Using AI to predict MRI no-shows is saving one radiology department nearly $180,000 per year

Outpatient appointment no-shows are a problem across the globe, with rates as high as 23.5% in North America alone. But one health system recently bucked this trend by using machine learning analytics to predict which patients were likely to miss their MRI exam.

Radiologists with Changi General Hospital in Singapore unveiled their predictive analytics platform and subsequent quality improvement project Wednesday in the American Journal of Roentgenology. The algorithm keyed in on 10 factors to make its determinations, which proved moderately accurate.

Harnessing those metrics, however, helped the team develop a simple phone call reminder tool for patients most at risk for not showing up for their scan. Ultimately, the intervention led to a 17.2% improvement from the institution’s baseline no-show rate.

And based on some simple math, Le Roy Chong and colleagues say their radiology department will save about $180,000 per year.

“Appointment no-shows are a multifaceted problem given the multitude of behavioral, social, medical, physical, logistic, and geographic factors that interact in a complex and unpredictable fashion to influence the outcome of appointment attendance; nonetheless, this problem may be tractable given the capabilities and successes of recent machine learning techniques…,” the authors wrote.

To develop their analytics tool, the group used nearly 33,000 anonymized outpatient MRI appointments from a two-year period, along with a set of 1,080 records from this past January. They also incorporated an open-source decision tree-based algorithm known as XGBoost to create the model.

The best performing approach identified 10 contributing factors to missed appointments, including patient age, appointment wait time and reschedule counts, and postal district, among others. Chong and colleagues reported that the AI performed “moderately well” at predicting no-shows.

But after deploying the model in the rad department’s workflow, the specialists were able to generate phone call reminders for the 25% of patients most at risk of missing their MRI.

Despite some limitations, including a relatively small amount of training data, Chong and colleagues believe other institutions should consider using similar technologies to target no-shows.

“State-of-the-art artificial intelligence predictive analytics can perform moderately well in solving complex multifactorial operational problems such as outpatient MRI appointment no-shows, using a modest amount of data and basic feature engineering,” the group concluded. “Such data may be readily retrievable from frontline information technology systems commonly used in most hospital radiology departments, and they can be readily incorporated into routine workflow practice to improve the efficiency and quality of healthcare delivery.”

Read more about the project here.