The following is a guest blog post by Inga Shugalo, Healthcare Industry Analyst at Itransition.
In contrast to legacy systems that are just algorithms performing strict tasks, artificial intelligence can extend the task itself, creating new insights from the information fed to it. Current healthcare AI is powerful enough to undertake such complex challenges as automated diagnosis, medical image analysis, virtual patient assistance, and risk analysis, supporting health specialists in making more swift and informed decisions.
In 2016, Frost & Sullivan predicted the healthcare AI market to reach $6.6 billion by 2021. Meanwhile, 2017’s Accenture report estimates AI saving $150 billion annually for the U.S. healthcare economy by 2026. “At hyper-speed, AI is re-wiring our modern conception of healthcare delivery,” researchers from Accenture say.
Standing in the middle of 2018, the industry already hints on its course regarding further AI expansion. Spoiler alert: as well as with blockchain AR, VR, and any other kind of innovative custom medical software, the adoption challenges persist.
Current and prospective AI directions in healthcare
One of the most fascinating and valuable directions for AI to evolve is its ability to help providers diagnose patients more accurately and at a higher pace. We are thrilled to see how 2018 erupts with many healthcare organizations adopting artificial intelligence and creating unprecedented cases of assisted diagnostics with it.
Geisinger specialists applied AI to analyze CT scans of patients’ heads and detect intracranial hemorrhage early. Intracranial hemorrhage is a life-threatening form of internal bleeding, affecting about 50,000 patients per year, with 47% dying within 30 days.
Geisinger was able to automatically pinpoint and prioritize the cases of intracranial hemorrhage, focusing the attention of radiologists on them and thus allowing for timely interventions. This approach reduced the time to diagnosis by 96%.
Mayo Clinic currently uses IBM Watson’s superpowers to match patients with fitting clinical trials. The clinic’s officials stated that only 5% of patients enrolled in trials in the U.S., which significantly hinders clinical research and innovation in cancer therapies. On the other side, manual patient-trial matching is a time-exhausting process.
Watson runs this process on the background, comparing the patients’ conditions with available trials and suggesting the appropriate trials for providers and patients to consider including in a treatment plan. Since its implementation in 2016, Watson was able to deliver about an 80% increase in enrollment to Mayo’s trials for breast cancer.
Patient risk analysis
“…Healthcare is one of the most important fields AI is going to transform,” Google CEO Sundar Pichai noted during the Google I/O 2018 keynote. Last year, the event presented Google AI, a “collection of our teams and efforts to bring the benefits of AI to everyone.”
In 2018, Google uses their AI to tap into critical patient risks, such as mortality, readmission, and prolonged LOS. Cooperating with UC San Francisco, The University of Chicago Medicine, and Stanford Medicine, they analyzed over 46 billion anonymized retrospective EHR data points collected from over 216 thousand adult patients hospitalized for at least 24 hours at two US academic medical centers.
The deep learning model built by researchers reviewed each patient’s chart as a timeline, from its creation to the point of hospitalization. This data allowed clinicians to make various predictions on patient health outcomes, including prolonged length of stay, 30-day unplanned readmission, upcoming in-hospital mortality, and even a patient’s final discharge diagnosis. Remarkably, the model achieved an accuracy level that significantly outperformed traditional predictive models.
According to Pichai, “If you go and analyze over 100,000 data points per patient, more than any single doctor could analyze, we can actually quantitatively predict the chance of readmission 24 to 48 hours earlier than traditional methods. It gives doctors time to act.”
Of course, researchers don’t claim that their approach is ready for implementation in clinical settings, but they are looking forward to collaborating with providers to test this model further. Hopefully, we will see field trials and, who knows, even early adoption in 2019.
EHRs “on steroids”
HIMSS18 was all about artificial intelligence and machine learning. Surprisingly, all major EHR vendors – Allscripts, Cerner, athenahealth, Epic, and eClinicalWorks – came up with a promise to include AI into upcoming iterations of their platforms.
At the event, Epic announced a new partnership with Nuance to integrate their AI-powered conversational virtual assistant into the Epic EHR workflow. Particularly, the assistant will enable health specialists to access patient information and lab results, record patient vitals as well as check schedules and manage patient appointments using voice.
Similarly, eClinicalWorks puts AI into work on voice control but also prioritizes telemedicine, pop health, and clinical decision support. According to the company’s CEO Girish Navani, “We spent the last decade putting data in EHRs. The next decade is about intelligence and creating inferences that improve care outcomes. We can have the computer do things for the clinician to make them aware of actions they can take.” The new EHR’s launch is expected in late 2018 or early 2019.
Athenahealth also added a virtual assistant into their EHRs to improve mobile connectivity and welcomes NoteSwift’s AI-based Samantha technology to enhance clinical workflows by introducing robust automation. Samantha can grasp free-text and natural language, process information, structure it, assign ICD-10, SNOMED or CPT codes, prepare e-prescriptions and orders.
Pre-existing challenges for healthcare AI adoption
Gartner predicted that 50% of organizations will miss AI and data literacy skills to gain business value by 2020. Certainly, a lot of healthcare organizations will get in this 50%, and there are two reasons for that.
Regulations and security concerns are the main pre-existing challenges that delay practically any technology adoption in healthcare and entail an array of new challenges along with them.
First, an AI application or device has to be approved by the FDA. The catch is that the existing process focuses on the hardware or the way that algorithms work, but not the data it should or would interact with.
Speaking of data, another challenge is security breaches. Safeguarding sensitive information is a must for healthcare because patient data is a constant target for identity theft and reimbursement fraud. In Accenture’s new report, nearly 25% of healthcare execs admitted experiencing “adversarial AI behaviors, like falsified location data or bot fraud.” While this doesn’t mean AI threatens patient data, such claims do increase the concerns related to its adoption.
Still, artificial intelligence is growing in healthcare and will continue to do so. Maybe not at rocket speed, but the most recent cases show consistent improvements in major care delivery gaps. Healthcare AI’s future appears bright.
About Inga Shugalo
Inga Shugalo is a Healthcare Industry Analyst at Itransition. She focuses on Healthcare IT, highlighting the industry challenges and technology solutions that tackle them. Inga’s articles explore diagnostic potential of healthcare IoT, opportunities of precision medicine, robotics and VR in healthcare and more.