Value-Based healthcare: Using predictive analytics to forecast risk and outcomes

Jason Gilder, PhD is the Senior Director of Informatics and Analytics at IBM Watson Health.

Predictive analytic models have been used to improve the understanding of healthcare delivery for decades. However, the push to advance value-based healthcare has intensified the need for predictive analytics to help clinicians and care managers anticipate problems before they develop, and mitigate health issues before they worsen.

With the use of big data analytics, predictive models can be designed to be:

• Incorporated in clinical workflows to facilitate care management and identify and address the needs of at-risk individuals
• Used to perform risk adjustment on quality measures to account for patient severity and allow benchmarking between providers
• Employed to understand the treatments with the potential for better outcomes

Healthcare delivery already relies on a variety of predictive analytics.  More advanced models, including cognitive computing, are now entering the market to help improve healthcare outcomes. Here are some of the models being used in healthcare today as well as advanced models and methods currently in development across the industry:

• Event and outcome prediction –identifies the likelihood of an avoidable readmission or individuals with a higher than average chance of developing a new chronic condition, such as diabetes. The goal of these types of predictions is to manage a population in such a way that undesired outcomes are avoided.
• Utilization and risk prediction – involves identifying the relative financial risk of an individual within a population of patients in an actuarial, meaningful way. The IBM Explorys Risk Model is a complex utilization prediction algorithm which considers a patient’s demographics, diagnoses, procedures, medications and prior healthcare costs to predict prospective utilization.
• Risk adjustment –uses an individual’s severity, risk or burden to normalize a reported outcome or quality metric. For example, risk adjustment is used to normalize the chance of readmissions after knee replacement based on the severity of the illness and demographics of the patients receiving those surgeries.
• Machine learning, deep learning and cognitive computing – are all new, next generation, cognitive technologies being used to take risk forecasting, analysis of healthcare utilization and outcomes analysis to new levels of understanding and insights. These newer technologies use significantly more data within a flexible modeling environment that can understand context, learn, and adapt to hundreds or even thousands of dynamic data inputs.

Clinical adoption of these predictive algorithms will grow as models gain in maturity, predictive power, and usable applications. Read more about how these types of models will continue to support data-driven value-based health in the Watson Health white paper:  Predictive Analytics in Value-Based Healthcare:  Forecasting Risk, Utilization, and Outcomes.