Introduction
The coronavirus disease 2019 (COVID-19) pandemic disrupted health care delivery systems globally, exposing vulnerabilities in hospital preparedness and straining the resilience of health care infrastructure. In the United States, academic medical centers (AMCs) were uniquely affected due to their roles as tertiary referral centers, teaching institutions, and leaders in clinical research and innovation. AMCs often serve as regional anchors in emergency response, intensive care delivery, and surge capacity coordination during public health emergencies.Reference Arsenault, Gage and Kim1
Initial pandemic response efforts led to widespread deferral of elective procedures, rapid reallocation of clinical resources, and significant alterations in patient behavior and hospital operations.Reference Hartnett, Kite-Powell and DeVies2, Reference Jazieh and Kozlakidis3 National analyses have reported disruptions in health care utilization, case complexity, and mortality across multiple pandemic phases.Reference Birkmeyer, Barnato and Birkmeyer4, Reference Cannata, Watson and Daniel5 However, most published studies have focused on single institutions or geographic regions, limiting their utility for national or regional policy planning.
Furthermore, while large-scale datasets have been employed to study pandemic trends,Reference Meille, Decker and Owens6 there remains a critical gap in the literature evaluating the differential impact of the COVID-19 pandemic on AMCs across distinct U.S. regions. To better reflect the evolving nature of the pandemic, we segmented the COVID period into early and late phases, corresponding to pre- and post-vaccine availability. This approach allows for more nuanced analysis of hospital performance and regional variation across distinct pandemic phases. Such regional analyses are essential for disaster response planning, as the progression and burden of pandemics vary by population density, health care infrastructure, and local policy responses.Reference Khullar, Bond and Schpero7
This study aimed to evaluate trends in inpatient hospital metrics across U.S. AMCs during the COVID-19 pandemic using a nationally representative dataset. By examining variation in discharge volume, length of stay (LOS), case mix index (CMI), ICU utilization, and mortality across pre-pandemic, pandemic, and post-pandemic periods, this work provides evidence to inform region-specific emergency preparedness strategies and health system resilience planning.
Methods
We conducted a retrospective observational study using inpatient discharge data from the Vizient® Clinical Data Base (CDB) encompassing 106 U.S. academic medical centers (AMCs) from October 2019 to December 2023; see Appendix 1. Vizient® CDB is an administrative database with standardized, de-identified patient-level billing data submitted by member institutions. It comprises approximately 97% of U.S. AMCs and is used for benchmarking and performance improvement. For this analysis, an AMC was defined as a tertiary care hospital serving as the principal teaching hospital for a U.S. medical school, as classified by Vizient.
The study period was divided into 4 phases: Pre-COVID: October 1, 2019-March 31, 2020; Early COVID: April 1, 2020-December 31, 2020 (representing the pre-vaccine phase); Late COVID: January 1, 2021-May 31, 2023 (representing the post-vaccine availability period); and Post-COVID: June 1, 2023-December 31, 2023. The COVID segmentation aligns with the timeline of vaccine rollout to the general population and allows for more granular analysis of pandemic impact.
AMCs were mapped to four standard U.S. regions (Midwest, Northeast, South, West) using the Beresford Research classification; see Appendix 2.8 These regional classifications reflect important differences in public health infrastructure, hospital system organization, population density, and state-level health policy response to COVID-19, which may have influenced care delivery and outcomes.
The study population included adult inpatients discharged during the study period. To ensure consistency, only the flagship hospitals of each AMC were included. We excluded affiliated community hospitals unless data were directly tied to the academic center’s CMS Certification Number (CCN).
Monthly, facility-level mean values for the following outcomes were provided by Vizient (Appendix 3): Number of discharges (encounters), Observed length of stay (LOS), Expected LOS (Vizient risk-adjusted), LOS index (observed/expected LOS), ICU admission rate, ICU LOS, Case mix index (CMI), Observed and expected mortality, Mortality index (observed/expected mortality), and Percentage of early deaths (within 48 hours of admission). Monthly encounters were calculated as the total number of adult inpatient discharges per AMC. These were not adjusted for population size or number of hospitals per region. Observed mortality was defined as in-hospital death during the index admission. The dataset did not include 30-day post-discharge mortality. Variable definitions followed Vizient’s standardized methodology, including risk-adjusted metrics derived from comorbidities and DRG-based severity algorithms.9, Reference Elixhauser, Steiner and Harris10
Descriptive statistics (mean, standard deviation, and 95% confidence intervals) were calculated for all outcomes across four periods. We employed mixed models to examine temporal changes within facilities, using the facility as a random effect and time period as a fixed effect. The Kenward-Roger adjustment was applied for F-tests, and compound symmetric correlation structures were assumed. For regional comparisons, we incorporated a fixed effect for region and a region-by-time interaction term to assess whether outcome trends varied across regions. Post hoc comparisons were conducted using Tukey-Kramer adjustments. Statistical significance was defined as a 2-sided P-value < 0.05. All analyses were performed using SAS version 9.4 (SAS Institute Inc, Cary, NC).
Ethics Review
This study used de-identified, aggregate data and did not involve direct patient contact. It was deemed exempt from Institutional Review Board oversight by Augusta University. No protected health information or minors were included.
Results
A total of 10 academic medical centers (AMCs) were analyzed on 13,527,471 monthly aggregated adult inpatient discharges. Table 1 summarizes descriptive statistics for each of the four defined periods. Mixed models demonstrated significant temporal changes in hospital metrics across all academic medical centers. Average monthly discharges per facility declined during the early COVID (2363.0) compared to late COVID (2482.7), pre-COVID period (2557.2), and post-COVID (2576.4) periods and late COVID showed lower discharges than pre-COVID and post-COVID. Observed length of stay (LOS) increased from 6.2 days pre-COVID to 6.4 days during early COVID, 6.7 late COVID, and 6.6 post-COVID, while expected LOS rose from 5.9 to 6.4 days over the same periods. The LOS index increased from pre-COVID (1.0) to late COVID (1.1) but declined post-COVID (1.0), and late COVID was higher than early COVID (1.1) and post-COVID. Regarding ICU metrics, the percentage of encounters involving ICU admission declined slightly from 19.6% pre-COVID and 19.8% early COVID to 18.8% post-COVID, whereas ICU LOS increased from 4.6 to 5.2 days during the early COVID, late COVID (5.2), and post-COVID (4.8) periods. The percent of cases with >1+comorbidities increased from pre-COVID (5.4%) to early COVID (5.8%), late COVID (5.8%), and post-COVID (5.9%). Observed mortality rose from 2.8% pre-COVID to 3.4% during early COVID, 3.2% late COVID, but returned to pre-COVID levels at post-COVID (2.8%). Percent expected mortality increased from pre-COVID (2.8%) to early COVID (3.4%), late COVID (3.4%), and remained higher post (3.3%). The mortality index remained stable from pre-COVID, early COVID, and late COVID (1.0) before decreasing slightly to 0.9 post-COVID. Case mix index (CMI) also increased from 2.1 to 2.2 during early COVID, late COVID, and post-COVID. The proportion of early deaths—defined as deaths occurring within 48 hours of admission—early COVID (0.92) had the highest percent early mortality compared to pre-COVID, late COVID, and post-COVID. The mixed model showed significant differences by time period for the number of encounters (P < 0.0001), observed LOS (P < 0.0001), percent of ICU encounters (P=0.0112), ICU LOS (P < 0.0001), observed mortality (P < 0.0001), and CMI (P < 0.0001) (Table 2).
Descriptive statistics by time
Abbreviations: SD = standard deviation, CI = confidence interval, LOS = length of stay, ICU = intensive care unit, CMI = case mix index
Mixed model results by time
Abbreviations: LS = least squares, SE = standard error, CI = confidence interval, LOS = length of stay, ICU = intensive care unit, CMI = case mix index
Regional Variation
There were statistically significant region×time interactions for number of encounters (P < 0.0001), expected LOS (P = 0.0442), percent of encounters with an ICU stay (P = 0.0249), and CMI (P = 0.0140). For encounters, the Midwest had the highest adjusted volumes across all periods, while the West had the lowest; see Table 3. From pre‑COVID to early COVID, encounters declined in all regions; from early COVID to post‑COVID, encounters increased in all regions. From pre‑COVID to late COVID, Midwest, Northeast, and South declined, whereas the West increased. Observed LOS increased during the pandemic across all regions, with the South generally highest (tied with the Northeast post‑COVID); however, the region×time interaction was not significant (P = 0.5422). Expected LOS showed differential trends (interaction P = 0.0442): the Midwest was highest (tied with the West pre‑ and early COVID) and, together with the Northeast, increased steadily across periods; the South also increased through post‑COVID, while the West plateaued from late to post‑COVID. Percent of ICU encounters exhibited regional differences over time (interaction P = 0.0249), with the Northeast lowest at each period. The Northeast was stable from pre‑ to early COVID then declined below pre-COVID levels by late and post‑COVID; the South increased pre‑ to early COVID and then declined below pre‑COVID; the West was stable pre‑ to early COVID and then declined; the Midwest rose from early to late and remained above pre/early in post‑COVID. Observed mortality increased during the COVID periods in all regions, with the largest early‑period increase in the Northeast; the interaction was not significant (P = 0.1375). Mean ICU LOS showed no significant region×time interaction (P = 0.7746). CMI increased in all regions (largest absolute increase in the Midwest), and its interaction was significant (P = 0.0140). Figure 1 demonstrates significant and non-significant regional differences across periods for all outcomes.
Mixed model results {adjusted LS mean (SE), [95% CI]} of region and time on outcomes
Abbreviations: LS = least squares, SE = standard error, CI = confidence interval, LOS = length of stay, ICU = intensive care unit, CMI = case mix index
Line graphs showing the adjusted least squares mean (SE) for all outcomes by time period and region.
Discussion
This national study of U.S. AMCs demonstrates that the COVID-19 pandemic significantly disrupted inpatient operations, resulting in decreased encounter volume, prolonged hospital stays, and increased case complexity across all regions. Despite differences in timing and magnitude, these effects were broadly observed nationwide and carry significant implications for health system preparedness, surge planning, and policy development during future public health emergencies.
The observed decline in inpatient discharges during the COVID period likely reflects deferred elective care, patient avoidance of hospitals, and resource reallocation toward COVID-specific response.Reference Mehrotra, Chernew and Linetsky11, Reference Elliott, Beckett and Cohea12 The rebound in post-pandemic volumes underscores the pent-up demand and the need for flexible capacity planning. Longer observed and expected LOS across all regions may signal operational strain, delays in discharge, and increased resource utilization—key challenges during disaster scenarios.Reference Khunti, Valabhji and Misra13
The increase in ICU LOS without a corresponding rise in ICU admission rates suggests heightened disease severity and complexity among ICU patients, possibly exacerbated by constrained triage capacity and limited bed availability during surges. While ICU length of stay increased during the pandemic, this may reflect greater clinical complexity and prolonged recovery among critically ill patients. However, the dataset does not include direct measures of ICU census, staffing shortages, or bed availability, and therefore, cannot definitively assess critical care system strain. While regional variations were modest, the South experienced the highest observed LOS, and the Northeast experienced the most significant increase in mortality, consistent with region-specific pandemic burden and policy responses.Reference Roy, Bollman and Carson14, Reference Danet Danet15
While CMI increased during and after the pandemic, this may reflect increases in clinical complexity and changes in documentation or billing practices.Reference Khullar, Bond and Schpero7 Post-pandemic periods saw a decline in mortality and ICU use despite elevated CMI, raising the possibility of administrative inflation in case severity rather than purely clinical deterioration. Similarly, the declining LOS index post-COVID may reflect improved risk adjustment and coding accuracy rather than operational gains.
Significant regional differences in encounters, ICU metrics, and CMI suggest that pandemic impact was not uniform across the U.S. These disparities may reflect differences in state policies, health infrastructure, urbanicity, and public health preparedness.Reference Arsenault, Gage and Kim1, 8 The findings support the need for regionally tailored surge strategies, ICU planning frameworks, and flexible staffing models adaptable to shifting case burdens. While this study did not directly incorporate vaccination rates, it is plausible that regional differences in vaccine uptake contributed to observed variations in hospital metrics. For example, lower vaccination rates in certain Midwest and Southern states may have influenced prolonged ICU stays or higher case complexity. Future studies should integrate immunization data to better contextualize regional disparities.
The segmentation of the COVID-19 pandemic into distinct phases—pre-COVID, early COVID, late COVID, and post-COVID—aligns with key milestones in vaccine availability and public health response. The early COVID phase (April-December 2020), prior to widespread vaccine access, was marked by peak mortality and operational disruption across academic medical centers. Vaccines first became available to health care workers and long-term care residents in mid-December 2020, followed by high-risk individuals and essential workers in early 2021. Widespread public access began in April 2021, marking the transition into the late COVID phase (January 2021-May 2023), which was characterized by increased ICU burden and case complexity. Post-COVID trends (June-December 2023) suggest partial normalization of hospital metrics, though elevated length of stay and case mix index indicate persistent system strain. These temporal distinctions provide critical context for understanding the evolving impact of the pandemic on academic medical centers and reinforce the need for phase-specific preparedness strategies.
Administrative datasets such as Vizient® CDB offer an important tool for real-time hospital surveillance, enabling health systems and policymakers to track evolving patterns during public health emergencies. Incorporating such data into regional and federal disaster response dashboards can support agile decision-making and more equitable resource allocation.
Policy Recommendations
To inform future disaster preparedness and health system resilience, we recommend:
-
• Investing in scalable inpatient and ICU surge capacity, particularly in high-burden regions.
-
• Supporting data infrastructure to enable timely benchmarking of hospital performance.
-
• Refining DRG coding and risk-adjustment models to account for novel diseases.
-
• Implementing regional mutual aid and telehealth triage networks to relieve localized surges.
This study has several limitations. First, administrative data may be influenced by changes in coding practices and billing priorities, particularly in response to financial strain. Second, we limited our dataset to AMCs, which limits generalizability to non-teaching or rural hospitals. Third, the study does not account for patient-level demographic or socioeconomic variables. Fourth, categorizing regions broadly may overlook intraregional heterogeneity. Fifth, vaccination rates were not directly included in the analysis; however, regional differences in vaccine uptake may have influenced observed trends and are discussed in the limitations. Sixth, we were unable to perform a sensitivity analysis to determine whether trends were disproportionately influenced by high-volume centers such as those in New York City. As a result, regional findings may be partially driven by outlier institutions, and this should be considered when interpreting the results. Lastly, we could not assess outpatient care shifts, quality of care metrics, or specific effects of COVID variants.
Conclusions
The COVID-19 pandemic led to significant disruptions in hospital operations across U.S. academic medical centers, with regional variation in intensity and recovery. These findings highlight the importance of adaptive, region-specific preparedness strategies, while acknowledging that this study does not directly measure hospital strain or access among underserved populations. Because the dataset is limited to academic medical centers, it may not fully capture the experiences of vulnerable populations who face barriers to accessing tertiary care.
Supplementary material
The supplementary material for this article can be found at http://doi.org/10.1017/dmp.2025.10217.
Acknowledgments
-
1. Contributors: Not Applicable
-
2. Funders: Not Applicable, no funding sources
-
3. Prior presentation: Wilkins T, Walsh D, Waller J, Xie D. From Crisis to Recovery: Evaluating the Evolution of Patient Care in Academic Medical Centers Through the COVID-19 Era. A Oral presentation on completed research presented at the North American Primary Care Research Group, Quebec City, Canada, November 2024.
-
4. Abstract citation for presentation: Wilkins T, Walsh D, Waller J, Xie D. From crisis to recovery: Evaluating the evolution of patient care in academic medical centers through the COVID-19 era. Ann Fam Med. 2024;22:5966.
Competing interests
The authors declare there are no conflicts of interest.
