Readmission Rates: Deep Learning Comes to the Table

A key component of U.S. healthcare reform is one of linking Medicare and Medicaid reimbursement rates to hospital readmission rates. The idea is to use money to motivate hospitals and their doctors to guarantee patients are treated properly before being released in order to reduce the likelihood that they will be readmitted again for the same condition.

Controlling readmission rates is not an exact science. Patients can receive very good care only to be readmitted a few days later due to issues related to diagnosis, scope of care, clinical conditions, etc. So what’s the solution? It might prove to be deep learning. If researchers at Penn State have anything to say about it, a deep learning system they have developed will become a key future indicator of patient readmission rates.

Introducing the REDD System

Penn State researchers teamed up with Geisinger Health System to come up with a deep learning platform they call REDD. The acronym comes from the three possibilities often seen when patients need more care in the days following hospital discharge: readmission, emergency department treatment, and death.

REDD is an analytical tool utilizing a ton of patient data, both past and present. The system’s initial data set was built on two years’ worth of clinical, socioeconomic, and administrative data supplied by Geisinger. New data is continually input in order to enable the deep learning components of the system.

Penn State researchers are convinced that REDD can predict whether a patient is likely to be readmitted to the hospital for the same condition within days of being discharged, based on that patient’s data as compared to a library of data already in the system. Accuracy should theoretically increase as the REDD data set grows.

A Big Financial Problem

According to a Penn State post from early in 2018, readmissions created big financial problems. They do not reflect very well on doctors either. Penn State says that 2014 numbers show the cost of readmissions to be somewhere around $41 billion annually. Medicare pays some $26 billion, of which approximately $17 billion is related to “avoidable re-hospitalizations.”

Medicare and Medicaid have both since implemented programs penalizing hospitals for readmissions within 30 days of discharge, when said readmissions are for the same condition. It should be clear that this becomes a big financial burden on hospitals.

The Penn State researchers took the 30-day threshold into consideration when building the REDD model. They say data indicates that readmissions after 30 days are not likely due to the original hospital stay or the care provided during that stay. Rather, they are due to things like home environment and patient support.

What It Means to Doctors

The REDD system is all well and good for hospitals attempting to avoid penalties, but what does it mean to doctors? It means a tool that can help doctors make more informed decisions about their patients and the treatments being provided. When statistical data shows that a certain course of treatment is likely to lead to readmission, a doctor can get with the healthcare team to discuss other options – and that’s just one example.

Locum doctors could benefit tremendously from the system as well. Locums often go into new contracts unprepared because they don’t have access to the same data their employed coworkers have. The REDD system could immediately bring them up to speed and provide the necessary guidance to properly treat patients they’ve never met before.

Here’s hoping the REDD system delivers as promised. If it does, the cost of hospital-based healthcare should be better contained in the next decade or so.