This study utilized the Medical Information Mart for Intensive Care (MIMIC-IV) electronic database (version 2.2), developed collaboratively by the Massachusetts Institute of Technology (MIT) and Beth Israel Deaconess Medical Center (BIDMC). The database contains relevant information on patients who underwent inpatient treatment at BIDMC between 2008 and 2019. The Institutional Review Board (IRB) of BIDMC waived the requirement for informed consent and allowed the sharing of research resources since all data were de-identified (Johnson et al., 2023). Before data extraction, the author Xiaomin Dong fulfilled all necessary requirements to access the database (Record ID: 57459186).

Medical records of all adult patients aged 18 years or older who were admitted to the ICU after coronary artery bypass grafting (CABG), valve surgery, or aortic surgery were analyzed. A comprehensive list of cardiac procedure codes is provided in Supplementary Table S1. For patients who underwent more than one cardiac surgery, only the data from their first cardiac procedure admission were included. The exclusion criteria were as follows: (1) patients with multiple ICU admissions for cardiac surgery, for whom only the data from the first admission were considered; (2) patients who did not undergo cardiac surgery; (3) patients with missing data on ondansetron exposure; (4) ICU stay ≤24 h; (5) patients aged under 18 years. After applying these exclusion criteria, a total of 7,170 patients were included in the final analysis (Figure 1).

Data collected included (1) demographic characteristics [gender, age (yr), race]; (2) physiological features and laboratory indicators [mean heart rate, mean blood pressure, mean respiratory rate,mean body temperature, preoperative creatinine, white blood cell (WBC) count, blood urea nitrogen (BUN), chloride, potassium, sodium]; (3) Charlson Comorbidity Index, Acute physiology score III (APS-III), Sequential Organ Failure Assessment (SOFA) score; (4) comorbidities (hypertension, myocardial infarct, sepsis, renal disease, cerebrovascular disease, diabetes, liver disease, chronic pulmonary disease, dementia, peripheral vascular disease, congestive heart failure; (5) treatment measures (mechanical ventilation, antibiotics, vasoactive drugs).

The follow-up period began at hospital admission and ended upon death. All laboratory variables and disease severity scores were obtained from data recorded at the first instance after the patient’s admission to the hospital. Covariates with missing values exceeding 20% were excluded. Covariates with less than 20% missing data were processed using the multiple imputation scheme of the Free Statistics software version 1.9 (Beijing, China) and the statistical software package R 3.3.2.

The primary outcome was postoperative AKI and POAF. AKI was defined by the Kidney Disease: Improving Global Outcomes (KDIGO) guidelines, involving an increase in serum creatinine (SCr) to ≥ 1.5 times the baseline within the prior 7 days; or a rise of ≥0.3 mg/dL in SCr within 48 h; or urine volume of <0.5 mL/kg/h for 6 h or more (Khwaja, 2012). Both serum creatinine levels and urine output measurements were utilized to diagnose AKI, ensuring high sensitivity in detecting cases. The secondary outcomes were 28-day mortality, defined as death within 28 days after surgery, ICU mortality.

Continuous variables were reported as median interquartile range (IQR), while categorical variables were expressed as percentages. Fisher’s exact test was used to assess statistical differences between groups for each variable. To evaluate influencing factors related to the risk of AKI and POAF, binary logistic regression analysis was performed.

We prioritized variables that are clinically significant in the context of postoperative cardiac ICU patients. For example, we included factors such as the severity of preoperative disease and the presence of comorbidities, which are critical in assessing patient risk and guiding clinical management in this population. Confounding variables were selected based on a combination of clinical expertise and previous research (Xiong and Xiong, 2022; Cheruku et al., 2023).

In the crude model, covariates were not adjusted. In Model 1, the covariates were adjusted for age, gender, and race. Model 2 and Model 3 were developed based on previous literature and common clinical knowledge to identify characteristic variables that significantly impact the prediction of AKI following cardiac surgery.

Model 2 adjusted for the variables included in Model 1, along with comorbidities such as hypertension, myocardial infarction, renal disease cerebrovascular disease, diabetes, liver disease, chronic pulmonary disease, peripheral vascular disease, and congestive heart failure. Model 3 extended the adjustments of Model 2 by including additional clinical factors such as the use of mechanical ventilation, antibiotics, and vasopressors. It also accounted for the first recorded levels of WBC, chloride, potassium, and sodium. Additionally, physiological parameters such as mean heart rate, mean arterial blood pressure, mean respiratory rate, and mean body temperature, along with the APS III and SOFA score, were included in Model 3.

To mitigate potential confounding, we employed Propensity Score Matching (PSM) to balance baseline characteristics between patients who received ondansetron and those who did not. A 1:1 nearest-neighbor matching algorithm, without replacement, was applied with a caliper width of 0.1. Additionally, a multivariate Cox proportional hazards model was used to assess the association between ondansetron use and mortality.

We performed a subgroup analysis to investigate whether ondansetron had any effect in different subgroups, including age, gender, Charlson comorbidity index, SOFA score, type of surgery, and diabetes. Additionally, we performed a sensitivity analysis by incorporating patients with an ICU stay of less than 24 h following cardiac surgery. For this analysis, we employed logistic regression (Model 3) to evaluate the relationship between ondansetron use and the primary outcomes.

All statistical analyses were performed using R software (version 3.3.2), and p < 0.05 was considered statistically significant.

A total of 7,170 patients who underwent postoperative cardiac surgery were included in this study (Figure 1). The median age of the cohort was 68.7 years (interquartile range [IQR]: 60.9–76.2 years). The incidence of AKI was 74.4% at 2 days and 76.7% at 7 days post-surgery. The 28-day postoperative mortality rate was 1.4%, while the ICU mortality rate was 1.0%. POAF occurred in 17.4% of the patients. Baseline characteristics and clinical outcomes, stratified by ondansetron exposure, are presented in Table 1.

Analysis revealed that patients who received ondansetron postoperatively were more likely to be female and had lower APS III scores. These patients also exhibited fewer comorbidities, including congestive heart failure, peripheral vascular disease, chronic lung disease, and sepsis. Additionally, they had a lower incidence of requiring vasoactive drugs. However, they had a higher prevalence of chronic renal disease and were more likely to need mechanical ventilation (Table 1).

The 1:1 propensity score matching algorithm applied to the dataset of patients receiving ondansetron resulted in 5,204 matched pairs. In the propensity-matched cohorts, the standardized mean differences for all baseline characteristics were below 0.10, indicating minimal imbalance between the groups (Supplementary Table S2).

Logistic regression analysis was conducted to evaluate the relationship between ondansetron use and 2-day AKI. Ondansetron was identified as a significant risk factor in all models. In the unadjusted model, the odds ratio (OR) was 1.41 (95% CI: 1.26–1.57; p < 0.001). After adjusting for multiple covariates in Model 3, the OR remained significant at 1.36 (95% CI: 1.21–1.52; p < 0.001) (Table 2). These findings were further corroborated in the propensity-matched cohort, where ondansetron exposure was associated with an OR of 1.28 (95% CI: 1.13–1.45; p < 0.001), reinforcing the robustness of the association between ondansetron use and increased risk of 2-day AKI (Table 3).

Similarly, logistic regression analysis for 7-day AKI revealed that ondansetron use was a significant risk factor in all models. In the unadjusted model, the OR was 1.36 (95% CI: 1.21–1.51; p < 0.001). In Model 3, after adjusting for multiple covariates, the OR was 1.32 (95% CI: 1.17–1.49; p < 0.001) (Table 2). Results from the propensity-matched cohort showed an OR of 1.25 (95% CI: 1.10–1.42; p < 0.001), further supporting the strong association between ondansetron exposure and an increased risk of 7-day AKI (Table 3).

Logistic regression analysis was also employed to examine the relationship between ondansetron use and POAF. There was no significant difference in POAF risk between patients exposed and unexposed to ondansetron in the unadjusted model (OR 1.11, 95% CI: 0.98–1.26, p = 0.087). However, in Model 3, after adjusting for multiple covariates, the OR was 1.19 (95% CI: 1.04–1.37, p = 0.01) (Table 2). Additionally, results from the propensity-matched cohort revealed an OR of 1.20 (95% CI: 1.04–1.39; p = 0.014), confirming the robustness of the association between ondansetron exposure and increased POAF risk (Table 3).

Cox proportional hazards analysis was used to assess the relationship between ondansetron and ICU mortality. No significant difference in ICU mortality risk was found between the ondansetron-exposed and unexposed groups in any of the models. In the unadjusted model, the OR was 1.24 (95% CI: 0.78–1.97; p = 0.370), and in Model 1, the OR was 1.19 (95% CI: 0.75–1.89; p = 0.467) (Table 4).

Similarly, no significant difference in 28-day mortality risk was found between the groups exposed and unexposed to ondansetron. In the unadjusted model, the OR was 0.91 (95% CI: 0.62–1.35; p = 0.638), and in Model 1, the OR was 0.88 (95% CI: 0.60–1.31; p = 0.539) (Table 4).

In the subgroup analysis for 2-day AKI (Figure 2), no significant interaction was observed between ondansetron exposure and any of the variables examined, including gender (female vs. male), age groups (≥65 years vs. <65 years), Charlson comorbidity index categories (<6 vs. ≥6), SOFA scores (<6 vs. ≥6), types of surgery (coronary artery bypass grafting [CABG], valve surgery, and aortic surgery), and diabetes status (all P for interaction >0.05).

However, stratified analysis for 7-day AKI (Figure 3) revealed significant interactions between ondansetron exposure and age groups. The association between ondansetron and the increased risk of 7-day AKI was notably stronger in patients aged ≥65 years (OR 1.51, 95% CI: 1.29–1.78; p < 0.001) compared to those younger than 65 years (OR 1.13, 95% CI: 0.94–1.36; p = 0.178). The interaction between age and ondansetron exposure was significant (P for interaction = 0.018). Similarly, the association between ondansetron and increased POAF risk was stronger in patients aged ≥65 years (OR 1.31, 95% CI: 1.12–1.53; p = 0.001) compared to those younger than 65 years (OR 0.94, 95% CI: 0.71–1.23; p = 0.653) (Figure 4), with a significant interaction observed (P for interaction = 0.020). No significant interactions were observed in other subgroups.

The final cohort included 7,170 patients, with 427 additional patients whose ICU stay was less than 24 h included in the analysis, bringing the total to 7,597 patients. Sensitivity analysis indicated that ondansetron use was associated with a higher risk of both 2-day AKI [OR 1.45 (95% CI: 1.30–1.62); p < 0.001] and 7-day AKI [OR 1.43 (95% CI: 1.28–1.61); p < 0.001]. Furthermore, ondansetron exposure was associated with a higher risk of POAF [OR 1.22 (95% CI: 1.06–1.39); p = 0.004] (Table 5).