The study was approved by Mayo Clinic Institutional Review Board (Approval No. 13-008906, Title: ‘Does Serum Osmolarity Affect Outcomes of Patients with Acute Lung Injury?’, Approved on 12/4/2013). All patients provided informed consent allowing their medical records reviewed for research purposes. All research activities were in accordance with the ethical standards of the Mayo Clinic ethics committee on human experimentation and with the Helsinki Declaration of 1975.
This is a retrospective cohort study of adult patients admitted to the medical, surgical, and multidisciplinary ICUs at two hospitals in the Mayo Clinic Health System from January 1, 2009, to January 1, 2019. Our inclusion criterion was adult patients with age ≥ 18 years old. The exclusion criterion was patients who did not provide authorization for use of their health records. Jail or prison detainees were also excluded according to Minnesota Research Authorization Regulations. Only the first ICU admission was included for review if a patient had more than one ICU admissions during one hospitalization. Patients who developed ARDS prior to ICU admission were excluded from the analyses.
The diagnoses, key events in the clinical course, laboratory tests, respiratory parameters, the severity of illness (represented by The Acute Physiology and Chronic Health Evaluation, APACHE III score), and outcomes were collected from the EHR. The definitions of the collected data items are listed in Supplemental Table 1. The first available serum sodium value and the most extreme sodium value within 24 hours after ICU admission were used for analysis. The most extreme sodium value was the value with the largest deviation from the laboratory normal range (135–145 mmol/L). The first available sodium, urea, and glucose values within 24 hours after ICU admission were used for the calculation of serum osmolarity.
Computable Phenotyping Of Ards
An EHR-based computable phenotyping strategy to identify ARDS has been previously published . According to this rule-based search strategy, patients who had a clinical diagnosis of ARDS after ICU admission were identified. A clinical diagnosis of ARDS was defined as ICD-9 code 518.52, or ICD-10 code J80, or having ‘ARDS’ or’ acute respiratory distress syndrome’ documented in the clinicians’ notes. In addition, for patients who did not have a clinical diagnosis of ARDS during the ICU admission, automatic EHR searching was applied. ARDS was identified if all these criteria were met:
(1) PaO2/FiO2 ratio ≤ 300
(2) PEEP ≥ 5cm H2O on invasive or noninvasive mechanical ventilation
(3) Bilateral infiltrates on chest radiographs
(4) The presence of at least one risk factor for ARDS (sepsis/septic shock, pneumonia, pancreatitis, trauma, aspiration, multiple transfusion, drug overdose, and shock) ≤ 7 days prior to the onset of 1), 2), and 3).
(1), (2), (3) must be met concurrently.
Patient characteristics were summarized descriptively with median (Interquartile range, IQR) for continuous variables and frequency counts and percentages for categorical variables. In the first analysis, the relationship between sodium/osmolality and the occurrence of ARDS was assessed using univariate and multivariable logistic regression. The multivariable models adjusted for admission ICU and lung injury prediction score (LIPS) using restricted cubic splines. The assumption of a linear function form was assessed by comparing model fit to a version using restricted cubic splines. The second analysis assessed the relationship between sodium/osmolality and outcomes of interest, among patients who developed ARDS in the ICU, using univariate and multivariable linear regression models with generalized estimating equations (GEE). Outcomes of interest were ICU-free days, ventilation-free days, and mortality within 28 days of ARDS diagnosis. The multivariable models adjusted for admission ICU, APACHE score, and PaO2/FiO2 ratio. A similar analysis considered the primary explanatory variable of most extreme sodium in the first 24 hours of ICU admission. Sodium values were analyzed on a scale of 10 mmol/L. Calculated osmolality values were analyzed on a scale of 10 Osm/kg.
In the first analysis, data were missing for sodium and calculated osmolality. Multiple imputations were used to handle missing data for these variables assuming data were missing at random. Twenty-five imputed datasets were created, analyses run on each and results pooled across imputations to account for uncertainty in missingness. In the second analysis data were missing for sodium, calculated osmolality, APACHE score, and PaO2/FiO2 ratio. Similar to the first analysis, multiple imputation was used and 25 imputed datasets were created. In all analyses two-tailed p-values of 0.05 or less were considered statistically significant. Data management and statistical analysis were performed in SAS Studio 3.8 (SAS Institute Inc, Cary, North Carolina).