This blog is written by the Summer 2021 Shift Scheduling Team: Dipra Debnath, Madelaine Emesden, Arlo Halpern, Yvonne Lin, and Shraddha Ramesh.
The first thing the Shift Scheduling team did this summer was learn how to use the scheduling tool. We did this by watching our staff lead and scheduling wizard, Billy Pozehl, build the Block 2 schedule on Zoom. When we receive inputs from the Chief Resident, we need to make edits so that the information is digestible for the code. He walked us through this process and then told us to “make a schedule we personally think is good.” This could be different for each team member based on what they prioritize. For example, what is worse: denying a resident’s vacation or having multiple bad sleep patterns? During the following meeting, we had to explain why we thought our schedule was good and should be used. This was not only an effective way to learn how to use the tool but also how to communicate the important aspects of the schedule to the Chief Resident.
As part of our learning process for how to build schedules, the team then built actual resident schedules for Mott Children’s Emergency Department. So far, the students have built schedules for three blocks. Something that we’ve learned through this process is that, even though the schedule generation is automated, there is still an empathetic human element in play when determining how to improve schedules and which schedules to ultimately use. For instance, we might get a schedule that has a very small number of total denied vacations and bad sleep patterns, but if the majority of those metrics are lumped onto one individual resident’s schedule, then we will try to do everything we can to more equitably distribute the metrics. We found that scheduling the blocks two at a time, with one long, two-block schedule, led to improved schedules with fewer bad sleep patterns and denied vacations per block.
One of the most important metrics in the Shift Schedule Building Tool is bad sleep patterns. This metric measures when a consecutive shift assignment for a given resident would result in them having an unhealthy sleep schedule. In practice, this often looks something like working a night shift one day, followed by a day shift on their next assignment. However, some of these bad sleep patterns that we measure are more severe than others. Going from an 11 pm shift to a 5 pm shift two days later is classified as a bad sleep pattern but is far less severe than going from an 11 pm to a 9 am two days later. This summer we have been working on a system of dividing bad sleep patterns into three separate tiers, ordered by severity, so that we can more effectively distinguish between these different types of bad sleep patterns when making schedules.
Documenting the process is also an important part of the project. To standardize the documents and processes in each project, the CHEPS Optimization Process team has come up with templates for the optimization projects to implement. As part of this effort, we have created the background slides and created a rules list that would help the new members of the team. Making the document on the rules we got to understand how each of the rules is implemented and used in the tool.
Lastly, non-IOE students or anyone who needed a refresher on IOE 310, Optimization and Computational Methods, were given the opportunity to take the course in a bootcamp format. This course is taught by Dr. Amy Cohn, and was split into 5 modules: Optimization Problems, Modeling, Graphical Approach and BFSs, The Simplex Method, and Integer Programming, all of which are very applicable to our SHIFT scheduling project! In addition to watching lecture videos and completing problem sets individually, we also had live sessions open to all of CHEPS to work through extra problems together and ask questions.