Predictors of Survival in Critically Ill Patients With Acute Respiratory Distress Syndrome (ARDS)

An Observational Study

Felix Balzer; Mario Menk; Jannis Ziegler; Christian Pille; Klaus-Dieter Wernecke; Claudia Spies; Maren Schmidt; Steffen Weber-Carstens; Maria Deja

Disclosures

BMC Anesthesiol. 2016;16(108) 

In This Article

Methods

This observational analysis was conducted at a 14-bed intensive care unit (ICU) of a national reference centre specialized on treatment of ARDS in adult patients with severely compromised medical conditions. On average, two thirds of all ARDS patients are transferred from other hospitals. Patients at our institution were treated according to a strong treatment algorithm.[7,8]

After written consent of the Ethics Commission at Charité - Universitätsmedizin Berlin (EA1/223/12), clinical routine data from all patients admitted for ARDS between January 2007 and December 2013 were extracted from the two electronic patient data management systems operated at the hospital (COPRA, Sasbachwalden, Germany and SAP, Walldorf, Germany). In addition to basic demographic data, we assessed length of stay and duration of prior mechanical ventilation in referring institutions, comorbidities (using Charlson comorbidity score[9]), ICU admission scores, and use of extracorporeal oxygenation in order to characterize the patient population. As major clinical causes leading to ARDS, we differentiated pneumonia, sepsis of extra-pulmonary origin, trauma, immunedeficiency and "acute on chronic", i.e. patients with an acute pulmonary disease on pre-existing chronic pulmonary disease (e.g.primary lung fibrosis, COPD >/= GOLD 4 or cystic fibrosis), because these are well known influencing factors of mortality in ARDS patients.[10,11]

Day 1 of study inclusion was defined as the first day with a median PaO2/FiO2 below 300 at our hospital. Patient-specific data that was extracted on a daily basis comprised SOFA score, ventilator settings/respiratory parameters (tidal volume (VT), tidal volume/predicted body weight (VT/PBW), Pmean, Ppeak, PEEP, static compliance, FiO2), gas exchange using arterial blood gas analyses (pH, PaO2, PaCO2, PaO2/FiO2), use of nitric oxide and positioning therapy. Status of ARDS was assessed according to the definition of AECC[1] and the Berlin definition[3] upon admission on our ICU.

For analysing data based on ventilator settings and arterial blood gas analyses, the following algorithm was applied: Each day was divided in four intervals of six hours each. In each interval, the combination of ventilator settings and results of blood gas analyses with the least difference in time was chosen. For each of these parameters, the median was calculated and transferred to the study database. Ventilator settings had been saved approximately every 30 min in the electronic patient records and were only considered when they were documented prior to lab results.

Predictive validity for the AECC and Berlin definition as well as for PaO2/FiO2 and FiO2/PaO2*Pmean (OI) regarding mortality was assessed with receiver operator curves (ROC) and corresponding results for area under the curve (AUC). Kaplan-Meier curves were used to illustrate differences in survival using these four mentioned parameters. In order to show differences for continuous variables (i.e. PaO2/FiO2 and FiO2/PaO2*Pmean), we selected the value that maximized the vertical distance between ROC curve and diagonal line (highest sum of sensitivity and specificity).[12] This cut-off value was used to attribute patients to one of two groups (i.e. above or below calculated cut-off) in order to analyse predictive validity.

Descriptive analyses and statistical testing were performed using the R Project of Statistical Computing 3.0.1 with a p value below 0.05 regarded as significant. When normal distribution was ruled out using the Kolmogorov-Smirnov test, results were given in median and interquartile range (IQR), otherwise mean ± standard deviation (SD). Qualitative observations were characterized by numbers with percentage. Statistical significance among groups was univariately analyzed by the exact nonparametric Kruskal-Wallis-test and (pairwise) with the exact Mann–Whitney U test. Exact Chi-Square tests were used for qualitative data. In order to test multivariately for influencing factors of mortality and survival, Cox regression was applied with stepwise backwards selection including variables that showed a statistically significant impact in univariate analyses. All tests should be understood as constituting explorative analysis, such as no adjustment for multiple testing has been made.

Comments

3090D553-9492-4563-8681-AD288FA52ACE

processing....