Geographic Cohorting – An Industrial Engineering Approach to Reducing Waste

Tze Chao Chiam, PhD, Lori Pelletier, PhD, and Richard Forster, MD

Download PDF


This study at UMass Memorial Medical Center (UMMMC) investigates various models and configurations to improve placement of general medicine patients from the emergency department (ED) and other sources into acute-care units in the hospital. After applying rules based on hard constraints, such as the needs of surgical patients and specialty patients, the general medicine patients are assigned randomly to available beds throughout the hospital. Physician teams are assigned to these patients without consideration of either their geographic location or the locations of the teams’ current patients. This practice introduces several potential inefficiencies and wastes to the system, including increasing physicians’ walk time to provide care, decreasing their “touch time” with patients, decreasing the team’s communication with nursing staff regarding patients’ plan of care, and delayed discharge planning.

Geographic cohorting involves placing general medicine patients and their physicians at specific geographic locations in the facility. It has been proposed as a possible solution to several challenges, while decreasing system waste. However, it could also potentially increase the use of the ED as a “holding spot” for several types of patients. The implementation of geographic cohorting poses several challenges, including assuring nursing competencies in the acute-care units, balancing admission quotas of physician teams in the teaching hospital, and managing bed and physician reassignment for patients needing transfers between acute-care units due to changes in their medical conditions.

This study investigates the effectiveness of the various levels of “cohortness” through the use of a discrete-event simulation which was subsequently implemented at the UMMMC site. Results from both the simulation and the implementation demonstrated promising trends including a decrease in physicians’ average walk time, a decrease in the average time between ED admission to arrival in acute care, and a decrease in patients’ length of stay.


Healthcare delivery is plagued with wastes that impact patient care. A Lean culture for process improvement has been pursued in nearly every industry and has generated remarkable gains in quality and efficiency by reducing unwanted variation and eliminating non-value added activities or wastes (Imai, 1986; Liker, 2004; Parks, 2003; Sharrock, 2007; Wood, 2004). Additionally, there is increasing recognition of the opportunity that industrial engineering and operations research tools may play in improving healthcare delivery by providing better decision support and data-based analysis (Benneyan, 1997; Khurma, Bacioiu & Pasek, 2008; Lawrence, 2005; Lee & Yih, 2010; Wu, Lehto & Yih, 2011).

This paper will focus on wastes associated with the processes of delivering care to general medicine patients being admitted to an inpatient setting, and the timeliness of care provided by hospital medical teams. A case study at UMass Memorial Medical Center (UMMMC) in Worcester, Massachusetts, has demonstrated that the inability to put the right patient in the right bed under the right caregiver is a contributing factor to extended inpatient wait times. General medicine patients, who constitute over 50% of the overall inpatient population, are geographically spread out across the large medical center for various reasons, including an inefficient algorithm for patient-bed assignment and patient-physician assignment. As a result, each physician team is responsible for patients who are scattered across various locations, and physician teams consistently need to walk long distances to provide treatment to the patients under their care. This geography barrier results in delaying treatment or care plan disposition and thus directly contributes to the problem of hospital overcrowding.

In this research, simulation is employed to evaluate different cohort levels and the effects on the wastes of unnecessary movement and waiting. It is believed that with a high level of cohort, patients belonging to the same general medicine team will receive more timely care and be less geographically scattered. The time saved from walking could potentially be allocated to increase patient-physician touch time or enable a more timely evaluation of patients boarding in the ED.


 As a result of overcrowding in hospitals there is often a delay in placing the right patient in the bed most appropriate for the patient’s medical needs. As a result, some patients become “boarders” in the ED while others are assigned to an available bed, regardless of whether the bed is appropriate for the patient’s diagnosis. Studies at a large multi-site acute-care teaching hospital in Ontario, Canada have shown that an ED length of stay greater than 12 hours is associated with a 12.4% longer inpatient hospital length of stay and 11% greater inpatient cost, translating into an estimated 2,183 additional inpatient days and a corresponding increase in inpatient costs (Huang, Thind, Dreyer & Zaric, 2010). Many studies have attempted to alleviate this patient flow issue through various solutions, including the addition of more hospital beds (Bazarian, Schneider, Newman & Chodosh, 1996; Moloney et al., 2006; Ross, Naylor, Compton, Gibb & Wilson, 2001) and more staff (Bucheli & Martina, 2004). However, with increasing demand to bring down costs and new reimbursement guidelines, hospitals are reluctant to acquire additional resources.

In this study, a computer simulation model was created to provide decision makers with an understanding of the extent of waste present in their system (in this case, waiting and unnecessary movement); different options to alleviate the waste and thereby provide more value to the patient; and a mechanism for continuous improvement after implementation.


This study used data provided by UMMMC, a private, non-profit integrated healthcare system in central Massachusetts.  The study was initiated in April 2011 as part of a Lean project with the goal to reduce waiting and unnecessary movement. A discrete event simulation was used to perform a feasibility study of geographically cohorting general medicine patients. Performance metrics that are important to the UMMMC include physicians’ walk time, discharge time, and the time between when a patient is admitted from the ED to an acute-care unit and when the patient is physically in the acute-care bed (referred to as “ED admit to head-in-bed” in this paper). These metrics are chosen because it is believed that with a reduction of walk time, physicians will be able to provide more value-added activities with their patients (i.e., treatment time). An earlier discharge time could also potentially indicate the ability to make a timely bed-assignment for a newly admitted patient. Also, the delay between ED admit to head-in-bed could have a direct impact on patient safety. (Fordyce et al., 2003; Kulstad et al., 2010). For this study, Arena Simulation package was used.

There are nine physician teams at the University campus. Each team typically consists of an attending physician, a resident and two interns. Each team has a patient panel that consists of a maximum of 20 patients, with approximately half of the patient panel belonging to each intern to ensure a balanced workload.

At the beginning of a typical day, residents and interns who work during the day shift attend a night sign-out session in a designated night sign-out room from 7 to 7:30 a.m., where hand-off of patients is performed between the night-shift physician and the day-shift physicians. From 7:30 to 8:15 a.m., residents attend morning report, which is an educational session where the night-shift resident presents a patient case from the night before. During this time, the interns start seeing their patients and executing the relevant tasks associated with each patient. At the beginning of each round, each intern is aware of his/her patient location and diagnosis as well as general information such as the demographics of each patient. From 8:30 a.m. to noon, attending physicians and residents join the interns on their team for rounds. At noon, residents and interns attend a conference session at a designated conference room until 1 p.m., after which they continue to round on their patients and follow-up on any necessary tasks. The geographical location of each activity as well as the patients directly impacts the walking time and distance of the physicians.

A three-week study was performed to examine the geographical locations of patients of all nine  physician teams. The typical percentage distribution of patients is shown in Table 1.

Unit Percent distribution
3E 28.13
3W 7.59
4Adm 9.38
4E 2.68
4W 0.89
5P 1.34
6E 19.64
6W 8.03
7E 10.71
7W 5.36
ED 6.25

Table 1: Percent distribution of patient locations for general medicine physicians based on three-week study.

Each unit has its own nurse staff, with the exception of 7E and 7W, which shares a common team. As a result 7E and 7W are represented as one unit in the simulation. Since adult general medicine patients typically are not placed in unit 5P as 5P is a pediatric floor, patients on this floor are removed in the simulation, and the percentage distribution of patient location is normalized as shown in Table 2. This distribution of patient locations is used as the base case in the simulation.

Unit Percent distribution
3E 28.51
3W 7.69
4Adm 9.51
4E 2.72
4W 0.91
6E 19.9
6W 8.14
7E 16.29
ED 6.33

Table 2: Normalized percent distribution of patient locations for general medicine physicians in simulation.


Patient type Priority level Percent distribution Associated tasks
Unstable patient 7 5 Calling consults, ICU transfer
Simple discharge patient 6 5 Gathering information, completing forms, family updates
Night admit in ED 5 10 Calling consults, gathering information, completing forms, performing history and physical assessments for admissions
Acute-care patient (regular) 4 39 Writing daily orders, calling consults, gathering information, completing forms, writing progress notes, family updates
Complex discharge patient 3 15 Gathering information, family updates, complex discharge paperwork
New morning admits 2 10 Calling consults, gathering information, completing forms, performing history and physical assessments for admissions
Acute-care patient (minimal care needs) 1 20 Writing progress notes

Table 3: Patient types, priority levels and associated tasks performed by physicians.

In the simulation, every patient is assigned a priority to be seen by the physician team. This priority and the percent distribution of each patient type are shown in Table 3. The percent distributions, indicating the volume of each patient type and the tasks performed by physicians associated with each patient type are obtained through interviews with physicians. Process times for the execution of each task are also obtained through interviews with physicians as well as through time studies.

Since the activities during rounds along with the geographic distribution of patients across all nine teams can be generalized, the simulation is set up to study the impact of the patient cohort on one team. The patient geographical locations are assumed to follow the distribution in Table 2.

A distance matrix is set up in the simulation using relative unit distances among various units based on actual physical locations of the units. It is assumed that the distance is minimal if a physician travels from one unit to the same unit (the case in which both patients are on the same unit) as shown in Table 4. There is no distance associated with traveling from night sign-out room to night sign-out room or from conference room to conference room as no patient care is provided in these rooms.

ED 3E 3W 4Adm 4E 4W 6E 6W 7E Night sign-out room Conference room
ED 2 6 6 8 8 8 12 12 14 12 12
3E 2 3 4 4 4 8 8 10 8 8
3W 2 4 4 4 8 8 10 8 8
4Adm 2 3 3 6 6 8 6 6
4E 2 3 6 6 8 6 6
4W 2 6 6 8 6 6
6E 2 3 4 3 3
6W 2 4 3 3
87E 2 4 4
Night sign-out room 0 0
Conference room 0

Table 4: Relative distances among units.

ED 3E 3W 4Adm 4E 4W 6E 6W 7E Night sign-out room Conference room
ED 1 3 3 4 4 4 6 6 7 6 6
3E 1 2 2 2 2 4 4 5 4 4
3W 1 2 2 2 4 4 5 4 4
4Adm 1 2 2 3 3 4 3 3
4E 1 2 3 3 4 3 3
4W 1 3 3 4 3 3
6E 1 2 2 2 2
6W 1 2 2 2
7E 1 2 2
Night sign-out 0 0
Conference room 0

Table 5: Time traveled between units.

A walk time matrix is also set up to capture the walk times by the physicians in the simulation as shown in Table 5. The walk times are average walk times captured through time studies, rounded to the nearest whole number.

Four scenarios are studied in the simulation against the base case. In order to be prepared for worst-case scenarios, all cases are set up to simulate the processes during a day when the hospital is operating at high census (all acute-care beds are occupied at the beginning of the day, i.e., 7 a.m.) and when the physician team is at a maximum load with a patient panel of 20. The simulation was run until all tasks were completed in each replication. A total of 30 replications were run, and the data was collected from each replication.

In the base case, patient geographical distribution is depicted in Table 2. Patient priority levels, percent distribution of each priority level, and associated tasks are described in Table 3. Relative distances among various units are shown in Table 4.

In all four scenarios it is assumed that 6.33% of the team’s patients are night admits who are still in the ED at the beginning of the day, as indicated by historical data. In Case 1, 93.07% of the patients are cohorted at unit 3E, whereas cases 2, 3 and 4 assume 80%, 60% and 40% of the patients are cohorted at unit 3E, respectively. The geographical distribution of the remaining patients in each case is obtained by normalizing the remaining percent distribution of each case against the base case distribution in Table 2. Unit 3E is arbitrarily chosen as the unit of choice to study the effect of cohorting patients in this paper. Table 6 shows the percent geographical distributions of patients in cases 1, 2, 3, and 4.

Unit Base case Case 1 Case 2 Case 3 Case 4
3E 28.51 93.67 80 60 40
3W 7.69 0 1.61 3.97 6.33
4Adm 9.51 0 2.00 4.91 7.83
4E 2.72 0 0.57 1.41 2.24
4W 0.91 0 0.19 0.47 0.75
6E 19.9 0 4.17 10.28 16.39
6W 8.14 0 1.70 4.21 6.7
7E 16.29 0 3.42 8.42 13.42
ED 6.33 6.33 6.33 6.33 6.33

Table 6: Percent geographical distribution of patients in cases 1, 2, 3 and 4.


Simulation Results

Simulation shows that a 93.67% successful patient cohort at unit 3E will result in several positive outcomes. The average daily walk time by the physicians decreased by 51.4%, from 37.67 minutes in the base case to 18.29 minutes. Sensitivity analyses show that there is a decrease in average walk time for cases 2 and 3 as well, as shown in Figure 1. Significance tests show that the decrease of average daily walk time in cases 1, 2 and 3 compared to the base case is statistically significant. Case 4, with 40% cohorted at 3E, did not show a statistically significant decrease compared to base case. Based on the analysis in this metric, the threshold above which geographically cohorting patients will bring positive outcome is a cohort of 40%.

Figure 2 shows the decrease of average time between an ED admit to head-in-bed at a pilot unit. Although the decreases in all cases compared to the base case are not statistically significant at the α=0.05 level, the data shows a desirable trend.

Implementation Results

Based on the simulation outcome as well as other analyses, the leadership team at UMMMC decided to implement the inpatient cohorting project on May 8, 2012.

Although the medical center experienced a few high-census days during the data collection period, the overall census during the preliminary study was not as high as the worst-case scenarios in the simulation. Thus, it is expected that the data collected from the preliminary study will appear more superior to the outcome from the simulation.

Data from implementation is analyzed. The outcomes of interest to this study include time between ED admit to head-in-bed at the pilot units, patients’ length of stay and discharge times.

Results show that there is a 20% decrease in the average time from ED admit to head-in-bed in an acute-care unit. This decrease also is statistically significant at α = 0.05 level. Discharge time is chosen as a metric because an overall earlier discharge will allow admitted patients to be transferred to the acute-care beds earlier in the day, thus reducing the time between admit to head-in-bed. Results show that the discharge time from acute-care units also has been moved earlier in the day as shown in Table 7. This improvement is also statistically significant at α = 0.05.

Changes in performance measure
Average time from admit to head-in-bed 20.0% reduction
Average discharge time 24 minutes earlier in the day
Average length of stay (day) 12.2% reduction

Table 7: Implementation results.

Figure 1: Average daily walk time.

Figure 2: Average time for ED patients getting an inpatient bed.

Analysis shows that there is a decrease in average length of stay by 12.2%. A decrease in length of stay also indicates freeing up beds to admit new patients. This decrease is attributable to the process changes that allow for efficient communication among care providers, specifically the physicians, nurses, and case managers.

Conclusion and Future Work

Industrial engineering models can provide both awareness of the magnitude of waste existing in processes as well as scenarios to maximize the trade-offs between value-added and non-value-added activities. In this study, the wastes of waiting and unnecessary movement by physicians were examined using discrete event simulation. In order to alleviate hospital overcrowding under the constraint of not adding additional capacity and incurring large capital expenses, UMMMC conducted a feasibility study to test various changes in hospital processes and configurations. One of the changes being proposed is to geographically place general medicine patients at specific units where these patients are assigned to a physician team geographically located at the same unit. It is believed that with such a configuration, there will be a reduction in physician walk time, thereby potentially increasing physician face time with patients, and improved physician communication with other staff, leading to a more efficient discharge process for current patients and a more efficient admission process for new patients.

A discrete event simulation is used as a tool to conduct this study. The simulation shows positive outcomes, including a decrease in the unnecessary movement waste of physicians’ average daily walk time and a decrease in the waiting waste evidenced by the average time between ED admit and patient head-in-bed.  Preliminary results from implementation show desirable outcomes, including a decrease in average time from ED admit to head-in-bed, an earlier average discharge time, and a decrease in patients’ length of stay.

In order to further utilize the simulation model and refine the processes implemented, several opportunities for future work have been identified by this study:

  1. The model can be further expanded to include all general medicine physician teams as well as subspecialty teams to study the global impact that cohorting the general medicine teams has on specialty teams.
  2. A patient safety measure can be defined that translates time measures, such as patient wait times to be seen by a physician, into a patient safety measure.
  3. In order to achieve a higher level of patient cohort, every patient has to be assigned to a physician, and the same patient has to be assigned to a bed where the majority of the physician’s patients are located. These assignments have to be made under systems constraints such as room availabilities, patients’ conditions, nursing competencies and staffing mix. A future study on such assignments can also provide greater insights into realistic levels of patient cohort based on organizational constraints, as well as their limitations and benefits.

Healthcare improvement is gaining momentum as lawmakers, healthcare organizations, insurers, and payers seek to design processes to deliver high-quality care at lower costs. This study has demonstrated the usefulness of one methodology to reduce waste by applying Lean and industrial engineering tools that have often been used in manufacturing and service industries to dramatically improve performance.


Bazarian, J. J., Schneider, S. M., Newman, V. J., & Chodosh, J. (1996). Do admitted patients held in the emergency department impact the throughput of treat-and-release patients? Academic Emergency Medicine: Official Journal of the Society for Academic Emergency Medicine, 3(12), 1113-1118.

Benneyan, J. C. (1997). An introduction to using computer simulation in healthcare: Patient wait case study. Journal of the Society for Health Systems, 5(3), 1-15.

Bucheli, B., & Martina, B. (2004). Reduced length of stay in medical emergency department patients: A prospective controlled study on emergency physician staffing. European Journal of Emergency Medicine: Official Journal of the European Society for Emergency Medicine, 11(1), 29-34.

Huang, Q., Thind, A., Dreyer, J. F., & Zaric, G. S. (2010). The impact of delays to admission from the emergency department on inpatient outcomes. BMC Emergency Medicine, 10, 16. doi: 10.1186/1471-227X-10-16

Imai, M. (1986). Kaizen (ky’zen), the key to Japan’s competitive success (1st ed.). New York: Random House Business Division.

Khurma, N., Bacioiu, G. M., & Pasek, Z. J. (2008). Simulation-based verification of Lean improvement for emergency room process. Proceedings of the 2008 Winter Simulation Conference, 1490-1499.

Lawrence, D. (2005). Bridging the quality chasm. In P. P. Reid, W. D. Compton, J. H. Grossman & G. Fanjiang (Eds.), Building a better delivery system: A new Engineering/Health care partnership. Washington D.C.: National Academy Press.

Lee, S., & Yih, Y. (2010). Analysis of an open access scheduling system in outpatient clinics: A simulation study. Simulation, 86(8-9), 503-518.

Parks, C. M. (2003). The bare necessities of Lean. Industrial Engineer: IE, 35(8), 39.

Ross, M. A., Naylor, S., Compton, S., Gibb, K. A., & Wilson, A. G. (2001). Maximizing use of the emergency department observation unit: A novel hybrid design. Annals of Emergency Medicine, 37(3), 267-274. doi: 10.1067/mem.2001.111519

Sharrock, R. (2007). Road to gold. Industrial Engineer: IE, 39(5), 44-48.

Wood, N. (2004). Lean thinking: What it is and what it isn’t. Management Services, 48(2), 8-10.

Wu, S. S., Lehto, M., & Yih, Y. (2011). Clinical decision support systems. In Y. Yih (Ed.), Handbook of healthcare delivery systems (pp. 48-1). Boca Raton, FL: CRC Press.

About the Authors

Tze Chao Chiam and Lori Pelletier, Department of Quality and Patient Safety, Center of Innovation and Transformational Change/Process Improvement, University of Massachusetts Memorial Healthcare

Richard Forster, Department of Medicine, University of Massachusetts Memorial Healthcare

Correspondence concerning this article should be addressed to Tze Chao Chiam, Department of Quality and Patient Safety, Center of Innovation and Transformational Change/Process Improvement, 11 Shattuck Street, SH-258, Worcester, MA 01605. Email: [email protected]