Ylan Kazi, Vice President, Data Science + Machine Learning, UnitedHealthcare
Machine learning is becoming more useful across many different industries, and within health care this also the case. There are a lot of challenges within the U.S. healthcare system, particularly with cost of services. We pay more for health care services than other countries yet have poorer overall quality outcomes.
Many of the costly diseases are things like diabetes and heart disease, which tend to be a product of our lifestyles. These diseases must be managed over a lifetime and are contributing towards higher costs for health care overall. For diabetes in particular, 30 million Americans have it yet 7 million of them are undiagnosed as of 2015.
Enter Machine Learning
Machine learning takes large amounts of data and essentially learns patterns and characteristics of the data that no human being could do in a short amount of time. From these insights, predictions can be made about almost anything. Generally, as more data is added, the machine learning algorithm improves.
In the health care world, there is a wealth of data from health claims, patient demographic information, and labs. Electronic medical record systems act as a repository for patient data instead of having to aggregate this information from paper charts.
Managing disease doesn’t end with predicting it. There must be feedback mechanisms in place that drive how a patient will receive their care after the prediction
Running machine learning algorithms against this data can vastly improve the disease predictions.
Many of these chronic diseases can be diagnosed much earlier in a patient’s life and can be managed more effectively. For diabetes, earlier diagnosis can lead to a higher quality of life and potentially even reversing the condition under the right circumstances.
Machine learning can help supplement the traditional diagnosis process. It won’t replace clinical judgment but can act as another recommendation for physicians and other clinical professionals. A machine learning solution could give a likelihood of specific disease states, as well as offer reasons for why it is predicting a disease. The ability to explain a prediction is key to having clinical professionals trust the systems as well.
Even the process of predicting disease is changing. It used to be where you would visit your doctor and over the course of time be diagnosed using a combination of different lab tests and an assessment of lifestyle. Now with the proliferation of wearable devices, it is possible to diagnose these diseases in real time. Companies are developing technology to test for things like blood sugar levels without drawing blood. This is adding further data that can be used to improve disease predictions for patients.
Beyond Disease Prediction
Managing disease doesn’t end with predicting it. There must be feedback mechanisms in place that drive how a patient will receive their care after the prediction. This type of cultural change within health care organizations will take longer but is just as important as a disease prediction. There is no point in identifying a disease if we are going to approach it in the traditional way.
With a combination of disease prediction and management, along with new technologies that allow for more real-time data, machine learning will continue to improve the way that patients receive their health care services and ideally slow down the trend of growing health care costs.