As we touched upon in Chapter 3, Machine Learning Foundations, healthcare is no stranger to complex risk factor assessments. For almost every major disease, one can find several risk-scoring models that are used widely by physicians to assess the risk of having a disease or suffering morbidity/mortality from that disease. When we use the term "risk score," we are largely referring to criterion tables, in which risk factors are allotted point values, and the points for all of the risk factors are summed to give an overall risk based on the total. These scoring systems are used widely in medicine; interestingly, many of them are based on research involving logistic regression models (similar to the one developed in Chapter 7, Making Predictive Models in Healthcare). The crucial question of the last several decades is whether machine learning can improve our ability to predict whether an individual has or will have a disease, how much care that disease will require, and whether the patient will die from the disease in a certain time period.

That question is addressed in this chapter. We organize this chapter by having sections on several leading causes of morbidity and mortality in developed countries for which risk scores have been developed. The covered entities include overall cardiovascular risk, congestive heart failure, cancer, and all-cause readmission. The first three entities are leading causes of mortality, and the fourth entity is a common way to measure quality in healthcare. We then explore the subsequent machine learning literature to see whether machine learning has improved the traditional risk assessment used for that disease. By the end of this chapter, you should have a thorough understanding of how machine learning can be used to improve disease prediction.