Using Machine Learning to Predict Hospital Readmissions

As we settle into this winter season and start to see COVID-19 cases on the rise once again, many hospitals have reached their capacity. 

Now, more than ever, it’s vital that hospitals lower readmission rates to create room for severe COVID-19 cases.

That’s where machine learning comes into play. When approached with the challenge of revamping a client’s readmissions-tracking tool to incorporate new technology and documentation, we happily accepted. 

Our goal was to make a Generalized Linear Machine Learning Model that could be expanded upon in the future—little did we know, that future would include COVID-19. But before we get ahead of ourselves, let’s cover some basics. 

How Machine Learning Works

There are several different models used in machine learning. The three most commonly used in practice are: 

For the requirements of this project, we used a Generalized Linear Model. This means the machine analyzes a binary classification, otherwise known as a yes/no problem. The question it’s trying to solve: Will a patient be rehospitalized within 90 days?

How to use Machine Learning to Predict Hospital Readmissions 

Using data which represents people who have been rehospitalized, the machine learning algorithm will analyze preset factors—such as type of illness, number of days in the hospital, body mass index, etc.—to determine which combination of factors will lead to rehospitalization. This process is called “training.”

After the model is trained, it can then begin “scoring” new data. In this case, looking at current patient hospitalization data to decide how likely it is that a patient will be readmitted within 90 days. For each patient whose data is analyzed by the algorithm, the machine learning system will provide a percentage-based score on whether or not that patient will be readmitted. 

Machine Learning Data: Real-Life Application

Medical staff can then use that score to develop potential additional procedures to prevent high-risk patients from being re-admitted. For example, perhaps the medical staff will follow up with the patient two days after discharge to see how they are adhering to their discharge plan.  At that point, the medical staff can call in additional medication or suggest other remedies to help the patient prior to a need for hospital readmission.

This process helps medical staff provide a more accurate, safe, and intentional quality of healthcare to their patients, even when the medical staff is stretched for time and resources due to, let’s say, a pandemic. 

While this machine learning system is certainly helpful for the doctor and the hospital, it can also be lifesaving for a patient—especially in the age of COVID-19 when medical professionals across the globe are wading through a sea of symptoms in order to accurately test treatments and save lives.

FortyAU Combines Data Science and Software Development at an Enterprise Level

Through our work with this major healthcare provider and other enterprise-level clients, we’ve been able to create software that directly impacts our world for the better, and that’s important to us. 

We understand the unique needs of large companies, and our team of developers are passionate about partnering with businesses at any stage during their project—working together to make software better. 

What important insights are hiding in your data? Our team can help you locate trends and create technology that changes lives. Let’s explore together.

Have an existing project that needs extra development help?