The amount of data produced by our health system is enormous. As lead data scientist of Kaiser Permanente, Drew Clausen candidly explained on a AIMed North America 2018 panel. At 44 petabytes, they require the storage space of 1.4 million smart phones. The number of floppy disks (if anyone still uses them these days) storing the data can circle the World 65 times.
Apart from quantity, health data are also comprehensive and diverse. They consist of complete pictures of individuals’ wellbeing deriving from decades ago. Unfortunately, most of these data stay where they are and minimum attention is paid to them after the collection process is complete. There is a big gap between the availability of information and how to use them.
“We need to take data out of the cycle and start using them,” Clausen said. He believes the artificial intelligence (AI) in healthcare industry put a heavy demand for machines which can build complex models at the moment. Yet, it’s still unable to transform information we have, into something that the computers can understand. This adds on to the existing problem: not all available data is obtainable.
Workflow integration and decision support
The fact that AI models may be built using a myopic set of data does not concern doctors and clinicians as much as whether they fit neatly into their workflow. Trust forms a huge component in workflow integration. If doctors and clinicians do not believe that the AI model will ease their workload, they will not use it.
“Most physicians do not have formal training in computer science. Most of the time they learn on their job. So we need to start thinking about what are the barriers to implement AI in clinical setting” said Dr. Sharief Taraman, chief of general pediatric neurology & medical informaticist at Children’s Hospital of Orange County (CHOC) and vice president of medical at Cognoa.
The debate naturally gave rise to yet another problem. Since there is not enough open sources and data, the reliability of AI models are questionable and the present workforce may not be capable of integrating new technology into their work, how can AI provide the necessary decision support which aims at easing doctors’ workload?
The solution is to change
That’s when some panel speakers believe it’s time to change the present medical education. There is a need to connect technology with medicine. “Nothing is static, even for traditional medicine. So awareness is the first step, we need to collaborate clinicians and data scientists” said Dr. Taraman.
Others feel that AI will be automatically adopted when there is a need. “Physicians don’t even learn statistics until they have to use it. Similarly, physicians may not use AI until they have to. Unless they are incentivized in some way” said Dr. Randall Wetzel, director of VPICU (virtual paediatrics intensive care unit), Children’s Hospital Los Angeles.
Nevertheless, Clausen said forming trust in AI is similar to forming trust in any new medical finding documented by journals. They may or may not create consistent results at the beginning and doctors are forced to choose between changing their practices or wait. “Most importantly, we need to ensure the relationship between doctors and patients are not disturbed when we introduced data science” Clausen added.
A science writer with data background and an interest in current affair, culture and arts; a no-med from an (almost) all-med family. Follow on Twitter.