Learning from Data Across the Alzheimer's Disease Spectrum

Earlier this summer, I had the opportunity to attend the Alzheimer's Association International Conference (AAIC), the largest gathering of people working on Alzheimer's Disease (AD).  I say "working on" because the breadth of effort is so stunning that it's hard not to use a catch-all term.  There were talks and posters from a strong corps of academics and booths from large pharma companies, small biotechs, consortia, and advocacy groups.  As my colleague Joy wrote in her recent blog post, AAIC is a great showcase of how these different stakeholders come together to tackle AD.

As a newcomer to AD research, one of the most striking features of the conference was the gap between academic work and clinical trials.  A significant fraction of academic work presented at AAIC focused on understanding the transition from normal to impaired cognitive function.  Clinical trials, on the other hand, have been largely focused on the later stages of AD, which are clinically much more clear-cut.  It's distressing to see this gap - billions are spent on clinical research on what's often called "mild to moderate" AD, while large swathes of the academic community have largely decided that it's already too late and drugs need to be developed earlier in the disease.  To make matters worse, the disparate data generated by these efforts risks leaving us with a fractured picture of the disease.

Of course, this observation is nothing new.  The question of what the right stage is AD treatments is one of the most prominent being asked of current drug programs.  In the media this manifests through opinion pieces stressing the need to move trials to earlier stages of AD (see here and here).  And the situation we find ourselves in now is a natural one given how the community has been thinking about AD over the past two decades.  Hypotheses about the cause of AD have risen and fallen over the years, but some of the most prominent theories suggest that cognitive decline is just the last - and depressingly, most obvious - signpost in a decades-long molecular-scale deterioration of the brain.  

Yet it isn't time to throw in the towel, so to speak. Progress is still happening: clinical trials are moving towards the preventive stage, and academia has made great advances in understanding the earlier stages of AD.

At Unlearn, we're also working towards creating solutions for treating earlier stages of AD. Specifically, we aim to model control subjects in clinical trials, so naturally we're interested in where clinical trials are going.  At AAIC, I watched the talks with this question in mind: how would we build a model of disease progression in the early stages of AD?

Let's take a quick step back and build a picture of AD progression, focusing on cognitive aspects of the disease.  The earliest stages are termed subjective cognitive decline (SCD), where impairment is just that - subjective.  A 60-year-old patient may have the cognitive function of a 75-year-old but pass a standard battery of cognitive tests.  At the next stage of the dysfunction, mild cognitive impairment (MCI), that patient may show cognitive decline on those tests.  Whereas SCD is subjective, MCI is objective.  Patients with MCI will exhibit symptoms of cognitive decline in areas like memory, language, or judgment while their ability to perform daily activities is not impacted.  This decline can be measured through cognitive tests although the ability to detect MCI remains a challenge.  In the later stages of progression, AD is characterized by more significant cognitive decline and functional impairment. 

The ability to track AD progression is complicated for a couple of reasons. Not every patient with SCD will transition to MCI, and not every patient with MCI will transition to AD (or other dementias).   Also due to natural cognitive decline as people age, most cases of SCD and MCI are not diagnosed, leading to the sobering fact that most diagnoses happen when a patient already has AD.

A major effort in understanding SCD and the transition to MCI is to build cognitive tests that are sensitive to the early phase of the disease.  I also saw a lot of great talks at AAIC on work aimed at figuring out who will progress towards dementia and how to connect this to molecular biology.  But as a data scientist - more specifically, someone who thinks about how to organize data and build useful resources for modeling - I focus on the question of how to use data from these studies.  One very powerful feature of clinical trials is that they are largely standardized, meaning the measurements collected across trials are generally quite similar.  

Because academics are thinking about how to detect the earliest signs of cognitive impairment and make them more objective, they often devise new tests.  This has led, over the years, to a profusion of tools for assessing cognitive decline.  Watching the talks, I definitely got this feeling.  There were more acronyms for tests than I could count!  Having a plethora of tools poses a real challenge to data science and analysis.  Because individual studies of SCD and early MCI are small and typically not very diverse, data scientists get a lot of benefit from combining datasets.  If the studies don't measure the same things, they produce a lot of fragmented datasets that are significantly more difficult to compare to one another and use together.

So when working with medical data, such as cognitive test results, standardization is a big deal.  At Unlearn, we work hard to build good tools that standardize data across clinical trials, and when we think about the larger ecosystem of research, standardization makes everyone's work more powerful.  It can be a tradeoff choosing what to measure requires compromise but the benefits to the common cause can be immense.  All of these great studies that do good science on their own can be pooled into a great resource and extend the science even further.

At AAIC, I saw some amazing examples of efforts to build resources that have high quality, standardized data.  At Unlearn we work with data from the Critical Path Institute's Critical Path for Alzheimer's Disease (CPAD) consortium.  CPAD has built a fantastic resource of control arm data from clinical trials, which has been used to create drug development tools open to the community.  Another great effort is the European Prevention of Alzheimer's Dementia Consortium (EPAD), a partnership of industry, medical centers, and academic groups carrying out a multi-center study that has taken a thoughtful approach to the data they are collecting and how they are using it.  It's going to be exciting to see what trials and results emerge from this effort.

On my flight home from AAIC, my thoughts turned to the long-term work in developing therapeutics and testing them on earlier stage populations at risk of developing AD.  What will we learn from these trials, and how can we get better at conducting these experiments?  How will ongoing academic efforts inform drug development?  These are long-term questions, but I only hope that by learning from the data from both of these groups we can move faster towards answers.

Photo credit: Jon Walsh

Enter your email address to download paper.

Click the link to begin download.
Oops! Something went wrong while submitting the form.

Enter your email address to watch the webinar.

Click the link to watch webinar.
Oops! Something went wrong while submitting the form.

Why can’t we agree on how to define digital twins in healthcare?

White Papers

Summary of the EMA September 2022 Qualification Opinion for PROCOVA™


Charles Fisher, Unlearn.AI: “now is the time to adopt AI-based solutions”

The potential for AI implementation in healthcare can barely be measured, as it can already do what humans do, just countless times better and more efficiently.
The European Medicines Agency has qualified Unlearn’s AI-powered method for running smaller, faster clinical trials.
Digital twins seem to be everywhere in healthcare now, but no one agrees on a single definition for them.