Intel explores how AI is changing healthcare
April 17, 2018
Healthcare in the USA is in crisis, according to Intel executive vice president Navin Shenoy (pictured), speaking at last month’s Solve Healthcare conference at the University of California, San Francisco. But artificial intelligence (AI) is the most promising remedy.
He said that healthcare in the USA made up a shockingly large 18 per cent of US GDP. The $3.3tn US healthcare system is larger than the national education and defence budgets combined.
Throughout the discussion, one thing became clear: artificial intelligence is the most promising remedy for an industry in distress. Jonathan Cohen of the Princeton Neuroscience Institute summed it up when he said: “We’re now in an AI spring, because we are doing things that neuroscientists knew 20 to 30 years ago but simply couldn’t bring to scale – and now we can.”
Cohen was one of 14 partners who joined Intel at the healthcare showcase to talk about the ways AI was changing healthcare for the better.
“I noted three ways this is proving true,” said Jennifer Esposito, general manager of Intel’s health and life sciences organisation. “AI shifting care out of the hospital, AI enhancing the work of providers, and AI providing new kinds of medicine.”
Parsa Mirhaji, chief technology officer at Montefiore Medical Center, said: “As an accountable care organisation in the Bronx, we want patients out of the hospital and to keep them out.”
Mirhaji and the team at Montefiore care for one of the most ethnically and socioeconomically diverse – and underserved – populations in the USA. It’s critical for them to deliver value, not by providing more services, but by rethinking how they care for that population.
Today, Montefiore clinicians use an AI system called Palm (patient-centred analytics and learning machine) to predict which patients in the ICU are most at risk for respiratory failure. Using this practice at scale can predict which patients are at risk for respiratory failure 24 to 48 hours in advance and can do that with high levels of sensitivity and specificity.
Montefiore caregivers hope to extend this prediction model to patients outside the hospital using a variety of data sources – including genomics and socioeconomic data – to determine who will develop chronic conditions as early as two years before they occur.
Xavier Urtubey, AccuHealth CEO, explained that most of the costs for chronic disease came from complications. But thanks to the predictive algorithms that analyse data captured from sensors in the patient’s home, or simply by using behavioural data entered in by the patient or their family, they’re able to foresee complications coming two to four days before they happen. They can now reach out to the patient hopefully to avoid the complication. By reducing emergency room visits for patients with chronic conditions such as diabetes by over 30 per cent, AccuHealth is able to achieve an outcome-based payment model for payers.
“We’re not going to replace physicians, but we can make lives better, and there is real opportunity to apply this,” said Rachael Callcut, a practicing trauma surgeon at the University of California.
Callcut said that helping bedside clinicians sift through alerts and other data in time-sensitive conditions or just fixing the efficiency issues that can reduce burnout would go a long way.
“The easiest portions of what each field does across medicine are the things that AI will help to solve,” she said. “It will allow us to focus attention on things that machine and computer learning won’t be able to do in terms of clinical context of a situation or intuition and clinical decision-making that goes into what transforms an OK outcome into a great outcome.”
In another example, Shreyas Vasanawala, Stanford Medical radiologist and founder of AI start-up Arterys, said he didn’t believe AI would replace radiologists. He talked about the field of radiology being a lot broader than just making a diagnosis and mentioned many areas where AI could make a radiologist’s life easier. These included using AI algorithms to predict how long scanning sessions would take, optimising scheduling to get more throughput, predicting which patients would show up late to schedule better and get more utility out of the scanner, and using AI to auto-determine which images needed to be reacquired.
“We estimated that a third or half of the healthcare costs that Navin talked about around the burden of health and disease to society costs is associated with mental health,” said Princeton’s Jonathan Cohen. “And 50 per cent of therapeutic interventions for depression are behavioural.”
Behavioural therapies can be more effective than drug therapies. The problem is that things like negative thoughts aren’t something doctors can easily observe. Over the past ten to 15 years, developments of brain imaging tools now allow for tracking internal brain states. In collaboration with Intel, neuroscientists at Princeton have taken two years of compute time down to real time.
“Something that began as collaborative research has become a full-on scientific endeavour,” said Cohen.