Researchers from Boston Children's Hospital and Yonsei University have developed a machine learning tool to predict no-shows in pediatric medical appointments.  

By adopting a data imputation method for patients with missing information in their records, developing an interpretable approach that explains how a prediction is made, and exploiting local weather information, the team created a model identifying 83% of no-shows at the time of scheduling.

"Our no-show prediction method may potentially be informative when identifying appropriate interventions to reduce no-shows," wrote the team in an article published this past week in npj digital medicine.  

WHY IT MATTERS  

As the researchers pointed out, so-called "no-shows" – when a patient schedules but doesn't attend an appointment – can have negative impacts on patient health, and hospital and clinics’ resource utilization.  

"As continuity of treatment, preventive services, and medical check-ups cannot be delivered when a patient misses an appointment, no-shows at appointments have been associated with poor control of chronic diseases and delayed presentation to care," they wrote.  

They noted that the increase in available data has bolstered no-show prediction possibilities, but that challenges remain. The team set out to address those challenges.   

They found that including patients’ records with missing information significantly improves the models' predictive accuracy when compared with an approach that can only be trained on patients with complete information.

In addition, researchers said, "our analysis suggested that inclusion of local weather information into predicting features improves model accuracy," especially atmospheric pressure.  

The team also identified potentially actionable ways for further studies to explore how to reduce patients' no-shows. They suggested that their model might be helpful when it comes to interventions, such as texting, emailing or calling patients.   

"It may also be useful to calculate the optimal frequency at which reminders should be sent for each patient and may help better allocate free transportation resources when needed," they wrote.  

"We acknowledge that clinics with limited resources, the interventions mentioned above may not always be available," they continued. "With this limitation in mind, we explored potential actionable items that could be implemented in the majority of clinics."   

The items included choosing the day of the week and time of day that would be easier for patients and their parents to come to their medical appointments; using a language service; and choosing a day with likely nicer weather.  

THE LARGER TREND  

For this study, researchers noted that their model showed minimal predictive performance differences across racial groups. But other experts have warned in the past that scheduling systems can lead to longer wait times for Black patients due to disproportionate hurdles they may face to care.  

"Currently, these scheduling systems are penalizing Black patients for not showing up based on socioeconomic issues that are out of their control," said Shannon Harris, an assistant professor at the Virginia Commonwealth University School of Business who co-authored an August 2021 study on the issue.

Other health system leaders have pointed out that no-shows can be reduced via other technologies such as telemedicine and patient engagement platforms.  

ON THE RECORD  

"A potential avenue of future research could be the development of no-show predictive models for an array of patients visiting different healthcare providers," wrote researchers in the npj digital medicine study. "We suspect that adding information on the different medical problems affecting patients may improve predictive performance."

Kat Jercich is senior editor of Healthcare IT News.
Twitter: @kjercich
Email: kjercich@himss.org
Healthcare IT News is a HIMSS Media publication.