Improving endoscopy attendance rates in high-risk patient groups that are identified using a machine-learning model

Fernandes, Denzil; Beigang, Fabian; Bilsborough, Richard; Wilcockson, Christopher; Stebbings, Clare; Nevens, Lisa; Lee, Tom

Improving endoscopy attendance rates in high-risk patient groups that are identified using a machine-learning model

Authors

Fernandes, Denzil

Beigang, Fabian

Bilsborough, Richard

Wilcockson, Christopher

Stebbings, Clare

Nevens, Lisa

Lee, Tom

Issue Date

2025

Type

Presentation

Language

en

Keywords

endoscopy , appointment attendance , machine learning , artificial intelligence , demographics

Abstract

Introduction: To determine if a pre-appointment phone call, targeted at endoscopy patients, who based on a machine learning model, are predicted to be at higher likelihood of not attending, will reduce overall non-attendance rates and reduce inequalities in the non-attendance rate. Methods: This is a single-trust pilot project which took place between October 7, 2024 and January 17, 2025. 868 calls have been made, to 540 patients, ahead of 653 endoscopy appointments, mostly gastroscopies, flexible sigmoidoscopies and colonoscopies. Each appointment was given an estimated DNA (did not attend) probability using a machine learning model trained on two years of prior endoscopy appointments across the trust. Factors contributing either positively or negatively to this probability include, but are not limited to: prior attendance history, demographic factors and procedure type. Most patients were called during working hours on weekdays, on average 3 days before their procedure. Calls were targeted at appointments with a 10% or greater probability of non-attendance – our ‘higher risk’ group. All calls were made by one person, using a PowerBI report built to collate and visualise all the information they needed to make the calls. Results: Patients were successfully contacted by telephone in 501 of 653 appointments (77%). The non-attendance rate for 501 ‘higher risk’ appointments with a successful contact was 4.7%, rising to 10.7% when including ‘higher risk’ appointments with no successful contact. The average non-attendance rate for ‘higher risk’ appointments in the four months prior to this project was 24.4%, based on 580 attendances and 189 non-attendances. In the 653 high-risk appointments due a phone call, our model estimated a total of 133 non-attendances, compared to 51 observed. 33 of these 51 were appointments with no successful contact. 427 ‘higher risk’ appointments resulted in an attendance. The non-attendance rate for 265 appointments for patients living in the 20% most deprived areas, was 6.2% for appointments with a successful contact, rising to 12.5% when including those not successfully contacted. The pre-intervention average for such patients was 14.3%. A chi-square test comparing the number of attendances and non-attendances in ‘higher risk’ patients between the intervention period, and the four months prior, shows a statistically significant reduction in the proportion of appointments not attended (p=0.00001). Conclusions: Pre-appointment phone calls targeted at patients with a higher estimated probability of non-attendance are associated with a statistically significant reduction in the proportion of appointments not attended.

Citation

Fernandes D. et al. (2025) 'Improving endoscopy attendance rates in high-risk patient groups that are identified using a machine-learning model', Gut, 74, A81.

Journal

Gut