Predicting the Future: Using Simulation Modeling to Forecast Patient Flow on General Medicine Units
BACKGROUND: Hospitals are complex adaptive systems within which multiple components such as patients, practitioners, facilities, and technology interact. A careful approach to optimization of this complex system is needed because any change can result in unexpected deleterious effects. One such approach is discrete event simulation, in which what-if scenarios allow researchers to predict the impact of a proposed change on the system. However, studies illustrating the application of simulation in optimization of general internal medicine (GIM) team inpatient operations are lacking.
METHODS: Administrative data about admissions and discharges, data from a time-motion study, and expert opinion on workflow were used to construct the simulation model. Then, the impact of four changes – aligning medical teams with nursing units, adding a hospitalist team, adding a nursing unit, and adding both a nursing unit and hospitalist team with higher admission volume – were modeled on key hospital operational metrics.
RESULTS: Aligning medical teams with nursing units improved team metrics for aligned teams but shifted patients to unaligned teams. Adding a hospitalist team had little benefit, but adding a nursing unit improved system metrics. Both adding a hospitalist team and a nursing unit would be required to maintain operational metrics with increased patient volume.
CONCLUSION: Using simulation modeling, we provided data on the implications of four possible strategic changes on GIM inpatient units, providers, and patient throughput. Such analyses may be a worthwhile investment to study strategic decisions and make better choices with fewer unintended consequences.
© 2019 Society of Hospital Medicine
Sensitivity Analysis
Overall, average time in system was most affected by the number of patient arrivals. This became particularly significant as the volume of patient arrivals approached and exceeded the capacity of the rounding teams. Adding a nursing unit had more impact on decreasing average time in the system than adding a medical team or aligning teams with NUs under the conditions defined by the model. However, under different conditions, such as increasing admission volume, the relative benefit of different approaches may vary.
DISCUSSION
Given that hospitals are large, complex systems,2 the impact of system-level changes can have unpredictable and potentially deleterious effects. Simulation provides a technique for modeling the impact of changes to understand the ramifications of these interventions more thoroughly.3 In this study, we describe the process of building a simulation model for the admission and discharge of patients from general medicine services in a tertiary care hospital, internally validating this model, and examining the outcomes from several potential changes to the system.
The outcomes for these what-if scenarios provided some important insights about the secondary effect of system changes and the need for multiple, simultaneous interventions. Given that hospitals often function at near capacity, adding a hospitalist team or nursing unit might be seen as a reasonable strategy to improve the system metrics, number of patient discharges, or average LOS. On the basis of our analysis, adding a nursing unit would have more benefit than adding a hospitalist team. Leaders who want to increase capacity may need to consider both adding a hospitalist team and a nursing unit, and model the impact of each choice as described with a simulation.
Additionally, assigning patients to medical teams aligned with NUs seems theoretically appealing to improve interprofessional communication and decrease the time spent in transit between patients by physicians. While our findings supported a decrease in rounding time and patient dispersion, the teams not aligned with a nursing unit (ie, the hospitalists) exceeded 80% utilization, the threshold at which efficiency is known to decrease.24 Potentially, benefits resulting from teams being aligned with NUs were offset by decrements in performance of the teams not aligned with NU. If medical teams and NUs become aligned, then a higher number of teams may be necessary to maintain patient throughput.
Simulation models identify these unexpected consequences prior to investing resources in a significant change; however, modeling is not simple. Simulation models depend on the characteristics of the model and the quality of the input data. For example, we used an expert approach to map physician workflow as an underpinning of the model, but we may have missed an important variation in physician workflow. Understanding this variation could strengthen the model and provide some testable variables for future study. Likewise, understanding nursing workflow and how variation in physician workflow shapes nursing workflow, and vice versa, is worth exploring.
Other data could also be added to, and help interpret, the outputs of this model. For example, the impact of various levels of team and unit utilization on diversion time for the hospital ED may help determine whether adding team capacity or unit capacity is more beneficial for the system. Likewise, aligning medical teams with NUs seems to hinder patient throughput on this analysis, but benefits in patient satisfaction or decreased readmissions might improve reimbursement and outweigh the revenue lost from throughput. Underpinning each of these types of decisions is a need to model the system well and thoughtfully choose the inputs, processes, and outputs. Pursuing a new strategic decision usually involves cost; simulation modeling provides data to help leaders weigh the benefits in terms of the needed investment.
The major limitations of the study stem from these choices. Our study focused on matching capacity and demand while limiting other changes in the system, such as changes in nursing unit LOS. Future work to quantify the relationship of other variables on parameters, such as the impact of decreased team dispersion on LOS, early discharges, and decreasing care variation, would make future models more robust. This model does not consider other strategies to improve patient flow, such as shaping demand, adaptive team assignment algorithms, or creating surge capacity. We also used only hospitalist time and motion data in our model; housestaff workflow is likely different. In addition, we modeled all patients as having a general level of nursing care and did not account for admissions or transfers to intensive care units or other services. These parameters could be added in future iterations. Finally, the biggest limitation in any simulation is the underlying assumptions made to construct the model. While we validated the model retrospectively, prospective validation and refinement should also be performed with attention to how the model functions under extreme conditions, such as a very high patient load.
