Clinical information of patients in our study were collected from MIMIC III database, which was illustrated by the Massachusetts institute of technology and had over 40,000 patients admitted between 2001 and 2012 [6]. Patients in the database were fully anonymized. One author(X J) gained access involves MIMIC III database (certification number 9195641) and extracted the data.

Acute myocardial infarction patients diagnosed with AMI according to the 9th revision of the International Classification of Diseases Code (ICD-9) were initially screened. Patients with treatment records indicating sedatives using after CCU admission were initially screened. Sedatives included midazolam, propofol, and dexmedetomidine. Patients who were < 18 years or > 90 years old were excluded. For patients who had more than once CCU inpatient record, only the first CCU inpatients record was collected.

Data on the patients’ characteristics, past medical history, vital sign, biochemistry, sedatives and other treatments were recruited from the database. Variables with missing data are very common in the database of MIMIC III. Serum tropoin and RASS score, with more than 30% missing, were removed from this analysis. For continuous variables with less than 5% missing, we used imputation method with linear regression.

The primary outcome was defined as 28-days mortality. The secondary outcomes included CCU mortality, hospital mortality, length of mechanical ventilation and CCU stay.

Data analyses were performed using StataMP software version 16. Numeric variables were summarized as the mean (standard deviation). Categorical variables were reported as counts (percentage). The student’s test, \({\chi}^{2}\) test, Wilcoxon rank-sum test was used, as appropriate. Univariate and multivariate logistic regression were to explore significantly factors for 28-days mortality. The log-rank test was used to assess differences in 28-days mortality between groups divided by midazolam, propofol, and dexmedetomidine. Subgroup analysis was utilized with \({\chi}^{2}\) test to detect any interaction between midazolam and 28-days mortality, and stratification was performed according to age (< 60, ≥ 60), gender(male, female), BMI(< 24, ≥ 24), WBC(≤ 10, > 10), beta-block (Yes, No), and revascularization (Yes, No). Propensity score matching (PSM) could decrease the influence of confounding factors. The propensity score was allocated based on the probability of a patient who receive midazolam therapy and estimated with using a multivariable logistic regression model. The nearest neighbor matching algorithm was applied using a caliper width of 0.02. There were variables selected to establish the propensity score: age, male, hypertension, creatinine (Scr), myocardial infarction (NSTEMI, AWSTEMI, NAWSTEMI), beta-blocker, stain, vasopressor, and revascularization. Graph of the p score were used to examine the PSM degree. Finally, 140 patients from midazolam groups and 192 from non-midazolam groups were selected and used to further analyses. We explored the potential for unmeasured confounding by calculating E-values [6]. The E-value quantifies the required magnitude of an unmeasured confounder that could negate the observed association between midazolam and 28-days mortality. Two-sided P values less than 0.05 were considered statistically significant.

Information about 427 AMI patients with sedative therapy were recruited. There were 143 patients with midazolam using, 272 patients with propofol using, 28 patients with the dexmedetomidine using, 53 patients with propofol and midazolam using, 22 patients with propofol and dexmedetomidine using, 13 patients with midazolam and dexmedetomidine using, and 8 patients with three seditives using. The overall 28-days mortality rate was 23.9%, and receiving mechanical ventilation was 93.4%. The comparisons of characteristics stratified by 28-days mortality are show in Table 1. There were no significant differences among the groups regarding sex, BMI, diabetes, PLT, using of clopidogrel and mechanical ventilation. Compared with 28-days survival, patients with 28-days mortality had older age, higher rate of hypertension and midazolam using and lower rate of using aspirin, betablocker, and revasucularization therapy(all p < 0.01).

Through logistic regression analysis, we founded that only midazolam using was significant association with 28-days mortality among three sedatives (Table 2). The Kaplan–Meier curves revealed that a increased 28-days mortality was significantly associated with midazolam using (Fig. 1). In Table 3, when compared with non-midazolam using, the rate of CCU and hospital mortality were significantly higher, and the time of mechanical ventilation duration and CCU stay were significantly longer in midazolam using. Subgroup analysis was performed according to the age, gender, BMI, WBC, beta-block, and revascularization (Fig. 2). The HR of midazolam use was significant in the age subgroups(< 60 years old: HR 3.44, 95%CI 1.71–6.94; ≥ 60 years old: RR 2.14, 95%CI 1.46–3.13), gender subgroups(male: RR 2.58, 95%CI 1.71–3.88; female: RR 2.07, 95%CI 1.16–3.68), BMI subgroups(< 24: RR 2.15, 95%CI 1.16–3.98; ≥ 24: RR 2.54, 95%CI 1.71–3.78), WBC subgroups(≤ 10: HR 3.12, 95%CI 1.50–6.52; > 10: RR 2.13, 95%CI 1.47–3.10), beta-block subgroups(Yes: RR 2.43, 95%CI 1.53–3.87; No: RR 1.53, 95%CI 1.05–2.22), and revascularization subgroups(Yes: RR 3.91, 95%CI 2.31–6.59; No: RR 2.17, 95%CI 1.61–2.91), and there were none significant interaction was observed.

Using PSM, 140 patients from midazolam groups and 192 from non-midazolam groups matched from each group were generated (Table 4). In order to assess the quality, we compared the standardized difference of the means and the ratio of the variances between pairs, and drawed the propensity scores (Fig. 3). None significant difference was founded between the two matched groups concerning all nine covariates. After PSM, we found that the rate of CCU mortality & hospital mortality & 28-days mortality and the length of CCU stay & mechanical ventilation were significantly higher or longer in the modazolam using.

We generated an E-value to assess the sensitivity to unmeasured confounding. We determined that the observed RR of 2.41 (As shown in Fig. 2) for 28-days mortality associated with midazolam could be explained away by an unmeasured confounder that was associated with both the exposure (midazolam) and outcome (28-days mortality) by an RR of at least 4.25, above and beyond the measured confounders, but not by weaker confounding.