In recent years, buzzwords such as machine learning, artificial intelligence (AI), predictive models, analytics and big data have dominated the conversation across many industries, including healthcare. However, while widely used, these terms are rarely explained, nor is their importance to the healthcare industry.
Predictive Modeling: You’re Doing it Without Even Realizing It
Simply put, a predictive model is a way to predict the future based on what has occurred in the past. The human brain creates predictive models every day. Just think about your morning or evening commute. When estimating how long it will take you to reach your destination, your brain will use several factors—the weather, the time you leave and which day of the week it is—to make a prediction. You know that if it’s raining or if you leave during the heart of rush hour, your commute will take longer. As you learn from experience and test different scenarios, your predictions become much more accurate. An accurate prediction of your commute home is important because it can impact other aspects of your day, such as picking the kids up from daycare or preparing dinner. Predictive models can help answer questions to guide your decision about what time on a given day you need to leave.
Predictive models can also be used to improve healthcare. As health plans, providers, employers and other at-risk organizations struggle to minimize the high cost of healthcare, predictive models can be used to better leverage resources and target engagement programs. For instance, predictive models can be used to analyze the care and cost patterns associated with your population’s medication adherence. Healthcare organizations can predict how different members will adhere to their medications and identify those who may need additional help to do so, even before they show lapses in adherence. Specifically, these insights can help to identify which of these members need support with medication adherence immediately in order to avoid adverse events in the future.
The Importance of Data
Another buzzword, big data, is often used in conjunction with predictive analytics. Put simply, big data refers to extremely large data sets that may be analyzed computationally to reveal patterns, trends, and associations relating to human behavior and interactions. Predictive models are only as good as the underlying data you use. Without concise and accurate data, they are not able to give meaningful predictions. If you miss inputting data from one of the key prediction factors, over a period of time, it will affect the results.
In healthcare, data points include medical and pharmacy claims, demographic information, social determinants of health, consumer information, and more. These data sets can help assess relevant insights about each person, such as clinical conditions, use of the healthcare system, medication adherence history, engagement preferences, ability to pay for care and more.
Bringing Machine Learning into the Mix
Predictive modeling preceded big data and relied on statistical techniques to ensure that the “small” or “limited” data sets available would generate accurate predictions. The availability of big data opened the door to additional prediction methods. This is where machine learning comes in.
Machine learning is a type of AI that has the ability to improve with more data. Machine learning techniques use several examples and data points to form an accurate prediction, improving these predictions when provided with new examples. This allows problems to be set up and solved in new ways.
Imagine that you had access to not only the data points you have collected about your commute time, but also had access to thousands of other people’s data points who share some portion of your commute. With this more comprehensive data set, you could create separate predictions for each portion of your commute and could include additional factors—sunlight angle, season, holidays, etc. By default, machine learning methods thrive on more examples and more data, resulting in better predictions.
Applying Machine Learning to Population Health
Machine learning is being used constantly, from teaching self-driving cars to determining if a person is likely to be readmitted into the hospital. The frequency of use in daily life has increased for a few reasons:
- More data is being collected than ever before
- Data can be processed at a quicker rate
- Increased improvements to algorithms have resulted in more accurate predictions
Health plans and providers can use machine learning to predict which members may develop a chronic illness, like diabetes. Using both clinical and non-clinical data from previous patients and current data indicating who developed diabetes and who did not, machine learning can learn what differentiates these two groups. Then, this knowledge can be applied to others. Members that look most similar to the ones who developed diabetes will be selected for diabetes prevention programs.
At Health Dialog, we use the Pathways Engine, our predictive analytics and machine learning platform, to help us prioritize resources and select the ideal members for interventions. As an example, with our Chronic Care Management solution, the Pathways Engine identifies members diagnosed or at risk for developing chronic cardiovascular and respiratory conditions, as well as diabetes. We then use our Care Pathways framework to determine what stage the member is in along their healthcare journey, from pre-diagnosed to onset of the condition to a late progressive stage. This allows our Health Coaches to tailor specific clinical interventions and action plans to address identified members’ needs, helping them make behavioral changes that delay disease progression.
With these machine learning tools, we are able to sift through a lot of data to identify a person who may be struggling with managing their condition before they suffer a major health event, providing earlier interventions to help improve the member’s whole health while reducing costs.
