Imagine: You are in your car, pull up Google Maps, punch in your directions, and get on your way. Halfway through the drive, the app begins to recalibrate, indicating heavy traffic and suggesting an alternate route to get you to your destination faster.
What if a similar platform could give you just as smart of a roadmap to better health? What if you were told you’re at risk for developing chronic back pain down the line, and could then get prescribed a few exercises to prevent the condition from progressing?
Equipping people with the right information, education, and incentives to achieve better health is beneficial in a number of ways, including overall population health and reducing overall healthcare spend. This is what the Data Science team at Evive is actively working to achieve.
Learning through development
Originally when we were developing and training our machine-learning algorithms, our team had to clean, standardize, filter, and analyze millions of data points based on biometric data, survey results, and insurance and medical claims. However, this manual process was laborious and took a lot of time, meaning we were only able to tackle 8-10 procedures in a year.
We knew there had to be a better way to do this—and there was. That’s how we developed our AFPS algorithm.
Automatic Feature Population Selection (AFPS)
The AFPS machine-learning algorithm is an intelligent and highly efficient way to analyze large amounts of complex medical codes (such codes that classify diagnoses and procedures). It can be used to predict someone’s risk of needing a future medical procedure based on early conditions diagnosed.
To illustrate how it works, let’s do a quick exercise. In the image of the busy Chicago street below, identify anything that could lead to an accident. For example, some might notice the yellow cab at an angle with a car right behind it—if the cab reversed without warning, that might lead to an accident. Someone else might focus on the traffic light—if it stopped working, it could lead to several accidents.
However, there are many other things in the image that may have nothing to do with causing an accident. A man on the sidewalk with a cup of coffee, flowers lining the road, the sign at the bus stop, and much more can easily get ignored. Our brains automatically scanned the picture, and in the blink of an eye, filtered out information that seemed unrelated. We focused on what we felt was significant.
Now imagine this kind of challenge on a much larger and more complicated scale. This is essentially what we trained our machine-learning algorithm to do with medical data. Say we want to find the medical conditions that indicate a likelihood of knee surgery in the future. Our algorithm can sort through 60,000 medical codes and identify 4,000 codes connected to knee surgery. Furthermore, it filters out 40 conditions that help to predict the occurrence of future knee surgery—conditions like osteoarthritis in the knee, obesity, lower limb muscle weakness, and arthritis injections.
How this impacts the individual
The AFPS approach helps us address thousands of medical conditions instead of just a handful. It carefully reads medical data and connects relevant information to a medical procedure. Finally, it’s highly accurate in its predictions. When we compared results from our manual approach of talking to doctors and poring through ICD and CPT codes to the newly automated one, the results were remarkable: a 95% accuracy rate in identifying the right diagnoses that predict future procedures.
At Evive, we pride ourselves in providing cutting-edge solutions to employers that, in turn, help their employees live better lives. Focusing on maximizing employee health, while lowering employer healthcare costs, will ultimately help people understand and achieve a better health roadmap.