Article Text
PDF
Abstract
Objective In-person healthcare delivery is rapidly changing with a shifting employment landscape and technological advances. Opportunities to care for patients in more efficient ways include leveraging technology and focusing on caring for patients in the right place at the right time. We aim to use computer modelling to understand the impact of interventions, such as virtual consultation, on hospital census for referring and referral centres if non-procedural patients are cared for locally rather than transferred.
Patients and methods We created computer modelling based on 25 138 hospital transfers between June 2019 and June 2022 with patients originating at one of 17 community-based hospitals and a regional or academic referral centre receiving them. We identified patients that likely could have been cared for at a community facility, with attention to hospital internal medicine and cardiology patients. The model was run for 33 500 days.
Results Approximately 121 beds/day were occupied by transferred patients at the academic centre, and on average, approximately 17 beds/day were used for hospital internal medicine and nine beds/day for non-procedural cardiology patients. Typical census for all internal medicine beds is approximately 175 and for cardiology is approximately 70.
Conclusion Deferring transfers for patients in favour of local hospitalisation would increase the availability of beds for complex care at the referral centre. Potential downstream effects also include increased patient satisfaction due to proximity to home and viability of the local hospital system/economy, and decreased resource utilisation for transfer systems.
- Health services research
- Simulation
- Efficiency, Organizational
Data availability statement
Data are available upon reasonable request. Data requests should be directed to the corresponding author.
This is an open access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited, appropriate credit is given, any changes made indicated, and the use is non-commercial. See: http://creativecommons.org/licenses/by-nc/4.0/.
Statistics from Altmetric.com
WHAT IS ALREADY KNOWN ON THIS TOPIC
Discrete event simulation is commonly used in healthcare operations to model patient flow within an institution.
WHAT THIS STUDY ADDS
This project applies simulation to patient movement within a multisite healthcare system to understand how potential changes in healthcare delivery may impact bed availability.
HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY
Using simulation in this way may represent a convenient approach to identify the effects of proposed changes on patients and resource utilisation.
Introduction/background
Rural healthcare systems represent an area of care disparity related to access to specialty services compared with their urban counterparts. Impact of this disparity has been measured in terms of outpatient care1 as well as inpatient specialty access and outcomes.2–5 Patients from less resource-rich areas frequently need to be transferred to specialty centres located distant from their home for a higher level of care and specialist interventions. Additionally, constraints on bed capacity—particularly staffing limitations—often prompt patient transfers to facilities outside the patients’ communities. Transfers are associated with high resource utilisation, creating additional stress realised as extended length of hospital stay (LOS) and a nearly doubled cost of care,6 ambulance use for long-distance trips rather than emergency care, coordination of care at dismissal to include transportation home, and detrimental effects on patient families/social environments due to distance from home.
Two major disruptions to healthcare in the USA at this time are staffing shortages and the rapid implementation of telemedicine services in the wake of the SARS-CoV-2 pandemic. Surveys have shown that nurses are considering leaving the profession at increasing rates,7 while demand for nurses is projected to continue to grow through 20318—indicating an even greater shortage of staff on the horizon. Meanwhile, the USA public health emergency declaration eased the path to providing care remotely using video solutions.9 Many health systems rapidly deployed telehealth while outpatient practitioners and patients remained at home, using video-based care for in-hospital consultation, and other novel uses to fill gaps created by the COVID pandemic.
In our regional health system, local staffing challenges—often more prominent at small facilities with small pools of staff, restricted specialty resources, limited equipment and expertise for specialised procedures contribute to limited ability to care for patients who reside in the area. This results in transfers to regional hubs. Identifying safe ways to keep patients in their home communities involves ensuring that there is adequate staffing and expertise available to meet their needs.
Discrete event simulation is an established tool within operations to investigate systems and flow. Healthcare discrete event simulation has been used to look at patient movement through hospitals and to specific services.10–13 These simulations step through time and allow for populations of patients to be modelled with different characteristics as they traverse the system. Simulation models are best constructed using institutional contextual knowledge to make the correct assumptions about the system.
Our study aims to answer the question of how to right-size and right-staff a rural hospital system to accommodate local patients using a test scenario based on the constructed simulation model. Using computer simulation, we gain insights into the effects of transfers on the receiving facility, and what resources are regularly dedicated to patients originating from the broader catchment area but outside the immediate community of the referral centre, and identify capacity needs to accommodate additional patients locally. While local contexts and conditions may vary, we posit that discrete event simulation is a valuable tool when coordinating care across a large geography.
Methods
Setting
Our study is set in a regional health system in the Midwestern United States, consisting of 1 academic referral centre, 3 regional hubs with specialty care, 4 mid-sized hospitals with limited specialty care and 10 critical access (CA) hospitals that offer only general medical services. The focus of this study is on patients presenting to the emergency department (ED) at any of the community hospitals who then were transferred to other facilities. Where possible, we adhere to the Strengthening Reporting of Empirical Simulation Studies (STRESS) guidelines for reporting empirical simulation studies.14 This project was initiated as a quality and practice improvement initiative and was exempt from Institutional Review Board (IRB) review.
We focus on general hospital internal medicine (HIM) and non-procedural cardiology patients, those who require only medical management (CVD-MM). These represent the largest volume of patients transferred and a population for which interventions could be implemented to impact care delivery. Regional hub hospitals have multiple subspecialty consult services available and interventional cardiac capabilities (eg, cardiac catheterisation laboratory). Mid-sized hospitals have few medical or surgical subspecialists available for consultation, whereas CA sites offer only general medical services. Often, patients are transferred due to capacity issues within the local facilities—not enough staffed beds or staff are available to accommodate all local patients who would be appropriate for the level of care offered.
Patient and public involvement
Patients were not directly involved in the development of this project; however, the experience of the patient is a strong consideration in this work. Being treated close to home and loved ones is a patient-oriented outcome of reducing patient transfers that can be safely mitigated. This reduces costs associated with driving, parking and other daily living expenses that would otherwise not be incurred in the home community.
Logic
Diagrams of the base model are demonstrated in figures 1 and 2. The model was populated using transfer data from each site. Transfer arrival distributions were created from the time intervals between arrivals. Data were normalised to reflect the most current data with stable arrival patterns and sufficient quantity to calculate statistically viable distributions. For the model, transferred patients were randomly assigned to one of 37 services based on the aggregate transfer data. 36 services had at least 35 observations and were retained as separate nodes for the model, whereas those services which had fewer than 35 observations were collapsed into one group. Transfer data were used to create statistically viable hospital LOS distributions for each individual service node. The model then tracked the number of patients in the academic centre for each service with a count at the end of each simulated day. This information was saved as output for each simulation entity at the end of each simulated day.
Baseline model in which patients are seen at an outlying hospital and are determined to need transfer to another facility for ongoing care either due to specialty or bed capacity constraint. Individual sites are coded with region number (1–4) and individual site within the region (a–d). Hub refers to a large community hospital with specialty services. CA, critical access; MS, mid-size.
Test scenario: hospital internal medicine (HIM) and non-procedural/critical cardiovascular patients stay within community hospitals, either the original site or a larger regional hospital. Individual sites are coded with region number (1–4) and individual site within the region (a–d). Hub refers to a large community hospital with specialty services. CA, critical access; MS, mid-size.
The test scenario explored modifications of patients being transferred versus remaining in local facilities and used the same distributions as the base model for services that were not tested in the scenario.
Algorithms used for LOS, arrival distribution and transfer of cardiac patients are presented in tables 1 and 2. Model entities include a patient entity for each site and a counter entity to record information at the end of each day. Entities arrive at each site and then transfer to either the academic centre or their regional hub. Once at another hospital they occupy a bed resource for a period of time before departing the simulation. There is no limit on the hospital bed capacity and there are no queues. Entry and exit points are present for all patient entities when arriving and departing the hospital bed. There is one additional entry point for the end of day record and one additional exit point for cardiac patients that did not transfer during the experiment simulation.
Hospital length of stay distributions by service
Baseline simulation results for all specialties transferring into the academic centre
Community site arrival distributions and cardiac transfer probabilities
Data
Data from the electronic health record were used, encompassing all patient transfers between June 2019 and June 2022. A sample set of data consisting of 10 patient encounters was manually evaluated to ensure appropriate variables were selected and anomalies would be detected. A dictionary of clinical services was created to consolidate similar services that use different notation across sites (eg, ‘Gastroenterology’ vs ‘Gastroenterology and Hepatobiliary’).
Regional practice leaders were engaged to guide the simulation development. Based on concerns for transfers occurring often due to hospital capacity/staffing rather than a need for specialty expertise, HIM and family medicine (FM) were identified as a high yield area to pursue, based on simulation results and in collaboration with regional practice leaders. Cardiovascular diseases (CVD) were also included to determine what the impact of developing a telehealth programme for cardiac patients not requiring procedures would be. In our system, most patients requiring subspecialty care or specialty-specific procedures are admitted by the relevant service. For example, patients with a congestive heart failure exacerbation are admitted to a CVD service, and those requiring a cholecystostomy tube are typically admitted to a gastroenterology team. Patients requiring general medical care are admitted to HIM or FM.
All HIM and FM transfers were included. A subset of patients with CVD was selected to remain local, based on their final diagnoses. Patients who had diagnoses that would be anticipated to need a procedure, such as pacemaker/defibrillator placement, or cardiac catheterisation, would still transfer.
Input parameters include distributions of transfers from each site. Patients were assigned one of 37 clinical services based on the transfer data. Bed time occupied, defined as the length of time that a patient was in a bed, was extrapolated from the transfer data. LOS distributions were built for each service type, enabling us to model viable natural distributions for each service. Arrival patterns were normalised for the most recent time period with stable arrival rates, but did not include growth of patients, thus the model is largely a reflection of patient movement during the first half of 2022.
There was no data manipulation for missing data. Some early arrival data were removed to represent a stable period of time.
The simulation assumes that the rate of transfers (per cent of patients who require transfer to another facility) for each service is the same regardless of site. There were not sufficient numbers to model each site separately; however, we were able to cohort regional hospitals together. The simulation model does not take into account time of day for discharge and does a census once a day for statistical reporting purposes. It is noted that this is not a real-world modelling of discharge behaviour, and thus the results are likely slightly undercounting the number of staffed beds being used by transferred patients, by a small amount.
For validation, a sample set of data was manually reviewed to ensure appropriate variables were selected and anomalies in location recording and patient record validity could be detected. Additionally, transfers detected with our system were compared with the records from our internal transfer coordination centre, which are tracked using a different system. We found that there was excellent correlation between the two data sets.
Experimentation
For initialisation, bed usage variables started at zero. A warm-up period of 300 days was used to stabilise the number of beds being occupied by the various specialties after which daily totals were collected. Initial model experimentation determined that 300 days of warm-up was sufficient to ensure that all specialties were represented within the simulation. The model is terminating, stopping after 3650 days with 10 iterations for a total of 33 500 days of observations after removal of the warm-up period. The model collects information at the end of each simulated day, collecting the number of patients for each specialty occupying a room on that day and noting the location of that patient. The CIs for the mean number of beds occupied were adjusted to account for autocorrelation relating to the number of beds being occupied on consecutive days. Patients in the specialties HIM, FM and CVD-MM were analysed to see the impact of them staying at their originating facility to determine the number of beds which would be required at those locations and the number of beds which could be potentially freed at the academic centre. There were no capacity constraints on the number of beds available. While typically there would be capacity constraints on the beds occupied at the regional sites, the purpose of this experiment was to give insight on the capacity that would be needed if these patients had not transferred. This output data were analysed using SAS V.9.4 (SAS Institute) and Excel 365 (Microsoft, Redmond, Washington).
Implementation
Arena V.14.0 Professional was used for the simulation. Arena’s input analyser was used to build arrival distributions and LOS distributions. Arena is an industry-standard simulation software tool15 and has an established use in healthcare-related problems.16 The model was executed as a discrete event simulation. Calculations were run on an Intel Core i5-10500 CPU @ 3.10 GHz Code access/Windows 10 Enterprise; run time of 3650 days with 10 iterations.
Ethical considerations
All data used to create the simulation were retrospective protected health information and secured as such. While this project was initiated in the context of practice improvement and exempt from IRB review, patient privacy and data security were undertaken similar to research.
Results
25 138 unique encounters resulting in a transfer that occurred during the inclusion period were incorporated into the model. Tables 1 and 2 demonstrate the probabilities calculated for LOS for all services, arrival rate at community hospitals and likelihood of transfer for patients with CVD. 36 services were found to have greater than 35 encounters and were individually assessed. Those with fewer than 35 encounters were collected as the ‘Rare’ category (table 3). There were 7138 HIM/FM and 602 CVD encounters.
The simulation was run for a total of 33 500 days to establish baseline performance with no modifications. This resulted in a mean of 122.48 (median=121, 95% CI 121.1 to 121.85) beds used by patients transferred from community sites to the academic centre and an observed range of 80–176 beds per day (table 4). The number occupied by patients originating in different regions showed a clear skew towards patients in the most immediate geographical locations, Region 1 (R1). The fewest patients came from the Region 3 (R3). Clinical services that used the highest number of beds per day were CVD, critical care medicine and HIM.
Daily beds occupied by all transferred patients by region, medically managed cardiovascular patients (CV-MM), family medicine patients (FM), hospital internal medicine patients (HIM) and a cohort of these three groups combined
When the test simulation was run, CVD-MM patients (table 4) occupied a mean of 9.21 (median=9, 95% CI 9.11 to 9.31) beds per day with an observed range of 1–24 beds per day. When broken down by region, R1 was again the highest user, with a mean of five beds (median=5, 95% CI 4.97 to 5.02, range=0–16) occupied per day at the academic centre by patients who originated in that region.
The FM service (table 4) did not use a meaningful number of beds in the baseline data and did not prompt any consideration of further investigation.
HIM bed utilisation (table 4) was again dominated by the R1 using a mean of 8.79 beds per day (95% CI 8.69 to 8.9), Region 2 used 5.55 (95% CI 5.462 to 5.63), Region 4 used 2.32 (95% CI 2.26 to 2.37) and R3 used 0.65 (95% CI 0.624 to 0.67). The majority of patients were transferred to the academic centre from the nearest geographical region, R1. A mean of 17.3 (95% CI 17.16 to 17.45) HIM beds were occupied by R1 patients per day (minimum 3, maximum 36).
The analysis of potentially replacing transfers with local services across HIM, FM and CVD-MM (table 4) resulted in a mean opening of 26.95 (95% CI 26.77 to 27.13) beds in the academic centre (minimum 8, maximum 50) per day. The effect of patients staying in their home facilities was spread among the 17 hospitals, requiring 0–31 additional staffed beds per region needing to be available to accommodate them. Of note, the facilities have the physical space and capabilities to accommodate this increase in beds. Staffing would need to be adjusted to accommodate opening additional beds in some cases, but no additional bed licences or space would be required.
Among the distributions of services there were similarities, particularly among the less used services. Services with a similar mean tended to have a similar maximum. Stochasticity, or randomness, played a role in the outcomes. As the mean increases, the variance increases. Adding to this finding is the Poisson distribution in these results. We use a count of a number of independent events at a point in time, yielding this result. As such, the Poisson distribution, the similarities of the means and maximum within our data, accounted for the similar distributions among some services.
Discussion
The number of bed-days occupied by patients who are being cared for outside their home area is a non-trivial challenge to health systems. There is a cumulative effect on hub sites, particularly tertiary care centres taking referrals from a large catchment area, such as the academic centre in our system. Delivery of complex care can only be provided at centres with robust subspecialty services. However, improvements in technology, shared electronic medical records and ease of video consultation may reduce the need for patient movement particularly when no specialty procedures or equipment are required.
Our investigation demonstrates accumulated bed-days for patients who do not need hands-on subspecialist input and is intended to be a surrogate for patients who have relatively low infrastructure needs. Our selection of non-procedural CVD-MM patients could be serviced using local monitoring and nursing skills, with remote access to cardiology consultation. Patients requiring critical care, those in need of any procedure are excluded. Incremental resources would be needed to provide these consultative telehealth services along with local resources to increase the capacity. The cost of introducing a consulting cardiologist to care for patients across multiple locations may be balanced by opening capacity for medically complex care at the referral centre. Often, complex procedures are not viable to provide in a small hospital as volumes at smaller sites may be insufficient to maintain high-level multidisciplinary skills yielding the best possible patient care and outcomes.17–20
In reducing the work of patient movement, we may find that other systems have increased capacity for innovative work. Considering the transportation requirements for all transferred patients, a reduction in the number of patients being moved between sites will enable ambulance and patient transportation services to maintain focus on emergency response. This opens an opportunity to bolster human and vehicle resources in an area that is notoriously difficult to staff.21 22 Shifting clinical duties away from transportation of hospitalised patients could allow paramedics to engage in novel aspect of out-of-hospital care such as community paramedicine.23
Consideration of the patient experience is also important when reallocating resources. Studies have identified a preference for patients to be transferred to a skilled nursing facility closer to home after they have been hospitalised in a distant referral hospital.24 Choosing when to bypass a local rural hospital is a complex decision. Subpopulations demonstrate varying preferences for larger more remote facilities versus small rural hospitals.25 Identifying the role of patient preference when making changes to hospital offerings will be important.
We believe our analysis provides a good starting point for exploring these ideas and identifying target populations for potential improvement by highlighting where shifts in care to remote consultation may be appropriate and where a shift in staffing/bed availability will make an impact.
Thinking differently about our hospital systems and being thoughtful by planning and modelling interventions rather than mounting a pilot programme to gather preliminary data can be a relatively inexpensive and versatile tool to innovate in regional health systems. Identifying ways that care can be delivered differently and using technology to improve access in rural settings can benefit patients and systems. Once built, a model like ours can be adjusted to look at different specialties and aspects of care. Additional work validating the output of simulation models with real-world comparisons will be needed to ensure that they translate accurately to real-world scenarios. Idealised systems of abundant space, staff and equipment are much more easily built in a virtual world than in the physical one.
Limitations
A major assumption of our model is the surrogacy of procedures as the primary indication for transfer between sites. This does not take into account the potential need for specialised equipment, imaging or in-person evaluation that may prompt a patient transfer and represents a limitation to the model.
Smaller sites, CA hospitals, are collected into regions rather than analysed individually due to smaller numbers of patients that were too low for reliable analysis. This remains useful, as the level of care able to be provided at these sites precludes even simple cardiac patients from remaining there, and they would be transferred to regional centres regardless. Therefore, cohorting them together results in a bed change that would be applied to the regional hub, or hospital with the highest level of care available. How each site is cohorted is in alignment with regional transfer patterns that exist currently. For example, CA3 is a critical access hospital that is located nearest to Hub 3, and they are in the same regional operating group. The model reflects this close relationship but cohorting them.
Patients who are transferred tend to have a longer LOS than those who stay locally.6 If this holds true for our patient population, it is conceivable that fewer bed-days would be needed in local facilities to accommodate the larger number of beds spared in the referral centres. Additional investigation would be needed to understand the reasons for this and how it affects our system.
Our model was based on transfers from one site to another. We know that some patients elect to present to a referral centre ED rather than a local one and are ultimately admitted. This analysis does not include individuals who self-select but are limited to patients who are transferred from outlying facilities.
The simulation model does not consider seasonal or weekly variation, but this could be done with more advanced work.
The simulation model did not place capacity limits on the number of beds available in the regional sites as the purpose was to estimate what capacity would be required if patients did not transfer to the academic centre.
We recognise that this model assumes that the regional centres would be able to upstaff to accommodate additional patients. The census at each site varies, some sites have immediately available capacity on most days (due to transferring patients) and others are often near capacity—generally limited by staffing. We believe that the benefit of this model is for planning and while this is a limitation in one sense, but in another is informative for strategic planning.
Conclusion
Using retrospective data and incorporating assumptions on clinical trajectories we are able to determine the effects of shifting hospitalisation location for patients with different hospital-based needs. This information can be used to allocate resources and emphasise either growth or contraction of hospital bed capacity, update staffing models, procedural planning and provide more patient-centric services. Identifying ways to keep more patients in local hospitals is an opportunity to continue to bolster smaller towns with economies that often are highly linked to availability of local medical care.26 Future investigations may include evaluation of surgical patients to determine if transfer is needed for specialised care at the academic centre versus ability to care for patients locally.
Data availability statement
Data are available upon reasonable request. Data requests should be directed to the corresponding author.
Ethics statements
Patient consent for publication
Ethics approval
This is a quality improvement study and as such is considered exempt by our Institutional Review Board.
References
Footnotes
Contributors LW conceived the study and drafted the manuscript and guarantor, is responsible for the overall content. KK served as a scientific advisor and contributed to the manuscript. MJ assisted with data collection and processing and contributed to the manuscript. TH devised and performed simulations and contributed to the manuscript.
Funding The authors have not declared a specific grant for this research from any funding agency in the public, commercial or not-for-profit sectors.
Competing interests None declared.
Patient and public involvement Patients and/or the public were not involved in the design, or conduct, or reporting, or dissemination plans of this research.
Provenance and peer review Not commissioned; externally peer reviewed.