by John R. Fischer
, Senior Reporter | September 15, 2020
Radiology departments could reduce the number of no-shows they see for scheduled MR appointments with the help of AI predictive analytics.
At least that’s what researchers at Singapore’s Changi General Hospital are saying in a new study which found the technology to be capable of solving multifactorial operational problems, including outpatient MR appointment no-shows, using only a modest amount of data and basic feature engineering.
"The ability to track appointment no-shows is not, in itself, useful unless one is able to intervene, which is really where the challenge lies," lead author Dr. Chong Le Roy, senior consultant for the department of diagnostic radiology at Changi General Hospital, told HCB News. "Reducing no-shows is complicated due to the myriad human and nonhuman factors (both known and unknown) that can potentially interact in complex ways to affect the likelihood of appointment no-shows. It is only with the recent advent of state-of-the-art machine learning algorithms and ready availability of increased computational power in the last few years that makes it possible to be able to predict and thereby allow us to address such problems with interventions."
Numed, a well established company in business since 1975 provides a wide range of service options including time & material service, PM only contracts, full service contracts, labor only contracts & system relocation. Call 800 96 Numed for more info.
Many large and academic facilities are staffed by subspecialty radiologists and a high number of no-shows can result in inefficient use of resources, staff rostering and workload distribution or balancing.
Le Roy and his colleagues trained and validated their model using records of 32,957 outpatient MR appointments scheduled between January 2016 and December 2018 from Changi General Hospital’s radiology information system. A further holdout test set of 1,080 records from January 2019 was also incorporated. The aim of their study was to produce a model that could be developed relatively quickly, would require minimal data processing, and would be readily deployable in workflow practice for quality improvement.
Finding an overall no-show rate of 17.4%, the team assessed various machine learning predictive models developed with widely used open-source software tools before finding a decision tree-based ensemble algorithm that used a gradient boosting framework, XGBoost, version 0.8. They then implemented telephone call reminders over a six month period for patients whom the model predicted were among the top 25% with the highest risk for not showing up to their MR appointments.
Following these six months, the no-show rate of the predictive model was 15.9%, compared to 19.3% in the preceding 12-month pre-intervention period. This corresponded to a 17.2% improvement from the baseline no-show rate. The no-show rates of contactable and noncontactable patients in the group at high risk of appointment no-shows were 17.5% and 40.3%, respectively, as predicted by the model.
"General radiology subspecialties providing high-volume imaging services utilizing advanced imaging modalities such as magnetic resonance imaging, CT and ultrasound scans where there is a large proportion of ambulant outpatients, patients with chronic nonlife-threatening ailments, and routine screening would benefit the most from reducing no-show rates," said Le Roy. "Some examples include musculoskeletal imaging for common internal derangements of various joints, spine imaging for cervical/low back pain and breast screening imaging."
The findings were published in ARRS’ American Journal of Roentgenology