Catching the Gorilla: Applying Machine Learning to Electronic Health Records

In this special guest feature, Ori Geva, Co-Founder and CEO of Medial EarlySign, discusses how the ue of machine learning can help create new opportunities for earlier intervention and delivery of improved, personalized care by allowing physicians and health systems to increase their scope of attention. Medial EarlySign is a developer of machine learning tools for data driven medicine. The company’s advanced algorithm platform accurately detects the likelihood of disease for subpopulations using basic medical information, such as blood test results, and other EMR data. Predictive tools provide physicians with actionable insight, while providing insurers with effective models to flag and focus on patients at risk, helping to prioritize resources, save money and improve care. Medial EarlySign’s platform addresses numerous potential clinical outcomes, including cancers, diabetes and other life-threatening illnesses.

Many of us are familiar with the Invisible Gorilla Experiment, the world-famous awareness test conducted by Daniel Simons and Christopher Chabris. If you haven’t seen it, here’s a quick refresher: Two groups of people wearing black and white T-shirts pass a basketball and study subjects are asked to count the number of passes made by one of the teams. In the middle of the drill, a man in a gorilla costume walks into the picture.

Surprisingly, more than 50% of the test subjects completely miss the gorilla. They were so engaged in the task of counting the number of passes that they simply didn’t see him. This experiment, originally conducted in 1999, still enthralls people to this day, shedding light on the psychological process of selective attention and inattentional blindness.

How does this relate to machine learning, and particularly to health? In our case, the gorilla is a progressing health condition, or development of a health complication. This may include diabetes, a GI disorder, any number of cancers, or other life-threatening disease. If so many people manage to miss the gorilla in a visual test with limited factors – one that should be fairly easy to pass – imagine just how easy it is to miss something involving dozens of different factors and variables that are not even visual, but are conveyed as numbers.

That’s where machine learning comes in.

Think about what may have happened if someone had told the viewer, ‘keep an eye out for the gorilla.’ The outcome would likely have been different. Additionally, with healthcare data, gorillas may come in a variety of forms, and providers and physicians use guidelines and thresholds to detect the obvious “symptomatic” gorillas.

By applying machine learning to analyze vast amounts of data, computers can help us identify the slightest changes and patterns, without being explicitly told where to look. Unlike traditional methods, which required us to provide a system of rules or guidelines in order to draw conclusions and insights, machine learning allows us to effectively ‘train’ the system by supplying it with data and outcomes.

For example, how does the machine know it needs to look for a gorilla, or a specific disease? How does it even know what a disease is? And how will the machine know that this is not a typical occurrence?

As implied above, it learns. The ability to analyze vast amounts of information, combining multiple factors, can show risk similarities between patients and identify potential outcomes. It can easily and automatically test millions of complex hypotheses and analyze the health operations and interventions that were successful, and those which were not.

In our case, applying machine learning to electronic health records (EHRs), which contain thousands of data points, can successfully uncover hidden signals of a disease. These signals may have otherwise remained undetected because no threshold was crossed, or results were borderline when considering only one or two parameters.

This is significant because instead of relying only on traditional predictors, such as family history and just one or two blood count parameters, machine learning evaluates dozens of factors to help physicians identify more people at risk of having health complications or diseases. In other words, machine learning takes into account all the subtle factors prior to, during and following the appearance of the gorilla – or the manifestation of disease – to increase the likelihood that it not be missed.

Use of machine learning can help create new opportunities for earlier intervention and delivery of improved, personalized care by allowing physicians and health systems to increase their scope of attention. Additionally, the ability to refine data and segment patients into high and low risk categories enables health providers to better allocate resources, prioritizing those at the highest risk and helping to close the care gap.

Returning to the gorilla test, the human eye alone can easily miss the gorilla. However, machine learning can serve as an important tool for health systems and physicians, to support their clinical expertise and remind them to constantly be on the lookout for any kind of gorilla.

As increasingly more global healthcare providers work toward identifying rising-risk patients and improving outreach to those individuals, machine learning can enable them to do so in the most effective way. Indeed, the combination of EHR data, physician expertise, and machine learning holds tremendous potential to positively transform the way healthcare is delivered – ensuring the gorillas will be caught, to enable earlier intervention and help reduce healthcare costs.

 

Sign up for the free insideAI News newsletter.