Doctoral Student Develops Machine Learning Model to Predict COVID-19 Hospital Stays

October 5, 2020 | COVID-19, News, Research, UToday, Engineering
By Tyrel Linkhorn



Early in the COVID-19 pandemic, Mohammadreza Nemati closely watched news reports of hospitals so overrun by patients that they struggled to triage the sick, let alone find them a bed.

Nemati, a doctoral student in The University of Toledo Department of Electrical Engineering and Computer Science, wondered if it might be possible to apply machine learning approaches to assist hospitals with managing the volume of patients.

“I knew this was a tough situation. I thought if we are able to predict the discharge time for hospitalized patients, it could be a really big help for the decision makers in the hospitals to use their equipment and facilities efficiently, he said. “I also thought it could give them a better insight to manage the overload of patients.”

UToledo doctoral student Mohammadreza Nemati

Nemati

With clinical data from nearly 1,200 patients, Nemati and Jamal Ansary, a mechanical engineering graduate student, built precise predictive models able to forecast with 70% accuracy the recovery time of hospitalized COVID-19 patients based on their age and sex.

The research was recently published in the CellPress journal Patterns.

“We all know that COVID-19 is a really, really complex disease, but our findings indicate that the patient’s discharge time from the hospital can be predicted even with the most basic features of the patient, including age and sex,” Nemati said. “The results of this work also provide a decent criterion for other studies that aim to measure the impact of clinical features on the patient’s length of stay in hospital.”

Because the study’s demographic data was limited to only age and sex, Nemati said by including additional demographic data or clinical history, the model might be able to provide even more precise predictions.

Nemati, who also earned a master’s degree in computer science from UToledo, is focused on the intersection of machine learning, statistical analysis and computational biology. His research is primarily concentrated on using those tools to explore how risk factors for kidney donors and recipients are related to post-transplant graft survival times.

However, the magnitude of the COVID-19 crisis and those images of overflowing hospital emergency rooms inspired him to shift his focus toward the pandemic.

“I was pretty familiar with how to use my engineering knowledge toward biomedical and medical problems,” he said. “We thought we could potentially write a paper and help the medical personnel who are trying to tackle the pandemic problem.”

More generally, the study’s findings mirrored previous research indicating females have shorter recovery times and a slightly higher probability of being discharged from the hospital following a COVID-19 admission. Older individuals are also more likely to have longer hospital stays and higher rates of death.

Click to access the login or register cheese