Simulating the effects of reducing transfer latency from the intensive care unit on intensive care unit bed utilization in a Korean Tertiary Hospital
Article information
Abstract
Background
Latency in transferring patients from intensive care units (ICUs) to general wards impedes the optimal allocation of ICU resources, underscoring the urgency of initiatives to reduce it. This study evaluates the extent of ICU transfer latency and assesses the potential benefits of minimizing it.
Methods
Transfer latency was measured as the time between the first transfer request and the actual ICU discharge at a single-center tertiary hospital in 2021. Computer-based simulations and cost analyses were performed to examine how reducing transfer latency could affect average hourly ICU bed occupancy, the proportion of time ICU occupancy exceeds 80%, and hospital costs. The first analysis evaluated all ICU admissions, and the second analysis targeted a subset of ICU admissions with longer transfer latency, those requiring infectious precautions.
Results
A total of 7,623 ICU admissions were analyzed, and the median transfer latency was 5.7 hours. Eliminating transfer latency for all ICU admissions would have resulted in a 32.8% point decrease in the proportion of time ICU occupancy exceeded 80%, and a potential annual savings of $6.18 million. Eliminating transfer latency for patients under infectious precautions would have decreased the time ICU occupancy exceeded 80% by 13.5% points, and reduced annual costs by a potential $1.26 million.
Conclusions
Transfer latency from ICUs to general wards might contribute to high ICU occupancy. Efforts to minimize latency for all admissions, or even for a subset of admissions with particularly long transfer latency, could enable more efficient use of ICU resources.
INTRODUCTION
Critical care is a scarce resource, and demand for it is increasing [1,2]. As healthcare spending on critical care rises and the need to control healthcare expenses grows, effective utilization of intensive care unit (ICU) beds is becoming more and more important [3,4]. However, delays in transferring patients from ICUs to general wards occur in up to 25% of transfers, hindering the efficient use of ICU resources [3,5].
Latency in patient transfers from ICUs to general wards increases ICU occupancy, which can adversely affect patient outcomes and increase hospital costs. Studies have shown that inappropriately high ICU occupancy is associated with suboptimal quality of care, increased unplanned readmissions, and increased inpatient mortality [6-8]. Johnson et al. [3] estimated that an additional $21,547 per week was spent due to transfer latency in a 20-bed surgical ICU. The most common reason for prolonged transfer latency was a lack of non-ICU beds, followed by a lack of beds for patients requiring infectious precautions [3,5].
The frequent occurrence of delays in transfers from ICUs to general wards and its negative consequences suggest the need for active reduction. However, no study has yet explored the potential effects of reducing transfer latency. Therefore, we assessed ICU transfer latency and quantified how reducing transfer latency could be expected to affect ICU occupancy and hospital costs.
MATERIALS AND METHODS
Ethics Statement
This study was approved by the Institutional Review Board of Samsung Medical Center (No. 2022-06-153-001). Informed consent was not required because only previously collected, de-identified data were used.
Setting
We conducted a retrospective cohort study using data from January 1 to December 31, 2021, at a single center tertiary hospital with a 1,985-bed tertiary hospital with 129 ICU beds in Seoul, South Korea, to evaluate the transfer latency of patients aged 18 or older discharged from ICUs to non-ICU beds. Our hospital has 102 adult ICU beds in three medical ICUs and five surgical ICUs in two interconnected buildings. The ICUs operate in a semi-closed model with a team of attending physicians, intensivists, critical care fellows, residents, and nurses. Medical ICU admissions are made by intensivists, and surgical ICU admissions are made by anesthesiologists and surgeons. Depending on the situation, an attending physician from the general ward or an intensivist can be responsible for the patient. When an attending physician is not a member of the critical care team, intensivists in charge of the ICU actively consult and support decisions. Decisions to transfer patients from ICU to general wards are made through discussions between intensivists and the physicians or surgeons who will care for the patient after transfer.
Transfer Process at the Study Hospital
Once the physician in charge of an ICU patient decides that the patient is ready to be discharged to a non-ICU bed, nurses fill out the transfer request form in the electronic medical record. Upon completion of each transfer request form, the date and time of the request are captured, and the physician in charge of the patient outside the ICU is asked to make a room assignment based on the patient’s care needs and bed availability in the general wards. When patients are discharged from ICUs, nurses click the ”discharge” button in the electronic medical record to capture the ICU discharge date and time. If a transfer is not completed on the same day the request is made, nurses will fill out a new transfer request form every day until the request is fulfilled.
Data Collection
For each ICU admission from January 1, 2021, to December 31, 2021, the first ICU-to-non-ICU transfer request date and time and the actual ICU discharge date and time were collected to measure transfer latency. Only consecutive requests made each day were considered valid. If requests were not consecutive, we assumed that the original request was cancelled. Note that when a patient had multiple ICU admissions, each admission was counted separately, as each occupies an ICU bed and has its own transfer latency. We also collected additional data: ICU admission date and time, hospital admission and discharge dates, receiving ward (i.e., the unit the patient was transferred to from an ICU, categorized into three groups—medical, medical/surgical, and surgical—depending on the department in charge of the unit), demographic data (age and sex), clinical information (comorbidities, Sequential Organ Failure Assessment [SOFA] score at ICU admission, use of mechanical ventilation, use of extracorporeal membrane oxygenation, use of continuous renal replacement therapy, and use of vasopressors), and the presence of infectious precautions at the time of ICU discharge.
We used the ICU admission and discharge dates and times to compute the ICU length of stay (LOS) for each patient. The hospital admission and discharge dates were used to compute each patient’s hospital LOS. We also recorded the capacity of each ICU at midnight of each day. Due to the changing coronavirus disease 2019 (COVID-19) pandemic status in South Korea, the number of ICU teams dispatched to COVID-19 wards fluctuated throughout 2021, which caused ICU capacities to fluctuate as well.
Transfer Latency
The length of transfer latency from an ICU to a general ward was defined as the time between the first transfer request timestamp and the ICU discharge timestamp for each patient. To examine whether patients who experienced a longer transfer latency differed from those who experienced a shorter transfer latency, we examined patient characteristics after dividing them based on whether their transfer latency was longer than 24 hours. To examine whether and how much transfer latency was affected by the origin ward, destination ward, time of transfer, and special needs, we computed summary statistics for transfer latency after grouping the patients by the sending ICU, the receiving ward (medical ward, medical/surgical ward, or surgical ward), the transfer request day of the week, the transfer request hour of the day (12:00 am–11:59 am or 12:00 pm–11:59 pm), and the number of infectious precautions.
Simulation Approach
We used computer-based simulations to study the effects of reducing transfer latency on the utilization of ICU beds. Using the ICU admission and ICU discharge timestamps of all patients during the study period, we derived the actual hour-by-hour occupancy level of each ICU during the study period, which served as the benchmark. We then considered three scenarios for simulation: (1) no transfer latency, (2) transfer latency capped at 12 hours, and (3) transfer latency capped at 24 hours.
To simulate the scenario with no transfer latency, we modified each patient’s ICU discharge timestamp to be equal to their first transfer request timestamp. To simulate the scenario with transfer latency capped at 12 hours, we modified the ICU discharge timestamps of patients whose transfer latency exceeded 12 hours to be equal to their transfer request timestamp plus 12 hours. Similarly, to simulate the scenario with transfer latency capped at 24 hours, we modified the ICU discharge timestamps of patients whose transfer latency exceeded 24 hours to be equal to their transfer request timestamp plus 24 hours. Supplementary Figure 1 indicates an illustration of the simulation approach. Matrix Laboratory was used to compute both the actual occupancy and occupancy simulated under the different scenarios.
To ensure the correct measurement of ICU occupancy, we focused on March 2021 to December 2021 to report our simulation results. When deriving the hour-by-hour occupancy level of each ICU, we used patient flow data from all 8,998 ICU admissions. However, when conducting the simulation analyses, we modified the ICU discharge timestamps of only the 7,623 ICU admissions in our final dataset (Figure 1). Our outcomes of interest for the simulation experiments were as follows: (1) change in the average hour-by-hour occupancy of each ICU compared with the benchmark and (2) change in the proportion of time that the utilization of each ICU exceeded 80% compared with the benchmark.
Patient cohort selection. ICU: intensive care unit.
Cost Analysis
Following the approach described by Johnson et al. [3], we estimated the costs associated with transfer latency by calculating the difference between the daily cost of an ICU bed and a non-ICU bed. We selected medical ICU A as representative of our hospital’s eight ICUs. Using itemized billing data for patients treated in Medical ICU A in 2021, we estimated the average daily cost of both ICU and non-ICU beds. We then multiplied the cost difference by the total number of transfer latency hours to estimate the total cost of transfer latency. For currency conversion, we used the 2021 average exchange rate of 1,114.60 Korean won (KRW) to 1 United States dollar (USD).
RESULTS
Patient Cohort
We used a dataset of 8,998 ICU admissions from January 2021 to December 2021. Figure 1 illustrates our patient cohort selection. Patients with inter-ICU transfers, in-ICU mortality, direct discharges to home or other hospitals, inaccurate or missing data, or discharges after December 2021 were also excluded. The remaining 7,623 ICU admissions composed the final dataset. Table 1 provides summary statistics for the patient characteristics of all ICU admissions in the final dataset, ICU admissions whose transfer latency did not exceed 24 hours, and ICU admissions whose transfer latency exceeded 24 hours. The mean age of patients admitted to the ICU was 61.9 years, and 59.4% of the admitted patients were male. Admissions to medical ICUs composed 18.9% of the admissions (1,444 admissions). In-hospital mortality was observed in 3.7% of admissions (283 admissions), and the ICU readmission rate was 8.1% (620 admissions). Patients who experienced a transfer latency longer than 24 hours had a higher SOFA score (P<0.001), a higher in-hospital mortality rate (P<0.001), and a longer hospital LOS (P<0.001) than those with a shorter latency. Additionally, medical ICU patients were more likely to experience delays over 24 hours compared to surgical ICU patients (P<0.001).
Characteristics of patients with transfer latency less than and more than 24 hours
Transfer Latency
Figure 2 illustrates that 82% of the first transfer requests occurred between 6 am and 10 am, and 82% of ICU discharges occurred between 11 am and 4 pm. First transfer requests reached their peak (41%) from 7 am to 8 am, and ICU discharges reached their peak (24%) from 1 pm to 2 pm, suggesting about a 6-hour transfer latency in usual practice. Table 2 provides the summary statistics for transfer latency. The mean transfer latency was 14.5 hours, and the median was 5.7 hours. The mean transfer latencies in medical and surgical ICUs were 35.9 and 9.5 hours, respectively. The mean transfer latency was longer than 20 hours on Sunday and Monday and longer than 40 hours if the first transfer request occurred in the afternoon or evening (12:00 pm–11:59 pm). The most pronounced difference was observed when the presence of infectious precautions was used to group the admissions. The mean transfer latency of patients requiring infectious precautions was 75.8 hours, more than six times longer than the mean of 12.0 hours for patients without infectious precautions. The most common reason for infectious precautions was vancomycin-resistant enterococcus, followed by carbapenemase-producing Enterobacteriaceae and tuberculosis.
Percentage of intensive care unit (ICU) admissions, ICU-to-ward transfer requests, and ICU discharges by hour of day.
Summary statistics for transfer latency, in hours
Simulated Effect of Shortening the Transfer Latency of All Patients
Figure 3 illustrates the capacity, actual occupancy, and simulated occupancy from the simulation assuming no transfer latency in medical ICUs and surgical ICUs. Note that sometimes the actual or simulated occupancy exceeds the capacity because the capacity was measured at midnight, and the daytime capacity can be greater than the midnight capacity. We observe that the anticipated effect of removing transfer latency on reducing ICU occupancy is more significant in medical ICUs than surgical ICUs. The capacity, actual occupancy, and simulated occupancy of the individual ICUs assuming no transfer latency are provided in Figure 4.
Simulation 1 results: actual occupancy versus simulated occupancy assuming no transfer latency in (A) medical intensive care units (ICUs) and (B) surgical ICUs.
(A-H) Simulation 1 results: actual occupancy versus simulated occupancy assuming no transfer latency in each intensive care unit (ICU).
Table 3 summarizes the simulation results. The mean simulated reduction in the hourly occupancy of all ICU beds was 14.7 beds with no transfer latency; the medical ICUs and surgical ICUs were shown to have 7.7 and 7.0 fewer occupied beds, respectively. The mean simulated reduction in the hourly occupancy of all ICU beds was nine beds with transfer latency capped at 12 hours and 7.4 beds with transfer latency capped at 24 hours.
Simulation 1 results: effect of shortening the transfer latency of all patients on the average occupancy and proportion of time ICU utilization exceeds 80%
The observed proportion of time that the hourly occupancy of all ICU beds exceeded 80% was 37.2%. Simulation results showed that it could be reduced to 4.4% with no transfer latency and 13.1% with transfer latency capped at 24 hours. As with the reduction in hourly occupancy, the proportion of time ICU occupancy exceeding 80% was greater in medical ICUs than surgical ICUs. In medical ICUs, it decreased from 81.6% to 12.2% with no transfer latency and 25.4% with transfer latency capped at 24 hours. In surgical ICUs, it decreased from 24.7% to 7.1% with no transfer latency and 16.5% with transfer latency capped at 24 hours.
Simulated Effect of Shortening the Transfer Latency of Patients with Infectious Precautions
Table 2 shows that patients with infectious precautions at the time of ICU discharge experienced markedly longer transfer latencies than other patients. Although ICU patients requiring infectious precautions accounted for only 3.9% of the 7,623 admissions (298 admissions) in our final dataset, they accounted for 20.4% of the total transfer latency hours experienced by the 7,623 admissions (22,588.4 of the 110,533.5 hours). If it is too challenging to reduce the transfer latency for all patients, hospitals might consider intervening for just the subset of patients whose transfer latency reduction would have the greatest effect. Therefore, we conducted simulation analyses in which we modified the ICU discharge timestamps of only the 298 ICU patients requiring infectious precautions at the time of ICU discharge.
Table 4 summarizes those simulation results. The mean simulated reduction in the hourly occupancy of all ICU beds was 3.9 beds with no transfer latency, and medical ICUs and surgical ICUs were shown to have 2.9 and 0.9 fewer occupied beds, respectively. Compared with the mean simulated reduction of 14.7 beds when the transfer latency of all admissions was removed (Table 3), the reduction of 3.9 beds suggests that removing the transfer latency of only patients with infectious precautions at the time of ICU discharge could achieve about 26% of the reduction in occupancy that could be attained by removing the transfer latency of all admissions.
Simulation 2 results: effect of shortening the transfer latency of patients with infectious precautions on the average occupancy and proportion of time ICU utilization exceeds 80%
Similarly, the simulation results in Table 4 show that the proportion of time that the hourly occupancy of all ICU beds exceeded 80% could be reduced from 37.2% to 23.7% with no transfer latency for patients requiring infectious precautions, which is 41% of the reduction expected when the transfer latency is removed for all admissions.
Estimated Costs of Transfer Latency
In our study hospital, the estimated cost of a day in an ICU bed was 2,676,666 won or $2,338.52 at the average KRW to USD exchange rate in 2021. The estimated cost of a day in a non-ICU bed was 1,139,997 won or $995.98. Thus, the difference in costs was $1,342.54 per day. The total number of transfer latency hours was 110,533.5 hours (4,605.6 days) for all ICU admissions and 22,588.4 hours (941.2 days) for ICU admissions with infectious precautions at the time of ICU discharge. Therefore, the cost of ICU transfer latency in the study hospital in 2021 was estimated to be $6,183,202.22 per year or $16,940.28 per day for all ICU admissions and $1,263,598.65 per year or $52,649.94 per day for ICU admissions with infectious precautions at the time of ICU discharge.
DISCUSSION
In this retrospective observational study, routine documentation of timestamps for transfer decisions and actual transfers enabled a detailed investigation of the transfer latency of patients transferred from ICUs to general wards. We found that more efficient use of ICU resources might be possible by reducing the transfer latency for all admissions or even just patients with infectious precautions at the time of ICU discharge, who had a particularly long transfer latency.
Previous studies have proposed suggestions to reduce transfer latency from ICUs, such as direct discharge from the ICU and active identification of discharge-ready patients in general wards [3,9]. However, to the best of our knowledge, this study is the first to conduct simulation experiments to quantify how reducing transfer latency could be expected to affect ICU occupancy.
In our study, the median transfer latency was 5.7 hours, and 14.4% of transfers took more than 24 hours. Different studies have reported a range of transfer latencies, from 13.6% to 24.8%, depending on how transfer latency was defined [3,5,10-13]. Ofoma et al. [13], who used the definition of transfer latency most similar to ours, observed a median transfer latency of 4.8 hours, with 13.6% of the transfers requiring more than 24 hours, and those findings are also comparable to ours.
Our simulation results suggest that by eliminating transfer latency for all ICU admissions, 14.7 more beds could be made available across 129 ICU beds, and the proportion of time that the utilization of all adult ICU beds exceeds 80% could be reduced by 32.8% points. Although pinpointing the optimal occupancy for all ICUs is challenging, multiple studies have suggested that ICU occupancy higher than 70 to 80% is not ideal [14-17]. High ICU occupancy can lead to the cancellation of elective surgeries, increased patient severity in ICUs, and premature discharges that in turn could negatively affect hospital mortality and ICU readmission rates [15,18-21]. Our simulation results suggest that reducing transfer latency can be a lever for improving patient outcomes by improving ICU occupancy.
Our study shows that the effect expected from reducing transfer latency was more pronounced in medical ICUs than surgical ICUs, possibly because medical ICU patients had longer transfer latency than surgical ICU patients. Unlike medical ICUs, where most admissions are unplanned, surgical ICU admissions are typically planned in the preoperative assessment. Additionally, surgical ICU admissions generally follow a predictable postoperative course after ICU discharge, barring unexpected complications. This predictability allows for more accurate forecasting of bed availability and timely bed preparation, potentially contributing to shorter transfer latency for surgical ICU admissions. Most previous studies have focused on transfer latency in surgical or mixed ICUs; further studies on the differences between medical and surgical ICUs could provide additional insights [3,5,13,22-24].
Several factors contribute to prolonged transfer latency from the ICU to general wards. Previous studies have identified the lack of appropriate beds on general wards as a major cause of transfer delays [3,5]. For example, Johnson et al. [3] reported that the most common reason for delays was the unavailability of rooms that met infectious contact precautions. Other factors included changes in the primary service (e.g., from surgery to medicine), lack of available patient attendants, and delays in transportation services. While our study does not directly analyze these causes, our simulation of reduced transfer latency for patients requiring infectious precautions indirectly suggests that limited room availability for such patients may be an important contributor.
Our study hospital follows the Centers for Disease Control and Prevention guidelines and requires that patients infected or colonized with pathogens be treated in isolated rooms [25,26]. Due to the scarcity of isolated non-ICU rooms, patients requiring infectious precautions experienced a transfer latency more than six times longer than patients who did not require such precautions. Consequently, although patients under infectious precautions represented only 3.9% of the total ICU admissions, they accounted for 20.4% of the total transfer latency hours. We showed that intervening with this specific subset of patients, who had the longest transfer latency, could lead to significant improvements overall. This finding suggests that hospitals can implement targeted strategies to manage transfer latency.
In many countries, including South Korea, the cost of critical care services is expected to increase while overall healthcare budgets remain constrained [27,28]. Our study suggests that measures to improve transfer latency would be associated with more efficient use of limited ICU resources, creating additional ICU beds without physically increasing capacity. Such measures can target only a subset of ICU admissions; our findings suggest that focusing on patients with the longest transfer latency can yield substantial improvement. Further studies are needed to identify effective strategies for reducing transfer latency and assess their effects on patient outcomes and medical resource utilization.
The following limitations should be considered when interpreting our results. First, we used data from a single tertiary hospital. Transfer processes and latencies can vary in different health care settings. Second, our data did not include information on the reasons for or appropriateness of ICU admissions, discharges, transfers, transfer delays, and cancellation of transfer requests. Evaluation of these factors—along with considerations such as unnecessary ICU admissions, life-sustaining treatment decisions, and general ward bed availability—could provide further insight into improving ICU utilization. Third, fluctuations in ICU and general wards capacity and staffing due to the COVID-19 pandemic may have influenced some transfers in our study, although the effect was likely limited given the separate operation of the COVID-19 wards. Fourth, the infectious surveillance strategy was not standardized across the ICUs in our study hospital. It was mandatory only for patients in medical ICUs. Fifth, the simulation in our study needs to be supported by actual implementation. Subsequent studies are needed to apply different strategies to reduce transfer latency and measure the real-world effects. Sixth, our study does not assess the impact of transfer delays on patient outcomes. Future research could examine whether and how delayed transfers affect outcomes such as ICU readmission rates and in-hospital mortality, which would provide additional insight into improving ICU and hospital resource utilization.
In conclusion, ICU transfer latency can contribute to high ICU occupancy. Efforts to reduce transfer latency for all admissions, or a subset of admissions with particularly long transfer latency, could enable significantly more efficient use of ICU resources.
KEY MESSAGES
▪ By eliminating transfer latency for all intensive care unit (ICU) admissions to 129 ICU beds, the proportion of time that ICU occupancy exceeds 80% could be reduced by 32.8% points, and hospital costs could be reduced by $6,183,202.22 per year.
▪ By eliminating transfer latency for patients requiring infectious precautions, the proportion of time that ICU occupancy exceeds 80% could be reduced by 13.5% points, and hospital costs could be reduced by $1,263,598.65 per year.
▪ More efficient use of ICU resources can be expected from efforts to reduce transfer latency for all admissions or a subset of admissions with long transfer latency.
Notes
CONFLICT OF INTEREST
No potential conflict of interest relevant to this article was reported.
FUNDING
None.
ACKNOWLEDGMENTS
This study was supportd by the Institute of Management Research at Seoul National University.
AUTHOR CONTRIBUTIONS
Conceptualization: JC, SHK, REK, CRC. Methodology: SHK, REK, GYS, CRC. Formal analysis: SHK, REK, JHY, CMP. Data curation: JC, SHK, GYS, CRC. Visualization: JC, SHK, REK, JC. Project administration: GYS, JHY, CMP, JC, CRC. Writing - original draft: JC, SHK. Writing - review & editing: JC, SHK, REK, CRC. All authors read and agreed to the published version of the manuscript.
SUPPLEMENTARY MATERIALS
Supplementary materials can be found via https://doi.org/10.4266/acc.002976.
Simulation approach with no transfer latency and latency capped at 24 hours. ICU: intensive care unit.
acc-002976-Supplementary-Fig-1.pdf