This retrospective study collected data from the MIMIC-IV database (19). The establishment of this database was approved by the Massachusetts Institute of Technology (Cambridge, MA) and Beth Israel Deaconess Medical Center (Boston, MA); consent was obtained for the original data collection. Therefore, the ethical approval statement and the need for informed consent were waived for this manuscript. The permission of access to the database is obtained by author Gang H (Certification number 39516115). These patients diagnosed with septic shock were included in this study. Septic shock was defined according to the several criteria. Firstly, the SOFA score ≥ 2; secondly, consistent low blood pressure (MAP < 65 mmHg) was observed in which vasoactive drugs (such as norepinephrine, dopamine, dobutamine, and vasopressin) were utilized to maintain the pressure, despite adequate fluid resuscitation; thirdly, the lactate level > 2 mmol/L (20). We excluded patients who were younger than 18 years old, not firstly admitted in the ICU, the SOFA score < 2, duration of ICU stay < 24 h.

In our study, the data regarding the patients’ age, gender, weight, comorbidities, type of ICU admission, mean arterial pressure (MAP), heart rate, temperature, respiratory rate, oxygen saturation (SpO₂), white blood cell (WBC) count, hemoglobin level, platelet (Plt) count, and sodium (Na⁺), potassium (K⁺), glucose, urea nitrogen, creatinine, albumin, and lactate levels detected at the admission to the ICU were included.

Serum osmolarity was calculated using the following equation (Na⁺ + K⁺) × 2 + (glucose/18) + (BUN/2.8). Only the values of each element measured at the same time were used in this study. In this study, 290–309 mmol/L was used as the normal range and reference group. Hypo-osmolarity was defined as the serum osmolarity at the first day was lower than 290 mmol/L; Hyper-osmolarity was defined as the serum osmolarity was higher than 309 mmol/L.

The primary outcome of this study was the ICU 28-day mortality. The secondary outcomes included the hospital mortality, duration of ICU stay, volume of fluids administered, volume of urine output, blood product intake, corticoids intake, and acute kidney injury (AKI) incidence.

Neither the patients nor the public were involved in the design, planning, or reporting of this study.

Continuous variables are presented as the mean and standard deviation (SD) or median and interquartile range (IQR). Categorical variables are presented as numbers and percentages. Comparisons between the groups were made using analysis of variance for continuous variables and χ²-test for categorical variables.

Multivariate logistic regression analysis was performed to characterize the relationship between serum osmolarity and ICU mortality. Baseline characteristics such as age, gender, weight, SOFA score, MAP, heart rate, hypertension, respiratory rate, temperature, SpO₂, WBC, Plt counts, hemoglobin, creatinine, lactate levels, serum osmolarity group were selected to enter the multivariate logistic regression model. The missing values of variables were replaced using multivariate imputation by chained equations (MICE); subsequently, the values would be abandoned if the proportion of missing values was larger than 20%.

Except for multivariate logistic regression, propensity score matching (PSM) was used to adjust the covariates to ensure the robustness of our findings. For PSM, the basic admission characteristics including age, weight, gender, comorbidities, vital signs, and laboratory examination values were selected into the analysis, and 1:1 nearest neighbor matching with a caliper width of 0.05 was used. After PSM, standardized mean differences (SMDs) were used to evaluate the balance of values between two groups and the matched number for each group was 392. Afterward, logistic regression analysis was performed on the matched cohort. Results with p-values < 0.05 value were considered as statistically significant. All statistical analyses were performed in SPSS (version 25) and STATA (version 14).

In the present study, the flowchart of the cohort selection is shown in Figure 1. Firstly, data for 13,604 patients diagnosed with sepsis from 2008 to 2019 were collected from this database. Then, we excluded patients who were not admitted into the ICU for the first time and those younger than 18 years old. Furthermore, we excluded sepsis patients without septic shock and 5,055 patients were preliminarily included in this study. Afterwards, according to the latest sepsis-3 criteria, 69 patients with a SOFA score lower than 2 points, and 279 patients with an ICU stay duration shorter than 24 h were excluded. Finally, 4,707 septic shock patients met the inclusion criteria and included in the final study.

The baseline characteristics of patients from each group are presented in Table 1. The patients in the hyper-osmolarity group were relatively older than those in the normal and hypo-osmolarity group (71.3 vs. 67.1 and 61.8 years, p = 0.001). Patients in the hyper-osmolarity group showed significantly higher sofa scores (10.4 vs. 8.7 and 8.8, p = 0.001) and lactate levels (4.6, 4.1, and 4.1 mmol/L, p = 0.001) than in the normal and hypo-osmolarity groups. In addition, the percentage of patients with a history of cardiovascular diseases (CAD), liver disease, renal disease, and diabetes were higher in the hyper-osmolarity group. Further, the vital signs and values of other laboratory tests at admission also show significant difference between three groups.

The serum osmolarity was categorized into three groups and the outcomes within these groups were compared using the chi-square test or Kruskal-Wallis test. As shown in Table 2, the ICU 28-day mortality was significantly lower in the normal serum osmolarity group than in the hypo-osmolarity and hyper-osmolarity groups, respectively (23.0% vs. 30.8 and 34.9%; p = 0.001). This difference was also observed in hospital mortality of septic shock patients (p = 0.001). However, the durations of ICU stay (LOS) in the four groups did not show a significant difference (p = 0.561).

In the present study, we also compared the albumin intake, fluid intake, urine output, blood products intake, and corticoid intake in these groups; the detailed results are presented in Table 2. The patients from the hypo-osmolarity and normal osmolarity groups were administered less amounts of fluid (days 1 and 2; p < 0.01) and produced greater amounts of urine (days 1, 2, and 3; p < 0.01) than the patients from the hyper-osmolarity group. Moreover, the proportion of patients who received albumin was statistically higher in the hypo-osmolarity group (p = 0.001), while corticoids intake within the ICU duration was significantly higher in the hyper-osmolarity group (p = 0.001).

Furthermore, the PSM was used to balance the baseline characteristics between three groups. As shown in Table 3, there were 392 patients in each osmolarity group, the differences of clinical characteristics such as age, gender, sofa score, vital signs, and laboratory results between three groups were not statistically significant (p > 0.05). However, the ICU mortality and hospital mortality in the normal osmolarity group were significantly lower than that in the hypo-osmolarity and hyper-osmolarity group (p = 0.009 and p = 0.026, respectively).

Univariate and multivariate logistic regression analysis were used to identify the relationship between serum osmolarity and ICU mortality. As shown in Table 4, the result of multivariate logistic regression showed that both hypo-osmolarity and hyper-osmolarity were significantly associated with ICU mortality (OR = 1.301, p = 0.007 and OR = 1.470, p = 0.003; respectively). After PSM, the logistic regression was conducted again to study the risk factors for ICU mortality, and the results indicated that both hypo-osmolarity and hyper-osmolarity were independent risk factors (OR = 1.796, p = 0.002 and OR = 1.754, p = 0.006; respectively) (Table 5).

In the present study, we also presented the dynamics of mortality and serum osmolarity within 3 days after ICU admission (Figure 2). As shown in Figure 2A, patients with hypo-osmolarity at admission transformed into those with normal osmolarity on days 2 and 3 after treatment did not show a significant decrease in mortality, compared with those that remained in the hypo-osmolarity state. However, this trend did not exist for patients from the normal osmolarity and hyper-osmolarity groups. In the normal osmolarity group, the ICU mortality of patients who maintained normal osmolarity or those that transformed into the hypo-osmolarity state was statistically lower than that of those that transformed into the hyper-osmolarity state on days 2 and 3 (p < 0.001) (Figure 2B). Obviously, in the hyper-osmolarity group, the patients who achieved normal osmolarity on days 2 and 3 showed a statistically lower ICU mortality (p < 0.001 and p < 0.01, respectively) than those who remained in the hyper-osmolarity state (Figure 2C).