The once futuristic potential for artificial intelligence in healthcare is coming to fruition, integrating with key operational and clinical aspects of healthcare to improve patient care. But how far will it go?
"The field of applying AI on healthcare data is developing rapidly, mostly as a follow-up to the explosion of new and highly efficient AI methods used in other fields such as image and voice recognition and natural language processing," said Theodoros Zanos, PhD, head of the Neural and Data Science Lab and assistant professor at the Feinstein Institute for Medical Research at New Hyde Park, N.Y.-based Northwell Health. "The use of cutting-edge machine learning methods on data of multiple modalities — for example lab, vital, demographic and imaging data — related to a variety of different diseases will no doubt heavily affect the way physicians diagnose and treat their patients."
Diagnostics is one of the ways AI is being tested in clinical care, and Dr. Zanos sees it enabling more accurate diagnoses, potentially earlier in the disease cycle. AI could also simplify some procedures and give physicians more time to focus on patient interactions and possibly reduce unnecessary exams and costs.
"Using AI to find data patterns that are hard for humans to detect can also provide new insights and a deeper understanding of certain diseases and their mechanism of progression," said Dr. Zanos. "Beyond diagnosis, it will enable evidence-based optimization and personalization of treatments."
Studies show it can be as good or better than physicians in small sample sizes. In the January 2020 issue of Nature, a study compared Google Health and DeepMind's technology to physicians when diagnosing breast cancer in women. The study included six U.S. radiologists as well as U.K.-based physicians.
DeepMind and Google Health's collaboration outperformed the six U.S. physicians, reducing the number of false positives among 500 mammograms by 5.7 percent. It reported similar results to the U.K. physicians. To develop the AI system, Google used 29,000 X-ray images to show the system how to detect abnormalities.
"The impacts are already improving current provider workflows resulting in improved detection. Google's early results are promising," said Sanket Shah, a clinical assistant professor of biomedical and health information sciences at the University of Illinois at Chicago. "They continue to enhance their AI models as more data becomes available, which is increasing accuracy and providing fewer false negatives and positives compared to human experts."
Potential benefits
While AI is still in its early stages, many are hopeful about the potential for this technology. CIO of Renton, Wash.-based Providence B.J. Moore thinks AI will continue growing as a key enabler for diagnostics in the future.
"Overall it will be good for medicine, as it will provide a level of precision and augmentation to doctors that can only be provided by advanced computer cognitive capabilities," he said. "In the early stages of AI, it will be an augmenting capability to doctors. As the technology improves and matures, it will be able to make a diagnosis as good or better than clinicians across a broad spectrum of clinical scenarios. I believe this will be empowering to doctors, freeing them up for more complex scenarios that only a human clinician can solve, while improving the accuracy of diagnosis for the more common diagnostic scenarios."
As EHRs and other data gathering systems have integrated with health systems, organizations have much more data at their fingertips. Clinicians no longer need to spend time looking through deep paper records or hand-written doctor notes from their colleagues to treat patients; however, with the increased efficiency and volume of data, it can be challenging to really understand and synthesize the information during the treatment decision-making process.
"The ability to leverage AI tools to cull through and synthesize important information, presenting it up to the physician and patient to help make the best personalized decision for them will help everyone translate volumes of data into rational action plans," said Amy Compton-Phillips, chief clinical officer of Providence St. Joseph Health. "Leveraging AI to enhance the skills and productivity of physicians at a time where we otherwise could feel that we are drowning in data will make life better for all."
In September, Rochester, Minn.-based Mayo Clinic and Google announced a 10-year strategic partnership to develop AI-based diagnostics and carry out medical research. Google plans to open a new office near Mayo's campus for better cross collaborations between the system's clinicians and IT staff and Google's data scientists.
"Both of our organizations have published research in the application of machine learning and AI to radiology, cardiology and many other areas of medicine. We will collaborate to accelerate that work," Mayo CIO Cris Ross told Becker's in a December 2019 interview.
Will AI replace physicians?
Not all clinicians see AI as it stands today a boon. James Tinsely, MD, of Newport News, Va.-based Lighthouse Direct Primary Care, questions whether to trust the technology with patient care. "If an AI can't drive a car without crashing would you trust an AI to diagnose your parents or children?" he said.
There are also concerns about whether technology would replace physicians or clinicians in the future. Dr. Zanos doesn't give credence to this notion.
"I don't believe that physicians and nurses can be replaced by algorithms any time soon but they can be tremendously assisted by these methods so they can focus more on the irreplaceable aspects of their work, like human interaction," he said. "For these algorithms to properly work, they need to be an integral part of the process of training, evaluating and using them, so they have a huge role to play in these advances. Finally, healthcare providers, with their unique experience, can guide the engineers and computer scientists that create these algorithms to solve problems that actually exist and could truly benefit from an AI-based solution. This is why it's very important for physicians and nurses to embrace this progress and become part of it so the field can advance in faster and proper ways."
Oscar Marroquin, MD, chief clinical analytics officer at UPMC, agreed with that sentiment. "[AI replacing doctors] is not a reality right now," he said. "Most agree that AI in diagnostics affords clinicians the opportunity to be more efficient in how they do their work. There are tasks that machines can do well and reliably well, and we are going to take advantage of those things to increase workflow while at the same time improving accuracy and efficiency of how we deliver care. I see what AI will bring on the diagnostic side as an enabler to what clinicians are doing, not necessarily as a replacement."
What the future holds
The future holds huge potential for AI, as companies are testing algorithms to detect conditions such as pneumonia, flagging those patients for physicians to take a more nuanced look. Mark Hoffman, PhD, chief research information officer at Children's Research Institute at Children’s Mercy Kansas City, anticipates the next step will be focused on methods to evaluate the accuracy of AI, as well as its value, in diagnostics. And as with any changes in healthcare, regulatory agencies will slow the process.
"It will take some time for regulators to determine their comfort level with these capabilities," he said. "This will be followed by work to develop AI that integrates multiple types of data sources to seek higher quality diagnostics. At the Children's Research Institute at Children's Mercy, we are doing foundational work to characterize large de-identified clinical data sets in order to ensure that the data feeding these algorithms is high quality and is well-understood by the developers of algorithms."
Mr. Shah also sees a need for a de-identified repository holding various images from disparate sources for researchers to analyze on a broader scale. "There also is going to be a focus on improving these AI models by integrating other data points such as medical history and genomic make-up," he said. "A lot more fine-tuning and tinkering. However, the arrow is certainly pointing in the right direction."