Association of Fluid Management with Outcome in Invasively Ventilated COVID–19 ARDS Patients – Insights From the PRoVENT-COVID Study: A National, Multicenter, Observational Cohort Analysis.

Background: Increasing evidence indicates the potential benets of restricted uid management in critically ill patients. Evidence lacks on the optimal uid management strategy for invasively ventilated COVID-19 patients. We hypothesized that the cumulative uid balance would affect the successful liberation of invasive ventilation in COVID-19 patients with acute respiratory distress syndrome (ARDS). Methods: We analyzed data from the multicenter observational ‘PRactice of VENTilation in COVID-19 patients’ (PRoVENT-COVID) study. Patients with conrmed COVID-19 and ARDS who required invasive ventilation during the rst 3 months of the international outbreak (March 1, 2020, to June 2020) across 22 hospitals in the Netherlands were included. The primary outcome was successful liberation of invasive ventilation, modeled as a function of day 3 cumulative uid balance using Cox proportional hazards models, using the crude and the adjusted association. Sensitivity analyses without missing data and modeling ARDS severity were performed. Results: Among 650 patients, three groups were identied. Patients in the higher, intermediate and lower groups had a median cumulative uid balance of 1.98 liters (1.27-7.72 liter), 0.78 liter (0.26-1.27 liter) and –0.35 liter (–6.52-0.26 liter), respectively. Higher day 3 cumulative uid balance was signicantly associated with a lower probability of successful ventilation liberation (adjusted hazard ratio 0.85, 95% CI 0.77-0.94, P = 0.0013). Sensitivity analyses showed similar results. Conclusions: In a cohort of invasively ventilated patients with COVID-19 and ARDS, a higher cumulative uid balance was associated with a longer ventilation duration, indicating that restricted uid management in these patients may be benecial.

organs, causing injury [10]. However, a restrictive uid strategy could lead to extrapulmonary organ dysfunction consequent to reduced cardiac output. Therefore, the uid balance being an adverse prognostic factor, yet also a potentially modi able risk factor, pose a unique dilemma in the management of critically ill COVID-19 patients. Current evidence is insu cient and constantly evolving to best address the optimal uid management strategy in invasively ventilated COVID-19 patients. Using the database of the multicenter observational 'PRactice of VENTilation in COVID-19 patients' (PRoVENT-COVID) study, we investigated cumulative uid balance in invasively ventilated COVID-19 and ARDS patients and factors associated with a higher positive cumulative uid balance. We aimed to test the association between the cumulative uid balance during the rst 4 calendar days of invasive ventilation and successful liberation of ventilation in these patients. We hypothesized that a higher cumulative uid balance is independently associated with a lower probability of successful liberation of invasive ventilation in COVID-19 ARDS patients.

Methods Design
PRoVENT-COVID is an investigator-initiated national, multicenter, observational cohort study that included COVID-19 patients with acute respiratory failure requiring invasive ventilation in 22 hospitals in the Netherlands in the rst 3 months of the international outbreak. The study protocol was approved by the local institutional review board of Amsterdam University Medical Center (location 'AMC') and registered at Clinicaltrials.gov (NCT04346342). The institutional review board waived the requirement for written informed consent at the participating sites. The original study protocol was pre-published elsewhere [11]. The proposed plan and statistical analysis of the current analysis were approved by the Core Steering Committee and published with the website of PRoVENT-COVID before data acquisition [12].
The protocol was revised to address the unanticipated severely zero-in ated distribution of ventilator-free days during the initial data acquisition (Supplementary Figure 1, Additional File 1). This analysis adheres to the Strengthening the Reporting of Observational Studies in Epidemiology statement.

Selection criteria
Invasively ventilated adult patients who met the criteria for ARDS using the Berlin de nition [13] and who had real-time polymerase chain reaction con rmed COVID-19 admitted to one of the participating ICUs were eligible for participation. The original PRoVENT-COVID study protocol had no exclusion criteria; however, for the current analysis, we excluded patients if they were not invasively ventilated beyond the rst 4 calendar days and patients who were transferred within the rst 4 days of ventilation from or to another ICU that did not participate in the PRoVENT-COVID study.

Exposure
The primary exposure of interest was the cumulative uid balance during the rst 4 calendar days of invasive ventilation, calculated by total uid input minus total uid output on a certain day of ICU admission.

Outcomes
The primary outcome was successful liberation from invasive ventilation at 28 days, de ned as the timepoint at which a patient was alive and extubated. Secondary outcomes were acute kidney injury (according to a modi ed Kidney Disease Improving Global Outcomes de nition) [14], the use of renal replacement therapy, duration of invasive ventilation in survivors and non-survivors, ICU and hospital length of stay in survivors and non-survivors, and 28-day mortality.

Statistical analyses
Descriptive statistics were used to describe the study population and uid management parameters. Data are presented as numbers and percentages for categorical variables and as means and standard deviation or median and interquartile range according to distribution. Where appropriate, statistical variability is expressed by 95% con dence intervals.
Using a mixed-effects model, we rst examined the crude association between cumulative uid balance and successful liberation from ventilation at day 28 with successful liberation of ventilation as a dependent variable, uid balance as ( xed effect) as an independent variable, and hospital as a random intercept effect. Cumulative uid balance from day 0 through day 1 was grouped as day 1, and the subsequent days were labelled as day 2, and day 3. Cumulative uid balance during the rst 4 calendar days is referred to hereafter as cumulative uid balance at day 3. To examine potential nonlinearity in the association, the cumulative uid balance was entered as a restricted cubic spline function with 3 knots distributed equally along the density. The complexity of the spline function was reduced in a stepwise fashion until minimization of the Akaike information criterion (AIC) (Supplementary Figure 2, Additional File 1). The exposure (cumulative uid balance at day 3) was divided into tertiles to facilitate interpretation.
The association between cumulative uid balance at day 3 and the probability of successful liberation from invasive ventilation was then adjusted for possible confounders by including these variables as ( xed effect) covariates in the mixed effects model. The set of prede ned adjustment variables included the following: sex, age, body mass index, serum creatinine, use of vasopressors (norepinephrine dose), tidal volume, arterial pH, positive end-expiratory pressure, partial pressure of oxygen to fraction inspired oxygen, dynamic respiratory system compliance and arterial lactate, all measured on the day of intubation.
Conditional on the assumption that the data were missing at random and the severity illness scores were collected differently by hospitals, before imputation, the percentage of missing data in the severity of illness scores in the rst 3 days were assessed and addressed by a multi-level multiple imputation method. We imputed 20 datasets using multiple imputation by chained equations [15]. No exposure (day 3 uid balance) or outcomes (survival and duration of ventilation) were imputed. All models described in the 'statistical analysis' section were reproduced in the 20 databases after multiple imputations and the results were pooled. We considered statistical signi cance at P ≤ 0.05.
No formal statistical power calculation was conducted before the study. The sample size was solely based on the available data from the PRoVENT-COVID database.

Sensitivity analyses
To assess the robustness of the ndings toward the missing data and imputation method, we re t the main regression model (i.e., the marginal effect of day 3 uid balance on the hazard of successful liberation from invasive ventilation) on cases with complete data only. To retain the largest possible sample size, only covariates that were signi cantly associated with the outcome in the main model were included in the sensitivity model.
We also estimated the main effect of different classes of ARDS severity by including this variable in the adjusted mixed effects model.

Patient population and characteristics
We identi ed 687 invasively ventilated COVID-19 and ARDS patients admitted to ICUs between March 1, 2020, and June 1, 2020. The study ow chart is summarized in Figure 1. Table 1 describes the baseline and ventilation characteristics of our study participants. The most prevalent comorbidities were hypertension and diabetes.

Cumulative uid balance distribution
The distribution of cumulative uid balance at day 3 was evaluated for the overall cohort (Supplementary liter [-6.52-0.26 liter], respectively. Patients in the higher cumulative uid balance group had a higher prevalence of chronic hypertension; worse baseline Simpli ed Acute Physiology Score, and higher lung compliance ( Table 1).

Association of cumulative uid balance with outcomes
The association between day 3 cumulative uid balance and the hazard of successful liberation from ventilation was most parsimoniously characterized by a 0-spline (linear) survival model. Models with 3-, 2-, or 1-knot-restricted cubic splines for cumulative uid balance had higher AICs (i.e., no better t) (Supplementary Figure 2, Additional File 1).
The resulting association between cumulative uid balance at day 3 and the probability of successful liberation from invasive ventilation is shown in Figure 2. In unadjusted analysis, there was a signi cant association between higher cumulative uid balance at day 3 and a lower probability of successful liberation from ventilation, with a hazard ratio per liter uid balance of 0.86 (95% CI 0.78-0.96, P = 0.005). After adjusting for a prede ned set of possible confounding variables (listed in Supplementary Table 1, Additional File 1), exposure to higher cumulative uid balance at day 3 remained signi cantly associated with lower probability of successful liberation of invasive ventilation. The adjusted hazard ratio for successful liberation from invasive ventilation associated with each liter increase in cumulative uid balance was 0.86 (95% CI 0.77-0.95, P = 0.0047).

Secondary outcomes and sensitivity analysis
The secondary outcomes are summarized in Table 2. Length of ICU stay, length of hospital stays, and duration of intubation were signi cantly shorter for surviving patients who were in the lower tertile of cumulative uid balance. Other outcomes did not differ signi cantly between uid balance tertiles.
Two sensitivity analyses were performed. First, the robustness of our ndings was assessed toward the missing data and imputation method (Supplementary Table 2, Additional File 1). In the complete-casesonly model, the sample size was reduced to 461 patients; the estimated association between day 3 cumulative uid balance and successful liberation of invasive ventilation was similar to the estimation with imputed data. The hazard ratio per liter uid balance was consistent with the primary analysis, 0.86 (95% CI 0.76-0.98, P = 0.0247).
Second, the models were re-analyzed according to ARDS severity, considering a possible interaction between the severity of ARDS and uid balance. The interaction between ARDS severity and day 3 cumulative uid balance did not improve the model t (AIC of interaction model 4336 vs. 4335 for reduced model), indicating that there was no signi cant interaction between day 3 cumulative uid balance and ARDS severity on the association with successful liberation from invasive ventilation ( Figure  3).

Discussion
The main ndings of this multicentric observational study of COVID-19 and ARDS patients include the following: (1) a higher day 3 cumulative uid balance was associated with a lower probability of successful liberation from invasive ventilation; (2) these results remained consistent even after adjustment for potential prede ned confounding factors and sensitivity analyses; and (3) reduction in duration of invasive ventilation, hospital and ICU length of stay was noted in patients who had lower cumulative uid balance.
Our results corroborated the growing evidence suggesting the unfavorable effect of higher positive uid balance on outcomes in critically ill patients [16][17][18][19]. However, compared to these studies, there are also some notable differences in our study. We speci cally evaluated the exposure of cumulative uid balance in COVID-19 and ARDS patients on the ventilation liberation irrespective of prior spontaneous breathing trials. Cumulative uid balance was calculated from hospital admission until day 3, whereas in other studies, it was calculated differently. Despite these differences, a signal of potential harm with excessive cumulative uid balance and weaning outcomes was consistently observed.
Evidence emanating from large trials of ARDS patients has led to an overall practice change that resulted in relatively less aggressive initial uid management. In 2006, the Fluids and Catheters Treatment Trial reported a causal effect between positive uid balance and duration of ventilation in ARDS patients [8]. The authors found that the conservative group had a shorter ventilation duration than the liberal-strategy group without an increase in non-pulmonary organ failure. Another study, performed by the ARDSnet, showed that negative cumulative uid balance was signi cantly associated with more ventilator freedays and lower mortality than positive cumulative uid balance [20]. A limitation of using 'ventilator-free days' in these reports is that a more frequently occurring component of the composite (such as survival or duration of ventilation) presumably drives the effect estimates and could in uence the results, even stronger when the components are oppositely affected by the exposure [1,21]. Our rationale for using 'ventilator-free days' was to compare our analysis to previously conducted studies readily; one of the challenges was disentangling the contribution of 'zero-in ated distribution' in ventilator-free days. However, it is possible that a greater-than-expected number of non-survivors had died within 24-hours of initiation of ventilation, and this could presumably drive the mean difference towards null. Or, because of unknown factors, certain patients might not have been able to present values other than zero.
Nevertheless, we addressed it by restricting our primary outcome to only 'successful ventilation liberation' instead of 'ventilator-free days'; however, our analysis suffered from model selection bias.
Several mechanisms may explain the association of higher cumulative uid balance and decreased odds of ventilation liberation. Higher positive uid balance increases the extravascular lung water, and inattention to uid overload may inadvertently promote counterproductive outcomes, such as pulmonary dysvasoregulation and alveolar edema, contributing to weaning failure. This risk is particularly high among patients with COVID-19 and ARDS because of relatively higher extravascular lung water and pulmonary vascular permeability indices, in distinct contrast to non-COVID ARDS [22]. Furthermore, alveolar uid clearance is perhaps slow or even impaired in ARDS pathophysiology. The combined processes of high vascular permeability and impaired alveolar uid clearance may therefore rapidly worsen the alveolar edema -even with a slight increase in intravascular volume [23,24]. Consistent with this view, we showed that even a one-liter increase in the dose of cumulative uid balance might signi cantly decrease ventilation liberation odds. For example, about 14% (hazard ratio of 0.86) lower rate of successful ventilation liberation was noted with each liter of uid addition to cumulative uid balance -implying a dose-response relationship. Importantly, our results do not imply a causal relationship, as causality can only be identi ed in a randomized trial; however, given the strength of association between cumulative uid balance and weaning outcome, a well-designed trial seems well justi ed. Taken together with the previous research, our results indicate a possible bene cial effect of restrictive uid management in invasively ventilated COVID-19 and ARDS patients.
Higher cumulative uid balance has also been suggested in previous studies of non-COVID-19 ARDS patients to be potentially associated with worsened outcomes, such as acute kidney injury and decreased survival [25]. In our study, no association was observed in the lower cumulative uid balance group with respect to our secondary endpoints, such as acute kidney injury, the requirement of renal replacement therapy, and mortality, with the caveat that our analysis was too small to evaluate these endpoints, and therefore, should be considered as only hypothesis-generating for future investigations.
The strengths of our study include the size of the multicenter cohort of 22 hospitals that comprised both academic and non-academic institutions, increasing the generalizability of our results. We took careful steps to prevent selection bias that could have been caused by patients who were transferred from other hospitals. Also, trained study coordinators performed careful data collection, and a pre-speci ed statistical analysis plan was prepared before data acquisition.
Our study is subject to several limitations. Although we adjusted for potential prespeci ed confounders, our results may be biased by unmeasured confounders. Our analysis did not adjust for imbalances such as hypertension and baseline Simpli ed Acute Physiology Score in the higher cumulative uid balance group. Furthermore, sensitivity analysis detected no heterogeneous effect. Multiple challenges to research existed during the pandemic that may have affected the clinical outcomes for the included patients, such as organizational issues to utilize resources to prevent future upheavals. Included data were derived from 22 collaborative hospitals that exhibited variation in practice; for example, weaning did not occur with a mandatory protocol, and healthcare provider-related bias could have affected the weaning outcomes. While the percentage of missing values was low, missing data-related bias due to the adoption of different severity illness scores by various centers were thoroughly handled by imputation approaches such as last observation carried forward and robust evaluation tools.

Conclusions
This multicenter study of invasively ventilated COVID-19 and ARDS patients suggest a strong association between higher day 3 cumulative uid balance and the duration of ventilation, even after adjusting for a prede ned set of possible confounding variables. Nevertheless, randomized clinical trials are required to con rm our ndings. To the extent that higher positive uid balance suggests harm and in uences weaning outcomes, maintenance of restrictive cumulative uid balance may improve weaning outcomes in invasively ventilated COVID-19 and ARDS patients. Flow chart of the study.