Purchase this article with an account.
Wei-Chun Lin, Isaac Goldstein, Michelle Hribar, Abigail E Huang, Michael F Chiang; Data Analytics for Prediction of Patient-Provider Interaction Time in Pediatric Ophthalmology Clinics. Invest. Ophthalmol. Vis. Sci. 2019;60(9):5510.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
The interaction time between healthcare providers and patients is one of the most important factors associated with patient satisfaction in outpatient clinics. However, efficiently scheduling clinic appointments to improve provider-patient interaction time is a challenging task, particularly in fields such as ophthalmology with complex workflows. Improving clinical workflow requires being able to predict which patients will have lower, average, or greater visit length, but there is no easy way to do this. Data analytics using secondary electronic health record (EHR) data is a possible way to address this gap. This study developed analytical models to predict provider-patient interaction time in pediatric ophthalmology clinics at Oregon Health & Science University (OHSU) Casey Eye Institute.
The study data was collected from office visits from 2015 to 2016 at OHSU. Time-stamps and office visits related data were abstracted from the enterprise-wide clinical warehouse. Also, audit log timestamp data was used to calculate the provider-patient interaction time. Multiple linear regression and random forest regression models were developed to predict the patient-provider interaction time. The random forest classification model was used to predict clinic complexity: these predictions were characterized as “short” (fastest 20%), “medium”, or “long” (slowest 20%), and were compared to predictions by an expert pediatric ophthalmology provider. We used 10-fold cross-validation to reduce over-fitting.
The R2 of the multiple linear regression was 21%, and 20% of the variability of the provider interaction time could be explained by the random forest regression model. Random forest classification had an accuracy of 65% in classifying patient-provider interaction time as short vs. medium vs. long, and accuracy of expert provider prediction was 41% (Table 1). Area under receiver operating characteristic curves was 0.722 in the multiple linear regression model and 0.723 in random forest regression model.
Secondary use of EHR data can be applied to support analytic models to predict patient-provider interaction time, and provided more accurate prediction results than an expert provider in this study. This has potential to improve clinical scheduling and efficiency in the future.
This abstract was presented at the 2019 ARVO Annual Meeting, held in Vancouver, Canada, April 28 - May 2, 2019.
Table 1: The random forest model classifies the clinic complexity more accurately than the provider
This PDF is available to Subscribers Only