AI: Automating healthcare workforce planning

The shortage of healthcare workers is a social challenge that must be properly addressed. Pradtke GmbH teamed up with the Bochum Institute of Technology gGmbH and contec GmbH in the research project titled "AI-powered healthcare workforce planning and management (KI-unterstützte Personaleinsatzplanung und-steuerung im Gesundheitswesen, KI-PEPS).

In this MEDICA-tradefair.com interview, André Pradtke talks about the basic concept of the research project, details the goals of the project participants, and highlights the project results.

Mr. Pradtke, what is the idea behind the KI-PEPS research project?

André Pradtke: We wondered how we could mitigate staffing shortages and ease the burden on strategic planners at the same time. That means we studied how we can retain and engage the existing staff, while meeting the needs and demand amid the drastic shortage of workers. The challenge is to optimize competencies and avoid tying up strategic planners in healthcare organizations unnecessarily with workforce planning processes. These specialists have very explicit skills profiles, and our goal is to ensure they can perform tasks that put their unique know-how to the best use.

Our current software tools already offer the support that’s needed in this setting. However, we strongly believe artificial intelligence (AI) can create added value and increasingly automate this process.

How can AI support healthcare workforce planning?

Pradtke: For now, the AI-powered process doesn’t achieve better results than the existing rule-based techniques we are currently using. However, I am still convinced the process has the potential to provide an enormous boost based on the results we have seen already.

I believe rule-based approaches have reached their full potential. Meanwhile, AI is only at the beginning of its development. We will quickly see how the method achieves similarly excellent results as linear optimization. And though it may take a few months to get there, AI technology has so much more potential to drive capacities.

The project's funding has ended. What were the objectives and project results?

Pradtke: The goal was to implement a pilot operation with a prototype in the facilities to measure the quality we can achieve with the new mechanism and continue to optimize it. We created a prototype, used it in the healthcare facilities and collected feedback from early users.

From this vantage point, we already reached an objective we intended to achieve. Having said that, I also want to be transparent and point out that we had hoped to be further along at this stage from an algorithmic perspective. One reason for the shortfall was the onset of the COVID-19 pandemic in 2020, which prompted clinical facilities to primarily focus on infection prevention for about three months. We lost this time in our research, causing us to miss our targets towards the end.

Yet this also made us realize we must lay a much firmer foundation to build a successful machine learning model. These are exciting but also challenging processes one must go through to gain the insights needed to progressively implement AI in this setting. You achieve success on this path because you learn things you would otherwise not have learned or realized you had to understand to move forward. We are very happy we have already progressed this far.

We are also determined to continue the project along this trajectory, preferably with full funding because we are certain to have laid the groundwork for excellent future systems.

What was the feedback from pilot users?

Pradtke: All users are very interested in the solution. I think it was important for them to see that the process garners actual results. Admittedly, we were also told that the created schedules cannot be fully executed in their current form. They still require too much rework at this stage compared to other algorithmic techniques we are presently using.

That said, our customers realize that this is a research project and not a development project in the traditional sense. They appreciate the difference and gladly participate in our endeavor. After all, only by experimenting were we able to understand the full complexity of the task at hand. From this perspective, we are delighted with our interim results, especially since it has shown the relevance of our tryouts, taking us one step closer to a successful solution.