Oleg S. Pianykh, Presenter: Nothing to Disclose

Daniel Ira Rosenthal MD, Abstract Co-Author: Nothing to Disclose

Through exhaustive predictor analysis we discovered an efficient and accurate PWT predictor formula, much more accurate that constant PWT estimate. It was used to implement waiting area displays, informing patients of their anticipated wait time. This also helped us identify the outliers – patients waiting significantly longer than predicted – and train our staff to assist these patients with their needs. As a result, our PWT provided us with an improved radiology workflow, more informed patients, and staff able to concentrate on more urgent questions.

“How long will it take?” is the most frequent question asked in the waiting rooms. Knowing Patient Wait Time (PWT) allows one to improve patient/staff satisfaction, load-balance departmental resources, develop long-term planning and strategies. Our project addressed the most fundamental question of accurate PWT prediction. 

To perform PWT analysis, we used our onsite X-Ray imaging facility with 9 X-Ray units, providing images for orthopedic outpatients. 8478 exams records, corresponding to 13 consecutive weeks of patient visits and 6452 patients, were captured from the RIS database. We defined 23 independent PWT predictor variables, including patient wait line size L, number of patients in exam rooms F, patient arrival/processing rates, patient exam type, patient age, etc. Our goal was to build the most accurate PWT predictor, using as few parameters as possible. Therefore we analyzed all possible linear and quadratic regression predictors with up 4 parameters selected from the original 23-parameter set.

We discovered that the most accurate PWT predictor has the following form: PWT = W0+k0L0+ktLt+K2tL2t, Where: W0, k0, kt, k2t are optimally-chosen coefficients Lt is the patient waiting line size t minutes ago t is the sampling rate. The optimal sampling rate t was found to be 5 minutes. As a result, our PWT formula used the same parameter L, sampled at 5-minute intervals. The value of L was found from RIS as the count of patients, who arrived but have not started their exams yet. The median error for this PWT predictor was only 3 minutes; using all 23 predictors would decrease the error only by a few seconds.  

Pianykh, O, Rosenthal, D, Can We Predict Patient Waiting Time? .  Radiological Society of North America 2014 Scientific Assembly and Annual Meeting, - ,Chicago IL. http://archive.rsna.org/2014/14014898.html