by Julia Pierrepont III
LOS ANGELES, Dec. 14 (Xinhua) -- While some see artificial intelligence (AI), user data mining analytics and machine learning as industry-redefining and highly desirable, U.S. business insiders in the healthcare field said they are way behind in finding ways to make use of it at all.
"Big data is hugely important in most fields, but that is largely not the case in healthcare," Dr. Juan Espinoza, a pediatrician and co-director of the West Coast Consortium for Technology and Innovation in Pediatrics at Children's Hospital Los Angeles, told Xinhua at an annual data science conference last week.
Xinhua caught up with several medical practitioners and healthcare experts during and after the Innovative Methods with Big Data and Artificial Intelligence (IM Data) event at Pasadena, California, to find out what has led to the sluggish adoption and application of big data analytics in the healthcare industry.
The biggest challenge they think confronting data analytics' application in healthcare is that the U.S. government's Health Insurance Portability and Accountability Act (HIPAA) strictly limits access to and sharing of patient records.
"As healthcare providers, as payers, we don't own the data. The patient does," Espinoza said.
Guarding medical records is essential to protecting the privacy of a patient, but the compartmentalization of data can inhibit the gathering of valuable health information critical to improving clinical outcomes and public health initiatives, he added.
"First and foremost, we protect patient data," Espinoza said. "But secondly, there is another stratum where we need to ask, what would be the benefit to society and for other patients if we shared the data? How do we create systems that allow for those emergent benefits of data while still protecting patient privacy?"
Like financial data, medical patient data is amongst the most closely-guarded data in the world. However, while financial data has established secure protocols that enable it to be shared from bank to bank, company to company, even country to country, medical data has not done so.
"Because of HIPAA laws, health data can be very siloed and this makes it extremely hard to track a patient across different institutions and doctors over time as they are growing up," said Noel Gomez, chief executive officer and co-founder of Convexa, a health informatics start-up that uses analytics and machine learning insights to deliver prescriptive and predictive insights to users.
Gomez said that in other countries where there is a national health service, like New Zealand or China, data is more centralized, easily accessed and analyzed, unlike in the United States where a patient's data is scattered all over the place with different providers and insurance companies.
Natasha Lepore, a Ph.D. in physics and principal investigator of Brain Organization Research Group Lab at Children's Hospital here, told Xinhua that "Siloing is a problem in the healthcare space due to privacy rules -- for good reason. So, in medical imaging we have very limited data sets. That's a big obstacle to using AI effectively."
The experts agreed that this obstacle can lead to significant room for error or inefficiency in diagnosis, developing comprehensive treatment protocols, patient compliance issues, avoiding drug and treatment conflicts and side-effects, and other important issues that can have a powerful impact on patient care and outcomes.
"We are at a point in molecular modeling and pharmaceutics that make use of enormous amounts of information from the Human Genome project," said David Dyer, executive director of the Master of Science in Biotechnology program at Azusa Pacific University.
"But what we really need right now is longitudinal, phenotypic data associated with disease states," he added.
Anonymized medical data, or medical data that has been stripped of a patient's name, address, ID, or other personal identifiers, is being touted as an appropriate data management compromise that many health and medical practitioners believe could enable them to share their data safely to bridge the divide between protecting patients and protecting society.
There are still other significant obstacles to the general acceptance of analytics in healthcare.
Aaron Lai, senior manager of Technology Consulting of Payer Analytics at the PwC Advisory Services, told Xinhua that even asking a group of doctors what it means when a patient says "I don't feel well" could result in a huge difference in opinions.
"Compounding that huge variance, when we use Natural Language Processing (NLP), those kind of nuanced human interpretations cannot be picked up by the machine," said Lai.
Dyer thought it's important to manage expectations about what big data can and cannot do yet.
"We all want self-driving cars to work perfectly and never crash. But these systems are like children -- they are still learning as they go. And for at least a decade, they will need humans to babysit them," said Dyer.
Dyer also highlighted the importance of using "clean" data, saying "we need to create data sets that are adequately prepped so you get a meaningful result."
He believed there are glimmers of hope when people and institutions are beginning to see the positive results of cooperation. "It's not always about the profit model for this. It's about really cooperating," he said.
Cross-institutional cooperation would allow regional and national health initiatives to use sophisticated machine learning in vast data pools to pinpoint public health issues early on before they become a crisis, identify bottlenecks in healthcare delivery, enhance treatment protocols, reduce length-of-stay in hospitals, etc.
Espinoza isn't against old school practices. "From a clinical perspective, one of the single most important innovations that has improved patient outcomes and saved lives is checklists."
Standardizing treatment processes and protocols and making sure everything gets done the same way every time are the bedrock on which good medicine must be grounded, he said.
"You are trying to apply this shiny new tool to a system that is still just trying to make sure that simple checklists are followed properly," Espinoza said.
Surprisingly, the last bastion holding back the adoption of large-scale analytics in healthcare is doctors.
"You have to remember, in healthcare, the stakes are really high and we are still having to deal with lots of missing or confusing data," said Omolola Ogunyemi, director of the Center for Biomedical Informatics at the Charles Drew University of Medicine and Science. "Data problems don't mean life or death in other sectors."
She said that while doctors can wrap their heads around things like decision trees, the more opaque software approaches, like computational neural networks, can look like an incomprehensible black box to them. "It's harder for them to accept."
"A doctor is never going to say to their patient, 'My computer tells me you have cancer,'" said Lai.
