Jonasz Staszek

AI can unlock the most potential when processes are stable and predictable. In practice, this is frequently not the case. This is why some people pair up the capabilities of AI with other techniques. In my today’s post, I would like to tell you a story described in a great paper by Alessandro Agnetis, Caterina Bianciardi, and Nicola Iasparra. You can find a free pdf here. They joined the forces of Lean Thinking and Mathematical Optimization to improve the life of patients of the hematology ward of the Policlinico Santa Maria alle Scotte of Siena. Their efforts resulted in an up to 94% reduction in patient waiting times, with significant reductions across all patient groups.

The hematology ward of Policlinico Santa Maria alle Scotte of Siena admits approximately 10,000 patients per annum for diagnosis and treatments of blood diseases, mainly neoplasms. The ward suffered from sinking patient satisfaction. A field survey pointed out that the most problematic aspects pointed out by patients were environmental comfort and long waiting times.

While the former requires a complex response, the latter appeared to be a particular fit for a lean-approach. Hence, the team decided to work on the waiting times, by a review and redesign of appointment procedures. On a side note – they also hoped for an improvement in environmental comfort resulting from shorter waiting times.

The approach

The team structured this challenge according to the classical PDCA (Plan-Do-Check-Act) cycle.

Know your problem – “Plan”

During the “Plan” phase, the team collected the data on the existing processes within the ward. They listed seven types of therapies performed in their ward and determined the daily capacity for each of them. They also used statistical techniques to determine the distributions of their duration. Additionally, they listed the staff availability and requirements per each therapy type. To accurately reflect the resource requirements, the team then translated each of the therapy types into seven distinct paths, each defined by a (fixed) set of activities. They also established the critical success measures for their efforts – patients’ lead time, value-added time, and percentage of waiting time over the entire lead time.

Design the future with Lean and AI – “Do”

In the “Do” phase, the team performed the Root Cause Analysis of the problems. As a result, they figured out that a precise plan of daily treatments is required to shorten the waiting times. To that end, they needed to change it from a “push” to a “pull” paradigm: instead of giving out the treatment slots to patients as they arrived, an individual appointment had to be given in advance. That in turn is a great use case for Mathematical Optimization. Its techniques allow for fast selection of slots for the incoming patients while respecting the schedules already handed out to other patients and the capacity constraints of the ward.

Compare and contrast – “Check”

Before implementing the mathematical optimization model into the planning practice, the team needed to perform a “Check”. To test the performance of the new method, the team used the data gathered in the “Plan” phase. They also employed advanced simulation techniques to compare the waiting times under the old regime with the newly developed solution. The results were astonishing: the average waiting time reduction amounted to up to 94% for all patients. Additionally, the lead times turned out to be much less variable.

The main reason for such improvement is the shift from push- to pull-type patient flow management strategy. It is significantly different from the typical batch-and-queue logic of most healthcare processes. Mathematical Optimization then comes in to unlock the potential which is difficult to grasp by a human planner.

Make it happen – “Act”

Finally, in the “Act” phase, the team needs to carefully plan the implementation. It will require a mindset shift from all the stakeholders: the management, the medical staff, and the patients alike. The team will need to implement the mathematical model into the appointment scheduling software. I recently wrote a guide to scoping such a project. You can find it here.

Conclusions

Based on the authors’ experience, we can draw some conclusions:

1. Changing the planning paradigm from push to pull unlocks new, previously unavailable possibilities
   
2. Lean Thinking and Mathematical Optimization enable synergies that are unavailable if each technique is used separately
3. Mathematical Optimization may bring surprising improvements even for simple planning problems
   

Thanks for reading! Stay tuned for future posts!