Association of Intensive Care Unit Patient Load and Demand With Mortality Rates in US Department of Veterans Affairs Hospitals During the COVID-19 Pandemic

This cohort study examines the association of patient caseload and demand with mortality among patients with coronavirus disease 2019 (COVID-19) in US Veterans Affairs (VA) intensive care units.

count at each facility. The COVID-19 ICU load was calculated at the patient level as the mean number of patients with COVID-19 in the ICU during the patient's hospital stay divided by the number of ICU beds at that facility. The number of ICU beds at each facility was a fixed number. The COVID-19 ICU load included only the number of patients with COVID-19 in the ICU, excluding patients with other critical illnesses. The COVID-19 ICU load ranged from 0% to greater than 100%; it exceeded 100% if the hospital increased critical care bed capacity during the pandemic (eg, by converting a sleep laboratory into an ICU) and if those beds were occupied by patients with COVID-19.
The population of patients with COVID-19 in the ICU varied over time, with peak prevalence rates occurring early (eg, March) at some hospitals and later (eg, July) at other hospitals. The COVID-19 ICU demand described the caseload of patients with COVID-19 in the ICU when a patient was treated compared with peak COVID-19 ICU caseload. It was calculated at the patient level as the mean number of patients with COVID-19 in the ICU during the patient's stay divided by the maximum number of patients with COVID-19 in the ICU at that facility during the study period. The COVID-19 ICU demand ranged from 0% to 100%.
For example, if a hospital had 60 ICU beds before the pandemic and the mean number of patients with COVID-19 in the ICU during a patient's stay was 20, then the COVID-19 ICU load would be 20 divided by 60, or 33%. If at that same facility the peak surge included 20 patients with COVID-19 in the ICU and a patient was treated during the period when the mean of number patients with COVID-19 in the ICU was 20, then the COVID-19 ICU demand would be 20 divided by 20, or 100%.

Statistical Analysis
We described differences over time in baseline characteristics and mortality among inpatients with COVID-19 using χ 2 and Wilcoxon rank sum tests. We used Cox proportional hazard models to analyze the time in days from admission to death, either in the hospital or within 30 days after discharge, among patients who were admitted to the hospital (overall, in the general ward, and in the ICU).
Patients who were still in the hospital or out of the hospital and alive at 30 days after discharge were treated as censored observations. We included a random effect for the facility to account for the correlation of mortality among patients in the same hospital. Analyses were performed using SAS Enterprise Guide statistical software version 7.11 (SAS Institute). P values were 2-sided, and statistical significance was set at P < .05. Data were analyzed from March to November 2020.
Patients with COVID-19 in the ICU treated during high COVID-19 ICU strain had increased risk of mortality. Table 2 provides the unadjusted results, and  for the 2 measures of COVID-19 ICU strain early in the pandemic (ie, March-May 2020) and later in the pandemic (ie, June-August 2020), and these data are consistent with overall study findings.

Discussion
In this cohort study of patients with COVID-19 in US VA hospitals, receiving treatment during peak COVID-19 ICU demand, with demand describing the caseload of patients with COVID-19 in the ICU when the patient was treated compared with peak COVID-19 ICU caseload, was consistently and independently associated with COVID-19 ICU mortality. In the extreme case, the adjusted hazard of death was 1.94 for patients with COVID-19 treated in the ICU during periods with greater than 75% to 100% of the peak COVID-19 ICU caseload. The finding that COVID-19 ICU demand was associated with increased mortality for patients with critical COVID-19 early in the pandemic (ie, March-May) and later in the pandemic (ie, June-August) supports the overall study results that suggested that strains on critical care capacity were associated with increased COVID-19 ICU mortality.
Tracking COVID-19 ICU demand may be useful to hospital administrators and health officials as they seek to implement interventions to optimize outcomes for patients with COVID-19. 8

COVID-19
ICU demand can be calculated only retrospectively (because the peak number of patients with COVID-19 in the ICU can be assessed only retrospectively). However, facilities can identify the peak surge caseload since the pandemic started, in March 2020, and prospectively monitor COVID-19 ICU demand. Facilities within a health care system or within a geographic region could collaborate to triage patients with critical COVID-19 to sites with greater ICU capacity to reduce strain on any 1 facility. 9,10 Future research is urgently needed to investigate the mechanisms by which COVID-19 ICU  demand may be associated with increased mortality; it is imperative that we understand the degree to which patient characteristics (eg, disease severity) or facility issues (eg, staffing) contribute to the association between COVID-19 ICU strain and poor patient outcomes among patients with critical COVID-19.
We did not have a formal measure of ICU capacity, because VA ICU bed availability is not fixed but instead depends on staffing availability; therefore, we calculated COVID-19 ICU load as the ratio of ICU COVID-19 occupancy to the maximum ICU bed number as a surrogate for COVID-19 ICU capacity. Although the association between COVID-19 ICU load and patient mortality was statistically significant, it was neither as consistent over time nor as robust as the association between COVID-19 ICU demand and mortality. We hypothesize that facilities increased their critical care capacity in response to the pandemic and that the degree of this augmentation varied across facilities.
Therefore, the comparison with a fixed number of patient beds was likely a relatively poor measure of ICU capacity during the pandemic. Given that hospitals are charged with caring for patients with non-COVID-19 critical illness as well as patients with COVID-19, future studies should seek to examine whether measures of critical care strain that include all patients in the ICU (not just those with COVID-19) are associated with patient outcomes. Future studies should also evaluate whether ICU load provides an adequate measure of strain across the broad spectrum of VA and non-VA hospitals, which vary greatly in prepandemic ICU bed number and the potential to augment capacity during a pandemic. Our overall study findings are supported by cohort studies from 2013 11 and 2018 12 demonstrating that as ICUs are strained, mortality increases. 11,12 It may be the case that during periods of peak ICU caseload, patients who would be admitted to the ICU under more typical conditions are instead admitted to the ward. 13 Our data did not allow us to examine this issue directly; however, we did examine outcomes associated with COVID-ICU strain separately among patients in the general ward and patients in the ICU. Although the association between COVID-19 ICU load and general ward mortality was statistically significant, it varied over time (ie, early vs later in the pandemic). Future research should examine how critical care strains may be associated with outcomes in the general ward for patients with COVID-19 and those without COVID-19.

Limitations
This study has several limitations. First, this study evaluated care of patients with COVID-19 at VA hospitals; future studies should examine the association between COVID-19 ICU burden and mortality in non-VA facilities. Second, this study focused on COVID-19 mortality; future studies should examine the potential associations of COVID-19 ICU load and demand with outcomes among patients without COVID-19. Third, the results of this study should not be interpreted as a statement on scarcity of critical care or mechanical ventilation; we have no data to suggest that patients needing critical care or mechanical ventilation did not receive this care. 14 Fourth, although the risk adjustment models included demographic and clinical characteristics, they did not include social determinants of health (eg, income or education), which may contribute to COVID-19 mortality. Fifth, we did not examine changes in ICU staffing during the study period. Sixth, we do not have a measure of the degree to which facilities expanded ICU capacity during the pandemic. Seventh, patients with COVID-19 who were admitted to the ICU service could have physically been in diverse settings, including locations designated as the COVID-19 ICU, such as surgical ICUs; some patients with critical COVID-19 were cared for by the ICU team but were physically located in the emergency department.
Eighth, related to the observed change in mortality over time, our results suggest that changes in patient characteristics and measures of COVID-19 ICU strain were associated with some of the variation in mortality over time; however, given the observational nature of these data, causality cannot be inferred. Other potential causes (eg, use of medications, such as remdesivir and dexamethasone; clinical practices, such as proning; and unmeasured changes in patient characteristics, such as susceptibility) may have contributed to changes in COVID-19 mortality. 15

Conclusions
In this cohort study of patients with COVID-19 in US VA hospitals, COVID-19 ICU demand-a measure of COVID-19 ICU caseload when a patient was treated compared with peak COVID-19 ICU caseload-was associated with mortality among patients with COVID-19 in the ICU. Public health officials and hospital administrators may seek to prevent high COVID-19 ICU demand to optimize outcomes for patients with COVID-19.