Design and ethical statement
This retrospective observational study was approved by the Institutional Review Board (IRB) of Seoul National University Bundang Hospital (IRB approval number: B-1806/474–105; approval date: 2018. 6. 11), which waived the requirement to obtain informed consent from the subjects. All data were collected by medical record technicians from the medical informatics team of the institution who were blinded to the purpose of the study.
The medical records of all adult patients aged ≥18 years who were admitted to the ICU between January 2012 and December 2017 were reviewed. If a patient was admitted to the ICU two or more times during the study period, only the last admission (which may have been the most severe) was included in the analysis. Patients with incomplete or missing medical records related to Cl− levels were excluded from the analysis.
Fluctuations in cl− during the 72-h period after ICU admission (main independent variable)
In this study, the Cl− level upon ICU admission (baseline Cl−) was defined as the first Cl− level measured within 24 h after ICU admission. Positive and negative fluctuations in Cl− were defined as the differences between the baseline Cl− and the maximum and minimum Cl− levels, respectively, measured within 72 h after ICU admission. If a patient died within 72 h after ICU admission, the fluctuation in Cl− was calculated using data recorded in the ICU prior to death.
Definitions of normo-, hypo-, and hyperchloremia on ICU admission
We used the baseline Cl− measured as defined above to diagnose dyschloremia at the time of ICU admission. Cl− statuses were defined as follows: normochloremia, 97–110 mmol L− 1; hypochloremia, < 97 mmol L− 1; and hyperchloremia, > 110 mmol L− 1.
Cumulative fluid balance (%) during the 72-h period after ICU admission
The weight-based cumulative FB (%) was calculated using the following formula suggested in previous studies [17, 18]: (Total fluid input – output in L) × 100% × (hospital admission weight in kg)− 1. Fluid input was defined as all types of intravenous and enteral fluids used for maintenance or resuscitation. Fluid output was defined as all types of eliminated and removed fluids (e.g., drainage, rectal, orogastric, nasogastric, and urine output). If a patient died within 72 h after ICU admission, the cumulative FB was calculated from ICU admission until death. All patients were subsequently categorized based on their cumulative FB into the following categories: positive FB (≥5%), even FB (0–5%), or negative FB (< 0%). In addition, the positive FB group was divided into 2 groups: mild or moderate positive FB (5–10%) and severe FB (> 10%), based on previous definitions .
Other measurements (potential covariates)
The patients’ physical characteristics [sex, age (years), and body mass index (kg m− 2)]; Acute Physiology and Chronic Health Evaluation II scores; comorbidities upon ICU admission [hypertension, diabetes mellitus, ischemic heart disease, cerebrovascular disease, chronic obstructive lung disease, liver disease (liver cirrhosis, hepatitis, fatty liver), dyslipidemia, chronic kidney disease, anemia, cancer]; hospital admission through the emergency department; and admitting department (internal medicine, neurologic center, post-cardiothoracic surgery, or post-other surgery) were obtained from the database. Packed red blood cell transfusion, renal replacement therapy, vasopressor infusion, and fluid administration (NaCl 0.9%, balanced crystalloid, and hydroxyethyl starch in ml) within 72 h after ICU admission were recorded as reflective of treatment. Additionally, the number of Cl− measurements within 72 h after ICU admission was also collected.
30-day mortality (dependent variable)
In this study, 30-day mortality was defined as death within 30 days of the ICU admission date. We obtained approval from the Ministry of the Interior and Safety in South Korea to determine the exact date of death of each patient, including those who were discharged from the hospital. We were able to obtain the exact dates of death for all patients as of May 16, 2018.
This study assessed associations of fluctuations (positive or negative) in the Cl− levels within 72 h after ICU admission with 30-day mortality after ICU admission. Additionally, we investigated whether this association might differ according to the cumulative FB status or dyschloremia status upon ICU admission.
Initially, the relationship between a positive or negative fluctuation in Cl− levels within 72 h after ICU admission (continuous variable) and 30-day mortality was first examined using a restricted cubic spline (Additional file 3: Figure S1 and Additional file 4: Figure S2, respectively). Next, we subdivided both positive and negative fluctuations in Cl− into two groups each, using a cut-off point of 10 mmol L− 1. First, as the end points of the U-shaped curve of 30-day mortality associated with both positive and negative fluctuations in Cl− occurred at approximately 10 mmol L− 1 in the cubic spline analyses, this stratification could maximize the impact of fluctuations in Cl− on 30-day mortality. Second, a previous similar study suggested that 30-day mortality was lowest among critically ill patients with systemic inflammatory response syndrome who exhibited a 0–10 mmol L− 1 increase in Cl− .
We next examined the independent associations of each of the covariates with 30-day mortality using a univariable Cox regression analysis (Additional file 1: Table S1). Covariates that received a P < 0.1 in the univariable analysis were included in the final multivariable Cox regression analysis for covariate adjustment. The main independent variable, fluctuation in Cl−, was included as both a continuous and categorical variable (> 10 mmol L− 1 or ≤ 10 mmol L− 1) in different models to avoid multicollinearity between variables. Then, we investigated the interactions of cumulative FB groups or dyschloremia upon ICU admission with the effects of fluctuations in Cl− on 30-day mortality. After confirming these interactions, we performed a subgroup analysis according to the previously described cumulative FB groups and dyschloremia upon ICU admission using the Bonferroni correction to reduce type 1 errors in multiple comparisons . The results of the Cox regression analysis are presented as hazard ratios (HRs) with 95% confidence intervals (CIs), and a log-minus-log plot was used to confirm that the central assumptions of the Cox proportional hazard model were satisfied in each multivariable model.
All data were analyzed using IBM SPSS version 24.0 (IBM, Armonk, NY, USA). A P < 0.05 was considered statistically significant. In the subgroup analyses, a P < 0.012 or < 0.017 was considered statistically significant in the analyses conducted according to FB groups or dyschloremia upon ICU admission, respectively, after applying the Bonferroni correction.