LOS ANGELES, CALIFORNIA - One of the most puzzling aspects of healthcare is trying to predict who is going to get sick, who will become very sick and end up in the hospital, and who benefits from interventions to reduce the burden of illness.

Most high-cost patients don’t stay high-cost for long, research reveals. In a study by the Commonwealth Fund on a population of individuals with multiple chronic diseases, only 2 percent remained high-cost for two years in a row. This shows high-cost patients are often replaced by other high-cost patients the following year, posing a significant challenge in predicting who is going to be high-cost at any given time. However, accurate prediction of who is going to sicken will not only give a leg up in preventing illness, but also reduce healthcare costs to the tune of billions of dollars a year. 

This prompts the question as to why, with all the recent advancements in technology, artificial intelligence (AI), and machine learning (ML), why we are still unable to perform. 

One of the most effective ways to reduce healthcare costs is to prevent hospital readmissions, which is much easier with the knowledge of which individuals are more likely to need readmission. Thus, all-cause readmissions have been a hot topic as the outcome measure for predictive algorithms for decades, but selecting from a pool of almost infinite variables to model and picking ones that