples, and cognitive tests—have already been collected and are largely pre-processed, most of the study time was used for the scientific analysis.
For this work, I used 4 main categories of data:
- Brain PET Imaging: Dedicated brain scans were used to measure the build-up of two key proteins involved in Alzheimer’s—amyloid and tau. These scans help us understand where in the brain these proteins are accumulating and how early they appear. I have performed visual read and quality checks of the available baseline scans, to prepare their use for analysis.
- Brain MRI Scans: Another type of brain imaging was used to look at signs of small blood vessel disease, which is common in older adults and can influence memory and brain health. To expand the available data on vascular burden, all baseline MR scans (FLAIR, T2, and T2* sequences) were re-assessed by an expert radiologists and myself to collect additional information on the location and extent of vascular burden, including microbleeds, lacunary infarcts, and regional White matter Hyperintensities (WMH). In addition, I facilitated a collaboration with Dr. Carole Sudre, who performed high quality segmentation of white matter lesions, providing unique regional information required for the main objective of the work.
- Fluid Samples (CSF and Blood): Samples from the participants’ spinal fluid and blood were collected and stored. These samples are being analyzed for biological markers that tell us how brain cells are functioning and how the immune system is responding. In my work, I investigated different essays that provide complementary information.
- Cognitive tests: All participants completed a series of tests that cover different areas of brain function like attention, memory, language, and problem-solving. As cognition remains the main outcome criteria, cognition was a key outcome in my work.
Statistical analysis and models:
The original proposal stated the use of non-negative matrix factorization (NMF) clustering to identify the different amyloid-vascular subtypes (objective 1). However, after discussing with several experts in the field, and taking into account the skewness of the data and our interest to create a longitudinal model, we applied the Subtype and Stage Inference (SuStaIn) instead. With this approach, we identified two subtypes of white matter hyperintensities or white matter lesions, one frontal and one posterior. This was in line with previous literature, which was encouraging, while our models provides a temporal aspect to this heterogeneity. These two subtypes are differentially associated with other vascular risk factors and risk of cognitive decline.
However, due to the inherently vastly different behavior of amyloid-PET and WMH data, it was not possible to create a integrated model of the two common pathologies. Instead, we switched our approach to a key trial stratification and outcome tool, Tau-PET imaging. By investigating patterns on amyloid-PET, MR-based WMH, and MR-based atrophy across disease subtypes based on Tau-PET, we observed that the regional involvement of the brain is across these key biomarkers. In addition, some subanalysis using CSF panels indicated that a particular subtype is associated with high levels of neurodegeneration, while others have more inflammatory aspects. We are currently investigated how to optimally integrate all the available data to predict cognitive decline.
Because the data—including brain scans, fluid samples, and cognitive tests—have already been collected and are largely pre-processed, most of the study time was used for the scientific analysis.
For this work, I used 4 main categories of data:
- Brain PET Imaging: Dedicated brain scans were used to measure the build-up of two key proteins involved in Alzheimer’s—amyloid and tau. These scans help us understand where in the brain these proteins are accumulating and how early they appear. I have performed visual read and quality checks of the available baseline scans, to prepare their use for analysis.
- Brain MRI Scans: Another type of brain imaging was used to look at signs of small blood vessel disease, which is common in older adults and can influence memory and brain health. To expand the available data on vascular burden, all baseline MR scans (FLAIR, T2, and T2* sequences) were re-assessed by an expert radiologists and myself to collect additional information on the location and extent of vascular burden, including microbleeds, lacunary infarcts, and regional White matter Hyperintensities (WMH). In addition, I facilitated a collaboration with Dr. Carole Sudre, who performed high quality segmentation of white matter lesions, providing unique regional information required for the main objective of the work.
- Fluid Samples (CSF and Blood): Samples from the participants’ spinal fluid and blood were collected and stored. These samples are being analyzed for biological markers that tell us how brain cells are functioning and how the immune system is responding. In my work, I investigated different essays that provide complementary information.
- Cognitive tests: All participants completed a series of tests that cover different areas of brain function like attention, memory, language, and problem-solving. As cognition remains the main outcome criteria, cognition was a key outcome in my work.
Statistical analysis and models:
The original proposal stated the use of non-negative matrix factorization (NMF) clustering to identify the different amyloid-vascular subtypes (objective 1). However, after discussing with several experts in the field, and taking into account the skewness of the data and our interest to create a longitudinal model, we applied the Subtype and Stage Inference (SuStaIn) instead. With this approach, we identified two subtypes of white matter hyperintensities or white matter lesions, one frontal and one posterior. This was in line with previous literature, which was encouraging, while our models provides a temporal aspect to this heterogeneity. These two subtypes are differentially associated with other vascular risk factors and risk of cognitive decline.
However, due to the inherently vastly different behavior of amyloid-PET and WMH data, it was not possible to create a integrated model of the two common pathologies. Instead, we switched our approach to a key trial stratification and outcome tool, Tau-PET imaging. By investigating patterns on amyloid-PET, MR-based WMH, and MR-based atrophy across disease subtypes based on Tau-PET, we observed that the regional involvement of the brain is across these key biomarkers. In addition, some subanalysis using CSF panels indicated that a particular subtype is associated with high levels of neurodegeneration, while others have more inflammatory aspects. We are currently investigated how to optimally integrate all the available data to predict cognitive decline.
Progressi oltre lo stato dell’arte e potenziale impatto previsto (incluso l’impatto socioeconomico e le implicazioni sociali più ampie del progetto fino ad ora)
My findings has provided valuable insights into disease heterogeneity and opportunities but also challenges to model them.
Firstly, we confirmed the presence of two WM lesion subtypes in the context of AD and their association with distinct risk factors. This work has been presented at the Clinical Trials in AD (CTAT) 2024 conference. However, due to the inherently vastly different behavior of amyloid-PET and WMH data, it was not possible to create a integrated model of the two common pathologies. Future work should focus on developing and testing algorithms that can deal with vastly different data. Potentially deep learning approaches should be investigated, which I am currently discussing with experts from University College London (UCL).
Secondly, I observed that pathological heterogeneity in terms of patterns in the brain is present across imaging biomarkers and are inter-related. Currently, I am collaborating with experts in the field to determine how to optimally model this information and predict risk of cognitive decline at the individual level. Interim results have been presented at CTAD 2024 and ADPD 2025 conferences. The completed work is aimed to be published in 2025.
Finally, I still want to perform highly novel proteomics analysis to investigate if there are distinct disease pathways underlying this heterogeneity I observed in the brain images.
Together, my findings add to the previous literature, illustrating the link between AD-associated pathologies in terms of their regional distribution and their effect on disease progression. This work will add to our understanding of different progression rates and profiles in patients and support the identification of the right patient for the right treatment. An interesting expansion of the work, would be applying this information to trial data and assess their association with risk of anti-amyloid therapy side-effects. Also, to further our understanding of disease pathways, determine if AD imaging subtypes are associated with different genetic markers would be highly valuable.
Firstly, we confirmed the presence of two WM lesion subtypes in the context of AD and their association with distinct risk factors. This work has been presented at the Clinical Trials in AD (CTAT) 2024 conference. However, due to the inherently vastly different behavior of amyloid-PET and WMH data, it was not possible to create a integrated model of the two common pathologies. Future work should focus on developing and testing algorithms that can deal with vastly different data. Potentially deep learning approaches should be investigated, which I am currently discussing with experts from University College London (UCL).
Secondly, I observed that pathological heterogeneity in terms of patterns in the brain is present across imaging biomarkers and are inter-related. Currently, I am collaborating with experts in the field to determine how to optimally model this information and predict risk of cognitive decline at the individual level. Interim results have been presented at CTAD 2024 and ADPD 2025 conferences. The completed work is aimed to be published in 2025.
Finally, I still want to perform highly novel proteomics analysis to investigate if there are distinct disease pathways underlying this heterogeneity I observed in the brain images.
Together, my findings add to the previous literature, illustrating the link between AD-associated pathologies in terms of their regional distribution and their effect on disease progression. This work will add to our understanding of different progression rates and profiles in patients and support the identification of the right patient for the right treatment. An interesting expansion of the work, would be applying this information to trial data and assess their association with risk of anti-amyloid therapy side-effects. Also, to further our understanding of disease pathways, determine if AD imaging subtypes are associated with different genetic markers would be highly valuable.