This study was reported in accordance with the STrengthening the Reporting of OBservational
studies in Epidemiology (STROBE) statement[15]. We collected data from Medical Information Mart for Intensive Care (MIMIC)-III v1.4[16] and the eICU Collaborative Research Database (eICU-CRD) v2.0[17]. Both are extensive, free, public databases containing hospitalization information. MIMIC covers 61,532 ICU admissions for 46,476 patients at the Beth Israel Deaconess Medical Center in Boston, MA, USA. The eICU-CRD covers 200,859 ICU admissions from 139,226 patients at 208 U.S. hospitals. We completed the required courses for the use of the database and obtained the corresponding certificate (researcher certification number 1605699 and record id 27752407).
Study cohort
We conducted a retrospective study of mechanically ventilated adult patients from medical intensive care units (MICUs) and surgical intensive care units (SICUs) based on the method established by Serpa Neto et al[18]. We used only the first ICU admission data for the first hospitalization and patients 16 years of age or older who had been continuously ventilated for at least 48 hours. Patients who had incomplete datasets were excluded. Patients who underwent echocardiography less than 24 hours before mechanical ventilation or within 24 hours after mechanical ventilation were classified as the early TTE group, and the remaining patients constituted the group without earlier TTE (non-TTE).
Data were extracted from the database using structured query language (SQL). The following demographic data (using data from the first 24 hours of admission) were collected: age, sex, weight, race, comorbidities (chronic obstructive pulmonary disorder [COPD], asthma, sepsis, acute respiratory distress syndrome [ARDS]), Sequential Organ Failure Assessment (SOFA) score, Oxford Acute Severity of Illness Score (OASIS), vital signs (mean arterial pressure [MAP], heart rate [HR]) and laboratory values (white blood cell [WBC] count, haemoglobin [Hb], blood urea nitrogen [BUN], pH, pO2, pCO2, lactate). In addition, we also collected management data for the first day of mechanical ventilation (total IV fluid; ventilator settings; use of dobutamine and norepinephrine).
Outcomes
The primary outcome of the study was in-hospital mortality. Secondary outcomes were 30-day mortality from the date of ICU admission; days free of mechanical ventilation and vasopressors 30 days after ICU admission; use of vasoactive drugs; total IV fluid; and ventilator settings during the first day of MV.
Statistical methods
To control for confounding factors, propensity score matching (PSM) was performed. The baseline characteristics of the original cohort were stratified by TTE. The propensity score for an individual was determined based on the covariates age, sex, weight, race, HR, COPD, asthma, ARDS, sepsis, SOFA score, OASIS, WBC, Hb, pH, pO2, pCO2 and lactate using a standard software package (matching package) with a PSM methodology. These variables were selected due to their clinical relevance. This method consisted of ranking the MV patients with TTE and non-TTE, then selecting the TTE patients who had the highest propensity score and finding the non-TTE patient with the closest propensity score (maximum calliper, 0.2). Both patients were then removed from consideration for matching, and the next highest patient was selected (matched 1:1 using the nearest-neighbor algorithm).
After matching, to assess the balance between the two groups, the standardized mean differences (SMDs) between the TTE cohort and the non-TTE cohort were calculated. SMDs eliminate not only the influence of the absolute values from a study but also the influence of the unit of measurement on the results[19]. Continuous variables are shown as the means and standard deviation, and categorical variables are represented as the total and proportion. For continuous variables, we used a nonparametric test or the Wilcoxon rank-sum test. For the categorical variables, we used a chi-square test or Fisher’s exact test.
Secondary outcomes were observed after matching as well. We used paired t tests for continuous outcomes and chi-square tests for categorical outcomes.
We used the random forest model to impute missing data (Additional file 1, eFig.1 and eFig.2)[20].
Sensitivity analysis
We conducted a series of sensitivity analyses with the cohort with missing data, the cohort after imputation, and the cohort after PSM to assess the outcomes. In addition, we used multiple logistic regression, the inverse probability of treatment weight (IPTW)[21] and the covariate balancing propensity score (CBPS)[22] to further validate the primary outcome. To adjust for these covariates, the doubly robust estimation method[23] was used to deduce the independent associations between TTE and in-hospital mortality and 30-day mortality (details about the IPTW and CBPS can be found in the Additional file 2). In addition, we used multiple logistic regression to analyse the impact of TTE during different time periods on the outcome (echo time I: patients who had TTE but the TTE time was not in echo time II; echo time II: TTE time >= MV time+24 hours and TTE<=MV time+24 hours). Finally, we carried out a sensitivity analysis through multivariate logistic regression focusing on patients with ARDS and sepsis.
Statistical significance was assessed to be determined by a two-sided p < 0.05. All statistical analyses mentioned above were performed using R version 3.5.3.