Sepsis is a lethal condition characterized by multiple organ dysfunction due to disrupted host responses to severe infections.^[1] Affected patients often have a Sequential Organ Failure Assessment (SOFA) score ≥2.^[2] Patients with a SOFA score < 2 and at least one of the following were considered as “suspected sepsis”: (1) quick SOFA (qSOFA) score ≥2; (2) SOFA score = 1; or (3) National Early Warning Score (NEWS) 4-6.^[3] Compared with studies on fluid resuscitation in sepsis patients, there are few studies on fluid management in patients with suspected sepsis. Therefore, we conducted a retrospective cohort study to evaluate the relationship between fluid management and disease progression in suspected sepsis patients.

The present work is a retrospective cohort study on suspected sepsis patients included in the Multi-parameter Intelligent Monitoring in Intensive Care IV (MIMIC-IV1.0) database. Adult patients (>18 years old) diagnosed with suspected sepsis and admitted to the intensive care unit (ICU) were included. Patients with a length of ICU stay of less than 48 h or with missing information regarding fluid management were excluded from the present study. In patients with multiple ICU admissions during one hospital stay, only the first admission was counted.

The MIMIC-IV database was reviewed to retrieve the following information: demographics, clinical comorbidities, SOFA score, qSOFA score, Simplified Acute Physiology Score II (SAPS II), NEWS, first-day laboratory results, first-day vital signs, fluid intake, fluid output, fluid balance, urine output, blood product transfusion, in-hospital mortality, acute kidney injury (AKI), and length of hospital and ICU stay. The worst SOFA, qSOFA, and NEWS values collected within 24 h after admission were used to identify patients with or without sepsis. Patients with the worst SOFA score < 2 were selected for further screening for suspected sepsis based on qSOFA, SOFA, and NEWS scores as described before.

Suspected infection was defined as the acquisition of a body fluid culture temporally contiguous to the antibiotic administration after ICU admission.^[4] The comorbidities analyzed in this study were determined based on the International Classification of Diseases 9 or 10 coding algorithms for Charlson comorbidity.^[5] Fluid management data were extracted from tables called “input events” and “output events” in the MIMIC-IV database.

First, patients were assigned to either the stable group or the deterioration group according to the highest SOFA score between 24 h and 72 h after ICU admission to identify the association between fluid management and disease deterioration. Specifically, patients with the highest SOFA score≥2 (within 72 h after admission to the ICU) were included in the deterioration group; otherwise, they were included in the stable group. Then, patients were assigned into different groups according to their levels of fluid management.

The primary outcome was disease deterioration into sepsis within 72 h after ICU admission. The secondary outcomes included in-hospital mortality, length of ICU stay, length of hospital stay, and incidence of AKI.

All statistical analyses were conducted using the R software package (version 4.1.2, R Foundation for Statistical Computing, Austria). A two-sided P<0.05 was considered statistically significant. Continuous data are presented as the mean with standard deviation (SD) or median with interquartile range (IQR). Categorical data are presented as proportions. Two groups were compared with Student’s t-test, Wilcoxon rank-sum test, and Pearson’s Chi-square test, as appropriate. Univariable logistic regression was used to identify covariates associated with disease deterioration. In multivariate logistic regression, a stepwise backward elimination method was used to keep the variables with a P≤0.2 until all variables that remained in the model were clinically and statistically significant. The fit of this final model was tested by the partial likelihood ratio test.^[6] Patients were assigned into different groups according to fluid management (the cutoff values of fluid management were chosen as the inflection point of restricted cubic spline [RCS] functions). Propensity score matching (PSM) was used to adjust the baseline characteristic differences between different volumes of fluid management. After PSM, standardized mean differences (SMDs) were used to evaluate the balance of characteristics between the two groups (variables used to balance characteristics between groups and SMDs are shown in supplementary Table 1). A variable with SMD greater than 0.1 would be viewed as imbalanced.

A total of 1,285 patients were included in the analysis. The flowchart of patient enrollment is shown in supplementary Figure 1. Patients were assigned into two groups, with 745 (58%) in the stable group and 540 (42%) in the deterioration group. Differences in the baseline characteristics and clinical outcomes between the two groups are shown in supplementary Table 2.

The association between the variables and disease progression was explored by RCS function, as shown in supplementary Figure 2. Platelet count, blood urine nitrogen, systolic blood pressure, fluid intake within 48 h, urine output within 48 h and fluid balance within 48 h had linearity Wald tests of P<0.05 and were transformed using the linear spline functions, with the relative cutoff values shown in supplementary Table 3.

Univariate logistic regression was used to identify factors associated with disease deterioration (supplementary Table 4). The results from multivariate logistic regression showed that within 24 h after ICU admission, fluid intake and fluid balance were not significantly associated with disease progression after adjusting for confounding factors (supplementary Table 5). Within 48 h after ICU admission, compared with low fluid intake (≤100 mL/[kg·48 h]), high fluid intake (>100 mL/[kg·48 h]) was significantly associated with disease progression (adjusted odds ratio [OR] 1.692, 95% confidence interval [95% CI] 1.913-2.400, P=0.003). Compared with low fluid balance (≤40 mL/[kg·48 h]), high fluid balance (>40 mL/[kg·48 h]) was significantly associated with disease progression (OR 1.464, 95% CI 1.045-2.052, P=0.027) (Figure 1 and supplementary Table 6).

According to the fluid intake within 48 h after ICU admission, the patients were further assigned into two groups: the low fluid intake group (≤100 mL/[kg·48 h], n=772) and the high fluid intake group (>100 mL/[kg·48 h], n=513). The characteristics of these two groups are reflected in supplementary Table 7. After the 1:1 PSM, each group had 376 patients. The high fluid intake group still had significantly higher incidences of disease progression (50.0% vs. 35.4%, P<0.001) and longer hospital stay (8.1 d vs. 7.1 d, P=0.001) than the low fluid intake group. No differences were detected in in-hospital mortality, the incidence of AKI, or the length of ICU stay between groups (supplementary Table 8).

According to the fluid balance within 48 h after ICU admission, the patients were assigned into two groups: the low fluid balance group (≤40 mL/[kg·48 h], n=734) and the high fluid balance group (>40 mL/[kg·48 h], n=551). The characteristics of these two groups can be found in supplementary Table 9. After PSM, patients in the high fluid balance group had a significantly higher incidence of disease deterioration (51.9% vs. 34.0%, P<0.001), AKI (57.9% vs. 47.0%, P=0.001), and in-hospital mortality (6.3% vs. 2.7%, P=0.015), as well as a longer hospital stay (8.3 d vs. 7.3 d, P<0.001) than those in the low fluid balance group (supplementary Table 10).

In this study, we report that fluid management within 48 h after ICU admission is significantly associated with clinical outcomes in suspected sepsis patients. Specifically, after PSM, we found that the high fluid balance group was more likely to have disease deterioration, AKI, and higher in-hospital mortality, as well as a longer hospital stay when compared with the low fluid balance group.

Suspected sepsis could be considered a transitional stage from a non-complicated infection to sepsis.^[7] During suspected sepsis, the clinical presentations of organ dysfunction and blood perfusion impairment caused by the infection may not occur, but the underlying pathophysiological changes can progress. This critical transition stage has not attracted adequate attention from physicians.

Fluid therapy is an essential treatment for critically ill patients^[8] and one of the most popular research topics in sepsis. However, excess fluid administration can cause fluid overload, which is associated with unfavorable conditions.^[9,10] There is no widely accepted definition of fluid overload for critically ill patients. Our study found that fluid management (within 48 h after ICU admission) was associated with disease deterioration and clinical outcomes. The exact causal relationship between fluid management and clinical outcomes is not clear. There exists a possibility that fluid overload can exacerbate clinical conditions and push patients from suspected sepsis into sepsis. It can also be interpreted that patients with more severe conditions were more likely to transition from suspected sepsis to sepsis, and they needed greater fluid administration. Nevertheless, our results might indicate the potential clinical application of fluid balance evaluation to predict the outcomes of suspected sepsis patients.

In the early years of clinical practice, the resuscitation targets of sepsis were to rapidly achieve adequate oxygen delivery by augmenting cardiac output to meet the demand of the patients, which paid little attention to the potential damage from excessive fluid administration.^[11] Most physicians considered a state of “constant dehydration” or “in need of fluid” in severely infected patients. These severe septic patients commonly received a large volume of intravenous fluid. In our clinical practice, we have noticed a potential association between unfavorable outcomes and fluid overload. Therefore, we tested whether a “U” shaped model can explain the relationship between the infused fluid volume and clinical outcomes in septic patients. However, in the present study, we did not detect such a pattern using RCS function analysis. Our results showed an elevated risk of disease deterioration with increased fluid intake. The same relationship was also observed between fluid balance and disease deterioration (at 24 h and 48 h, supplementary Figure 2). Therefore, fluid interventions, even those initially considered to be essential to save lives, may cause adverse reactions,^[12] such as edema,^[13] dilutional coagulopathy,^[14] and multiple organ dysfunction.^[15] It was reported that even the “physiological” range of central venous pressure could be more harmful than beneficial in critically ill patients.^[16]

Nevertheless, we believe that there is no “one size fits all” during fluid resuscitation in patients with sepsis. “No harm” should be the first criterion guiding clinical practice. Our results might prompt clinicians to pay closer attention to the volume of fluid infusion during the resuscitation of suspected sepsis patients.

Our study lacked a standardized definition of suspected sepsis. The study population represented patients admitted to the ICU, which could limit the generalizability of the results. To obtain adequate information for analysis, we also excluded patients with a length of ICU stay of less than 48 h. This could exclude patients who had rapid deterioration and death or improvement and discharge from the ICU within 48 h. Future prospective clinical trials are needed to confirm our results and further investigate the relationship between the volume of fluid resuscitations and patient outcomes. Although we identified the associations between fluid management and short-term outcomes in suspected sepsis patients, the causal relationships between fluid management and clinical outcomes need further investigation.

In suspected sepsis patients admitted to the ICU, a high volume of fluid intake may be associated with disease deterioration and longer hospitalization. A higher volume of fluid balance may be associated with increased risks of disease deterioration, AKI, in-hospital mortality, and longer hospitalization.

Funding: None.

Ethical approval: The right of this database was approved by the Massachusetts Institute of Technology (Cambridge, MA) and Beth Israel Deaconess Medical Center (Boston, MA) and consent was obtained for the original data collection. Patients’ information in the MIMIC-IV database was anonymized; therefore, informed consent was not required.

Conflicts of interest: The authors have no financial or other conflicts of interest regarding this article.

Contributors: MB and ZHW contributed equally to this work. MB proposed the study and wrote the paper. All authors contributed to the design and interpretation of the study and to further drafts.

All the supplementary files in this paper are available at http://wjem.com.cn.