How Predictive Analytics in Nurse Staffing Saves One Hospital $1M Each Year

But first, exhaustive and painstaking data-gathering had to be done, and Tower needed help, Agnew says. They worked with management consulting company Kaufman Hall to do the initial exhaustive and painstaking data gathering.

“You need to have the right partners, and the right plan, and the right expertise,” Agnew says. “This is a heavy lift.”

Gathering the data

Tower Health began its work with Reading Hospital, gathering data from payroll, staffing grids, the historical census—including the hourly census—for the previous three years for each individual unit, bed capacity, nonproductive time, turnover rate, vacancy rate, hours used for FMLA (Family and Medical Leave Act), nurse-to-patient ratio, and all other essential information to put into a workforce optimization engine, she says.

“It was time intensive and very comprehensive. You really have to gather everything,” Agnew says. “Your staffing is impacted by dozens of other variables other than hours per patient day.”

With the goal to level over- and under-staffing, Tower collected hourly censuses at different hours of the day seven days a week, she says.

“The average hourly [census] in the ER is very different on a Sunday morning than on a Friday evening,” she says. “You can’t rely on the ‘flaw’ of averages; it doesn’t give you the data you need.”

After putting three years of historical data for each unit into the optimization model, the tool created a model for each unit, and identified what core staffing the hospital system would need to fulfill scheduling demands, Agnew says.

The next part of the equation was deciding what portion of their workforce would be flexible.

“You have your core workforce, which is lean, and then your flexible workforce, which is there for unpredicted variations,” she says. “If a bus [full of patients] pulls up or [there’s] some other spike in census not anticipated, you have the ability to deploy a flexible workforce. That model gets rid of the variation that causes over- and under-staffing.”

Previously, subsets of time were not delineated by shifts, but rather in terms of two- and four-hour increments, Agnew says, offering a heat map illustration to explain how staffing patterns were off balance.

“Let’s say the heat map is green and you’re good [with staffing] and yellow is on the edge, and red is not enough staff,” she says. “We would see green, orange, red, and yellow all in the same day. On a Monday morning, it might be green all morning and orange at noon, which means we’re overstaffed in the morning and understaffed in the afternoon.”

Predictive analytics provided a completely different way of scheduling, Agnew says.

“It took a lot of work away from nurse manager or director who previously constructed that schedule,” she says.

“It was great and also problematic. Nurse directors wanted to have that control and put extra people where they thought they needed them or to make deals where some nurses would work just certain days,” she says. “Special deals are not part of this. That went away.”

When managers and directors saw on the heat map how the new patterns improved staffing, they became believers, she says.

“When [they] saw the heat map, it was a real eye-opener,” she says.

An economical staffing model

The new staffing model decreased the number of core staff, but grew the flexible staff “a great deal,” Agnew says.

To be most economical and efficient with their new staffing model, they needed to increase the number of “point 6,” or part-time, positions.

That required some creativity, Agnew says.

“When we tried to entice people to take those point 6 positions, a lot said, ‘We would love to work point 6 … but from a financial standpoint, we can’t afford to,’ or the biggest issue was they would have higher premiums for health benefits, which was a deal breaker,” Agnew says.

“So, we created a flexible schedule for work-life balance. We paid an additional incentive, an hourly incentive, that defrayed the cost of … higher premiums for part-time benefits,” she says. “Many people took advantage of it and appreciated it.”

Filling those point 6 positions resulted in $1 million a year in savings at Reading Hospital, Agnew says.

The new staffing model has resulted in increased job satisfaction, both for nurses who wanted a more flexible work schedule and for core staff, called the Tower Select team, who wanted more variety in their work.

“When we created the Tower Select team … some of them wanted to work on other units to have something new and different,” she says. “They received incentive to do that because they are going to different units and receiving different education and training.”

Turnover rates improved and Tower has received positive employee feedback, Agnew says.

That positive feedback can partly be attributed to the fact that Tower employees were part of the staffing project from the beginning, she says.

“We didn’t want this to be top down,” she says, so the project included a group of peers making decisions with health system leaders about work rules, processes, and scheduling.

“It was a great exercise in building collaboration and teamwork and that was the reason for the success of the project: engaging people early on,” she says.

Continually tweaking the plan

The new staffing model has been worth the data-gathering and hard work required to implement it, Agnew says.

“It was just a much more objective, data-driven process; more efficient and not based on emotion,” she says. “It gave people predictability so they knew the number of hours they were getting.”

“As we improved and tweaked the models the last few years and gotten data down, we’ve gotten more precise in our staffing,” she says. “Capacity may change, so you have to consistently remodel every year and continue to keep it current.”

“If you have a bus that pulls up with 30 people, the flex staff is not going to be able to cover it,” she says. “But on daily predictions, we have the right number and a system of deploying them.”