Driving Digital Discovery
We live in a digital world.
Almost everyone has a supercomputer in their pocket. With a click of a button, you can have a package delivered to your doorstep. You can ask Siri where the nearest restaurant is.
But what about your health care?
“Health care isn’t yet present in your day-to-day digital life,” said Edmondo Robinson, MD, MBA, chief digital innovation officer at Moffitt Cancer Center. “Health care has a ways to go to catch up to the rest of the digital world.”
There has been a major push to digitize health care, with health care organizations harnessing new technology to improve business operations, communications and direct patient care. Many of these technologies use artificial intelligence and their underlying machine learning algorithms to transform workflows and improve decision-making. In order to develop these solutions, you must have high-quality data.
Moffitt began investing heavily in its data ecosystem about a decade ago when it established its first enterprise data warehouse, the Health and Research Informatics platform. In order to take advantage of cutting-edge technology, the cancer center would need data — electronic medical records, biobanking systems, patient information, cancer registry data, survival information — and ways to mine that data for research and clinical purposes.
Dana Rollison, PhD, vice president and chief data officer
“Over time we realized we needed to work more with unstructured data, like the text of an electronic medical record,” said Dana Rollison, PhD, vice president and chief data officer. “These reports are generated from dictation so we needed more sophisticated tools to pull information out of them.”
In 2016, the Enterprise Wide Analytics Strategy Steering Committee was formed to take a broader look at all of the institution’s data needs, from payer strategies to clinical care. The committee determined a priority initiative was utilizing AI to obtain more timely information on cancer diagnosis and staging from the text of pathology reports, using cancer registry data to train the algorithms.
Artificial Intelligence
AI is a wide-ranging branch of computer science concerned with building computers that are programmed to act and think more like humans.
“A lot of the technology used to extract or mine information is based in AI,” said Rollison. “We recognized AI also has applications across the research and clinical spectrum, like in drug discovery, health outcomes and behavior research, imaging analytics and prediction of treatment outcomes.”
At the end of 2018, Moffitt hired its first AI officer, J. Ross Mitchell, PhD. Mitchell began working on the Nvidia DGX-1 supercomputer.
J. Ross Mitchell, PhD, poses in front of Moffitt's supercomputer.
“It has the same power as the No. 1 supercomputer in the world in 2008 called the Roadrunner installed in Los Alamos,” said Mitchell. “That contained 20,000 processors, 300 racks of equipment, took up 6,000 square feet and cost $100 million to build.”
Moffitt’s supercomputer is about the same size as a typical personal computer tower.
Once the infrastructure was in place, Mitchell had to train the supercomputer for deep learning, a field within AI that deals with algorithms inspired from a human brain to aid machines with intelligence without explicit programming. Using a powerful new natural language processing tool called Bidirectional Encoder Representations from Transformers, or BERT, Mitchell’s goal was to extract information from pathology reports to help identify clinical trials and treatment pathways for cancer patients.
“If you want to put a patient on a clinical trial, someone has to look through the pathology reports manually and try to figure out what they might be suitable for,” said Mitchell. “That is an incredible drain on resources, time, treatment, everything. For years, it’s been a dream to have a computer use its power to do this, but that hasn’t really been possible until now.”
Mitchell trained the Nvidia DGX-1 supercomputer on Wikipedia pages, books, abstracts found in biomedical search engines and discharge notes from intensive care unit encounters. He brought in almost 14,000 Moffitt pathology reports and taught the machine how to answer questions. It was now possible to search the pathology report database for 51 different organs and 26 different types of solid cancers.
The program can determine the histology of a tumor with 96.7% accuracy and the tumor site with 92.9% accuracy. The next step is to improve the performance and training of BERT using 470,000 pathology reports to extract even more information from health records to better facilitate personalized medicine.
“A patient can come in and literally within minutes we could search their pathology reports and then we can start suggesting clinical trials they may be suitable for rather than relying on someone to go and look,” said Mitchell. “If we get the criteria for the clinical trial and we can extract this, then we are a huge step toward automating finding patients their clinical trials.”
Not only can the technology speed up the process of identifying clinical trials, it can also be used to further valuable research on treatment pathways. When a patient is diagnosed with a certain type of cancer, there is usually a standard treatment protocol for that cancer. Moffitt’s Clinical Pathways program has established treatment protocols and tracks patients’ progress with the goal of better preventing, detecting and treating cancer.
“This system will help enormously with tracking how patients do, compared to how on-protocol they were,” said Mitchell. “This will be huge for tracking progress over time.”
Machine Learning
In July, Moffitt became one of the first in the world to launch a dedicated machine learning department, which focuses on accelerating scientific discovery in cancer research and translating these powerful tools from the computer memory to the bedside.
“Moffitt is ahead of the game with this department and its prospect of making personalized medicine a reality,” said Issam El Naqa, PhD, chair of the Machine Learning Department. “I never thought there would be a machine learning department at a medical institution at this early stage.”
Machine learning can enable researchers to identify new complex patterns from data that can be used to diagnose cancer earlier, identify novel drug targets for treatment, predict which patients will respond to certain therapies and optimize personalized care plans. The department will also focus on automating tasks to reduce human error and better allocate scarce medical resources.
“As humans, we can only process about four to five variables at the highest level,” said El Naqa. “A computer can process thousands of variables and give clinicians guidance and help optimize decision-making.”
El Naqa is one of three researchers in the department and plans to add three more faculty members and three software engineers in the next five years.
“We are not only doing research, but also trying to get deployment into the clinic,” said El Naqa. “We are giving priority to work that can be directly translated, not just ‘pie in the sky’ ideas. We want ideas that are translatable to the clinic and that will have an impact on patients’ care.”
Digital Innovation
A core pillar of Moffitt’s strategic plan is digital care and discovery, and to execute that plan Moffitt needs to prioritize ideas and create a digital roadmap for the future.
Robinson was hired in 2019 as Moffitt’s Chief Digital Innovation Officer. He works hand in hand with Health and Data Services, IT and the new digital teams to advance technology in business operations, clinics, research, education and consumer areas.
“To truly ‘become’ digital and not just ‘do’ digital requires strategic planning, design, development and implementation focused on the end game of preventing and curing cancer,” said Robinson. “If you are going to truly transform how we do business, you have to wrap it around your mission and strategic goals.”
Earlier this year, Moffitt formed a committee of clinicians with data science backgrounds and a passion for digital innovation. The committee, chaired by Mitchell, will operate as a think tank to prioritize which ideas should move forward. Projects that have a direct benefit to patients will always move to the top.
“Moffitt’s uniqueness is that we have an eye for translational research,” said Rollison. “The research mission and clinical mission are so well aligned that we drive the science to be applied to the bedside and look to answer research questions that are very clinically relevant.”
The combination of advanced analytics and AI with digital innovation will drive Moffitt’s patient-centered care into the future. There are more than 500,000 unique patients in the system, and the machine learning and AI teams will continue to develop new algorithms to mine data and recognize patterns that impact care. Those projects will inspire digital innovation and build the framework to push Moffitt onto another level of the digital world.
“We are in this for the long haul,” said Robinson. “The goal is for Moffitt to be the most digitally enabled cancer center in the world.”
