|
Abstract:
|
On average, 23% of outpatient clinic visits result in missed appointments (i.e. ‘no-shows’). ‘No-shows’ lead to decreased clinic productivity, lost clinic revenue, increased healthcare costs, and decreased access to healthcare for patients. To offset these costs, clinics will often overbook some scheduled appointments with a 2nd appointment at the same time. This can lead to decreased patient satisfaction, longer wait times, and increased overtime-pay for staff. The strategy of “predictive overbooking” attempts to predict patients with the highest ‘no-show’ risk for whom to overbook in order to maximize clinic efficiency and minimize costs. In this context, we propose a probabilistic approach for optimizing the classification of predicted ‘no-shows’ in prediction models, taking into account the predicted probability of no-shows, scheduling collisions, and costs associated with each. Using electronic health record data from 14 pediatric sub-specialty clinics, we develop a Random Forest model predicting ‘no shows’ for each clinic and demonstrate our method for classification leads to improved clinic utilization and decreased costs compared to other common classification thresholds.
|
