Developing an algorithm to accurately predict nurse-treatment-room staffing.

620 Background: Tennessee Oncology (TO), a community practice with over 300 providers at almost 70 sites, relies on registered nurses (RN) to staff infusion-treatment rooms. Traditional paper or spreadsheet-based staffing models and retrospective data reviews have proved inadequate. TO partnered with Smirta Inc. to develop a data-driven, predictive algorithm for RN staffing. Methods: Reference data were collected from 30 clinics with treatment rooms ranging from 4 to 44 chairs. Inputs included electronic-medical-record (EMR) infusion length, acuity scores, and daily patient volumes. Cancer versus non-cancer visit ratios were used to adjust for expected acuity differences. Time-and-motion studies captured non-patient-facing RN tasks at two high-volume clinics on four separate days. The algorithm was trained on three months of data (Sep 1 - Nov 30, 2024) and its predicted staffing levels were compared with actual staffing needs. Five clinics of varying size were selected for prospective model testing. Performance was assessed by how often predicted RN-to-acuity ratios met or exceeded target staffing levels; “acceptable deviation” was defined as any prediction that erred toward over- rather than under-staffing, as long as the over-staffing was within established reasonable bounds. Results: Across all clinics, predicted staffing matched target acuity ratios in 12 of 25 comparisons (48 %) and met the acceptable-deviation threshold in 19 of 25 (76 %). Large clinics (>15 chairs) performed best: 12 of 15 (80 %) exact matches, rising to 15 of 15 (100 %) with acceptable deviation. Small clinics (<15 chairs) met targets in 1 of 10 comparisons (10 %) and reached 4 of 10 (40 %) with acceptable deviation. None of the tested scenarios resulted in predicted understaffing. Five successive recalibrations produced similar findings. Conclusions: The algorithm reliably predicted RN staffing needs in larger clinics but was less accurate in smaller sites. Two factors likely explain the discrepancy: (1) time-and-motion studies were conducted only in large clinics, and (2) small-sample variability (“law of small numbers”) amplifies deviations when nurse counts are low. Additional time studies at smaller clinics and mathematical adjustments are underway to improve performance across all practice settings.