Study: AI staffing cuts costs, keeps access
Columbia Business School published a study finding AI‑driven nurse staffing can better match staffing to demand while cutting costs and preserving patient access. The finding frames AI here as an operational forecasting tool—not a bedside decision replacement—and points toward staffing optimization as a near‑term, high‑impact application. That makes staffing analytics a realistic project area for clinicians who want to show measurable operational results. (business.columbia.edu)
Hospitals usually set nurse schedules weeks ahead, a little like stocking a grocery store before knowing how many shoppers will show up on Saturday. Columbia researchers tested whether a forecasting system could do that job better inside an emergency department. (business.columbia.edu) The study looked at emergency department staffing, where patient arrivals can swing sharply by hour and day. In that setting, too many nurses on shift raises labor costs, and too few nurses on shift pushes up waits, walkouts, and burnout. (nature.com) Most emergency departments already use a two-step staffing system. They set “base” schedules in advance, then add “surge” nurses later when the department gets busier than expected, often at higher incentive pay. (nature.com) The Columbia team built a two-stage prediction model for that exact workflow. The model forecasted patient volume first for the base schedule and then again for the surge decision, using real-time information instead of relying only on historical averages. (business.columbia.edu) Then they implemented it in a large adult emergency department and compared results before and after the change. The paper measured door-to-evaluation time, active treatment time, boarding time, length of stay, left-without-being-seen rate, and hourly nurse staffing cost. (nature.com) The cost result was concrete: hourly staffing costs fell by $162.04. Columbia’s April 7, 2026 press release said that works out to roughly $1.4 million in annual savings for a single emergency department. (business.columbia.edu) The access result was just as important: the study found no negative effect on throughput after the new system went live. In plain terms, the department spent less on staffing without slowing the flow of patients through the emergency room. (nature.com) The paper also tested what happens when staffing drops below the model’s recommendation. Cutting one nurse per hour added about two minutes of wait time, and falling more than 20 percent below the recommendation added another 2.3 minutes. (nature.com) That is why this study is about operations, not a robot replacing bedside judgment. The system predicts how many patients are likely to arrive, and managers still decide how to staff the floor around those forecasts. (business.columbia.edu) The authors were Yue Hu, Carri W. Chan, Jing Dong, Alice Kazekjian, Chayapol Ophaswongse, Gregory Sugalski, Joseph P. Underwood, and Rimma Perotte, and the article was published in May 2025 in npj Health Systems. The press release arrived almost a year later, on April 7, 2026, as hospitals across the United States keep looking for tools that can cut premium labor spending without closing the front door to patients. (nature.com) (business.columbia.edu)
