Abstract
Background
Discharge delays are difficult to quantify without standardized indicators for when patients are medically ready for discharge. We aimed to estimate the proportion of increased hospital length of stay attributable to discharge delays, as proxied by increases in ‘avoidable’ days.
Methods
We conducted a retrospective cohort study of Veterans Health Administration hospitals in the continental United States with emergency departments between 1 March 2019 and 28 February 2023. We included Veterans who were discharged from an acute medicine service without a COVID-19 diagnosis. We used standardized utilization management criteria to count ‘avoidable’ days, defined as hospital days when acute care was no longer required. Our primary outcome was geometric mean length of the discharging stay (the final acute medicine segment prior to discharge), which reflects the time most susceptible to discharge delays.
Results
During the study period there were 868,031 eligible hospitalizations. Adjusted geometric mean length of discharging stay increased 9.3% (95% CI, 8.7% to 9.9%) from the pre-pandemic year to the third pandemic year, with the largest increase among discharges to facility-based post-acute care (23.3% [95% CI, 21.6 to 24.9%]). However, among all hospitalizations only 16% (95% CI, 15 to 17%) of the increase in discharging stay was attributable to an increase in avoidable days.
Conclusions
Most of the increase in length of hospital discharging stay was not explained by discharge delays and may instead reflect longer periods of acute care delivery. Improving acute care processes may more effectively reduce hospital capacity strain than bolstering post-acute care availability.
Supplementary Information
The online version contains supplementary material available at 10.1186/s12913-025-13682-w.
Keywords: Length of stay, Hospitalization, Patient discharge, Patient transfer, Aftercare, Continuity of patient care
Introduction
Shortages of hospital beds—exacerbated by the onset of the COVID-19 pandemic [1]—have renewed attention to discharge delays as targets for improving inpatient capacity [2, 3]. Yet the impacts of discharge delays are difficult to quantify [4]. Most electronic medical records lack standardized indicators for when acute care delivery is complete or patients are “medically ready for discharge” [5]. How much reductions in discharge delays would improve hospital capacity strain (or “gridlock”) [6] is not well known [7].
To address this gap, we leveraged standardized utilization review data from Veterans Health Administration (VHA) hospitals to assess whether rising discharge delays contributed to longer hospital stays. Specifically, we examined trends in “avoidable” days—hospital days for which, according to utilization reviewers, acute-level care was no longer needed. We tested whether the rise in “avoidable” days, as a proxy for the rise in discharge delays, could explain the observed increase in length of discharging stay, the final stay on the acute medicine service before discharge and the portion of hospitalization most sensitive to discharge delays.
Methods
Cohort creation
After the Veteran’s Institutional Reviewer Board of Northern New England granted exemption from informed consent (VINNE 1,641,396), we used the methods of Vincent et al. To establish start and stop times for hospitalizations within the VHA Corporate Data Warehouse (CDW) [8]. Patients’ movements within and across VHA facilities are indexed as “bedded stays”—durations of time when a patient is provided a bed at a particular location. A bedded stay may be a complete stand-alone hospitalization or just part of one. As patients move bed locations, multiple bedded stays combine to form hospitalizations [9]. In addition, bedded stays themselves may be comprised of shorter specialty stays: unique combinations of locations and specialties (also known as services or specialty services). We delineated hospitalizations by first extracting the timestamps to split all bedded stays into specialty stays (Additional File 1). We then removed Extended Nursing or Housing specialty stays, as defined in Additional File 2, and spliced together adjacent specialty stays. We developed rules to deal with overlapping stays, which sometimes occur when a patient administratively occupies more than one bed at a time. We validated our method by comparing weekly national medicine and surgery inpatient census totals against a wholly separate database that is manually compiled by bed flow coordinators at all VHA hospitals. As described in Additional File 3, the census totals from both methods were remarkably similar across the study period, validating our approach.
We identified eligible hospitalizations of VHA enrollees at one of 109 VHA hospitals in the continental US with an emergency department throughout the 4-year study period—from 1 year before to 3 years after February 29, 2020, the date of the first COVID-19 related death in the US (Fig. 1). We restricted hospitalizations to those in which patients were discharged from an acute medicine service, and we excluded hospitalizations in which patients signed out against medical advice or had a COVID-19 diagnosis.
Fig. 1.
Cohort creation and data linkages. Abbreviations: VHA = veterans health administration affairs medical center; COVID-19 = coronavirus disease 2019. For the definitions of criteria used in this figure, see Additional file 4
We then linked to the Geriatrics & Extended Care Data Analysis Center Residential History File, which combines enrollment, utilization, and assessment data from VHA-care, VHA-purchased care, Medicare, and Medicaid to determine veteran movements between health care settings [10].
We used this file to exclude veterans who were enrolled in hospice, transferred to a non-VHA hospital for ongoing acute care, or discharged to a domiciliary. Further details about the definitions used to define our cohort are listed in Additional File 4.
Utilization review data
At VHA hospitals, trained nurses conduct concurrent utilization management reviews to determine if hospitalized patients meet standardized InterQual® criteria for continued stay at an acute level of care [11]. Utilization management reviewers are expected to conduct continued stay reviews for at least 80% of bed-days (1 hospital bed occupied for 1 day comprises 1 bed-day unit; we use the terms “bed-day” and “day” interchangeably); this expectation does not include days of hospital admission (which are evaluated against admission criteria), days of discharge, days beyond 30-days on the same specialty, or days enrolled in hospice care.
We retrieved archived reviews from the National Utilization Management Integration (NUMI) database and included any that were conducted during a patient’s final acute medicine stay. Given that such reviews are not conducted daily, we generated 2 rules to impute identical reviews across bed days within hospitalizations. First, we interpolated reviews when they were identical for bed-days before and after a gap. Second, we carried forward the last review that was conducted during a hospitalization if, during a 7-day period that ended with that last review, all of 2 or more reviews were identical. A day was classified as “avoidable” if a reviewer determined that 1) the patient no longer met continued stay criteria and 2) care could be provided at a lower level of acuity, such as at home or in a skilled nursing facility [12].
Continued stay criteria were based on disease-specific InterQual® criteria, which the VHA has been using to conduct utilization management reviews since 1993. Generally, patients meet criteria for continued stay when interventions are still needed for continued clinical improvement or for addressing barriers to a safe discharge. For example, a patient with severe pneumonia who, on hospital day 4, no longer requires intravenous antibiotics or supplemental oxygen, tolerates oral medication, breathes comfortably, and maintains stable vital signs, may nevertheless meet continued stay criteria if he continues to lack functional independence. If barriers to a safe discharge are then addressed on day 7, for example, by the patient regaining functional independence , he would no longer meet continued stay criteria. And if reviewers also determined that care could be provided at a lower level of acuity, then day 7 and any additional days until hospital discharge would meet our definition of avoidable.
Statistical analysis
Our primary outcome was the length of discharging stay, defined as the number of days on the final acute medicine service before discharge; we transformed the length of discharging stay (measured in days) to the natural log scale (loge(days)) to reduce the influence of extreme values and improve model convergence. We chose this segment of hospitalization to focus on the impact of delays related to hospital discharge, rather than the impacts of other factors such as transfers from intensive care or between other acute services [13]. Our exposures of interest were 1) year of discharge, 2) discharge destination (home, at-home post-acute care, or facility-based post-acute care), and 3) avoidable days. Potential confounders were defined at 3 levels: hospital-level were geographic region, rurality, and whether a VHA nursing home (a Community Living Center [CLC]) existed on campus (Additional File 5); patient-level were sex, race and ethnicity; and hospitalization-level were age at discharge, comorbidity at discharge, pre-hospital care, admission from a non-VHA hospital, admission to observation status, stays on specialties other than acute medicine, post-acute facility type, and similarity of pre- and post-hospital care. Further details about how exposures of interest and potential confounders were defined are listed in Additional File 6.
Patients are often hospitalized more than once, but they do not always return to the same hospital. To correctly adjust for the clustering of unique combinations of patients and hospitals, we built a nonhierarchical mixed effects model with additive crossed random effects for both patients and hospitals using a maximum likelihood estimator with the mixed command in Stata 18.0. For fixed effects, we entered discharge destination (an exposure of interest) into our model as both a main effect and as an interaction with year to allow its impact to vary by year. We entered avoidable bed days (another exposure of interest) and all potential confounders as both main effects and as interactions with year and discharge destination singly (as two-way interactions) and combined (as three-way interactions) to allow their impacts on the outcome (discharging length of stay) to vary by year, discharge destination, and the combination of year-by-discharge destination. While inclusion of three-way interactions makes our model less parsimonious, our aim was to capture the full complexity of the underlying phenomenon. Moreover, we present predictions from the entire model, not stand-alone estimates of individual coefficients. And so, the challenge of interpreting three-way interactions was not a drawback.
We used likelihood ratio tests, Akaike’s information criterion, and Bayesian information criterion to assess our fully ‘saturated’ model (with two- and three-way interactions, as described above) against simpler, less saturated versions. In all cases, models saturated with more interactions had improved fit over less saturated models, and so we retained the fully saturated model. Finally, we assessed the normality assumption of both random effects, as described in Additional File 7.
To assess the impact of avoidable days on discharging length of stay, we used the margins command [14] in Stata 18.0 to generate two sets of model-based predictions (‘conditional margins’) [15], while holding all confounders at their pre-pandemic values. In the first set, avoidable days were held constant at their pre-pandemic year average; in the second, they were allowed to vary by discharge year.
One benefit of log-transforming our outcome variable (length of discharging stay) is clear interpretation of predictions. The arithmetic mean of a log-transformed variable is, upon back-transformation, equal to the geometric mean of the variable in the original scale. And so, because our model predictions are arithmetic means in the natural log scale (loge(days)), back-transformation creates geometric means of our outcome variable, length of discharging stay. In addition, because length of stay variables often approximate log-normal distributions [16], a geometric mean will approximate a median. Another benefit is that differences between our predictions (arithmetic means in the natural log scale) will equal ratios of geometric means upon back-transformation [17]. These ratios are the relative (or ‘fractional’) changes in geometric mean length of discharging stay (eg, the relative increase between pandemic year 3 and the pre-pandemic year). We used the margins command with the contrast option to generate these differences.
Finally, we used the nlcom command in Stata 18.0 to calculate the proportion of the increase in adjusted length of discharging stay attributable to avoidable days by comparing two sets of predictions in which confounders were held at pre-pandemic values. In the first set, avoidable days were also held at pre-pandemic values; in the second, they were allowed to vary by discharge year. The proportion of the increase in adjusted length of discharging stay attributable to avoidable days was calculated by comparing these two sets of predictions for each pandemic year relative to baseline. Both the margins and nlcom commands in Stata 18.0 use the delta method to estimate confidence intervals, which were calculated based on the standard normal distribution.
Results
Our cohort included 868,031 hospitalizations among 481,695 unique veterans, most of whom were elderly, non-Hispanic white men (Table 1). Annual hospitalizations were 47,703 lower during the pandemic years in comparison to the pre-pandemic year (Table 2); exclusion of veterans with COVID-19 explained less than half of that difference (Fig. 1). Discharge destinations were home (70.1%), at-home post-acute care (16.7%), and facility-based post-acute care (13.2%). In 77.6% of cases, the discharge destination matched the patient’s pre-hospital care setting.
Table 1.
Characteristics of hospitalizations and veterans in study cohort
| Hospitalizations | ||||
|---|---|---|---|---|
| Overall (n = 868,031) |
Discharge Destination | |||
| Characteristic | Home (n = 608,078) |
Home with post-acute care (n = 145,367) |
Post-acute facilitya (n = 114,586) |
|
| Unique patientsb | 481,695 | 382,595 | 98,897 | 82,453 |
| Age, mean (SD), y | 70 (13) | 68 (13) | 74 (11) | 76 (10) |
| Men, % | 94.5 | 94.0 | 95.6 | 96.4 |
| Race, % | ||||
| White | 71.4 | 70.6 | 72.6 | 73.8 |
| Black | 25.4 | 25.8 | 24.9 | 23.9 |
| Otherc | 2.1 | 2.2 | 1.9 | 1.8 |
| Unknown | 1.1 | 1.4 | 0.5 | 0.5 |
| Hispanic, % | 5.1 | 5.4 | 4.9 | 4.1 |
| Geographic region, %d | ||||
| Northeast | 12.8 | 12.4 | 12.0 | 15.6 |
| Midwest | 22.7 | 23.0 | 20.3 | 23.9 |
| South | 42.9 | 42.9 | 45.4 | 39.5 |
| West | 21.7 | 21.7 | 22.3 | 20.9 |
| Elixhauser Mortality Index, median (IQR) | 9 (−2 to 22) | 7 (−3 to 19) | 13 (1 to 27) | 15 (2 to 29) |
| Transferred from a non-VA hospital, % | 3.0 | 2.5 | 3.3 | 5.3 |
| Initially admitted as observation status, % | 16.0 | 16.0 | 17.2 | 14.5 |
| Previous stay, % | ||||
| Intensive care | 15.7 | 14.3 | 16.7 | 21.6 |
| Acute mental health | 0.39 | 0.43 | 0.19 | 0.42 |
| Acute surgery | 1.2 | 0.8 | 1.5 | 2.7 |
| Any intensive care, acute mental health, or acute surgery | 16.7 | 15.2 | 17.8 | 23.5 |
| Rural hospitale | 2.7 | 2.4 | 3.4 | 3.0 |
| VHA nursing home on campusf | 70.3 | 70.3 | 69.1 | 71.7 |
| Pre-hospital careg | ||||
| Home | 85.8 | 97.1 | 59.1 | 60.0 |
| Home with post-acute care | 9.7 | 2.1 | 39.5 | 12.5 |
| Post-acute facility | 4.5 | 0.8 | 1.4 | 27.5 |
Abbreviations: IQR = interquartile range; SD = standard deviation
a Of 114,586 hospitalizations discharged to nursing homes, 41.2% were to VHA nursing homes (Community Living Centers), 40.3% were to skilled nursing facilities, 13.6% were to long-term care facilities (also known as contract nursing homes), and 4.9%% were to State Veterans Homes
b Of 481,695 overall unique patients, 63,968 (13.3%) were discharged to 2 different destinations throughout the study period, and 9,141 (1.9%) were discharged to all 3
c The ‘other’ race category included American Indian, Alaska Native, Asian, Native Hawaiian, and other Pacific Islander
d Geographic region defined by the Centers for Disease Control
e Based on rurality classification of discharging VHA hospital (Additional File 5)
f VHA nursing homes are also known as Community Living Centers (Additional File 5)
g Of the entire cohort, 77.6% (673,875 hospitalizations) had the same types of pre-hospital care and discharge destination, including the same type of post-acute care facility. Among the 31,531 hospitalizations in which veterans were admitted from and discharged to a post-acute facility, 5,333 were discharged to of post-acute care facility that was different from the type on admission
Table 2.
Length of stay by discharge destination
| Year | Annual Hospitalizations | Length of Stay (days) | ||||
|---|---|---|---|---|---|---|
| (no.) | Change | Discharging Stay | Hospitalization | |||
| geometric mean (IQR) |
arithmetic mean (SD) | geometric mean (IQR) |
arithmetic mean (SD) | |||
| Discharged to Post-Acute Facility | ||||||
| Pre-Pandemic Year | 35,307 | — | 5.4 (3.2 to 8.8) | 7.6 (10.4) | 6.7 (4.0 to 10.8) | 9.4 (15.2) |
| Pandemic Years 1–3 | 26,426 | −25.2% | 6.4 (3.9 to 10.9) | 9.4 (14.3) | 7.9 (4.8 to 13.1) | 11.4 (17.4) |
| Pandemic Year 1 | 24,159 | −31.6% | 6.1 (3.7 to 10.7) | 9.0 (14.4) | 7.6 (4.6 to 12.9) | 11.0 (18.8) |
| Pandemic Year 2 | 28,497 | −19.3% | 6.3 (3.9 to 10.7) | 9.1 (13.8) | 7.8 (4.8 to 12.8) | 11.0 (16.7) |
| Pandemic Year 3 | 26,623 | −24.6% | 6.8 (4.0 to 11.7) | 10.1 (14.6) | 8.4 (5.0 to 13.8) | 12.0 (16.8) |
| Discharge to Home with At-Home Post-Acute Care | ||||||
| Pre-Pandemic Year | 35,369 | — | 3.3 (1.9 to 5.8) | 4.6 (5.8) | 4.2 (2.7 to 6.8) | 5.6 (6.6) |
| Pandemic Years 1–3 | 36,666 | +3.7% | 3.6 (2.1 to 6.3) | 5.2 (7.0) | 4.6 (2.8 to 7.5) | 6.3 (7.8) |
| Pandemic Year 1 | 33,992 | −3.9% | 3.6 (2.0 to 6.1) | 5.1 (7.9) | 4.5 (2.8 to 7.1) | 6.2 (8.6) |
| Pandemic Year 2 | 38,643 | +9.3% | 3.6 (2.1 to 6.3) | 5.1 (5.9) | 4.6 (2.8 to 7.4) | 6.2 (7.0) |
| Pandemic Year 3 | 37,363 | +5.6% | 3.7 (2.1 to 6.7) | 5.4 (7.3) | 4.7 (2.8 to 7.7) | 6.5 (8.0) |
| Discharged Home | ||||||
| Pre-Pandemic Year | 182,109 | — | 2.5 (1.6 to 4.3) | 3.7 (5.4) | 3.2 (2.0 to 5.0) | 4.5 (6.8) |
| Pandemic Years 1–3 | 141,990 | −22.0% | 2.7 (1.7 to 4.7) | 4.1 (7.3) | 3.4 (2.0 to 5.6) | 4.9 (9.6) |
| Pandemic Year 1 | 136,988 | −24.8% | 2.6 (1.6 to 4.3) | 3.9 (7.9) | 3.3 (2.0 to 5.4) | 4.8 (9.7) |
| Pandemic Year 2 | 146,046 | −19.8% | 2.7 (1.6 to 4.7) | 4.0 (6.3) | 3.4 (2.0 to 5.5) | 4.8 (8.0) |
| Pandemic Year 3 | 142,935 | −21.5% | 2.8 (1.7 to 4.8) | 4.3 (7.8) | 3.5 (2.0 to 5.7) | 5.1 (10.8) |
| All Hospitalizations | ||||||
| Pre-Pandemic Year | 252,785 | — | 2.9 (1.8 to 5.1) | 4.4 (6.5) | 3.7 (2.2 to 6.0) | 5.3 (8.7) |
| Pandemic Years 1–3 | 205,082 | −18.9% | 3.2 (1.8 to 5.8) | 5.0 (8.7) | 4.0 (2.4 to 6.8) | 6.0 (10.9) |
| Pandemic Year 1 | 195,139 | −22.8% | 3.1 (1.8 to 5.6) | 4.7 (9.1) | 3.9 (2.3 to 6.6) | 5.8 (11.3) |
| Pandemic Year 2 | 213,186 | −15.7% | 3.2 (1.8 to 5.8) | 4.9 (7.9) | 4.0 (2.4 to 6.7) | 5.9 (9.7) |
| Pandemic Year 3 | 206,921 | −18.1% | 3.3 (1.9 to 5.9) | 5.2 (9.1) | 4.1 (2.5 to 6.9) | 6.2 (11.6) |
Abbreviations: IQR = interquartile range; SD = standard deviation
Unadjusted lengths of stay, both for the discharging stay and the entire hospitalization, increased across the study period. After holding potential confounders to their pre-pandemic levels, the largest increases were among discharges to post-acute care (Fig. 2, Additional File 8). From the pre-pandemic year to pandemic year 3, for example, adjusted geometric mean length of discharging stay for discharges to at-home post-acute care increased by 12.2% (95% CI, 11.0 to 13.6%) and to facility-based post-acute care by 23.2% (95% CI, 21.6 to 24.9%), while those discharged home without post-acute increased by only 8.9% (95% CI, 8.3 to 9.5%).
Fig. 2.
Relative increase in geometric mean length of discharging stay. Both panels depict the relative increase in geometric mean length of discharging stay for each pandemic year in comparison to the pre-pandemic year. The top panel includes all hospitalizations, while the bottom panel stratifies hospitalizations by discharge destination. (absolute values for geometric mean length of discharging stay are depicted in Additional file 8). The circles represent observed values unadjusted for potential confounders; the sizes of the circles are proportional to the number of annual discharges. The dashed line represents the relative increase in geometric mean length of discharging stay when both avoidable bed days and potential confounders are held to mean values from the pre-pandemic year. In contrast, the solid line represents the relative increase in geometric mean length of discharging stay when potential confounders are held to mean values from the pre-pandemic year while avoidable bed days are set to the mean value for the corresponding year of hospital discharge. The error bars represent 95% confidence intervals. The relative increase for all 3 pandemic years combined (not shown), holding potential confounders to pre-pandemic levels, was 6.1% (95% ci 5.7 to 6.5%) for all hospitalizations; when grouped by discharge destination the increases were 5.8% (5.3 to 6.3%), 9.3% (8.2 to 10.3%), and 17.2% (16.1 to 18.4%) for discharges to home, to at-home post-acute care, and to post-acute facilities, respectively
The mean number of avoidable days per hospitalization rose from 0.64 in the pre-pandemic year to 1.05 in pandemic year 3 (Table 3). Had this increase in avoidable days not occurred, the adjusted geometric mean length of discharging stay would have increased from 2.91 days (95% CI 2.85 to 2.97 days) during the pre-pandemic year to only 3.13 days (95% CI 3.07 to 3.19 days), instead of 3.18 days (95% CI 3.11 to 3.24 days) during the third pandemic year; the modest difference between these predictions reflects only 16% (95% CI 15 to 17%) of the increase in discharging stay above baseline. This suggests that most of the increase (84%, 95% CI 83 to 85%) was attributable to something other than increases in avoidable days (Fig. 3).
Table 3.
Avoidable bed days by discharge destinationa
| Year | Hospitalizations | Hospitalizations with at least 1 avoidable bed day during discharging stay | Total Avoidable Bed Days | |
|---|---|---|---|---|
| N | No (%) | No. | Mean No per hospitalization | |
| Discharged to Post-Acute Facility | ||||
| Pre-Pandemic Year | 35,307 | 15,651 (44.3) | 68,095 | 1.93 |
| Pandemic Year 1 | 24,159 | 11,327 (46.9) | 63,187 | 2.62 |
| Pandemic Year 2 | 28,497 | 14,200 (49.8) | 79,953 | 2.81 |
| Pandemic Year 3 | 26,623 | 13,912 (52.3) | 92,327 | 3.47 |
| Discharged to Home with At-Home Post-Acute Care | ||||
| Pre-Pandemic Year | 35,369 | 6,143 (17.4) | 18,596 | 0.53 |
| Pandemic Year 1 | 33,992 | 5,686 (16.7) | 20,336 | 0.60 |
| Pandemic Year 2 | 38,643 | 6,733 (17.4) | 23,537 | 0.61 |
| Pandemic Year 3 | 37,363 | 6,710 (18.0) | 29,000 | 0.78 |
| Discharged Home | ||||
| Pre-Pandemic Year | 182,109 | 23,350 (12.8) | 74,887 | 0.41 |
| Pandemic Year 1 | 136,988 | 16,910 (12.3) | 63,466 | 0.46 |
| Pandemic Year 2 | 146,046 | 18,892 (12.9) | 75,975 | 0.52 |
| Pandemic Year 3 | 142,935 | 18,935 (13.3) | 95,915 | 0.67 |
| All Hospitalizations | ||||
| Pre-Pandemic Year | 252,785 | 45,144 (17.9) | 161,578 | 0.64 |
| Pandemic Year 1 | 195,139 | 33,923 (17.4) | 146,989 | 0.75 |
| Pandemic Year 2 | 213,186 | 39,825 (18.7) | 179,465 | 0.84 |
| Pandemic Year 3 | 206,921 | 39,557 (19.1) | 217,242 | 1.05 |
a Avoidable bed days were only included in our analysis if they occurred during the discharging stay
Fig. 3.
Proportion of overall increase in length of discharging stay attributable to avoidable days. The heights of the bars represent the annual increase in geometric mean length of discharging stay in comparison to the pre-pandemic year for all veterans discharged from acute medicine services. Colored bars represent the fractions of annual increases that are attributable to increases in avoidable bed days accrued before discharge, grouped by eventual discharge destination. Clear bars represent the fractions of annual increases that were not attributable to increases in avoidable bed days. For example, 6.4% (95% CI 6.0 to 6.8%) of the increase in length of discharging stay for all 3 pandemic years combined when compared to the pre-pandemic year is attributable to the increase in avoidable bed days accrued from veterans who were discharged home; 4.0% (3.7 to 4.3%) from veterans discharged to at-home post-acute care, and 5.2% (4.9 to 5.6%) from veterans discharged to facility-based post-acute care; the remaining 84.4% (83.4 to 85.3%) of the increase was not attributable to increases in avoidable bed daysci
Conclusion
We found that only a small fraction of the increase in adjusted length of discharging stay over 3 pandemic years compared to the pre-pandemic year was attributable to an increase in avoidable days (Fig. 3). Since avoidable days ought to increase alongside discharge delays, this small fraction implies that most of the increase was not related to more discharge delays, suggesting that more time may have been spent on acute care delivery—the period before a patient is medically ready for discharge.
Our findings should be interpreted with 3 important limitations. First, VHA utilization management reviewers, working alone, may have overcounted or undercounted avoidable days. A multidisciplinary team that also included physicians, nurses, and social workers would likely be better equipped to determine whether a patient’s complex social, cognitive, and physical dimensions warrant ongoing hospitalization [5]. Nevertheless, when the purpose of measurement is to assess longitudinal change, stable systematic errors tend to cancel each other out, if that measurement is a reliable one [18]. Because the methods of VHA utilization reviewers are standardized and repeatedly evaluated to maintain high reliability [19], changes in avoidable days (Table 3) were likely to reflect actual change, not mere measurement error. Second, the COVID-19 pandemic may have altered hospital operations and patient behavior. To mitigate this, we excluded veterans diagnosed with COVID-19 and defined our study years relative to the onset of the pandemic. The fact that length of stay had been stable within VHA hospitals in the decade leading up to the pandemic [20], together with the consistent upward trends over 3 consecutive years, suggests that our findings are not merely a transient phenomenon. Third, we did not include some potential confounders in our model because they were not available. For example, nurses’ diagnoses upon hospital admission, such as risk for infection, impaired skin integrity, or disturbed sleep patterns [21], are associated with length of stay [22] and may have increased alongside the complexity of hospitalized patients [23]. While our inclusion of a comorbidity index, along with 15 other potential confounders, did address temporal changes in some of the determinants of length of stay (as evidenced by the difference between the circles and the solid lines in Fig. 2), this was likely incomplete given the vast number of other potential determinants [24].
An ideal measure would be if hospitals adopted a medically-ready-for-discharge indicator, which would allow changes in length of stay to be directly attributed to care delivery or discharge delays [5]. Until then, our findings suggest that targeting discharge delays may have a smaller overall impact on hospital capacity strain than anticipated [25]. Improving acute care delivery, perhaps by bolstering hospital staff and procedural care availability—both of which were diminished by the pandemic at a time when patient complexity was already on the rise [26]—may be more fruitful [26].
Electronic supplementary material
Below is the link to the electronic supplementary material.
Acknowledgements
We thank Christy McCay for her helpful contributions retrieving and transferring data.
Author contributions
BPL contributed to project administration, funding acquisition, visualization, writing (original draft), formal analysis, methodology. DMW contributed to project administration, funding acquisition, supervision, formal analysis, writing (review & editing). AAV contributed to project administration, funding acquisition, supervision, writing (review & editing). JS. CK contributed to investigation, formal analysis, writing (review & editing). ACJ contributed to project administration, funding acquisition, supervision, writing (review & editing). BN contributed to resources, supervision, data curation, writing (review & editing). LD contributed to project administration, funding acquisition, supervision, writing (review & editing).
Funding
This material is based upon work supported by the Department of Veterans Affairs, Veterans Health Administration, Office of Research and Development, Health Services Research and Development, The Disrupted Care National Project, Award C19 21-287, PI: Davies. Support for VA/CMS data provided by the Department of Veterans Affairs, VA Health Services Research and Development Service, VA Information Resource Center (Project Numbers SDR 02–237 and 98-004).
Data availability
US Department of Veterans Affairs regulations and our ethics agreements require that the analytic data sets used for this study do not leave the VA firewall without a data use agreement. However, VA data are made freely available to researchers with an approved VA study protocol. For more information visit https://www.virec.research.va.gov.
Declarations
Ethics approval and consent to participate
We received approval from the Veterans Institutional Review Board of Northern New England in accordance with the Declaration of Helsinki.
Consent for publication
Not applicable.
Disclaimer
The contents of this publication do not represent the views of the U.S. Department of Veterans Affairs (VA) or the United States Government..
Author information
BPL has been a practicing adult hospitalist (hospital-based) physician at the White River Junction VAMC, White River Junction, VT, since 2013.
Competing interests
The authors declare no competing interests.
Footnotes
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
References
- 1.Janke AT, Venkatesh AK. Understanding and addressing the us hospital bed shortage—build, baby, build. JAMA Netw Open. 2025 Feb 3;8(2):e2460652. [DOI] [PubMed] [Google Scholar]
- 2.Meo N, Liao JM, Reddy A. Hospitalized after medical readiness for discharge: a multidisciplinary quality improvement initiative to identify discharge barriers in general medicine patients. Am J Med Qual. 2020;35(1):23–28. [DOI] [PubMed] [Google Scholar]
- 3.Burke RE, Chatterjee P. Breaking hospital discharge gridlock through policy reforms. J Hosp Med. 2025 Apr;20(4):414–17. [DOI] [PubMed] [Google Scholar]
- 4.Landeiro F, Roberts K, Gray AM, Leal J. Delayed hospital discharges of older patients: a systematic review on prevalence and costs. Gerontologist. 2019;59(2):e86–97. [DOI] [PubMed] [Google Scholar]
- 5.Bann M, Rosenthal MA, Meo N. Optimizing hospital capacity requires a comprehensive approach to length of stay: opportunities for integration of “medically ready for discharge” designation. J Hosp Med. 2022;17(12):1021–24. [DOI] [PubMed] [Google Scholar]
- 6.Readlynn JK, Zagursky J, Burke RE, Lee JSG, Olson APJ. Gridlock in hospital medicine. J Hosp Med. 2024;19(8):736–38. [DOI] [PubMed] [Google Scholar]
- 7.Hughes AH, Horrocks D, Leung C, Richardson MB, Sheehy AM, Locke CFS. The increasing impact of length of stay “outliers” on length of stay at an urban academic hospital. BMC Health Serv Res. 2021;21:940. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Vincent BM, Wiitala WL, Burns JA, Iwashyna TJ, Prescott HC. Using veterans affairs corporate data warehouse to identify 30-day hospital readmissions. Health Serv Outcomes Res Method. 2018;18:143–54. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Gonsoulin M. VIReC factbook: corporate data warehouse (CDW) inpatient 3.0 domain (part I – inpatient). Hines IL: U.S. Department of Veterans Affairs, Health Services Research & Development Service, VA Information Resource Center; 2017 August. [Google Scholar]
- 10.Intrator O, Li J, Gillespie SM, et al. Benchmarking site of death and hospice use: a case study of veterans cared by department of veterans affairs Home-based primary Care. Med Care. 2020;58(9):805–14. [DOI] [PubMed] [Google Scholar]
- 11.Mitus AJ. The birth of InterQual: evidence-based decision support criteria that helped change healthcare. Prof Case Manag. 2008;13(4):228–33. [DOI] [PubMed] [Google Scholar]
- 12.Neugaard B, Politi R, McCay C. Level of care appropriateness in VA inpatient surgery cases. Prof Case Manag. 2023;28(3):98–109. [DOI] [PubMed] [Google Scholar]
- 13.Nowak M, Pfaff H, Karbach U. Does value stream mapping affect the structure, process, and outcome quality in care facilities? A systematic review. Syst Rev. 2017;6:170. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.StataCorp. Stata 18 base reference manual. College Station, TX: Stata Press; 2023. p. 1636–98.
- 15.Williams R. Using the margins command to estimate and interpret adjusted predictions and marginal effects. Stata J. 2012;12(2):308–31. [Google Scholar]
- 16.Marazzi A, Paccaud F, Ruffieux C, Beguin C. Fitting the distributions of length of stay by parametric models. Med Care. 1998;36:915–27. [DOI] [PubMed] [Google Scholar]
- 17.Bland JM, Altman DG. The use of transformation when comparing two means. BMJ. 1996;312(7039):1153. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Kirshner B, Guyatt G. A methodological framework for assessing health indices. J Chron Dis. 1985;38:27–36. [DOI] [PubMed] [Google Scholar]
- 19.Veterans Health Administration. VHA publications. VHA directive 1117. Updated October 8, 2020. https://www.va.gov/vhapublications/index.cfm. Accessed 25 Apr 2025.
- 20.Yoon J, Phibbs CS, Ong MK, et al. Acute hospitalizations and outcomes in veterans affairs hospitals 2011 to 2017. Medicine (Baltimore). 2024;103(30):e38934. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.D’Agostino F, Sanson G, Cocchieri A, et al. Prevalence of nursing diagnoses as a measure of nursing complexity in a hospital setting. J Adv Nurs. 2017;73:2129–42. [DOI] [PubMed] [Google Scholar]
- 22.D’Agostino F, Vellone E, Cocchieri A, et al. Nursing diagnoses as predictors of hospital length of stay: a prospective observational study. J Nurs Scholarsh. 2019;51:96–105. [DOI] [PubMed] [Google Scholar]
- 23.Naik H, Murray TM, Khan M, et al. Population-based trends in complexity of hospital inpatients. JAMA Intern Med. 2024;184:183–92. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Lu M, Sajobi T, Lucyk K, Lorenzetti D, Quan H. Systematic review of risk adjustment models of hospital length of stay (LOS). Med Care. 2015;53:355–65. [DOI] [PubMed] [Google Scholar]
- 25.Bann M, Meo N, Lopez JP, et al. Medically ready for discharge: a multisite “point-in-time” assessment of hospitalized patients. J Hosp Med. 2023;18(9):795–802. [DOI] [PubMed] [Google Scholar]
- 26.Zagursky JM, Burke RE, Olson APJ, Readlynn JK. Gridlock: what hospitalists and health systems can do to help. J Hosp Med. 2024;19(11):1066–70. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
US Department of Veterans Affairs regulations and our ethics agreements require that the analytic data sets used for this study do not leave the VA firewall without a data use agreement. However, VA data are made freely available to researchers with an approved VA study protocol. For more information visit https://www.virec.research.va.gov.