Beyond the Hype: AI and Healthcare
Tamara StClaire, COO, BaseHealth
Artificial intelligence (AI) has tremendous potential to improve both care and cost efficiency in healthcare, but it’s incredibly complex and continues to innovate at a rapid pace, making it hard to separate its real-world value today from its aspirational value of tomorrow. This is where most AI technology falls victim to the hype.
In its simplest form, AI is the next stage in automation. But in a crowded marketplace, it can be hard to determine which solution is right for your organization and needs, making it critically important to understand the algorithms and decision trees upon which its recommendations are based.
Most AI technologies today use an associative model, which draws correlations between data points to predict possible outcomes. While it has shown success, the model is purely associative, which means correlations don’t necessarily indicate causation. This means we can’t know the “why” or the “how” of what the model is telling us. For example, an associative model may identify a cluster of patients with a certain type of cancer that all live in the same zip code. That information is interesting and somewhat useful, but it isn’t immediately actionable because you may not know the underlying cause.
At BaseHealth, we start with a causal model, which is overlaid with an associative model. Causality is the process that connects one thing to another, where the first is partly responsible for the second, and the second is partly dependent on the first. It’s essentially establishing cause and effect. What we’ve done with our causal model is examined these types of linkages — in some 150 million published, peer-reviewed research studies — to map the cause and effect pathways for over forty (40) of the most common diseases and health conditions globally, that essentially make up 95 percent of US healthcare costs. The linkages we are modeling for these diseases and conditions are the thousands of risk factors that have been identified by evidence-based scientific literature.
Our review of those 150 million peer-reviewed published medical studies, containing data on 70 million people’s data , is our way of tapping into the knowledge of researchers and clinicians to tell us all the disease paths they know. The end result is our causal model, which understands all those known pathways in aggregate, including the complete set of risk factors that influence disease progression, and how those risk factors interact over the course of disease progression.
An easier way to think of it might be to imagine our causality model as the Google Maps for disease progression. As a navigation tool, Google Maps knows all the possible routes to take you from point A to point B. There are a set number of routes — by freeway, surface streets, and more — between these two points. Just like there are a set number of routes to get from point A to point B, there are a known set of paths that take a person from healthy to the point of a heart attack based on previously reported studies. A person who is the picture of health will not have a heart attack tomorrow. There are a finite number of paths that lead an individual from health to the point of a heart attack based on what we know today, just like there is a set number of routes I can take to the office in the morning when I walk out my door.
What’s more, just like Google knows to re-route me based on external factors like construction or traffic, when we examine an individual with our model, it knows how an individual’s particular health status (BMI, physical activity, diet etc.) influences their path and their current location on the road map. Understanding these roads, or disease pathways, is what gives us the power to understand not only what’s most likely to happen next, but also how and why it will happen, and, perhaps most importantly, what we can do to change course effectively.
Beyond improving care, our solution can also provide a nearly immediate ROI by providing a cost/benefit analysis for each potential clinical intervention. This enables health systems and ACOs to dedicate their outreach efforts and intervention programs toward a very specific set of patients where the return on investment and patient outcomes will be the greatest. Instead of offering a pricey, underutilized, one-size-fits-all wellness program to an entire population, they can intervene and help the specific group of patients within the population who need it most.
Our comprehensive causal model took us years to build, and is unlike anything else available today. Combining a causal model with an associative model is a new approach that delivers actionable insights that are financially sound for health systems and ACOs across the country.
