Hospital physicians can’t get no long-term satisfaction – an indicator for fairness in preference fulfillment on duty schedules

Abstract

Physicians are a scarce resource in hospitals. In order to minimize physician attrition, schedulers incorporate individual physician preferences when creating the physicians' duty roster. The manual creation of a roster is very time-consuming and often produces suboptimal results. Many schedulers therefore use model-based software to assist in planning. The planning horizon for duty schedules is usually a single month. Many models optimize the plan for the current planning horizon, without taking into account data on preference fulfillment and work load distribution from previous months. It is therefore possible that, when looking at a longer time horizon, some physicians are disadvantaged in terms of preference fulfillment more often than their peers, simply because this generates better results for the individual months. This may be perceived as unfair by the disadvantaged physicians. In order to eliminate this imbalance, we introduce a satisfaction indicator for preference fulfillment in physician scheduling. This indicator is computed for each physician on each monthly plan and is then used to inform decisions regarding preference fulfillment on the current and future plans. As a result, a more equal distribution of preference fulfillment among physicians is achieved. We run a computational study with three different update strategies for our satisfaction indicator. Our study uses 24 months of data from a German university hospital and derives additional generated data from it. Results indicate that our satisfaction indicator, combined with the right update strategy, can achieve an equal distribution of satisfaction over all physicians within a peer group, as well as stable satisfaction levels for each individual physician over a longer time horizon. As our main contribution, we identify that our satisfaction indicator is more effective in creating equal distribution of long-term satisfaction the higher the rate of conflicting preferences is.

Long-term workload equality on duty schedules for physicians in hospitals

“…For more literature on the topics of this work, we refer to respective literature reviews on inequity averse optimization (Karsu and Morton, 2015), staff scheduling (Ernst et al, 2004), nurse scheduling (Cheang et al, 2003), and physician scheduling (Erhard et al, 2018). Further literature on fairness in preference fulfillment can also be found in Gross et al (2018a).…”

“…In terms of preference fulfillment, it is possible that some physicians are repeatedly denied their duty requests whereas other physicians are repeatedly granted their requests. Gross et al (2018a) show that using a model that does not equalize preference fulfillment over several planning horizons creates unequal preference fulfillment between physicians. With these models, the fulfillment of preferences for an individual physician -and therefore this physician's satisfactionis based on solver implementation details.…”

“…With these models, the fulfillment of preferences for an individual physician -and therefore this physician's satisfactionis based on solver implementation details. Gross et al (2018a) propose a satisfaction indicator for preference fulfillment to mitigate this problem and equalize preference fulfillment over all physicians over several planning horizons. However, they do not take into account duties which have not been requested via a preference but must still be assigned to ensure adequate staffing of the hospital.…”

“…This work has two main contributions. First, we propose a workload indicator, modeled similarly to the satisfaction indicator by Gross et al (2018a). Second, we evaluate the impact of our workload indicator.…”

“…We incorporate both the existing satisfaction indicator and our proposed workload indicator into a physician scheduling model. The model is then compared to the model with only the satisfaction indicator and to a model with only the workload indicator by applying it to the data used by Gross et al (2018a). We generate additional data with a varying number of preferences and find that the effectiveness of the workload indicator is tied to the number of preferences when used together with the satisfaction indicator.…”