As discussed earlier in this chapter, healthcare is primarily concerned with producing better outcomes at a lower cost. Often, we try to determine which patients are at a high risk of a poor outcome directly, without necessarily focusing on the specific cause of their signs and symptoms. Popular outcomes for which machine learning solutions are being applied include predicting which patients will likely be readmitted to a hospital, which patients will suffer death, and which patients will be admitted to the hospital from the emergency room. As we will see in Chapter 6, Measuring Healthcare Quality, many of these outcomes are actively monitored by governments and healthcare organizations and, in some cases, governments even provide financial incentives to improve specific outcomes.

Often, instead of dividing outcomes into two classes (for example, readmission versus non-readmission), we can attempt to quantify a patient's chances of survival in terms of a specific time period, given the characteristics of the patient's disease. For example, in cancer and heart failure patients, you can attempt to predict for how many years the patient is likely to survive. This is referred to as prognosis, and it is also a popular machine learning problem in healthcare.