In healthcare, diseases are often classified as being acute or chronic (Braunstein, 2014). Both types of disease are important targets for predictive modeling. Acute diseases are characterized by a sudden onset, are usually self-limited, and patients often experience a full recovery after the appropriate treatment. Also, risk factors for acute conditions are often not determined by patient behavior. Examples of acute diseases include influenza, kidney stones, and appendicitis.

Chronic diseases, in contrast, typically have a progressive onset and last for the lifetime of the patient. They are influenced by patient behavior, such as smoking and obesity, and also by genetic factors. Examples of chronic diseases include hypertension, atherosclerosis, diabetes, and chronic kidney disease. Chronic diseases are particularly dangerous because they tend to be linked and cause other serious chronic and acute diseases. Chronic diseases are also costly to society; billions of dollars are spent annually on preventing and treating common chronic conditions.

Acute-on-chronic diseases are particularly popular in healthcare predictive modeling. These are acute, sudden onset diseases that are caused by chronic conditions. For example, stroke and myocardial infarction are acute conditions that are by-products of the chronic conditions hypertension and diabetes. Acute-on-chronic disease modeling is popular because it allows us to filter the population to a high-risk group that has the corresponding chronic condition, increasing the yield of predictive models. For example, if you were trying to predict the onset of congestive heart failure (CHF), a useful starting place would be patients that have hypertension, which is a major risk factor. This would lead to a model with a higher percentage of true-positives than if you were to randomly sample the population. In other words, if we were trying to predict CHF onset, it wouldn't be very useful to include healthy 20-year-old males in our model.