Foundation for Precision Medicine: Detecting Alzheimer’s disease earlier with big data

Alzheimer’s disease is heartbreaking for those afflicted, and perhaps even more so for their families and friends. As mortality from leading killers such as heart disease and cancer declines, deaths due to Alzheimer’s are sharply on the rise, increasing 123% between 2000 and 2015, according to the 2018 Alzheimer’s Disease Facts and Figures published by the Alzheimer’s Association. Alzheimer’s is now the sixth leading cause of death in the United States, and about one-third of all Medicare beneficiaries who die in a given year have been diagnosed with dementia.

Unfortunately, despite decades of research, there is no cure. Clinical treatments are marginally effective, and only in the early stages of the disease. To save time and lives, diagnosis prior to the dementia stage is currently the only hope. Alzheimer’s is already the most expensive single disease in the United States, claiming 20% of Medicare dollars. Annual payments for caring for individuals with Alzheimer’s or other dementias have already surpassed $270 billion, and cumulative long-term care costs will push into the trillions. Unless progress is made—and made quickly—Alzheimer’s could bankrupt the existing healthcare financing system.

After losing his mother to Alzheimer’s, Silicon Valley restaurateur Robert Tabz decided to take action. He joined a world-class team of scientists, medical doctors, and philanthropists to create the Foundation for Precision Medicine. The Foundation for Precision Medicine is helping to lead the fight against Alzheimer’s and other brain diseases by using artificial intelligence and big data to enable clinical diagnoses before loved ones regress and it is too late to intervene.

The Foundation for Precision Medicine’s team of top data scientists from Harvard, MIT, and Johns Hopkins studies one of the biggest Alzheimer’s patient datasets available—70,000 electronic health records for Alzheimer’s patients from hospitals and medical facilities across the United States—to develop machine learning (ML) algorithms that can help detect the disease before symptoms appear. The faster the Foundation for Precision Medicine can develop those algorithms, the more hope there is for early detection.

“By building a community and platform with novel methodologies, we are discovering new patterns allowing us to predict Alzheimer’s disease and guide patients and families in modifying risk factors that could delay the onset or limit the progression of this devastating disease,” says Mark Ereth, MD, co-founder and board chair at the Foundation for Precision Medicine, and emeritus professor at the Mayo Clinic.

A head start with Google

The Foundation for Precision Medicine’s “precision” approach is complex. It must consider individual variability in genes and lifestyle while measuring molecular, environmental, and behavioral factors that may trigger or contribute to the disease. However, as a nonprofit research startup, the Foundation for Precision Medicine has limited funding for staffing and technology. To prove the viability of its science and attract funding, it needed to develop a proof of concept showing that it could predict the onset of Alzheimer’s months in advance. After a slow start using an in-house server and laptops, the Foundation for Precision Medicine decided to analyze datasets and develop its algorithms entirely on Google Cloud.

“As part of its Data Solutions for Change program, Google provided us with free Google Cloud credits and gave us a good head start on our research to accelerate Alzheimer’s detection,” says Ayin Vala, co-founder and Chief Data Scientist at the Foundation for Precision Medicine. “With Google’s help, we gained traction quickly and completed our proof of concept in four months instead of a year.”

Revolutionizing data analysis

Before detection algorithms can be developed, datasets must be cleansed, prepared, and analyzed. Previously, making large datasets available to scientists and researchers was difficult. The Foundation for Precision Medicine did not have access to enough processing power to analyze entire datasets, so scientists had to down-sample the data and analyze small portions at a time. This approach lead to accuracy issues because hypotheses gleaned from the analyses could not be verified against a larger dataset.

Moving to Google Cloud solved this problem while supporting compliance with the US Health Insurance Portability and Accountability Act (HIPAA). The Foundation for Precision Medicine now uses Cloud Dataprep, an integrated partner service operated by Trifacta, to prepare the data for analysis. It then uses BigQuery to store and analyze entire datasets on demand. Accuracy immediately improved, and Area Under Curve (AUC—a widely used measure of the effectiveness of ML predictions) jumped from .7 to .8, a huge improvement for a disease-detection algorithm.

“BigQuery is revolutionizing our work,” says Ayin. “It’s 100x faster than the in-house server we used before, allowing us to analyze 100x larger datasets. That means we can query full datasets, test against all kinds of variables, and ultimately make our algorithms more accurate. Migrating from flat files to BigQuery was easy.”

“Detecting patterns and relationships within large datasets is critical in advancing medical science, especially in highly complex disease states such as Alzheimer’s," says Dr. Ereth.

Now that the Foundation for Precision Medicine can analyze larger datasets, Ayin hopes that its healthcare partners will provide more electronic health records, moving beyond the current control group of 4 million records to as many as 80 million. This could give the Foundation for Precision Medicine data on more than 1 million Alzheimer’s patients, helping it make more accurate predictions and study more population subsets.

70% more time for science

Previously, data scientists spent as much as 90% of their time on data engineering and cleansing tasks leading up to the analysis phase. Today, they spend only 20%, leaving the rest of their time available for more meaningful work.

“Since moving to Google Cloud, we’ve reclaimed 70% of our data team’s time for scientific discovery,” says Ayin.

To visualize the data and share reports, scientists use Looker Studio, which works seamlessly with BigQuery to keep the entire process more securely within Google Cloud and supports HIPAA compliance. Because scientists no longer need to spend hours transferring datasets between environments to obtain the necessary visualizations, they can move on to the machine learning phase in 25% less time.

“Our researchers love Looker Studio because it empowers them to visualize billions of rows of data in seconds without being computer science experts, and without leaving the secure cloud environment,” says Ayin. “It’s also constantly being enhanced and improved, which we love about all Google Cloud services.”

Accelerating machine learning

The Foundation for Precision Medicine’s machine learning environment is based on Compute Engine virtual machines (VMs), using both CPUs and GPUs for fast processing to develop algorithms 5x to10x faster. The Foundation for Precision Medicine also uses Cloud Datalab, a free interactive service, for scalable model training on preconfigured VMs. Moving forward, the foundation will build ML models using a combination of custom coding and BigQuery ML.

“BigQuery is empowering our data analysts and statisticians to build advanced analytics solutions,” says Ayin. “BigQuery ML expands our workforce to come up with new and innovative ideas when developing machine learning models. In our organization, it is now the fastest way to build an ML model, and the fastest way to run it on our large datasets. The automatic one-hot encoding works very well and makes it easy for handling categorical features. BigQuery ML is the quickest way to examine whether there is signal in our data, so we can allocate resources to develop more advanced models.”

More collaborative research

Working in Google Cloud allows the Foundation for Precision Medicine’s researchers to be much more collaborative, sharing datasets and bringing in volunteers from academia and industry. When new scientists and researchers come on board, they receive notification via Gmail and can instantly begin working with data in Google Cloud instead of spending hours downloading what they need. When researchers from around the world need to exchange ideas face-to-face, they use Google Meet to conduct high-quality video meetings.

“It’s so much easier to share data more securely and do collaborative research using Google Cloud and Google Workspace” says Ayin. “We’re also collaborating with Google engineers, who donate their time to our cause.”

Fighting a brutal disease

The Foundation for Precision Medicine is currently developing an Alzheimer’s detection mobile app that it plans to make available for free for Android and iOS devices. The app’s backend will be hosted on Kubernetes Engine for easy scalability, helping to democratize early detection across the globe and empower families to take action. Soon, the Foundation for Precision Medicine plans to study other brain diseases, such as epilepsy, to drive new discoveries that will help victims manage their condition. It’s also researching pharmacogenomics to discover which drugs are most likely to work well for certain patients.

“We’re a young foundation and, with help from Google, we’ve quickly made a lot of progress,” adds Dr. Ereth. “With the ability to rapidly analyze giant datasets and build machine learning models, the world is wide open for medical science.”

Data Solutions for Change

The Foundation for Precision Medicine participated in Data Solutions for Change, a Google data analytics program for nonprofits to achieve their missions at scale. As a participant, the Foundation for Precision Medicine received Google Cloud credits, self-training resources, and enablement support.