Making More Precise Health Care Decisions with Machine Learning
The health care industry, with its great variety and richness of data sources, is a natural area for application of machine learning algorithms in comparative effectiveness research (CER).
Especially within the context of CER in health care, machine learning algorithms are being used to improve productivity, evaluate alternative interventions, and develop new treatments. They often are used to discover intricate relationships between inputs and outputs that are hard to anticipate in advance. They are therefore particularly well suited for predictive tasks, such as predicting the future onset or progression of a disease, or the treatment to which an individual is most likely to respond.
Precision medicine relies on such tools to support joint decision making by patients and their providers regarding the best treatment plan given the patient’s individual characteristics, including lifestyle, environment, and genetics. The size and complexity of the health data required are daunting, as precision medicine ideally relies on information from all available sources, including electronic health record systems, patient-reported data, and genomic data. The potential upside from being able to use all these data for developing a treatment plan can be game changing, or even life saving.
Machine learning algorithms can also support the creation and revision of treatment guidelines, providing a deeper understanding of which genetic markers are associated with which side effects, for example, or how patients who followed the treatment guidelines fared relative to those who did not. These new approaches to data analysis may also help provide insight where treatment effects are heterogeneous, allowing researchers and practitioners to identify specific subsets of patients who might benefit from a treatment even when no benefit is discernible on average. ■
“The potential upside [for machine learning] can be game changing, or even life saving.”