Online Program

Healthcare Staffing Schedules: An Application of Studying Variation Instead of the Mean to Answer a Real-World Question (305345)

*Evan Carey, Colorado School of Public Health 
Luke William Patten, Colorado School of Public Health 
Michelle Feller, UCHealth Resource Management Center 
Sharon Pincus, University of Colorado School of Medicine 
Sylvie Novins-Montague, University of Colorado School of Medicine 

Keywords: Bayesian regression, simulations, interactive dashboard, decision aid, float pool, contingency pool, staffing

Sometimes, studying only the mean does not cut it. While the mean is still informative, it may not provide the most appropriate answer for real-world problems. High-value analytics provide the ‘right information’, to the right person, at the right time. We present the development of an analytic tool to optimize scheduling for a nurse staffing contingency pool, where we used probabilistic programming conditioned on prior data to estimate and provide the ‘right information’. A contingency pool of staff is used to cover staff shortages (PTO, call-ins, etc.), which are a stochastic process and prior data may inform future staffing decisions. The ‘right information’ to represent staff shortages includes both the center and the dispersion of this process. We used a Bayesian approach to fit a negative binomial regression model to identify trends with respect to monthly and weekday variability. Using MCMC simulations to identify the variability in the shortages, we estimated the contingency pool size needed to cover the expected demand 50%, 80%, and 90% of the time. We designed and implemented an interactive decision tool to effectively communicate these results to hospital management.