A new study from Pennsylvania State University shows that AI and machine learning can accurately predict primary care no-shows and slow cancellations.
Overall, the machine learning model got 0.85 on the 0-1 scale to predict reservation numbers and 0.92 to predict late cancellations, the research author wrote in Family Medicine Chronicles.
These findings are due to the efforts of healthcare professionals and practice managers to improve operational efficiency. Later-term cancellations and appointments no-shows can be costly as they represent a healthcare provider missed the opportunity to earn revenue.
“Not able to attend a primary care visit will disrupt the essential care of the patient, disrupt the clinical workflow and stake in medical resources,” the researchers explained. “A delayed or missed appointments for essential care can lead to serious health effects and increased disparities among people already experiencing health inequality.”
Healthcare organizations can deploy a number of technologies, including appointment reminders and self-scheduling options, to help reduce the number of late cancellations and no-shows. However, while these tools helped to promote better patient engagement, the researchers acknowledged that they did not completely close the appointment gap they missed.
Researchers say AI, particularly machine learning, can be effective.
Examining the impact of machine learning on no-shows, late cancellations
Using geolinked clinical, care use, socioeconomic and climate data from over 1 million appointments at 15 family medicine clinics in Pennsylvania, researchers were able to test different learning models to predict late cancellations and no-shows of appointments.
The tested models included gradient boost, random forest, neural networks, and logistic regression, predicting reservation outcomes. They also conducted a characterization significance analysis to identify factors that could lead to endings and late cancellations at the population and patient level.
According to the researchers, the gradient boost model was ideal for predicting no-shows and slow cancellations, with around 85% predicted quality for no-shows and 92% predicted quality for late cancellations. Importantly, this model did not predict biased outcomes against gender or race/ethnicity.
Regarding what contributed to late cancellations or missed appointments, the researchers noted that lead time is important. If there were too many lead times before an appointment – which means patients had to wait a long time to book an appointment and actually attend – they were likely to cancel at the last minute and didn't show up at all.
This insight will help clinics revamp their booking strategies.
“Given the strong effect of lead times, clinics may prioritize short waiting times for high-risk patients,” the researchers recommended.
However, reducing the time to booking is just one important strategy. Even when lead times decreased to less than 30 days, certain sociodemographic characteristics affected no-shows or late cancellations of bookings.
That's where customized patient engagement technology can emerge. Researchers said machine learning may provide deeper insight into how it predicts the needs of a particular patient.
For example, factors such as female gender, younger age, limited English proficiency, number of chronic illnesses, and distance to the clinic contributed to no-shows and late cancellations. Providing specific systems or assistance to patients adapting these profiles can reduce the outcome of shortages of appointments.
And to streamline this more personalized approach, clinics can deploy machine learning. AI could help identify patients at high risk of last-minute cancellations, or identify patients who have missed out on appointments entirely.
“By integrating personalized predictive models with system-wide initiatives such as automated reminders, patient navigation and slow cancellation fees, we provide the opportunity to help care teams create more targeted, effective, personalized interventions for better appointment compliance,” the researchers concluded.
Sara Heath has been reporting news related to patient involvement and health equity since 2015.
