Sources of data
We enrolled a cohort of patients admitted into ICU, treated with and without OND, from a real-world and publicly available clinical database named Medical Information Mart for Intensive Care Database IV (MIMIC-IV version 1.0), and maintained by Beth Israel Deaconess Medical Center in Boston, MA, USA from 2012 to 2019. We were permitted to extract data from the database, and all reporting followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines.
Population selection criteria
The medical records of all adult patients aged at least 18 year admitted to ICU were analyzed. We chose the first ICU admission for patients who were enrolled into the ICU more than once. Those who discharged or died within 48 h after ICU admission was excluded. Patients who were encountered with missing variable data (medication information) and outcome data (in-hospital mortality) were removed.
Confounding Variables and definitions
Data collected included (1) demographic characteristics (sex, age[yr], ethnicity); (2) the admission type; (3) Sequential Organ Failure Assessment (SOFA) score, Simplified Acute Physiology Score II (SAPS II) score; (4) comorbidities (myocardial infarct, congestive heart failure, peripheral vascular disease, cerebrovascular disease, dementia, chronic pulmonary disease, rheumatic disease, peptic ulcer disease, liver disease, paraplegia, renal disease, malignant cancer, metastatic solid tumour, and acute respiratory distress syndrome (AIDS)); (5) treatment measures (vasopressors, mechanical ventilation, and renal replacement therapy).
The data were obtained from MIMIC-IV using Structured Query Language (SQL) with pgAdmin (version 4). The Simplified Acute Physiology Score (SAPS) II and Sequential Organ Failure Assessment (SOFA) scores were calculated within the first 24h after ICU admission. Early application of OND refered to the OND application from 24h before ICU admission to 48h after ICU admission (-24h to 48h). A daily dose of early OND application refered to the average dose of three days (-24h to 48h). Treatment measures were collected on the first day admitted to ICU.
Main exposure and study endpoints
Low-dose OND was defined as >0mg per day and ≤8mg per day. Moderate-dose OND was defined as >8mg per day and ≤16mg per day. High-dose OND was defined as >16 mg per day. The endpoint of this study was in-hospital mortality.
Statistical analysis:
Continuous variables were expressed as the median and interquartile range (IQR) because of their non-normal distribution. Categorical variables were described as the number and percentage (%). Two-group comparisons (with OND vs. without OND group) were conducted with Manne Whitney U test or Chisquared test as appropriate.
Multivariate logistic regression analysis was conducted to assess the association between early ondansetron use and outcomes, with the results expressed as odds ratios (ORs) and corresponding 95% confidence intervals (95%CIs). The mortality outcomes adjusting for confounding variables, shown in Table 1, were selected based on p-value < 0.05 in univariate analysis and potential confounders decided by previous studies and clinical expertise.
Propensity score matching (PSM) and propensity score-based inverse probability of treatment weighting (IPTW) were utilized to ensure the robustness of our findings [11,12]. Logistic regression analysis was carried out in the cohort to perform OR assessment in in-hospital mortality between early OND users and non-OND users, and the confounding variables included age, gender, SOFA score and SAPS II score. The results were described as ORs with 95% CIs. In the PSM model, one-to-one nearest neighbour matching with a calliper width of 0.1 was applied in our study. For the IPTW model, a pseudo-population was generated according to the propensity score. Standardized mean differences (SMDs) were computed to evaluate the efficiency of an unadjusted cohort, PSM and IPTW. Notably, baseline profiles were well balanced between the two groups with SMDs that were less than 5% for all variables (Additional file 1: Figure A1).
As for the PSM cohort, subgroup analysis was conducted to explore whether the association between early OND administration and in-hospital mortality was modified by age, sex, ICU admission and primary diagnosis. Primary diagnosis was classified into the circulatory system, injury or poisoning, infectious diseases, digestive system, respiratory system and nervous system.
The relationship between the daily dose of OND and in-hospital mortality was also evaluated by multivariable logistic regression analysis in the entire population, PSM cohort and circulatory system group after PSM, and the confounding variables included age, gender, SOFA score and SAPS II score. Statistical analysis was performed using R 3.5.3 software for windows and Python 3.7.3. A p-value < 0.05 was considered statistically significant.