AI Adoption in Healthcare: A Balancing Act

A handful of headlines have emerged recently casting doubt and displaying concerns regarding the implementation of artificial intelligence (AI) in healthcare. Most notably, the IBM Watson for Oncology platform was targeted as recommending “unsafe and incorrect” cancer treatment suggestions for its hospital system and physician users by a STAT News report based on internal IBM reports. IBM Watson for Oncology is a cognitive computing platform that utilizes AI algorithms to interpret clinical information and provide treatment recommendations for multiple cancer types. According to STAT News, internal presentations from IBM indicated that Watson for Oncology was generating recommendations that did not adhere to national treatment guidelines and participating physicians at member institutions were not finding the tool useful in most cases.

These findings are hardly a ringing endorsement for this exponential technology. In its defense, IBM countered that they incorporated a great deal of continuous feedback from multiple sources in a persistent effort to improve their offering. The company cited eleven separate software releases in the past year aimed at improving functionality and incorporating new guidelines for cancer treatment. Watson for Oncology training partner Memorial Sloan Kettering Cancer Center also responded, expressing that the cited recommendations were given as part of system testing and not during actual use on patients.

Regardless of the specifics in this case, the headlines understandably stoke concern for the adoption of AI in healthcare. At Singularity University, we believe it is critical to understand emerging technologies and their future impact. This awareness should span from the wonder and amazement many of these advances offer, to the growing pains and negative implications, and everything in between. A well-rounded view is key to understanding how these forces will impact your industry. There is invariably a testing period that must occur in the progression of exponential technologies. They must endure a trough of disappointment, evidenced here, as they move on the exponential curve and cross to true acceleration and eventually exponential advancement. On that note, let’s look a little deeper at these considerations.

A Closer Look

Much can be made of the concerns raised regarding Watson for Oncology. First, it is important to note that the adoption of AI in healthcare will inevitably have its breakdowns which will result in negative clinical outcomes—the same as any traditional diagnostic or treatment implementation. This is a difficult pill to swallow but a reality of the situation. Next, the efficacy of an AI platform is strongly dependent on the input provided and access to a rich and diverse dataset. Some of the potential concern in the Watson case is that only a limited amount of clinical information and input from a handful of specialists was utilized to train in treatment algorithms. This is primarily human error that can be corrected with a more diligent arrangement and increasing access to rich healthcare datasets.

Furthermore, it is important to consider the context in which this technology should be implemented. The best application to capitalize on the power of AI is in a complementary role with human intelligence. The object in the short-term is not to rely on AI entirely for oncology or other treatment suggestions, but to provide physicians with another beneficial stream of information to enhance their treatment decisions. AI can help augment current decision-making by absorbing and integrating massive amounts of data the world over and can do so with unrelenting attention and a continuous workflow. This is a great asset in the clinical realm.

This also presents considerable opportunity for the enterprise healthcare company that can navigate the developing landscape. A good place to focus attention is on the frontline deployment of these applications. Physicians and other care providers are understandably apprehensive when they see the potential for questionable clinical implications and uncertain impact on their current format of care delivery. The last significant technology push was the widespread implementation of the electronic health record which led to disdain amongst this group as computer screen and documentation time surpassed face-to-face patient interaction. Uncovering the solutions that reverse this trend and improve system-wide efficiency will offer great potential. Key metrics for evaluation in the clinical realm include speed and specificity. Solutions that help strengthen the conviction behind a diagnosis or treatment plan and the time it takes to arrive at both are indispensable.

A look at current successful AI use cases in healthcare typically centers around more specific and limited tasks such as image classification or prediction of algorithmic diagnosis that are based on well defined and widely captured datasets. The transition to more sophisticated abilities will improve as advancements in healthcare data collection and interoperability improve, as well as AI methods such as speech interpretation and sensor integration. In that regard, we highlighted a company fulfilling this powerful complementary role and demonstrating the great impact this technology can have.

Spotlight on Viz.ai

Doctor or radiologist in hospital looking ct scan image of the brain on 2D and 3D rendering background.

“Time is brain.” This adage in the care of stroke patients expresses the simple truth that with each passing minute following a stroke, additional nervous tissue is lost and the opportunity for time-dependent medical intervention decreases. In fact, a previous quantitative study estimated that the typical patient loses 1.9 million neurons each minute that an ischemic stroke is left untreated. Viz.ai is using applied artificial intelligence to help combat this problem.

Their Viz LVO product utilizes AI and deep learning algorithms to automatically analyze computed tomography angiography (CTA) images and identify suspected large vessel occlusion strokes. Once identified, the necessary specialists for intervention are alerted via a messaging notification. This system serves to help triage stroke patients more expeditiously and help mobilize the care teams to take action.

Viz.ai received FDA clearance earlier this year. Most impressively, a study comparing Viz LVO to the standard of care demonstrated that the time from CT scan to specialist notification decreased from a median of 66 minutes to a median of 6 minutes with the Viz LVO platform. This complementary platform means that stroke care teams can augment current practice with a new powerful tool to improve clinical care and produce outcomes not realized without this technology.

Stroke is a devastating and prevalent medical condition. A stroke (or cerebrovascular accident) is a cerebrovascular event with focal or global disturbances of cerebral function secondary to decreased cerebral perfusion of a vascular origin. Stroke is a leading cause of death and disability worldwide. Stroke survivors are frequently faced with permanent impairments, including cognitive dysfunction, speech, and language deficits, and diminished motor function. The healthcare costs and economic toll of stroke are also tremendous. A solution like Viz.ai can make a demonstrable clinical impact on such a catastrophic condition.

Additionally, the implementation of this and like-minded offerings will have far-reaching consequences across the healthcare industry. Learn how you can future-proof your company to tap into these opportunities (and avoid disruption).

– Dr. Kev

Sources include STAT News, Advisory Board, mskcc.org, Stroke, and Viz.ai.