Improving Nurse Staffing Planning: A Data Driven Discovery of LPCH Pediatric ICU Nurse-to-Patient Assignment Rules 

In the LPCH PICU, current staff level planning is based primarily on the annual average midnight patient census, which does not account for variations due to seasonality, day-of-the-week peaks, and midday churn. Consequently the shift level sees last-minute adjustments that are in check with nurse-to-patient ratios (above and over California’s minimum 1:2) based on staffing guidelines, patient procedures, and acuity. To improve staffing methods we interviewed charge and floor nurses, PICU leadership, and staffing office managers to map out the current staffing process and nurse-derived guidelines, validate historical safe nurse staffing, and potentially uncover “hidden” staffing guidelines. 

We identified patient characteristics in the electronic health record, and used admits, discharges, and staff timecard data to simulate staffing levels with current guidelines. We devised three models using 1) monthly budgeted nurses to determine staffing, 2) simulating nurse-derived staffing guidelines, and 3) a combination of staffing guidelines and regression modeling. By performing regression on the error between historical staffing and current guidelines, we were able to identify specific patient procedures that are driving nurse staffing that warrant additional staffing and shared our recommendations to improve staffing methods with PICU clinicians and administration. Notably, 9 of the top 10 coefficients identified in regression analysis were related to patient churn.

The hybrid staffing guidelines and regression simulation notably performed better than the other two modeling strategies. We saw that overall goodness of fit improved when we tested this regression model generalized to new data from July 2019 to December 2019. The resulting predictions were added to previous simulation results, and we saw an overall improvement in the simulation error on the unseen testing data.

The results show that though charge nurses in the PICU were factoring in patient and unit level factors such as patient churn into their staffing decisions, our method of retrospectively applying those expert judgment guidelines to the patient census and then analyzing the difference with actual staffing levels will allow nurse managers to identify “hidden” patient and unit characteristics that better estimate nurse workload and staffing needs. Using patient acuity and historical trends can also improve patient safety and staff satisfaction along with better outcomes for patient flow throughput, cancellation frequency, and under/over-staffing costs. 

Our next steps include building a forecasting tool to better anticipate PICU staffing needs at the shift level. For LPCH, our method can allow analysts and nursing staff to gather data from units regarding staffing decision rules to determine how well predicted staffing needs based on existing rules meet actual staffing availability. 

We would like to thank Dr. David Scheinker, Dr. Timothy Cornell, Dr. Andrew Shin, Dr. Christos Vasilakis, Saied Mehidan, and the leadership, administration, clinicians, and staff of the Lucile Packard Children’s Hospital who provided us with data and extensive feedback through interviews.

Jonathan Ling 

MS in Management Science and Engineering, Operations Research and Analytics

Shira Winter 

Postdoctoral Fellow, Stanford School of Medicine and VA Palo Alto

Nicolai Ostberg

MS in Biomedical Informatics

MD student at Stanford School of Medicine

Sreeroopa Som 

BS in Management Science and Engineering