The Joint Commission Journal on Quality and Patient Safety 2025; 000:1–10

Reducing Hospital Length of Stay: A Multimodal

Prospective Quality Improvement Intervention
Joseph Walker Keach, MD; Mara Prandi-Abrams; Allison S. Sabel, MD, PhD, MPH; Romana Hasnain-Wynia, PhD;
Jonathan M. Mroch; Thomas D. MacKenzie, MD, MSPH

Background: Prolonged hospital stays beyond medical necessity pose avoidable risks and costs. Reducing length of stay
(LOS) without compromising patient outcomes is a national priority for hospitals. The authors aimed to systematically and
safely improve LOS and LOS index (LOSI) at an urban academic safety-net hospital.
Methods: The research team conducted a multifaceted quality improvement initiative in a 550-bed hospital, focusing
on improving LOSI and reducing LOS. Interventions included institutional investment in an LOS reduction program,
development of rigorous internal LOS data analytics, and multiple Diagnosis Related Group (DRG)–focused LOS reduction
initiatives (specifically, sepsis, obstetric, and psychiatric DRGs). Initial interventions occurred in mid-2019, with subsequent
iterative improvement through 2023. Statistical analyses assessed pre- and postintervention outcomes.
Results: Mean LOSI significantly improved postintervention from 1.15 to 1.02 (-0.13, p < 0.0001), with sustained per-
formance for more than three years. The average LOS demonstrated a non–statistically significant but clinically relevant
improvement from 6.24 to 5.91 days (-0.33 days, p = 0.45). Excluding outlier long-LOS encounters, the LOS demon-
strated a statistically significant improvement in the postintervention slope change (-0.02 per month, p = 0.04), indicating
a delayed improvement to LOS postintervention. There were no adverse effects on readmissions or mortality.
Conclusion: A multifaceted approach to LOS and LOSI improvement in a safety-net hospital yielded clinically significant
and sustained results, showcasing the value of integrated strategies and organizational commitment.

E xcess time patients spend in the hospital beyond what

is medically necessary heightens the risk of hospital-
acquired conditions, adds avoidable costs, and disrupts pa-
Benchmarking and comparing absolute LOS requires
adjustment for patient and disease factors using a wide ar-
ray of potential risk adjustment factors. There are many
tient flow throughout the hospital. Consequently, hospital ways to normalize the LOS across institutions, none of
length of stay (LOS) is a key target of quality improve- which is clearly superior to others.8 The Centers for Medi-
ment initiatives.1 Decreasing LOS, however, is challenging care & Medicaid Services (CMS) publishes an annually up-
due to nonclinical factors such as organizational culture, dated mean LOS (arithmetic mean and geometric mean)
bed capacity, local community resources, and patient so- for each Medicare Severity Diagnosis Related Group (MS-
cial risk factors.2 A common fear that shorter LOS will re- DRG). The Colorado Department of Health Care Policy
sult in higher readmission rates and higher mortality has and Financing publishes a table of severity-adjusted aver-
been refuted in several large investigations.3–5 For exam- age LOS for each All Patient Refined Diagnosis Related
ple, Kaboli et al.’s analysis of 129 Veterans Affairs hospitals Group (APR-DRG) using the national hospital-specific rel-
over 14 years showed that targeted efficiency efforts led to ative value version 33 from 3M (3M Health Information
shortened LOS, decreased 30-day readmissions, and fewer Systems). In addition, organizations such as Vizient incor-
deaths from any cause at 30 and 90 days after admission.4 porate proprietary risk adjustment methods to calculate av-
In addition, shorter LOS can increase hospital bed capac- erage LOS, which are available only to member institutions.
ity, potentially decreasing overnight stays in the emergency LOS index (LOSI) is a ratio of actual LOS to expected
department (ED), which have been linked to higher mortal- LOS. The actual LOS is the time a patient is hospitalized,
ity.6 With hospitals accounting for 37% of $3.7 trillion in often measured in days or hours. The expected LOS is cal-
US personal health care expenditures, addressing inefficien- culated based on the documentation in the patient’s medi-
cies and improving quality are top priorities for all hospitals cal record, which is translated by hospital coders into a Di-
and health care systems.7 agnosis Related Group (DRG). Severity of illness, risk of
mortality, significant comorbidities, and other patient- and
disease-specific factors lend weight to the DRG and affect
1553-7250/$-see front matter the expected LOS based on national averages for other pa-
© 2025 The Joint Commission. Published by Elsevier Inc. All rights are re-
served, including those for text and data mining, AI training, and similar tech-
tients with the same DRG. An LOSI of 1.0 indicates that
nologies. the actual LOS was equal to the expected LOS. When an
https://doi.org/10.1016/j.jcjq.2025.01.012
2 Joseph Walker Keach, MD, et al. Reducing Hospital Length of Stay

LOSI is greater than 1.0, it can indicate that there are inef- This project was reviewed by the Quality Improvement
ficiencies in the way care is provided, or that patient com- Review Committee (QuIRC) of Denver Health, which is
plexity was underestimated by the identified billing codes. authorized by the Colorado Multiple Institutional Review
Our institution chose to internally calculate the LOSI using Board (COMIRB) at the University of Colorado, Denver,
an individual patient’s actual LOS divided by the patient’s and was determined not to be human subjects research. As
expected LOS (using the billed DRG’s annually updated such, this project does not require IRB review. We used the
arithmetic mean LOS from CMS for MS-DRGs and the SQUIRE 2.0 publication guidelines to format this report.12
published severity-adjusted average LOS from Colorado for
APR-DRGs). Outcomes
Improving overall LOSI is anchored on efficient hospi- The primary outcomes of interest were (1) LOSI (mean
talwide patient flow—the right care, in the right place, at ratio of observed to risk-adjusted expected LOS), and (2)
the right time—to avoid putting patients at risk for subop- average (mean) LOS. Analyses were stratified by complex
timal care or potential harm. At our large academic safety- vs. noncomplex patients (described below). Balancing mea-
net hospital, despite prior concentrated efforts to improve sures included 30-day readmission rates and in-hospital
timely discharge,9 we had an overall LOSI of 1.11 to 1.16 mortality.
for several years, indicating that we were keeping similar
patients 11% to 16% longer than our peers. Despite 18 Statistical Analysis
months of continuous focus from leadership, this metric The preintervention period was defined by discharges oc-
failed to improve. Given mounting challenges to smooth curring from January 1, 2017, through June 30, 2019.
hospital flow and the financial health of our institution, im- The intervention period is defined as July to September
proving the LOSI became a top priority of our organization. 2019. The postintervention period began October 1, 2019,
The primary purpose of this initiative was to systematically and ended December 31, 2023. Over the entire period,
improve the overall LOSI by integrating external evidence 140,334 inpatient discharges occurred (Figure 1). Newborn
with internal data and experience, a practice that led to our nursery discharges and organ donation discharges were ex-
recognition as a learning health system in 2019.10 , 11 cluded consistent with standard LOS metric tracking. As
inpatient admissions for only emergent dialysis are uncom-
METHODS mon nationally and occurred at our institution secondary
to the state’s payment arrangement with hospitals during
Setting and Participants the first four years of this analysis, we excluded these emer-
The quality improvement initiative was conducted at a gent dialysis–only discharges. We also excluded eating dis-
550-licensed-bed academic safety-net hospital in Denver. order specialty unit discharges (managed separately from
The facility (at the time of this publication) has 263 adult the acute care hospital and not part of this intervention)
medical/surgical, progressive care, and intensive care beds; and discharges from long-stay contracted beds on the psy-
57 adult psychiatric beds; 21 psychiatric adolescent beds; chiatric unit where discharge decisions are determined by
44 pediatric/neonatal beds; 46 obstetrics beds; 30 eat- the state. After these exclusions, 133,311 discharges were
ing disorder unit beds; and 19 correctional care beds. It included in the analysis (approximately 95% of total). Ob-
serves primarily Medicaid patients (58.9% of total inpatient servation hospital encounters were not included in any part
discharges), with the remainder being Medicare (19.2%), of this analysis consistent with national standards for LOS
commercial insurance (15.0%), uninsured (5.9%), and metric reporting (observation encounters are not assigned
other (1.0%). The ED has 110 adult/pediatric/psychiatric a DRG nor an expected LOS, and are measured in hours,
beds and sees an annual volume of greater than 100,000 not days).
visits. Staff physicians (such as hospitalists, surgeons, obste- We compared the impact of the interventions before
tricians, and psychiatrists) manage hospitalized patients in and after the intervention period. We used statistical pro-
conjunction with advanced practice providers, medical stu- cess X-bar charts to analyze the mean LOSI, mean LOS,
dents, and resident house staff. 30-day all-cause readmission, and inpatient mortality by
Our study took place during the COVID-19 pandemic, month. The outcome analyses used segmented regression
which meaningfully affected hospitalizations in 2020 and autoregressive error models. The Durbin-Watson statistics,
2021—particularly among adult medical/surgical patients, autocorrelation function, and partial autocorrelation func-
and less so for the obstetric and psychiatric groups. Hospi- tion were used to determine residual covariance lag order.
tal volumes at our institution were stable at approximately Stationarity was determined using the Augmented Dickey-
1,500 discharges per month during the preintervention pe- Fuller test. Interrupted time series analyses account for secu-
riod. During the postintervention period, including the lar trends through the preintervention and postintervention
pandemic effects, volumes steadily increased to approxi- slopes. Control charts were created in QI Macros v2020,
mately 1,800 discharges per month in the final year (Sup- and statistical analysis was performed using SAS Enterprise
plemental Figure 1, available in online article). Guide 7.13 (SAS Institute Inc., Cary, North Carolina).
Volume 000, No. , February 2025 3

Figure 1: Shown here is the number of discharge encounters included in length of stay analyses.

Interventions DRG’s arithmetic mean LOS, depending on the payer).

Institutional Investment in an LOS Reduction Pro- This normalized LOS method was pursued, as our insti-
gram. In 2019 the hospital CEO identified improving tution serves a majority Medicaid population (billed under
LOS as a top priority for the organization. Executive, nurs- APR-DRGs) and a smaller population of commercially in-
ing, and physician leaders aligned on this goal. Two posi- sured and Medicare patients (billed under MS-DRGs). By
tions were created to specifically focus on LOS improve- dividing each patient’s observed LOS by the expected LOS
ment: a 0.5 full-time equivalent (FTE) medical director (APR- or MS-DRG arithmetic mean LOS), we calculated
of patient flow and a 1.0 FTE operations manager of pa- each patient’s specific LOSI. We were also able to stratify
tient flow. These two positions acted as a single point of by specific patient-level variables (for example, homeless-
accountability for improving LOS and were tasked with ness, established with primary care in our system, medical
implementing and monitoring various LOS improvement comorbidities) and particular DRGs. Furthermore, we di-
projects across the organization. These positions received vided our patients into two distinct groups: complex (or
extensive support from executives and leaders across the or- outlier) patients (LOSI greater than or equal to 3 times ex-
ganization. pected) and noncomplex patients (LOSI less than 3 times
expected). Approximately 96% of our discharges fell in the
Development of Rigorous Internal LOS Data Ana- noncomplex group, and the remaining 4% were complex,
lytics. An early focus of this effort was optimizing internal or outlier, discharges. We focused our internal process im-
LOS data analytics. Various external organizations, such as provement efforts on the noncomplex LOSI patients, while
Vizient, can aggregate and display an organization’s LOS the complex LOSI patients were thought to require exten-
metrics; however, there is often a several month delay in re- sive coordination and health care policy improvement out-
ceiving data and those data are not granular enough to ana- side of our organization.
lyze a specific patient’s LOS and patient-level variables. We Creation of DRG–Focused LOS Reduction Inter-
used Tableau (Salesforce, Inc., San Francisco) to aggregate ventions. We analyzed discharge DRGs by volume and
and display our data, which allowed for near real-time LOS LOSI. We created an opportunity score for each DRG
data monitoring, for groups and individual patients. To cal- by multiplying the DRG encounter volume by the LOSI
culate LOSI, we used the patient’s actual LOS (in hours) di- (Figure 2). This allowed us to focus our LOS improve-
vided by their expected LOS (using either the APR- or MS- ment work on six high-volume, high-impact DRGs specific
4 Joseph Walker Keach, MD, et al. Reducing Hospital Length of Stay

Figure 2: A real-time data display (Tableau, Salesforce, Inc.) Diagnosis Related Group (DRG) opportunity score (DRG en-
counter volume × length of stay index [LOSI]) prior to the intervention, shown here as a screen capture, was used to decide
which DRGs to target for intervention. The far left column includes the names of the DRGs. The second column displays the
opportunity score (calculated based on volume and actual LOSI). The middle column is the volume of patient encounters
for the DRG with an LOSI < 1 (in blue) and > 1 (in red/orange). On the far right is the actual LOSI for each DRG. Apr Drg,
All Patient Refined Diagnosis Related Group.

Table 1. List of Diagnosis Related Group–Focused Interventions.

Area Diagnosis Related Groups - Interventions
APR-DRG
Obstetrics 560 – Vaginal delivery • Expanded lactation consultation from 12 hours per day to 24 hours per
540 – Cesarean section without day
sterilization • Communicated to providers expected LOS post delivery
Psychiatry 751 – Major depressive disorders • Educated care team about expected LOS
750 – Schizophrenia • Focused discharge rounds and planning on expected LOS and
753 – Bipolar disorders remaining time in stay
• Optimized attending staffing on weekends to enable weekend
discharges
Sepsis 720 – Septicemia • Focused improvement work on top three diagnoses: cellulitis, urinary
tract infection, pneumonia
• Created an antibiogram with guidelines for treatment, clinical stability
criteria, and oral step-down therapy
• Optimized timing of intravenous antibiotics (ceftriaxone) to facilitate
timely discharge

APR-DRG, All Patient Refined Diagnosis Related Groups; LOS, length of stay.

to our safety-net institution’s patient populations and ar- provement.13 There is a focus on continuously adding in-
eas of inefficiency. DRGs chosen for rapid LOS improve- cremental value vs. needing a perfect solution from the be-
ment projects were sepsis (largely internal medicine), un- ginning. Throughout the improvement cycle, we frequently
complicated vaginal and cesarean delivery (obstetrics and monitored our LOS metrics for each intervention group. As
pediatrics), and decompensated bipolar, schizophrenia, and time progressed, we were able to analyze what was working
depression (psychiatry). We created three working groups to (and what was not) and provide real-time feedback to those
focus on these DRGs, consisting of frontline representatives doing the actual work of patient care. At first the changes
from physicians, nursing, pharmacy, care management, and implemented were novel, but over time, with continuous
others. Lean methodology was used to root-cause analyze improvement and data monitoring, we were able to oper-
the longer-than-expected LOS in each group. We used our ationalize all our improvements into organization standard
internal data analytics team to validate various hypothetical work.
root causes and target our interventions. We rapidly imple- Additional Institutional Interventions. In addition
mented several DRG–focused interventions (Table 1) and to our targeted DRG approach during the intervention pe-
closely monitored for improvement. riod of 2019, we implemented numerous other interven-
Continuous and Iterative Improvement. Lean is a tions over time (2020–2023), which we believe contributed
practice that looks at ways to reduce waste in a process, to the durability of our LOS improvement. These interven-
ways to increase value to the customer, and continuous im- tions were rolled out gradually, in an overlapping fashion
Volume 000, No. , February 2025 5

over years, with intermittent uptake. Thus the impact in discharge medical respite beds. This allowed us to provide
further improving, or sustaining our improvement, is diffi- better post-acute care to our patients experiencing home-
cult to ascertain. However, these additional strategies would lessness and likely allowed these unhoused patients to dis-
not have been possible without the aforementioned inter- charge sooner. Finally, we improved our utilization review
ventions (namely, investment in an LOS reduction pro- workflows to appropriately capture more patients meeting
gram, rigorous data analytics, and Lean continuous im- medical necessity as inpatients.
provement strategies).
We created a postdischarge outpatient clinic that could
remotely monitor patients discharged from the hospital RESULTS
with common medical diagnoses, such as respiratory fail- LOSI and LOS data are available for January 2017 through
ure, heart failure, and acute kidney injury. This was partic- December 2023. During this time, there were 133,311 in-
ularly important during the peak of the COVID-19 pan- patient discharges contributing to LOS and LOSI calcula-
demic, and we believe it allowed us to discharge some pa- tions. Figure 1 details patients included for analysis. The
tients earlier and prevent readmissions. We implemented quality improvement project is ongoing, but the initial tar-
Epic (Epic Systems Corporation, Verona, Wisconsin) elec- geted intervention period lasted from July through Septem-
tronic health record (EHR) tools such as expected discharge ber 2019. There was no evidence of autocorrelation for
dates, discharge milestones, and standard workflows to re- LOSI or average LOS for the entire cohort, but autocor-
port on discharge barriers daily. This was combined with relation did occur in both outcomes when complex pa-
creating a patient flow command center to bring together tients were excluded. The models with noncomplex patients
teams involved in minute-to-minute patient flow control. were fit with a one lag error model to adjust for the auto-
By better tracking expected discharges and discharge bar- correlation in the errors. All four models were stationary,
riers, we were better able to manage our bed capacity and so no correction was needed for seasonality or stationarity
ensure that we addressed common discharge barriers to en- (Table 2).
sure that discharges were not unnecessarily delayed. We cre- Mean LOSI prior to the observation period was
ated an ED–led clinical decision unit to pull observation 1.15, and following the intervention it significantly im-
pathway patients to one unit instead of hospitalizing on proved to 1.02 with an intervention effect of -0.13 (p
inpatient units. Geographically cohorting observation pa- < 0.0001) (Figure 3). For all outcomes, the preinterven-
tients in a protocolized, pathway-driven unit managed by tion trends were nonsignificant, indicating no month-to-
dedicated advanced practice providers both improved our month changes prior to the intervention. The postinter-
observation hours (which do not count toward standard vention trends were also nonsignificant for all outcomes,
LOS or LOSI calculations) and allowed us to treat more which demonstrates that the intervention effect was sus-
inpatients in our inpatient-licensed beds.14 We partnered tained. The intervention also significantly improved the
with local homeless organizations to lease additional post- LOSI when complex patients were excluded. Noncomplex

Table 2. Segmented Regression Models for Primary Outcomes.

Estimate 95% CI p Value∗
Length of Stay Index
Intervention effect -0.1275 -0.1845 to -0.0705 < 0.0001
Preintervention slope (/month) -0.0003 -0.0029 to 0.0023 0.82
Postintervention slope change 0.0003 -0.0026 to 0.0031 0.86
Length of Stay Index, Excluding Complex Patients
Intervention effect -0.0771 -0.0958 to -0.0584 < 0.0001
Preintervention slope -0.0002 -0.0011 to 0.0006 0.61
Postintervention slope change -0.0004 -0.0014 to 0.0005 0.41
Length of Stay
Intervention effect (days) -0.1768 -0.63 to 0.28 0.45
Preintervention slope (days/month) 0.0028 -0.0182 to 0.0237 0.80
Postintervention slope change (days/month) -0.0104 -0.033 to 0.013 0.38
Length of Stay, Excluding Complex Patients
Intervention effect (days) 0.0258 -0.3037 to 0.3553 0.88
Preintervention slope (days/month) 0.0039 -0.0119 to 0.0198 0.63
Postintervention slope change (days/month) -0.0196 -0.0377 to 0.0015 0.04
∗ A significant p value for the intervention effect indicates a significant change in the outcome immediately following the intervention. A
significant p value for the preintervention slope indicates a consistent change in the outcome even before the intervention. A significant
p value for the postintervention slope change indicates that the intervention had a different impact on the outcome over time.
CI, confidence interval.
6 Joseph Walker Keach, MD, et al. Reducing Hospital Length of Stay

Figure 3: This graph shows length of stay index pre- and postintervention. UCL, upper control limit; LCL, lower control
limit.

Figure 4: This graph shows noncomplex length of stay index (< 3 [roughly 96% of all discharges]) pre- and postintervention.
UCL, upper control limit; LCL, lower control limit.

LOSI prior to the observation period was 0.95; following vention (Figure 6). This improvement has been particularly
the intervention, noncomplex LOSI significantly improved pronounced since spring 2022 as the COVID-19 pandemic
to 0.85 with an intervention effect of -0.08 (p < 0.0001) hospitalization rates waned.
(Figure 4). For noncomplex LOSI, the postintervention LOSI data for targeted DRGs are presented in Table 3.
trends also demonstrated a significant and sustained inter- The targeted obstetric DRGs included 19,846 discharges,
vention effect. the vast majority of which were noncomplex (19,740). For
The average LOS demonstrated a non–statistically sig- the noncomplex discharges, a statistically significant reduc-
nificant but clinically relevant improvement from 6.24 to tion in LOSI was noted (p < 0.001). The targeted psychi-
5.91 days (-0.33 days, p = 0.45) (Figure 5). Excluding com- atric DRGs included 9,239 discharges, 8,752 of which were
plex LOS encounters, the noncomplex LOS demonstrated noncomplex. For the non-complex discharges, we also saw
a statistically significant improvement in the postinterven- a statistically significant reduction in LOSI (p < 0.001),
tion slope change (-0.02 per month, p = 0.04), indicat- as was the postintervention slope change (p = 0.02). The
ing a delayed improvement to noncomplex LOS postinter- sepsis DRG included 7,877 discharges, 7,620 of which
Volume 000, No. , February 2025 7

Figure 5: This graph shows average length of stay pre- and postintervention. UCL, upper control limit; LCL, lower control
limit.

Figure 6: This graph shows noncomplex average length of stay (< 3 [roughly 96% of all discharges]) pre- and postinter-
vention. UCL, upper control limit; LCL, lower control limit.

were noncomplex. No statistically significant impact was (LOSI ≥ 3) and noncomplex (LOSI < 3) populations, with
detected for LOSI. clear improvement in our noncomplex LOSI compared to
When analyzing LOS data, we saw statistically signif- no improvement (or worsening) in our complex LOSI.
icant improvement in the noncomplex LOS for our psy- Statistical process control analyses of in-hospital mor-
chiatric and obstetric targeted DRGs (p < 0.001 for both tality and 30-day readmissions showed only random cause
DRGs). The sepsis DRG LOS did not show a statistically variation over the entire study period (not shown).
significant improvement after the intervention. Our inter-
ventions to date have primarily focused on the noncomplex
patient population, and as such this is primarily where our DISCUSSION
LOS and LOSI improvements have occurred. Figure 7 is a Clinicians and leaders at acute care hospitals face pres-
screen capture of our live working tool (with the interven- sure from many stakeholders to reduce unnecessary costs
tion period superimposed) showing LOSI for our complex while maintaining high-quality outcomes. The results of
8 Joseph Walker Keach, MD, et al. Reducing Hospital Length of Stay

Table 3. Length of Stay Index for Targeted Diagnosis Related Groups.

All Discharges Noncomplex Discharges†

Discharges Estimate p Value∗ Discharges Estimate p Value∗

Obstetrics Length of Stay Index 19,846 19,740
Intervention effect -0.0703 0.18 -0.0997 < 0.001
Preintervention slope -0.0042 0.09 0.0005 0.72
Postintervention slope change 0.0040 0.14 -0.0006 0.71
Psychiatric Length of Stay Index 9,239 8,752
Intervention effect -0.6646 < 0.001 -0.3259 < 0.001
Preintervention slope -0.0026 0.58 0.0009 0.61
Postintervention slope change 0.0123 0.02 0.0011 0.57
Sepsis Length of Stay Index 7,877 7,620
Intervention effect -0.0130 0.90 0.0281 0.43
Preintervention slope 0.0031 0.50 -0.0013 0.45
Postintervention slope change -0.0043 0.39 -0.0008 0.67
∗ A significant p value for the intervention effect indicates a significant change in the outcome immediately following the intervention. A
significant p value for the preintervention slope indicates a consistent change in the outcome even before the intervention. A significant
p value for the postintervention slope change indicates that the intervention had a different impact on the outcome over time.
† Noncomplex discharges are defined as a length of stay index < 3 (roughly 96% of all discharges).

Figure 7: This screen capture of the live tool shows LOSI trends for complex (defined as LOSI ≥ 3) and noncomplex (defined
as LOSI < 3 [roughly 96% of all discharges]) patients. LOSI is denoted on the y-axis; time (months) is denoted on the x-axis.
The top (blue) line is complex LOSI discharges. The bottom (orange) line is noncomplex LOSI discharges. LOSI, length of
stay index; Drg, Diagnosis Related Group.
Volume 000, No. , February 2025 9

this quality improvement initiative suggest that a multi- plex patients between 2020 and 2022 seen in Figure 6 is
modal effort to reduce excess hospital days at our institution mirrored by similar increases in expected LOS during those
led to clinically significant and sustained improvement in years. The pandemic is one reason we focus our results on
LOS and LOSI with no measurable increase in readmissions the LOSI, which should mitigate the influence of changes
or mortality. In our resource-constrained safety-net hospital in patient severity.
caring for patients with high social and medical complex- We did not have the ability to effectively analyze
ity, we believe our strategy and experience will be of interest our data stratified by patient discharge disposition sec-
to similarly situated safety-net hospitals and to institutions ondary to our internal data collection limitations. An un-
with less complexity. Our approach reflects the high value derstanding of our interventions’ impact on LOS and
our clinical and administrative leaders place on the princi- LOSI by patient discharge disposition (for example,
ples and practices of a learning health system.10 As such, the home, skilled nursing facility, acute rehab) could have
described intervention is novel in its application while in- highlighted interactions between disposition and inter-
corporating components of other successful published LOS vention that we were unable to detect in the current
improvement approaches.15–17 analysis.
The spark that kicked off a cascade of linked and increas- Although improved coding could partially explain a
ingly complex interventions was a challenge from the CEO drop in LOSI, we believe increases in expected LOS were
to the executive leadership team to target improvement in largely driven by severity of illness as expected LOS re-
LOSI within 100 days. Paired with the challenge was a com- turned to preintervention levels by mid-2022 as hospital-
mitment to allocate resources to the effort for 100 days and, izations for COVID-19 became much less frequent. This
if successful, indefinitely. This resulted in a small patient quality improvement initiative was carried out at an ur-
flow team with dedicated time to design, implement, iter- ban, safety-net, teaching hospital, thus the DRGs cho-
ate, and monitor interventions. The CEO’s challenge and sen for intervention and the specific interventions em-
resource commitment ensured the engagement of the en- ployed at our institution may not be generalizable to other
tire organization, including many of the C-suite executives, institutions.
frontline providers, and staff. The other necessary infras-
tructure piece was a performance dashboard that allowed
the patient flow team to rigorously monitor and analyze CONCLUSION
near real-time LOS data by service line, DRG, patient type, Our quality improvement initiative demonstrated an im-
and location. mediate and sustained improvement to LOSI at a large,
Much of our work prior to this initiative had focused on academic, safety-net hospital. Key aspects of our initiative
complex adult patients with challenging psychosocial bar- included strong organizational support for LOS improve-
riers to discharge, the vast majority of whom would have ment and the creation of dedicated positions, rigorous in-
fallen into the complex group described herein.9 This group ternal data analytics, focused interventions with continuous
represents many of our outlier LOS patients, and the so- monitoring and iterative improvement, and ongoing work
lutions often relied on action from individuals or groups to improve internal hospital flow coordination and transi-
external to our organization. Conversely, for this initiative, tions of care. We believe this approach should also be ef-
we chose to focus on common conditions with typical LOS fective in improving LOS at other similarly situated insti-
values in which the locus of control was often within the tutions.
walls of the hospital. This represented a fundamental devi-
ation from prior efforts and may explain why we were able Conflicts of Interest. All authors report no conflicts of interest.
to achieve results.

Limitations
SUPPLEMENTARY MATERIALS
As with any quality improvement effort, our interventions
Supplementary material associated with this article can be
were implemented in an environment of dynamic unavoid-
found, in the online version, at doi:10.1016/j.jcjq.2025.01.
able variation and of changes in staff, patients, guidelines,
012.
policies, and leadership. Although we believe that our de-
scribed interventions are responsible for the improvement Joseph Walker Keach, MD, is the Medical Director of Hospital Care
Management and Patient Flow, Division of Hospital Medicine, at Den-
in key outcomes, it is possible that unmeasured changes in ver Health Medical Center, and an Associate Professor at the Univer-
the hospital environment, patient population, or staff may sity of Colorado School of Medicine Mara Prandi-Abrams is the Op-
explain the improvement in our results. erations Manager of Patient Flow at Denver Health Medical Center
Allison S. Sabel, MD, PhD, MPH, is the Medical Director of Clinical
We faced a global pandemic during our postinterven- Data and Analytics at Denver Health Medical Center, and a Visiting
tion period, which undoubtedly influenced the severity of Associate Professor at the University of Colorado School of Medicine
Romana Hasnain-Wynia, PhD, is the Chief of Academic Affairs and Public
illness in our hospitalized patients and hence their expected Health at Denver Health Medical Center, and a Professor at the University
LOS. In fact, the rise in observed LOS among noncom-
10 Joseph Walker Keach, MD, et al. Reducing Hospital Length of Stay

9. MacKenzie TD, et al. A discharge panel at Denver Health,

of Colorado School of Medicine Jonathan M. Mroch is the Director of
Enterprise Data at Denver Health Medical Center Thomas D. MacKen-
focused on complex patients, may have influenced decline
zie, MD, MSPH, is an Internal Medicine Physician and Senior Advisor of in length-of-stay. Health Aff (Millwood). 2012;31:1786–1795.
Health Systems Learning at Denver Health Medical Center, and a Pro- 10. Agency for Healthcare Research and Quality. Denver
fessor at the University of Colorado School of Medicine. Please address Health: How a Safety Net System Maximizes Its Value, Apr
correspondence to Joseph Walker Keach MD., [email protected]. 2019. AHRQ Publication No. 19-0052-3Accessed Feb 3,
2025 https://www.ahrq.gov/sites/default/files/wysiwyg/lhs/
lhs_case_studies_denver_health.pdf .
11. Borsky AE, et al. AHRQ series on improving translation of
REFERENCES evidence: perceived value of translational products by the
1. Lingsma HF, et al. Evaluation of hospital outcomes: the re- AHRQ EPC Learning Health Systems Panel. Jt Comm J Qual
lation between length-of-stay, readmission, and mortality in Patient Saf. 2019;45:772–778.
a large international administrative database. BMC Health 12. Goodman D, et al. Explanation and elaboration of the
Serv Res. 2018 Feb 14;18:116. SQUIRE (Standards for Quality Improvement Reporting Ex-
2. Brasel KJ, et al. Length of stay: an appropriate quality mea- cellence) guidelines, V.2.0: examples of SQUIRE elements
sure? Arch Surg. 2007;142:461–465. in the healthcare improvement literature. BMJ Qual Saf.
3. Han TS, et al. Evaluation of the association of length of stay 2016;25:e7.
in hospital and outcomes. Int J Qual Health Care. 2022 May 13. Gabow PA, Goodman PL. The Lean Prescription: Powerful
2;34:mzab160. Medicine for Our Ailing Healthcare System. Boca Raton, FL:
4. Kaboli PJ, et al. Associations between reduced hospital CRC Press, 2014.
length of stay and 30-day readmission rate and mortality: 14. Rothman RD, Peter DJ, Harte BJ. Improving healthcare
14-year experience in 129 Veterans Affairs hospitals. Ann In- value: managing length of stay and improving the hospital
tern Med. 2012 Dec 18;157:837–845. medicine value proposition. J Hosp Med. 2021;16:620–622.
5. Brown CM, et al. Trends in length of stay and readmissions 15. Jweinat J, et al. The safe patient flow initiative: a collabora-
in children’s hospitals. Hosp Pediatr. 2021;11:554–562. tive quality improvement journey at Yale-New Haven Hos-
6. Roussel M, et al. Overnight stay in the emergency depart- pital. Jt Comm J Qual Patient Saf. 2013;39:447–459.
ment and mortality in older patients. JAMA Intern Med. 16. Kutz A, et al. Association of interprofessional discharge
2023 Dec 1;183:1378–1385. planning using an electronic health record tool with hospital
7. Hartman M, et al. National health care spending in 2022: length of stay among patients with multimorbidity: a non-
growth similar to prepandemic rates. Health Aff (Millwood). randomized controlled trial. JAMA Netw Open. 2022 Sep
2024;43:6–17. 1;5:e2233667.
8. Lu M, et al. Systematic review of risk adjustment models 17. Siddique SM, et al. Interventions to reduce hospital length
of hospital length of stay (LOS). Med Care. 2015;53:355– of stay in high-risk populations: a systematic review. JAMA
365. Netw Open. 2021 Sep 1;4:e2125846.