Database introduction
Our data source was the Medical Information Mart for Intensive Care III (MIMIC-III, version 1.4), an open international database. The MIMIC-III database includes comprehensive, time-stamped information of over 46,000 unique patients from over 60,000 ICU stays at Beth Israel Deaconess Medical Center (BIDMC) in Boston, Massachusetts, between 2001 and 2012. Data were extracted by the author Chen, and access to the database was approved by the institutional review boards of the Massachusetts Institute of Technology after completing the National Institutes of Health’s web-based course and passing the Protecting Human Research Participants exam (certification number: 37147539).
Inclusion and exclusion criteria
Patients with ARDS who were 16 years or older, used mechanical ventilation during the ICU stay and stayed in the ICU for at least 48 consecutive hours were screened for inclusion. To screen the patients with ARDS accurately, the diagnostic information recorded in the MIMIC-III database and the Berlin criteria[15] were considered simultaneously, and the following conditions were proposed: the onset of ARDS is acute, and patients must have PaO2/FiO2 values equal to or less than 300 mmHg when positive and expiratory pressure (PEEP) was at least 5 cmH2O in the first 24 h of entering the ICU. Patients were excluded if they had no data on the EOS within the first 72 h of entering the ICU.
Data extraction
Structured query language (SQL) was used to extract the following data: age, sex, weight (kg), body mass index (BMI), comorbidities (COPD, asthma, diabetes, and sepsis), disease severity score (simplified acute physiology score (SAPS II)), laboratory outcomes (white blood cell count (WBC), red blood cell count (RBC), platelet count (PLT), pH, EOS, and characteristics of mechanical ventilation (tidal volume (mL/kg PBW) and minute ventilation (L/min)). The extracted data were obtained within 72 h after ICU admission. Corticosteroid use was also extracted from the database.
Stratification and definition
According to the cut-off of 2%, the maximum values of EOS within 72 h after ICU admission were used to divide the patients into EOS≥2% and EOS<2% groups. Corticosteroids can cause blood eosinophils to fall at least 50% after the administration of the first 4 h and then back to baseline within 24 h[16]. Therefore, in the subgroup analysis, all patients were assigned to two subgroups based on the usage of any corticosteroid drugs except the external administration route within 24 h before ICU admission to 72 h after ICU admission, including dexamethasone, hydrocortisone, and methylprednisolone. The primary endpoint was the 28-day mortality, defined as death within 28 days from ICU admission. The secondary endpoints included ICU mortality, hospital mortality, length of ICU stay, and length of hospital stay. For patients with more than one ICU stay, only the first ICU stay of the first hospitalization was considered.
Propensity score matching
Because confounders such as comorbidities and characteristics of mechanical ventilation may bias the outcomes, PSM was used to reduce this effect. A multivariable logistic regression model was used to evaluate the propensity score according to the probability of patients being divided into EOS≥2% and EOS<2% groups. The propensity score was calculated by age, weight, BMI, COPD, asthma, sepsis, SAPS II score, WBC, RBC, pH, tidal volume and minute ventilation. A 1:1 nearest-neighbour matching algorithm was used with a calliper of 0.1. We also used kernel density plots of the p score to test the matching level.
Statistical analysis
Continuous variables were summarized as the mean ± standard deviation or median (interquartile range) when appropriate, and categorical data were summarized as proportions. The characteristics of patients with ARDS were compared using a Student’s t-test, Wilcoxon rank-sum test, and χ2 test according to the distribution of the data. The Kaplan-Meier method and log rank tests were used to compare 28-day mortality among the EOS≥2% and EOS<2% groups in patients with ARDS. Cox regression models were used to assess the relationship between blood eosinophil counts and 28-day mortality. A stepwise backward method with p < 0.05 was used to build the model. Twelve potential confounders with a p value <0.10 in the univariate analyses were included in the Cox regression models: age, weight, BMI, COPD, asthma, sepsis, SAP II score, WBC, RBC, pH, tidal volume and minute ventilation. The variance inflation factor (VIF) was used to test multicollinearity, and VIF≥10 indicates multicollinearity between variables. The proportional hazards assumption was tested using Schoelfeld residuals, with p<0.05 constituting evidence for non-proportionality. Subgroup analyses were also performed separately in patients who used corticosteroids and those who did not. PSM was used to balance the cofounders between the EOS≥2% and EOS<2% groups. All p-values were two-tailed, and p < 0.05 was considered statistically significant. Statistical analyses were performed using STATA (Version 16; Stata Corp., College Station, TX, USA).