ICYMI: HIMSS NorCal Sacramento Meetup — “Machine Learning & Data Analytics to Solve Real World Issues”
What: “Machine Learning & Data Analytics to Solve Real World Issues”
When: 9/5/2017, 6–8pm
Where: UC Davis Health System Cancer Center
What is Machine Learning/AI?
Dr. Larry Ozeran, Physician and Health Technologist, moderated the panel discussion on Tuesday night and kicked it off by asking the panelists to define in simple terms what exactly Machine Learning is and how it applies to the healthcare field.
Prashant Natajaran, Stategy Leader/Innovation at Oracle, said the best definition is Arthur Samuel’s: “Machine learning gives computers the ability to learn without being explicitly programmed.”
He went on to say that, “Machine Learning and AI are not going to take your dog for a walk or slice your bread or take your job from you. However, there is a lot to be gained from existing data by gaining new insights through automated algorithms.”
Mitesh Rao, System and Patient Safety Officer at Stanford Health agreed and said that AI is about searching a large amount of hypotheses for one solution.
Alex Go, VP at Freed Associates added that learning algorithms are especially important in this day and age because they increasingly are “what can drive change.” The goal of AI is to “make health care simpler and easier for overburdened staff.”
Rao looks toward the future: “The larger vision of Machine Learning and AI is healthcare that is easier, more efficient, and more accurate. It should empower physicians…The current focus is clinical outcomes.”
The example he gave was patients that succumb to sepsis. Eventually AI should be able to predict this negative outcome when a patient first enters the hospital (or even before) and provide time for healthcare professionals to intervene before their health deteriorates to that point. It is about creating better care plans that reduce readmission rates.
Go added that AI will hopefully become more and more effective for determining a patient’s need for supplemental care such as home care, transportation, and operational changes (like staffing). Using AI to transform a patient’s health before they require a hospital stay can improve patient as well as physician satisfaction.
What real potential is there for replacing people with machines?
“Depends on the use case,” Go said. He gave the example that a large Med Center could spend an enormous amount of time cleaning data, putting it into a data warehouse, and transferring it into tableau for reporting. There is so much data that there are not enough people with the ability or bandwidth to review it all. That is where machine learning algorithms come in and do the busy work so that actionable data can be reviewed by healthcare professionals to drive clinical, operational, and revenue solutions in a timely and efficient manner.
Rao concurred, saying that Data is only as good as the people interpreting it. “The goal is to use data and AI to improve outcomes, efficiency, and accuracy”; rather than replace workers, it should supplement their knowledge and give them confidence when implementing solutions.
How has Machine Learning and AI evolved over the years?
The consensus of the panel was that Machine Learning has brought evidence-based healthcare solutions and care plans to the forefront of the medical field. Hospitals are under enormous amounts of pressure to increase efficiency and to cut length of stay and readmission rates. The focus of AI is data driven decision making, and to get ahead of the curve for better patient outcomes, Rao said.
But to make AI work, “The user base must be firm from the get-go” and AI firms must gain the confidence and trust of physicians as well as patients, Natarajan asserted.
The impact of “Dirty Data” on advancing AI
Go added to the mix the issue of patient generated data and “dirty data”; that is, data that is rapidly generated and not necessarily accurate. He gave an example of a patient letting their dog try on their wearable for fun, or even the questionable accuracy rate of wearable technologies like fitbit’s “heart rate” function. He said it causes concern for maintaining data confidence.
Dr. Ozeran added that professional in home medical devices may be more accurate but are often too expensive for average consumers and patients may not understand their importance.
Still, the general panel consensus was that the future is bright. “We live in a data-rich and information poor world,” Rao said, “These are efforts to make data work for us.”
- Go: Health Systems should prioritize appointing data stewards and focus on Data governance to prep for future transitions to AI workflows.
- Rao: Any AI solution needs to be evaluated for clear and actionable implementation plans before adopting the new tech
- Natarajan: It is imperative to ask for ROI when considering new technology solutions; Data is only useful if it is actionable. We must continue to share information and have conversations like this to constantly be reevaluating and improving.