Forecasting Urgent Care Patient Volume using Traditional Techniques and Machine Learning to Improve Resource Allocation
Open AccessLengthy patient wait times in an urgent care clinical setting can likely be attributed to a number of factors such as the complexity of the patient case, the length of time required for one or more procedures, the speed and skill of the provider, and the volatility in “walk-in” patient arrivals. Given the relatively extreme variability in patient visits to an urgent care clinic, balancing loosely anticipated quantities of patients on an hourly, daily, or even weekly basis with expensive medical staff and material resources to serve patients quickly and safely has been mostly ineffective. Additionally, historical datasets have been limited given their focus on one or few clinics, one primary geography, or have applied analysis techniques to datasets reflective of hospital emergency department (ED) visits rather than those at urgent care clinics. Consideration of independent predictors for patient demand have also been limited to fairly well reviewed features such as calendar and weather variables.This research focuses on leveraging a dataset that includes daily patient volume for a major urgent care organization with clinics in many geographies across the United States. It intends to present an evaluation of the relationship between patient demand and new external data features such as population density, average income, online patient reviews, and unique web-based keyword search terms. First, to evaluate the potential usefulness of identified external features (e.g., density, income, reviews, search terms, weather, weekdays, holidays), a simple correlation analysis served as a preliminary screening and down-select of features. Subsequently, the remaining features were modeled as independent variables against patient demand using a variety of traditional (e.g., multiple linear regression, autoregressive integrated moving average [ARIMA]) and nontraditional (e.g., machine learning) multivariate analysis techniques to determine accuracy and viability as a beneficial forecasting construct.The application of both traditional and nontraditional techniques led to a comparative performance assessment with research using similar independent calendar variables. The result was a demonstration of strong improvement to model accuracy with the introduction of select Google keyword search terms. Given the easily accessible nature of web search data via such avenues as Google Trends and other time- and location-tagged datasets, there is potential for leveraging thoughtfully selected search terms as predictors of patient demand at an urgent care clinic for primarily short-term planning. Such forecasting can then be used by clinic administrators and team members in a more effective and efficient allocation of staff and material resources to meet the variable patient demand levels seen on a daily, weekly, or seasonal basis.
- Correlation Regression Autoregressive Integrated Moving Average (ARIMA) Random Forest and Artificial Neural Networks (ANN) for Patient Demand Forecasting
- Demand Forecasting Using Machine Learning
- Reducing Patient Wait Times with Forecasting
- Traditional and Machine Learning Approaches to Demand Forecasting
- Urgent Care Patient Forecasting
- Improving Urgent Care Clinic Resource Allocation with Demand Forecasting
- All rights reserved
Notice to Authors
If you are the author of this work and you have any questions about the information on this page, please use the Contact form to get in touch with us.