The Joint Commission Journal on Quality and Patient Safety 2018; 000:1–12

A Lean Six Sigma Quality Improvement Project Improves

Timeliness of Discharge from the Hospital
Mithu Molla, MD, MBA; Duncan S. Warren, MA, CSSBB; Susan Leroy Stewart, PhD; Jacqueline Stocking, PhD,
MBA; Hershan Johl, MD; Voltaire Sinigayan, MD

Background: Hospital overcrowding has become a widespread problem, with constrained bed capacity and admission
bottlenecks having far-reaching negative impacts on quality and safety. Focus on timing of discharge may be the least dis-
ruptive and most effective way to address constrained bed capacity, yet there may be significant institution-specific barriers
to implementation.
Methods: With the creation of a “Value Team,” a 627-bed, tertiary care academic medical center embarked on a quality
improvement (QI) project using Lean Six Sigma process improvement methodology. After defining the problems around
timeliness of discharge, the team went through the steps in the Define, Measure, Analyze, Improve, Control (DMAIC)
framework. Interventions, which were implemented on the basis of an in-depth analysis of barriers to the discharge process,
included geographic cohorts of internal medicine physicians on specific hospital units and multidisciplinary huddles one day
before anticipated discharge.
Results: After accounting for the concurrent trends in the control group, the percentage of discharge orders released by
10:00 a.m. increased by 21.3 points (p < 0.001; adjusted odds ratio [OR] = 2.62; 95% confidence interval [CI] = 1.91–
3.59), and the percentage of patients discharged by noon increased by 7.5 points (p = 0.001; adjusted OR = 1.70; 95% CI
1.15–2.51). There were no significant changes in the 30-day readmission rate or length of stay.
Conclusion: A QI program shaped by Lean Six Sigma principles and reinforced by clinician huddles and geographic
cohorting was associated with earlier posting of discharge orders and physical discharge by noon.

tem costs.8,9 A number of studies suggest that early dis-

H ospital overcrowding has become a widespread prob-
lem across the United States1 and can have far-
reaching deleterious effects. Constrained bed capacity and
charge initiatives are achievable and sustainable.10–13
At the University of California, Davis (UC Davis), late
admission bottlenecks leave many admitted patients to afternoon discharges had been recognized as one of the pri-
linger in emergency departments (EDs) for hours to days, mary barriers to movement of patients out of the ED and
with negative impacts on quality and safety. Poor patient a major factor for lack of bed availability. For UC Davis
satisfaction levels, medication administration delays, and Medical Center, a tertiary medical center and a regional
increased mortality, as a result of overcrowding, have all referral center for specialty services, lack of bed availabil-
been documented.2,3 In addition to exacerbating this prob- ity has significant implications for both hospital finances
lem in the ED, delayed discharges from the floor may result and community access. Multiple prior attempts to promote
in longer length of stay (LOS), resulting in adverse hospital- early discharge were seen as largely unsuccessful and subse-
related complications. In principle, hospital overcrowding quently abandoned. The argument was made that rushed
can be mitigated in three ways: (1) decrease admissions, (2) discharges without adequate preparation can lead to higher
shorten LOS (that is, decrease the number of “midnights” readmissions, and an excessive focus on meeting a super-
in the hospital), or (3) hasten discharge (without decreasing ficial discharge metric will cause physicians to hold on to
the number of “midnights”) by focusing on timing of dis- patients until the next day, leading to longer LOS.
charge. Of these, changes in discharge timing are arguably Lean Six Sigma (LSS) is a set of techniques and tools
the least disruptive. In addition, timing of discharge from for process improvement, with a focus on defects, varia-
the hospital has been shown to have an impact on the need tion reduction, and customer satisfaction, which in other
to board admitted patients in the ED,4 and there have been sectors has been shown to better serve consumers, re-
studies that suggest that late afternoon hospital discharges duce costs, and improve safety.14 In the health care set-
contribute to crowding and increased LOS.5–7 Discharges ting, LSS can be used to reduce medical errors while
that occur early in the day reduce ED wait times, improve maintaining patient and provider satisfaction through the
patient and provider satisfaction, and decrease health sys- development of a sustainable patient care model.15–17 A sys-
tematic problem-oriented approach such as LSS may help
1553-7250/$-see front matter focus on institution-specific barriers. We embarked on a
© 2018 The Joint Commission. Published by Elsevier Inc. All rights reserved. quality improvement (QI) project using LSS process im-
https://doi.org/10.1016/j.jcjq.2018.02.006
2 Mithu Molla, MD, MBA, et al

provement methodology to affect hospital throughput by rapid decision making and implementation. Attention fo-
discharging inpatients from our medical services earlier in cused on the nonteaching hospitalist service for three rea-
the day. We evaluated the effort using a quasi-experimental sons: (1) hospitalists are long-term employees rather than
design in which we evaluated discharge timing, as well as rotating house staff, ensuring continuity of commitment;
secondary outcomes such as LOS and readmission rate. (2) attending hospitalists on this service drive the medicine
culture at our institution; and (3) the majority of medicine
patients are admitted to this service. To evaluate the impact
METHODS of the interventions, we chose two medical/surgical units
Setting (East 4 [E4] and East 8 [E8]) to serve as our intervention
group, while all other medical/surgical units served as our
The Morning Discharge Quality Improvement initiative control group for comparison. E4 and E8 each consist of
was conducted from June 2015 through February 2016 on 35 medical and surgical beds and admit patients attended
the nonteaching hospitalist service at UC Davis Medical by different services. E4 primarily boards patients on the
Center, a 627-bed, tertiary care academic medical center. adult internal medicine service. E8 is a ward that accom-
Physicians on this service rotate on an attending-only ser- modates adult internal medicine patients as well as chronic
vice without house staff or learners and admit patients to 1 conditions patients. Our control group comprised 12 other
of 14 combined medical/surgical wards. Physician rotations medical/surgical units throughout the hospital, each also
are week on, week off, with a capped census of 15 patients. consisting of 35 medical/surgical beds. Some of the units
LSS methodology was used to identify barriers to early in our control group were focused on certain patient con-
discharge and develop an intervention targeted to those ditions, such as orthopedics. To create a fair and accurate
barriers. After defining the problems around timeliness of comparison, we decided to analyze discharges from both
discharge, the team went through the steps in the De- the intervention and control groups that were discharged
fine, Measure, Analyze, Improve, Control (DMAIC) frame- by only nonteaching hospitalists.
work. DMAIC methodology provides a systematic ap- With the work group in place, and our intervention and
proach to improve an organization’s efficiency and effective- control groups defined, the DMAIC process began. We
ness. now provide the details and time line of each step of the
DMAIC process specific to this intervention. This was a
Planning the Intervention QI intervention that was endorsed by the Medical Staff Ex-
The UC Davis Health Performance Excellence (Px) De- ecutive Committee. The UC Davis Institutional Review
partment was created in 2011 at our medical center to Board (IRB) Administration determined that the project
champion hospital-based initiatives. The department con- was not human subjects research and did not require IRB
sists of an LSS Black Belt in addition to experts in data review.
analysis. Using the Vizient® consortium of reporting hos-
pitals, we learned that our institution ranked last—out of Define (May 1–7, 2015)
54 academic medical centers—on LOS in the ED (specif- Undertaking a QI initiative involves clearly defining the
ically, Core Measure ED-1b),18 with a median wait time problem. Using Vizient18 and our electronic health records
of 466 minutes in the ED before arrival at the admitting data for the nine-month period prior to implementation,
unit. the work group determined that the time of day when dis-
In December 2014, to address the ED LOS opportu- charge orders were written and released peaked between
nity, the Px Department created a value stream work group 12:00 p.m. and 1:00 p.m., and the time of physical discharge
(“value team”) that used process stream mapping to em- from the hospital was dependent on the time of order re-
bark on the patient flow improvement project. The health lease. On the basis of these findings, the value team defined
system chief medical officer, chief operations officer, and two primary goals: (1) Improve the rate of discharge or-
chief nursing officer were named as executive sponsors of ders written and released by 10:00 a.m. to 40% before the
the project. The executive sponsors were responsible for fa- next calendar year (January 2016); and (2) Improve the rate
cilitating access to resources, such as data, and removing of physical discharge by noon (DBN) to an absolute 12%
barriers so the work group could be successful. The hospi- over the same period. DBN was chosen as our outcome
talist section chief was named as the owner of the process. metric based on literature review that showed discharging
The process owner was responsible for the overall success of a majority of inpatients by noon decreased boarding hours
the project. Three physician champions from the hospital- in the ED.4 A two-hour lead time between discharge or-
ist service were appointed to the work group, primarily to der release and DBN was deemed adequate by our value
serve as clinical work flow and subject matter experts, and team as needed for preparation for physical discharge. As
to disseminate agreed-on interventions to other providers described above, the scope of the project was limited to the
in the group. While input was sought from all stakeholders nonteaching hospitalist service, and only to patients admit-
in the process, the work group was kept small to facilitate ted to E4 and E8.
Volume 000, No. , May 2018 3

Figure 1: The process map facilitated identification of several steps that took place in the physician work flow that led
to a delay in discharge order entry. DC, discharge; Med, medication; SNF, skilled nursing facility; DME, durable medical
equipment; UCDMC, UC Davis Medical Center.

With the problem statement, scope, and goals of the ini- Five main categories of delay were identified: (1) the dis-
tiative defined, the work group constructed a process map to charge process itself (failure to arrange for outpatient intra-
define each step in the actual process to place and release dis- venous [IV] antibiotics or home total parenteral nutrition
charge orders for the nonteaching physician teams. There prior to discharge), (2) discharge equipment delays (physi-
was a group brainstorming phase, during which sticky notes cal therapy–recommended mobility aids not ordered until
were placed on a whiteboard to represent each step in the late in the evening), (3) post-acute care need delays (inser-
current process. Figure 1 depicts the final computerized tion of peripherally inserted central catheter not performed
“current state” document representing the group’s work. until late in the evening), (4) medication reconciliation de-
lays (medication reconciliation not done on admission),
Measure (September 2014–May 2015) and (5) delays in rounding (physicians seeing discharge-
The next step of the process involved collecting and validat- able patients last). Each theme was explored and unpacked
ing trend data on the average time of day and distribution until a list of actionable root causes amenable to improve-
for discharge orders to be released by the hospitalist services. ment was identified. The work group found that hospital-
We reviewed the trends that existed during the baseline pe- ists are required to complete a large number of activities in
riod (September 2014–May 2015) and used them as the the morning hours. These activities limited the time avail-
basis for improvement. The distribution was normal with a able to see patients and write discharge orders. In addition,
range of 8:00 a.m. to 7:00 p.m., peaking at 12:30 p.m. Fur- communication with the other disciplines was fragmented
ther analysis of the hospitalist service on E4 and E8 revealed with a lack of adequate preparation for discharge.
that 15.6% of discharge orders were written and released by
10:00 a.m. DBN rate averaged 10.5% of patients on those Improve (June 1–15, 2015)
two units combined. On the basis of our analysis, we decided that we needed
to help the hospitalists plan and coordinate their morning
Analyze (May 8–30, 2015) discharge activities by way of multidisciplinary huddles and
The work group next conducted a root cause analysis geographic cohorts. Hospitalists were cohorted to each unit
(RCA), including constructing a fishbone (Ishikawa) di- (one specific hospitalist to E4 and one specific hospitalist to
agram, to identify and thematically cluster causal factors E8) during their week on service, which were the only two
leading to delays in releasing discharge orders. Major cat- units geographically cohorted to the hospital medicine ser-
egories and individual factors are depicted in Figure 2. vice. Our control group consisted of the other hospitalist
4 Mithu Molla, MD, MBA, et al

Figure 2: An Ishikawa diagram was constructed to identify and thematically cluster causal factors leading to delays in
releasing discharge orders. DC, discharge; Med rec, medication reconciliation; AMA, against medical advice; UCDMC, UC
Davis Medical Center; ED, emergency department; O2, oxygen; Psych, psychiatric; PT, physical therapy; DME, durable
medical equipment; Picc, peripherally inserted central catheter; IV, intravenous; Abx, antibiotics; TPN, total parenteral
nutrition.

services that were non-cohorted, where physicians could be efficient use of the hospitalist’s time. With a majority of pa-
responsible for overseeing care for medical patients located tients on their rounding list on the unit, they could discuss
in as many as 14 different wards and did not participate in their patients with the other disciplines, coordinate patient
the multidisciplinary discharge huddles. Group policy was needs, and communicate the plan in one sitting, and with
to cap a hospitalist at 15 patients and to keep all hospitalists all providers involved in their patients’ care. We began co-
relatively even in terms of distribution of patients. This re- horting hospitalists to E4 and E8, starting in January 2015,
sulted in little change in workload or census associated with several months before the other components of the inter-
the geographic cohort. Because physicians were cohorted to vention bundle, as it was felt that this was a necessary pre-
units according to their weekly schedule, these cohorts ex- requisite and needed to be perfected for the other parts of
tended through the weekend, even when multidisciplinary the bundle. Cohorting physicians to specific units was con-
huddles were not taking place. sidered a major practice change for many within the group;
Early on, through consensus and recommendations from several iterations were attempted, and multiple challenges
the work group, we decided to cohort hospitalists to our needed to be worked through in order to perfect the pro-
intervention units to provide consistency in staff and allow cess. With a lack of hospital beds, trying to redirect patients
a platform by which the other interventions could be im- to these units was not feasible, as patients from the ED were
plemented. We believed that a critical number of patients admitted to the first available bed in as many as 14 differ-
was needed to optimize efficiency for the interventions. For ent units. Rather, we cohorted hospitalists to those units,
example, if only two patients were on the unit, than any and assigned patients to that hospitalist’s list after their bed
interventions involving just those patients would not be an had been assigned during the admission process. The co-
Volume 000, No. , May 2018 5

horts were not perfect, in that the hospitalist on this ser- (Hospital Medicine section chief, nurse unit manager) also
vice would occasionally see some patients on other units attended to ensure adherence to our institutional goals and
either to maintain continuity or to help with distribution checklist. Attendance was monitored with formal sign-in
of patients when the other teams were capped by census. sheets taken during the intervention period. Goals for dis-
However, efforts were made to ensure that the majority of charge order release time were reiterated at every monthly
patients on each of these units were part of the hospitalist’s business meeting. Leadership attendance and engagement
cohorted list. Hospitalists who worked on the two interven- were critical to maintaining integrity and structure of the
tion floors crossed over to the control floors in subsequent huddle.
weeks and vice versa.
With the geographic cohorts in place, an intervention Sidebar 1. Multidsciplinary Huddle Checklist∗
bundle was assembled with the other components: medica- 1. Physician anticipate the patient to be medically cleared for
tion reconciliation assistance, a checklist to be used during discharge:

mandatory multidisciplinary huddles, performance audit, Today

and feedback to participating hospitalists. With the antici- Tomorrow
pated needs cared for the day before during huddles, hospi-
In 1-2 days
talists could then quickly see patients to be discharged the
In 3-5 days
morning of, and release their discharge orders before see-
ing other patients on their lists. With the exception of ge- In > 5 days

ographic cohorts, the intervention bundle was constructed 2. Discharge barriers:

and implemented over a period of two weeks at the start None
of our intervention period. The other components of the
Medications (prior authorization, etc.)
bundle are described in more detail below:
The multidisciplinary discharge huddle was designed to Placement

focus on causal factors for delay identified in our RCA. Prior DME
to implementation, there were no multidisciplinary hud- Home IV antibiotics, TPN or home O2
dles on any of the units. The work group decided that 2:00 Psycho-social issues
p.m. was the optimal time when physicians could participate
Transportation
with minimal interruptions, yet interventions could still be
executed in anticipation of discharge the next day (for ex- Pain

ample, order durable medical equipment for the follow-

3. Choosing wisely / Restful Night’s Sleep
ing day based on the physical therapist’s assessment). Hud-
dles would take place Monday through Friday, and mem- Frequency of vitals check

bers were required to sign in—specifically, charge nurse, Appropriateness of labs

discharge planning, hospitalist, pharmacy, nutrition, and Medication times at night
physical therapy. Participants were chosen based on per-
∗
DME, durable medical equipment; IV, intravenous; TPN, total par-
ceived needs as defined by the process map and RCA. For enteral nutrition; O2, oxygen.
example, lack of accurate medication reconciliation on ad-
mission was identified in our RCA. This led to delays down-
stream, as efforts had to be made to update and correct or Audit and feedback was also an integral part of the
perform medication reconciliation on discharge. Other de- huddle process: A discharge order release time and DBN
lays resulted from the prior authorization process for new dashboard was created and e-mailed weekly to huddle par-
prescriptions, or determination of the patient’s outpatient ticipants and to the hospital medicine group (Appendix
pharmacy, and discharge counseling on new prescriptions. 1, available in online article). The dashboards were con-
A pharmacy technician attended the huddles and identi- structed by the Px Department and included retrospective
fied which patients were targeted for discharge the next day. review of performance on a weekly basis. Individual physi-
The pharmacy technician would then work with the hospi- cians could review their performance and were provided
talist and complete medication reconciliation the day be- feedback from the previous week’s discharge order release
fore discharge, perform discharge counseling, and work on rate. We also reviewed the dashboards at our monthly busi-
the prior authorization process, so that these would not be- ness meetings and provided feedback to the entire hospital-
come barriers on the day of discharge. The hospitalist could ist group as a whole. Goals were reiterated that discharge or-
then quickly see the patient in the morning and write the ders should be released by 10:00 a.m. 40% of the time, and
discharge orders by 10:00 a.m. Participants were brought our DBN rate should be greater than 12%. The dashboards
together and used a huddle checklist (Sidebar 1), which en- were also reviewed with the other disciplines (discharge
couraged a proactive multidisciplinary approach to recog- planning, patient care services, pharmacy) every Tuesday
nized discharge barriers. Physician and nursing leadership morning with executive leadership during value team hud-
6 Mithu Molla, MD, MBA, et al

Figure 3: Patient ride home and ambulance transport were identified as common barriers leading to delay in discharge.
SNF, skilled nursing facility; OT/PT, occupational therapy/physical therapy; D/C, discharge.

dles. Here, participants were recognized for their success at Vizient18 case profile data with data from the hospital’s elec-
meeting goals, and problems were brought to the attention tronic health records system.
of executive leadership.
Statistical Analysis
Control (July 2015–February 2016)
Our primary objectives were to show that our interventions
To sustain our results, the team reviewed control chart data on E4 and E8 improved the process (as measured by dis-
on a weekly basis. Shortly after implementation, the team charge orders released by 10:00 a.m.) and that the process
noted that on E8, physicians were meeting their goal of re- improvement had the intended beneficial effect on an out-
leasing discharge orders by 10:00 a.m. approximately 40% come (as measured by patients discharged by noon). To
of the time, but the percentage of DBN was below target. ensure that the process change did not cause any harmful
E8 huddle team representatives were brought together to spillover effects, we assessed the performance of two addi-
discuss reasons potentially contributing to delays. We con- tional outcomes: patient readmissions (as measured by pa-
structed a Pareto chart (Figure 3) to identify factors lead- tients readmitted within 30 days) and patient LOS (as mea-
ing to delay of physical discharges from this unit. Lack of sured by LOS index). Historically, individual physicians
transportation (no patient ride or ambulance transport) was could not achieve early discharge or influence discharge
identified as a common cause of delayed discharge. As a re- time on a regular basis; therefore, we did not account for
sult, more focus was placed on this aspect of the daily hud- clustering of patient outcomes by physician in our analyses.
dle. For both of the nursing units on which discharge hud-
dles were implemented (the treatment group), study out-
Outcome Measures comes and patient characteristics were compared to the con-
Primary Outcomes. The primary outcomes were the per- trol group during a nine-month baseline period (September
centage of discharge orders released by 10:00 a.m., and the 2014–May 2015) and a nine-month intervention period
percentage of DBN. All discharges that took place on the (June 2015–February 2016). Patient characteristics were
nonteaching hospitalist service were analyzed, both before summarized by group and period using frequencies and
and after the intervention began. Discharge order entry percentages for categorical variables, and means and stan-
times and actual discharge from the hospital on these two dard deviations for numeric variables; differences between
units were compared with historical controls as well as with the groups in each period, and between periods within
our control group. each group, were assessed using chi-square tests or z-tests
Secondary Outcomes. Our secondary outcomes were to compare proportions and t-tests to compare means. A
the ratio of observed to expected LOS, and 30-day read- difference-in-differences (DID) approach was employed to
mission rates. All measurements were based on a merger of compare the outcome measures between the baseline and
Volume 000, No. , May 2018 7

intervention periods, and between the treatment and con- us to estimate how much of the change in an outcome
trol groups. This technique estimates the effect of a treat- measurement was attributable to the process intervention
ment on an outcome by comparing the change over time in after accounting for changes that would have happened
the treatment group with the change over time in the con- in the absence of the intervention, such as the trend over
trol group, thereby accounting for temporal trends as well time, a change that also occurred in the control group. The
as baseline differences between the groups. For binary out- DID estimates confirmed that the intervention was asso-
come variables, we computed the difference between the in- ciated with increases in both the percentage of discharge
tervention and baseline periods in the proportion with the orders released by 10:00 a.m. and the percentage of pa-
outcome (for example, discharge orders released by 10:00 tients DBN. After accounting for the concurrent trends in
a.m.) in each group, subtracted the difference in propor- the control group, the percentage of discharge orders re-
tions in the control group from the difference in propor- leased by 10:00 a.m. increased by 21.3 points (p < 0.001;
tions in the treatment group, and used a z-test to deter- adjusted OR = 2.62; 95% CI = 1.91–3.59), and the per-
mine statistical significance. We analyzed LOS index using centage of patients discharged by noon increased by 7.5
two-way analysis of variance (ANOVA) with main effects points (p = 0.001; adjusted OR = 1.70; 95% CI = 1.15–
for group and period, as well as a group-by-period inter- 2.51) (Table 2 and Table 3).
action to determine whether the period effects differed sig-
nificantly between treatment and control. To adjust for im- Secondary Outcomes: Observed to Expected
balances in patient characteristics, we also developed a mul- Length of Stay and 30-Day Readmission Rate
tivariable model of each outcome variable as a function of The intervention did not have a deleterious effect on ei-
group, period, a group-by-period interaction, and patient ther the 30-day readmission rate, or LOS index. There
characteristics (age, gender, race, and presence/absence of were no significant changes in the 30-day readmission rate
major illness). We used logistic regression to estimate ad- (DID estimate, 1.6 points; p = 0.492, adjusted OR = 1.13;
justed odds ratios (ORs) with 95% confidence intervals 95% CI = 0.79–1.61) or LOS index (DID estimate, 0.08;
(CIs) for binary variables, and linear regression to estimate p = 0.153, regression coefficient = 0.09; 95% CI = -0.02–
coefficients with 95% CIs for LOS index. Statistical signif- 0.21) (Table 2 and Table 3).
icance was assessed at the 0.05 level (two-sided).
Association Between Patient Characteristics and
Outcomes
EVALUATION RESULTS As one might expect, patients with major illnesses were less
Patient Characteristics likely to have their discharge orders by 10:00 a.m.—and less
likely to be discharged by noon. They were also more likely
Table 1 compares demographic characteristics of interven- to be readmitted and to experience a longer LOS.
tion and control group patients. Age, race, sex, severity of
illness, and expected mortality (calculated using Vizient’s Statistical Process Control Results
2015 AMC risk model) were analyzed for both the inter- The percentage of discharge orders written and released and
vention and control groups. A total of 4,134 patients were the percentage of physical discharge from the hospital were
included in the analysis, with 1,471 patients in our inter- analyzed using a statistical process control chart to distin-
vention group and 2,663 patients in our control group dur- guish between special-cause and common cause variation,
ing both the baseline and intervention periods. The mean on the basis of a merger of Vizient’s case profile data with
age of patients was higher in the control group than in the data from the hospital’s electronic health records system
study group, as was the proportion of patients age 65 and (Figure 4 and Figure 5). There was marked improvement in
over. Patients with major illnesses were more likely to be in discharge order release time and DBN in the intervention
the control group in both time periods. The proportion of group, with sustained results throughout the intervention
patients with major illnesses in the control group increased period (ending February 2016). An RCA of dips in perfor-
from the baseline to the intervention period, and there was mance revealed that new hires were rotating on our inter-
an increase in the expected mortality rate. There was not a vention units (September and February 2016). Discharge
significant change in proportion of patients with major ill- order release time and DBN rate were analyzed using statis-
ness in the intervention group, and there was no significant tical process control, beyond the intervention period, and
change in expected mortality rate between the baseline and show sustained results through December 2016.
intervention periods.
ED Throughput as Measured by Core Measure
Difference-in-Differences Analysis ED-1b
The four response measurements were compared between Core Measure ED-1b was not significantly affected by our
the study and control groups across the baseline and in- intervention. In the postintervention period (June 2015–
tervention periods (Table 2). These comparisons enabled February 2016), the median wait time from presentation
8 Mithu Molla, MD, MBA, et al

Table 1. Comparison of Demographic Data in the Study and Control Groups During the Baseline (September
2014–May 2015) and Intervention Periods (June 2015–February 2016)
Demographic Characteristics

Characteristic Baseline Period Intervention Period P Values

Study Group Control Study Group Control Baseline Intervention Study Control
(n = 675) Group (n = 796) Group Study vs. Study vs. Intervention Intervention
(n = 1,393) (n = 1,270) Control Control vs. Baseline vs. Baseline
Age, mean (SD) 55.4 (18.8) 60.7 (18.5) 54.9 (18.8) 58.8 (18.6) < 0.001∗ < 0.001∗ 0.582∗∗ 0.007∗
65 and Over, 214 (31.7) 597 (42.9) 258 (32.4) 513 (40.4) < 0.001† < 0.001† 0.772† 0.197†
n (%)
Male, n (%) 353 (52.3) 680 (48.8) 421 (52.9) 642 (50.6) 0.137† 0.300† 0.820† 0.371†
Race, n (%) 0.007‡ 0.060‡ 0.765‡ 0.567‡
Asian 56 (8.3) 138 (9.9) 67 (8.4) 123 (9.7)
Black 124 (18.4) 332 (23.8) 140 (17.6) 277 (21.8)
White 377 (55.9) 677 (48.6) 433 (54.4) 628 (49.5)
Other 118 (17.5) 246 (17.7) 156 (19.6) 242 (19.1)
Major Illness, 341 (50.5) 771 (55.3) 420 (52.8) 751 (59.1) 0.039† 0.005† 0.390† 0.048†
n (%)
Expected 0.031 (0.075) 0.028 (0.064) 0.029 (0.067) 0.038 (0.086) 0.373∗ 0.004∗ 0.591∗ < 0.001∗
Mortality, mean
(SD)
SD, standard deviation.
∗ Two-sample t-test;
† Two-proportion test;
‡ Chi-square test.

Table 2. Difference-in-Differences Analysis

Study Group Control Group Difference-in-
Differences Estimate
(percentage points)

Measurement Baseline Intervention Baseline Intervention Period

Period Period Period (Difference-in-Differences
(N = 675) (N = 796) (N = 1,393) Analysis) N = 1,270)
Discharge orders by 105 (15.6) 375 (47.1) 227 (16.3) 336 (26.5) 21.3 (p < 0.001)
10:00 A.M., n (%)
Patient discharged by 71 (10.5) 164 (20.6) 130 (9.3) 151 (11.9) 7.5 (p = 0.001)
noon, n (%)
30-day readmission, n (%) 100 (14.8) 121 (15.2) 241 (17.3) 204 (16.1) 1.6 (p = 0.492)
Length of stay index, 1.04 (0.96) 0.98 (0.88) 1.11 (1.03) .97 (0.77) 0.08 (p = 0.153)
mean (SD)
SD, standard deviation.

Table 3. Parameter Estimates for the Four Multivariate Models

Discharge order by Patient discharged by noon 30-day readmission Length of stay index
10:00 A.M. (N = 4,134) (N = 4,134) (N = 4,134) (N = 4,134)

OR 95% CI OR 95% CI OR 95% CI Coeff. 95% CI

Group 0.94 (0.73–1.21) 1.09 (0.80–1.48) 0.87 (0.67–1.12) -0.08 (-0.16–0.01)
Period 1.87 (1.54–2.26) 1.32 (1.03–1.70) 0.90 (0.73–1.11) -0.16 (-0.23– -0.09)
Group Period 2.62 (1.91–3.59) 1.70 (1.15–2.51) 1.13 (0.79–1.61) 0.09 (-0.02–0.21)
65+ 1.02 (0.88–1.19) 0.72 (0.59–0.89) 1.00 (0.84–1.19) -0.12 (-0.17– -0.06)
Male 0.94 (0.82–1.09) 1.08 (0.90–1.30) 1.08 (0.92–1.28) 0.03 (-1.03–0.09)
Asian 0.76 (0.57–1.00) 0.95 (0.67–1.36) 0.93 (0.68–1.27) -0.02 (-0.12–0.08)
Black 0.97 (0.80–1.17) 1.19 (0.94–1.50) 1.49 (1.21–1.83) -0.07 (-0.15– -0.01)
Other 0.96 (0.79–1.17) 1.07 (0.83–1.38) 0.96 (0.76–1.21) -0.08 (-0.16– -0.01)
Major Illness 0.82 (0.71–0.95) 0.67 (0.55–0.81) 1.70 (1.43–2.03) 0.26 (0.20–0.31)
OR, odds ratio; CI, confidence interval; Coeff., regression coefficient.
Volume 000, No. , May 2018 9

Figure 4: This statistical process control chart demonstrates marked improvement in discharge release time by 10:00 A.M.
after implementation of interventions. Geographic cohorts were implemented on January 1, 2015, and multidisciplinary
huddles were implemented on June 1, 2015. Root cause analysis for dips in performance in September 2015 and February
2016 revealed that new physician hires were rotating on these units.

Figure 5: This statistical process control chart demonstrates marked improvement in the percentage of patients physi-
cally discharged by noon. Geographic cohorts were implemented on January 1, 2015, and multidisciplinary huddles were
implemented on June 1, 2015.

in the ED to admission was 481 minutes (vs. 466 minutes We used LSS DMAIC process methodology as a structured
during the baseline period). framework to design a discharge initiative to overcome bar-
riers unique to our institution. This allowed us to focus
DISCUSSION on problems at our institution, such as medication rec-
Despite reports of successful and sustainable early discharge onciliation. Our results support this approach, as we saw
initiatives in the literature, many hospitals continue to have significant improvements in discharge order release time
institution-specific barriers that prevent early discharge ini- and physical discharge by noon, with marked improvement
tiatives from taking hold. A “one size fits all” approach may when compared to those services not taking part in the in-
not be fruitful, and the strategies that prove effective for one tervention.
institution may not lead to meaningful results for another.
10 Mithu Molla, MD, MBA, et al

Our primary process metric was the discharge order en- macy, were already stationed on the units (unit-based). Co-
try time, and physician and staff engagement were integral horts made it much easier for these disciplines to collaborate
drivers of this process. Staff members who are given encour- with one hospitalist responsible for many of the patients lo-
agement, training, and time to make meaningful improve- cated on the unit. We do not feel that other elements of the
ments in how the work is done are unlikely to want to re- intervention bundle could have interacted in an efficient
treat to an earlier period when formalized effort to improve and effective way without cohorting. We saw this when
existing processes was outside their domain of responsibil- hospitalists crossed over to the control floors (where they
ity.17 One reason our discharge initiative had been unsuc- probably took their early discharge practices with them),
cessful in the past was a lack of physician engagement, and but without the other elements of the huddle we saw a far
one of the challenges we experienced was integration of new less robust improvement in the DBN.
hires who had not been involved in the original process. We did not see deterioration in upstream measures such
This was supported by the observed dips in discharge order as LOS nor downstream measures such as readmission rates.
release time noted at the time of new physician onboard- Our results were based on medicine patients admitted al-
ing. These findings parallel those reported by Patel et al. most exclusively through the ED and are in contrast to a
(who also showed dips in performance when new house staff more recent analysis that demonstrated that the DBN rate,
picked up the service)12 and underscore the importance of specifically among medicine patients admitted emergently,
physician and staff engagement as primary drivers of this was associated with a longer LOS.19 Our analysis did show
particular process metric. that patients with major illnesses were less likely to have
Others have shown that increasing early discharges us- their discharge orders by 10:00 a.m., and less likely to be
ing a structured framework for QI is achievable and sus- discharged by noon. Counterintuitively, patients 65 years
tainable.10–12 For example, Beck et al. used Lean meth- of age and over experienced a shorter LOS, even though
ods to modify work flow and attain early discharges,10 and they were less likely to be discharged by noon. Older pa-
Patel et al. used an Institute for Healthcare Improvement tients may require a greater lead time to arrange for dis-
PDSA (Plan-Do-Study-Act) framework to achieve early dis- charge, with resultant discharge later in the afternoon, but
charges12 . A feature common to all these successful inter- still meet discharge criteria by midnight, thereby not affect-
ventions is a standardized forum for communication such ing the overall LOS. Younger patients probably have fewer
as a multidisciplinary huddle. Other elements central to needs on discharge, such as extensive medication reconcil-
the success of these interventions are accurate measurement iation or equipment needs, or they may have been able to
with data analysis, use of checklists, feedback, and identi- transport home without the need of an ambulance or other
fication of patients the day before discharge. These initia- barriers. These are the patients who likely benefit most from
tives were focused on house staff teams, whereas our study a focus on efficient discharge and the DBN rate.
involved attending physicians without house staff. In both Although our interventions were focused on two spe-
cases the morning hours were deemed unsuitable for the cific units of the hospital, we also saw modest improvement
kind of preparatory work needed to facilitate discharges in the discharge order release time and DBN rate in our
early in the day; preparation occurred the day before. Our control units. The early discharge initiative at its inception
intervention also relied on notable systems changes. Geo- was hospitalwide, rolled out to the entire health system. All
graphically cohorting physicians to our intervention units the disciplines—from social services to hospital medicine—
facilitated effective use of the huddle and brought team were aware of the desire to discharge patients earlier in the
members together who were familiar with the process and day, and the hospitalist group, as a whole, received feed-
had worked with each other. Without a critical number of back on the process during the monthly business meetings.
patients on the unit (achieved only through geographic co- Hospitalists crossed over to the control units in subsequent
horts), physicians would not be able to appreciate efficien- weeks, perhaps taking their early discharge practices with
cies created through this type of multidisciplinary commu- them. This general awareness and institutional desire may
nication. This was particularly important when starting the have led to some improvement, although far less than with
intervention, as efficiency and commitment to keeping the our interventions in place. The early DBN initiative at our
huddles short were necessary prerequisites to physician par- institution had not achieved significant results over the years
ticipation. With geographic cohorts resulting in a critical because of an emphasis on physician practice alone. By pur-
volume of patients to discuss, a platform was created with suing a comprehensive structured approach, with an em-
which to bring in other elements of the huddle—pharmacy phasis on staff engagement and systems-level changes, and
technicians to address medication reconciliation problems, an established QI framework, we were finally able to achieve
physical therapy to address delays in ordering durable med- significant results.
ical equipment, case management to address postdischarge Satisfaction with the geographic cohorts and multidisci-
needs—all of which occurred at one place and in one sitting. plinary huddles was high and contributed to the initiative’s
The other disciplines were brought in on the basis of prob- sustainability postintervention. Hospitalists came to appre-
lems identified in our RCA. These disciplines, such as phar- ciate the efficiency of huddles where they could commu-
Volume 000, No. , May 2018 11

nicate to multiple disciplines about their patients in one cused only on medicine patients on the nonteaching hospi-
sitting and prepare for the discharge process during a less talist service, a relatively small proportion of the total num-
hectic period of the workday, in the early afternoon. There ber of patients admitted through the ED. Conceivably, if
were anecdotal reports that they received far fewer pages as discharge huddles and geographic cohorts were widespread
a result of better communication and collaboration, which and utilized by every unit, a significant improvement in ED
also contributed to high satisfaction. The other disciplines throughput could be achieved. Indeed, at our institution,
also felt that communication was greatly improved and co- attempts are now being made to expand geographic cohorts
ordination of care elevated. In addition to developing famil- of physicians to other units with multidisciplinary huddle
iarity with the hospitalist, it was also much easier for nurs- structures in place.
ing, case management, or the other involved disciplines to Our control and intervention groups consisted of non-
relay information about their patients to one physician who teaching hospitalists without house staff at a single-site
covered many of the patients on their unit. tertiary university medical center. This type of practice
is reflective of nonteaching community-based hospital
LIMITATIONS medicine work; however, staffing and workload may vary
Although positive results were obtained from both our pro- in university-affiliated and nonaffiliated practices. Further
cess metric (discharge order release time) and outcome met- studies and replication will need to demonstrate the ef-
ric (DBN), there was a significant difference between the fectiveness of this approach in community-based hospitals
two. Our intervention bundle focused on the time orders with different staffing models. In addition, all our hospi-
are written, as well as the downstream discharge process talists were experienced clinicians receiving structured sup-
from the unit. Discharge orders can be placed without hav- port within a flexible framework. It is not clear how effec-
ing final discharge-ready processes in place, and there are tive these interventions would be with learners, or in a more
many more factors involved in physical discharge from the rigid learning-centered framework, where, for example con-
unit. We saw this when we constructed a Pareto chart and ferences or other structured teaching activities may interfere
learned that we needed to target our attention to facilitat- with huddles. There was considerable investment at the in-
ing rides home. There may have been more opportunities stitution level in the Px Department. This project required
to analyze this discrepancy and delineate further areas of hiring of additional staff, including an LSS Black Belt and
increasing engagement with the other disciplines. This also a team to perform real-time data analysis. These efforts are
underscores a weakness in focusing on the discharge order resource intensive and require large capital investments. In
release time alone: Ultimately, it may not lead to physical addition, our interventions were heavily dependent on ge-
emptying of beds. Because discharge orders can be placed ographic cohorts of physicians to a specific unit to be effec-
by the clinician without other aspects of discharge in place, tive, and this may not be achievable on other units/services
a focus on DBN rate as the primary goal would likely be or other institutions.
more meaningful in affecting the ED throughput time or
in designing incentive programs.
CONCLUSION
Our analysis revealed that our control and intervention
groups had important differences in their demographic pro- We developed a successful strategy of interventions by
files. Although we controlled for patient characteristics in using LSS DMAIC, a QI process with a proven track
our multivariable analysis, there may have been other un- record in other industries. Our experience suggests that
measured confounders that affected our results. Optimiza- LSS process improvement methodology has potential
tion of medical comorbidities and treatment, or the pa- to improve care and efficiency—including timeliness of
tient’s willingness to leave before noon, are examples of discharge—in a variety of complex health care settings.
very difficult to control confounders that could have an Our success in achieving our discharge goals was sustained
impact on the DBN rate. The hospital environment tends in part by a physician-centric approach, as well as systems
to be fluid and dynamic, and it can be difficult to create changes that were implemented on the basis of an in-depth
conditions in which precise comparisons can be made. Al- analysis of barriers to the discharge process. Although our
though statistical modeling was employed, there are limita- project successfully achieved results in discharging patients
tions with using observational data to draw conclusions. earlier in the day, further analysis with more widespread
Prior research indicated that higher DBN rates were as- application needs to be conducted. This may help to
sociated with admissions from the ED earlier in the day, clarify and support the idea that early discharges lead to a
decreasing boarding time in the ED.4,9 The ultimate goal reduction in ED throughput time.
of our initiative was to decrease ED throughput time. How-
ever, we did not see any improvement in Core Measure ED- Acknowledgements. The authors thank the Performance Excellence De-
1b (ED throughput time) during or after the intervention. partment at UC Davis Medical Center. They also thank Drs. Greg Maynard
and Richard Kravitz for their contributions.
Our intervention involved two units of the hospital and fo-
12 Mithu Molla, MD, MBA, et al

4. Powell ES, et al. The relationship between inpatient dis-

Conflicts of Interest. All authors report no conflicts of interest.
charge timing and emergency department boarding. J
Emerg Med. 2012;2:186–196.
5. Khanna S, et al. Impact of admission and discharge peak
times on hospital overcrowding. Stud Health Technol In-
SEE THE ONLINE VERSION OF THIS ARTICLE
form. 2011;168:82–88.
FOR APPENDIX 1. DISCHARGE HUDDLE 6. White BA, et al. Boarding inpatients in the emergency de-
DASHBOARDS FOR PERCENTAGE OF PATIENTS partment increases discharged patient length of stay. J
DISCHARGED BY NONTEACHING HOSPITALISTS Emerg Med. 2013;44:230–235.
ON UNITS EAST 4 (E4) AND EAST 8 (E8) BY 7. Derlet RW, Richards JR. Overcrowding in the nation’s emer-
NOON AND 10:00 A.M. gency departments: complex causes and disturbing effects.
Ann Emerg Med. 2000;35:63–68.
SUPPLEMENTARY MATERIALS 8. Chaiyachati KH, et al. Discharge rounds: implementation
Supplementary material associated with this article can be of a targeted intervention for improving patient through-
put on an inpatient medical teaching service. South Med J.
found, in the online version, at doi:10.1016/j.jcjq.2018.02.
2016;109:313–317.
006. 9. Wertheimer B, et al. Discharge before noon: ef-
Mithu Molla, MD, MBA, Health Sciences Associate Clinical Professor and
fect on throughput and sustainability. J Hosp Med.
Section Chief, Hospital Medicine, Department of Internal Medicine, Uni- 2015;10:664–669.
versity of California, Davis (UC Davis) Medical Center. Duncan S. War- 10. Beck MJ, Gosik K. Redesigning an inpatient pediatric ser-
ren, MA, CSSBB, Data Analyst, Department of Performance Excel- vice using Lean to improve throughput efficiency. J Hosp
lence, UC Davis Medical Center. Susan Leroy Stewart, PhD, Asso- Med. 2015;10:220–227.
ciate Adjunct Professor, Division of Biostatistics, UC Davis School of
11. Wertheimer B, et al. Discharge before noon: an achievable
Medicine. Jacqueline Stocking, PhD, MBA, Quality and Safety An-
alyst, Department of Internal Medicine, UC Davis Medical Center.
hospital goal. J Hosp Med. 2014;9:210–214.
Hershan Johl, MD, Health Sciences Associate Clinical Professor, Sec- 12. Patel H, Morduchowicz S, Mourad M. Using a systematic
tion of Hospital Medicine, Department of Internal Medicine, UC Davis framework of interventions to improve early discharges. Jt
Medical Center. Voltaire Sinigayan, MD, Associate Physician Diplo- Comm J Qual Patient Saf. 2017;43:189–196.
mate, Section of Hospital Medicine, Department of Internal Medicine, 13. Jweinat J, et al. The safe patient flow initiative: a collabora-
UC Davis Medical Center. Please address correspondence to Mithu Molla,
tive quality improvement journey at Yale-New Haven Hos-
[email protected].
pital. Jt Comm J Qual Patient Saf. 2013;39:447–459.
14. Neufeld NJ, et al. A Lean Six Sigma quality improvement
project to increase discharge paperwork completeness for
REFERENCES admission to a comprehensive integrated inpatient rehabil-
itation program. Am J Med Qual. 2013;28:301–307.
1. Barish RA, McGauly PL, Arnold TC. Emergency room crowd- 15. Linderman K, et al. Six Sigma: a goal-theoretic perspective.
ing: a marker of hospital health. Trans Am Clin Climatol As- Journal of Operations Management. 2003;21:193–203.
soc. 2012;123:304–310. 16. Arnheiter ED, Maleyeff J. The integration of Lean manage-
2. Forster AJ, et al. The effect of hospital occupancy on emer- ment and Six Sigma. TQM Magazine. 2005;17:5–18.
gency department length of stay and patient disposition. 17. Toussaint JS, Berry LL. The promise of Lean in health care.
Acad Emerg Med. 2003;10:127–133. Mayo Clin Proc. 2013;88:74–82.
3. Bernstein SL, et al. The effect of emergency department 18. Vizient(TM). 2018. https://www.vizientinc.com/.
crowding on clinically oriented outcomes. Acad Emerg 19. Rajkaomar A, et al. The association between discharge be-
Med. 2009;16:1–10. fore noon and length of stay in medical and surgical pa-
tients. J Hosp Med. 2016;11:859–861.