Publication
Article
Author(s):
Improvements and greater familiarity with artificial intelligence will lead to systems of checks and balances that will create trust and fulfill the promise of computer-assisted precision medicine.
Sarah Alwardt, PhD
Healthcare news is filled with stories about technological advances in artificial intelligence (AI) and machine learning. Some of the biggest and most respected technology companies are backing healthcare data initiatives with a heightened focus on oncology.
Because the purpose of AI is to review and assess enormous quantities of information, in oncology, AI’s power and potential lie in its ability to quickly cull relevant data from clinical studies on traditional and emerging therapies. However, machine learning on its own falls short due to the significant complexities of oncology that still require human intellect for successful outcomes. Knowing a patient’s diagnosis, prior treatments, comorbidities, and, importantly, desires (ie, adverse effects vs quality-of-life considerations) is critically important. As these data are often unique to the individual and not recorded in datasets, AI cannot factor them all into its computations. The collection, analysis, and personalization associated with treatment decisions continue to require human intervention by physicians and clinically trained staff.
IBM’s Watson for Oncology has made a significant investment in AI, and although it has been introduced in various healthcare settings, “no published research shows Watson improving patient outcomes,” concluded a recent Wall Street Journal article by Daniela Hernandez and Ted Greenwald.1 They said that the computerized oncology tool often comes up with recommendations for treatment that don’t augment what doctors already know and, further, the complexities and rapid advances in oncology frustrate efforts to keep the AI tool up-to-date. Although Watson has been abandoned by some, other oncology centers contend it is positively influencing a portion of therapeutic decisions at their institutions, and there remains a coterie of scientists and physicians who maintain that Watson for Oncology will improve.1-3
Beyond performing research tasks in the clinical setting, do cognitive tools such as Watson have the potential to deliver personalized treatment recommendations that encompass knowledge of an individual patient’s needs and wishes and the doctor’s assessments? AI is limited in that it is not a replacement for physician judgment and shared decision making. A patient’s previous diagnoses, prior treatments, comorbid conditions, desires, and treatment goals must be weighed along with clinical trial data and real-world experience. The high hurdle is that this information must be incorporated into the delivery of personalized care. The knowledge and care of the physician must be combined with the wants and needs of the patient, along with all that the world knows about a particular cancer type.
However, every decision we make is based on a series of assumptions, and those assumptions must be supported by concrete data. New treatments and clinical data for oncology are accumulating at an ever-faster pace. More than 2000 oncology clinical trials are underway or in the planning stages for new molecular entities and combination therapies.
Most organizations are not equipped to handle this tsunami of data, and technology has the potential to solve a large part of this challenge. AI can distill huge amounts of data into succinct findings. However, many of our healthcare systems don't have the technology they need to share information, something that AI algorithms rely on for fine-tuned treatment guidance.
In my opinion, there will always be 1 extra assumption and 1 additional data point to consider. In addition, there is the prediction that human beings will behave unpredictably. These difficulties create the need for improved synergy and expertgroup review. Critical to this collaboration are scientists who can combine clinical experience with data expertise. Because clinical decisions will be challenged by assumptions and bias, scientists can ensure that we are interpreting data with the highest level of confidence. As new data are generated, scientists, analysts, and clinicians need to sit down and go over it together. AI, such as Watson, is one way to bring these disparate minds and datasets together.
Another example of a technology that helps oncologists synthesize information for clinical decision making is McKesson’s Clear Value Plus, which integrates National Comprehensive Cancer Network clinical practice guidelines for 20 disease states a with multiple electronic health records, including iKnowMed, iKnowMed Generation 2, Epic, and Varian. Since its inception, the system has aided more than 375,000 clinical decisions from US community oncologists.
To date, Watson for Oncology has been employed in many locations outside the United States, including Thailand, India, and China, which have different healthcare systems—each with its own demographics, costs, and access to care. In regions where advanced training may be rare or unobtainable, Watson can provide valuable support for clinical decisions that involve uncommon cancers or cases that require complex decision making.
As we continue to advance personalized medicine, we find that the ability to capture robust, structured, detailed data—including biomarkers, dosing, duration, histology, and claims information (ie, Internation Classification of Diseases, 10th Revision codes)—in electronic systems is very powerful. This is the natural synergy of technology taking its rightful place in the clinic. Having this structured data available to us is clearly supporting breakthroughs in many types of cancers.
Furthermore, AI will be able to manage these growing volumes of data and find patterns and outcomes that we didn’t know to look for. Improvements and greater familiarity with AI will lead to systems of checks and balances that will create trust and fulfill the promise of computer-assisted precision medicine. However, for each advance in technology we will still need experienced physicians to question the algorithms, analyses, and underlying assumptions. Physician collaboration with data scientists will build the understanding and fine-tune the research.
Sarah Alwardt, PhD, is vice president of data evidence and insight operations for McKesson Life Sciences.