This is a retrospective, single-center, observational study. Patients admitted to the neurosurgical ICU in a tertiary, referral hospital (Samsung Medical Center, Seoul, Korea) from January 2013 to December 2019 were eligible. We included patients 1) who were hospitalized in the ICU for more than 5 days or died within 5 days after ICU admission and 2) whose serum sodium concentrations were obtained during the ICU admission. We excluded patients 1) with hypernatremia (serum sodium >150 mEq/L) on ICU admission, 2) with insufficient medical records, 3) those who were admitted to departments other than neurosurgery, and 4) those who were transferred to other hospitals and with unknown prognoses (Fig. 1).

In this study, baseline characteristics of comorbidities, ICU management, and laboratory data were collected retrospectively using Clinical Data Warehouse. Our center constructed the “Clinical Data Warehouse Darwin-C” designed for investigators to search and retrieve de-identified medical records from the electronic archives. It contains data pertaining to more than 4 million patients. The clinical and laboratory data were extracted from the Clinical Data Warehouse Darwin-C after finalizing the patient list in this study. The levels of serum sodium were measured at least once every morning in all patients. Additional laboratory tests were performed if patients underwent hyperosmolar therapy or when attending physicians or neurosurgeons needed additional tests throughout the day. Hypernatremia was defined as the highest serum sodium level exceeding 150 mEq/L during ICU stay [26]. We also classified the patients to determine the association between clinical outcomes and severity of hypernatremia. Patients were divided into four subgroups : no hypernatremia (≤150 mEq/L), mild hypernatremia (151-155 mEq/L), moderate hypernatremia (156-160 mEq/L), and severe hypernatremia (>160 mEq/L) [26]. Acute Physiology and Chronic Health Evaluation (APACHE) II score was calculated with worst values recorded during the initial 24 hours in the ICU admission [6,14]. If the patient was intubated, the verbal score of Glasgow coma scale (GCS) was estimated using the eye and motor scores as reported previously [21]. The primary endpoint was in-hospital mortality and the secondary outcome was 28-day mortality.

All data are presented as means±standard deviations for continuous variables and frequencies and proportions for categorical variables. Data were compared using Student’s t-test for continuous variables and the chi-square test or Fisher’s exact test for categorical variables.

Propensity score matching was used to control the selection bias and the confounding detected in this observational study. Each patient with hypernatremia was matched to one of the control patients with the nearest neighbor matching within calipers determined by the propensity score. A caliper width of 0.2 of the standard deviation of the logit of the propensity score was used for the matching [3]. To determine the effectiveness of propensity score matching in controlling the differences between patients with and without hypernatremia, the standardized mean differences (SMDs) were calculated for each variable before and after matching. SMDs less than 10% were considered successful propensity scores matching and balancing the groups.

To conduct doubly robust estimation to improve causal inference, we combined the propensity score matching and regression methods. The propensity score-matched population was subjected to multiple logistic regression analysis with stepwise variable selection. The variables included in the propensity score estimation and the other multivariable analyses were age, sex, comorbidities, cause of ICU admission, utilization of organ support modalities, including mechanical ventilators, continuous renal replacement therapy and vasopressors, invasive ICP monitoring devices, hyperosmolar therapy, GCS, and APACHE II score on ICU admission. The cumulative incidences of mortality were calculated by Kaplan-Meier estimates and compared using a log-rank test. All tests were two-sided and p-values less than 0.05 were considered statistically significant. Statistical analyses were performed with R Statistical Software (version 4.0.2; R Foundation for Statistical Computing, Vienna, Austria).

A total of 12743 patients were admitted to the neurosurgical ICU during the study period and 1146 patients were included in the analysis. In the study population, 353 patients (30.8%) had hypernatremia (Fig. 1). Mean age of all patients was 50.0± 22.9 years. The study included 597 male patients (52.1%). Malignancy (56.4%) and hypertension (34.1%) were the most common comorbidities. Brain tumors (39.1%) and intracerebral hemorrhage (18.2%) were the most common reasons for ICU admission. Age and APACHE II score on ICU admission were higher in patients with hypernatremia compared with those without hypernatremia (both p<0.001). However, malignancy was more common in the group with hypernatremia than in the group without (p<0.001). Vasopressors and mechanical ventilators were more commonly used in patients with hypernatremia compared with those without hypernatremia (both p<0.001). Although the frequency of mannitol use was similar between the two groups (p=0.293), glycerin was more frequently used in the hypernatremia group (p<0.001). The mean value of maximum serum sodium level was higher in the hypernatremia group than in the group without hypernatremia (160.2±8.6 vs. 141.9±5.8 mEq/L, p<0.001). The mean value of minimum serum sodium level was also higher in the hypernatremia group than in the nonhypernatremia group (144.3±12.9 vs. 135.1±5.8 mEq/L, p<0.001). Compared with neurocritically ill patients without hypernatremia, those with hypernatremia had increased blood urea nitrogen, glucose, and osmolality, and decreased glomerular filtration rate (Table 1). However, there were no significant differences in the length of stay in the ICU and hospital between the two groups (p=0.970 and p=0.232, respectively). After propensity score matching, 290 pairs of data were generated by 1 : 1 individual matching without replacement. No significant imbalance was found in the baseline characteristics between the matched data pairs (Table 1).

In the overall study population, the rates of in-hospital mortality and 28-day mortality were higher in the hypernatremia group compared with the non-hypernatremia group (45.6% vs. 15.5% and 43.6% vs. 13.5%, both p<0.001) (Table 1). Clinical outcomes in the propensity score-matched population were similar with those of the entire population. In the propensity score-matched population, the rates of in-hospital mortality and 28-day mortality were also higher in the hypernatremia group compared with the non-hypernatremia group (40.3% vs. 26.9% and 37.9% vs. 24.8%, both p=0.001) (Table 1).

In multivariable analysis of propensity score-matched population, the clinical outcomes, including in-hospital mortality and 28-day mortality, were not significantly different between non-hypernatremia and mild hypernatremia groups (p=0.720 and p=0.690, respectively) (Table 2). However, moderate and severe hypernatremia were significantly associated with inhospital mortality (adjusted odds ratio [OR], 4.58; 95% confidence interval [CI], 2.15-9.75 and adjusted OR, 6.93; 95% CI, 3.46-13.90, respectively) and 28-day mortality (adjusted OR, 3.51; 95% CI, 1.54-7.98 and adjusted OR, 10.60; 95% CI, 5.10- 21.90, respectively) compared with the absence of hypernatremia (Table 2). Finally, multivariable logistic regression analysis revealed that moderate and severe hypernatremia, use of vasopressors (adjusted OR, 3.07; 95% CI, 1.68-5.61), continuous renal replacement therapy (adjusted OR, 5.49; 95% CI, 1.80- 16.70), and GCS on ICU admission (adjusted OR, 0.63; 95% CI, 0.59-0.68) were associated with in-hospital mortality (Table 2).

In survival analysis, the mortality rates of patients with moderate and severe hypernatremia were significantly higher compared with those with mild hypernatremia and without hypernatremia in the overall population and propensity scorematched population (Fig. 2). Especially, the mild hypernatremia group of matched population showed the best survival rate in the Kaplan-Meier curve. However, the mortality rate of patients with mild hypernatremia was not significantly lower compared with those without hypernatremia (16.1% vs. 27.6%, log-rank test, p=0.163).