The study cohort was retrieved from the database of our center. The including criteria was as follows: (1) the patients were diagnosed STEMI; (2)there was no cardiogenic shock when admitted in emergency room; (3) the patients accepted PCI after emergency angiography. The exclusion criteria was as follows: (1)the patients presented cardiogenic shock when arrived emergency room;(2) the patients rejected emergency angiography; (3) the patients did not need emergency revascularization or need emergency coronary artery bypass graft (CABG) surgery; (4)the patients were deployed prophylactic IABP before revascularization.

From January 2010 to May 2017, total 1250 STEMI patients were admitted in our hospital. 250 patients were excluded because they did not accepted emergency PCI due to over the time window of emergency revascularization. 59 patients were excluded due to cardiogenic shock when admitted in emergency room. 37 patients were excluded for sake of prophylactic use of IABP. 23 patients were excluded because of referral to emergency CABG or referral to elected procedure. 11 patients were excluded due to refusal of emergency angiography. Therefore, the remaining 870 patients were eligible for the study cohort. The study population consisted of 2 series: 1 analysis series (625 consecutive patients for analysis and identification of predictive capability) and 1 validation series (245 consecutive patients for validation the predictive capability).

All the patients with acute chest pain in emergency room accepted ECG within 10 min. STEMI was defined as new onset of ST segment elevation at the J point in at least 2 contiguous leads of more than 2 mm in men or more than 1.5 mm in women in V2 and V3 lead and/or of more than 1 mm in other leads. The presentation of new left bundle branch block was considered equivalent to STEMI [12]. Cardiogenic shock was defined that the systolic blood pressure of the patients is below 90 mmHg more than 30 min or the inotropic agents are needed to maintain the systolic blood pressure above 90 mmHg accompanied with pulmonary congestion and/or peripheral perfusion impairment [13].

The patients ready to accept primary PCI were administered a loading dose of aspirin 300 mg and ticagrelor 180 mg before the procedure. Clopidogrel 600 mg was given if ticagrelor was contraindicated or unavailable. After a radial or femoral artery puncture, a 6F sheath was inserted. Heparin was administered at a dose of 70–100 IU/kg, while tirofiban, urokinase or argatroban were used if necessary. Thrombus aspiration catheter was used if it was considered high burden of thrombus under angiography. After blood flow recovery of IRA, the stent was deployed immediately or delayed according to the discretion of coronary lesions and interventionists’ experience. If CS occurred after the reperfusion of IRA, rescue IABP support was transfemorally placed preferentially. If the patients were not suitable for IABP, inotropic agents were alternative. All of the procedures were accomplished by experienced and qualified interventionists.

Continuous normally distributed variables were shown as mean ± standard deviation (mean ± SD) and were compared using T-test between two groups. While those that were not normally distributed were presented as median (M) and interquartile range (IQR) and compared using Wilcoxon rank-sum test between two groups. Categorical variables were shown as frequencies and percentages and were compared with χ² test or Fisher exact test. In the regression analysis, the following variables in the analysis series were set in the univariate regression analysis initially: age, gender, pre-procedural systolic blood pressure(SBP), pre-procedural heart rate(HR), Killips classification, total ischemic duration, multiple vessel disease, extensive anterior myocardial infarction(MI), infarct related artery(IRA), hypertension, diabetes, dyslipidemia, prior MI, prior stroke, smoking hobby, serum creatinine. The variables significant in univariate analysis were subsequently set in the multivariate analysis. The variables were selected using backwards method. The regression models were calibrated with Hosmer-Lemeshow χ² test for the goodness of fit. Thereafter, the significant covariates were tested for the accuracy with receiver-operating characteristics (ROC) analysis. The area under curve (AUC) was calculated to compare the diagnostic capability of the predictors. Specificity, sensitivity and Youden index (specificity+sensitivity-1) were calculated for identification of a reasonable cutoff value. In the validation series, the predictors were analyzed with ROC in order to identify the predictive value. The statistical analysis was performed by Stata version 12.0 (StataCop., College Station, Texus, USA). All the tests were 2 sided. Values of P < 0.05 were considered statistically significant.

Total 870 patients were valid for the current study. The median age was 65y with interquartile range of 55y-74y. There were 686 male patients (78.9%) and 184 female patients(21.1%). 769 patients were hemodynamic stable during pPCI(Non Shock Group), whereas 101 patients had developed CS during the procedure(Developed Shock Group). Compared with Developed Shock Group, Non Shock Group had lower proportion patients with prior stroke (P < 0.001), higher EF value (P = 0.03), higher pre-procedural SBP(P < 0.001) and lower pre-procedural HR(P = 0.008). Furthermore, more patients had left circumflex branch(LCX) or obtuse marginal branch(OM) as IRA and less patients had right coronary artery(RCA) as IRA in the Non Shock Group(P < 0.001). As regard to pharmacological therapy, there were more patients taking β-blocker in Non Shock Group in comparison with that in Developed Shock Group(P < 0.001), while other medication was no different between two groups (Table 1).

In the analysis series, 72 patients were subjected to the CS during primary PCI. We took CS as dependent variable, the following factors as independent variables: age, sex category, pre-procedural SBP and pre-procedural HR, Killips classification, total ischemic duration, multiple vessel disease, extensive anterior MI, IRA, renal function, prior related history including hypertension, diabetes, dyslipidemia, MI, stroke and smoking habit. On univariate analysis, pre-procedural SBP and pre-procedural HR, Killips classification were the significant variables. Prior history of hypertension had a trend toward to statistical significance (Table 2).

In multivariate analysis, the above statistical significant variables together with some clinical significant variables including age, total ischemic duration, multiple vessel disease were set in the multivariate regression analysis. Table 3 showed the pre-procedural SBP, pre-procedural HR and Killips classification were the independent predictors. Hosmer-Lemeshow test demonstrated the model was well fitted (χ² = 6.43, P = 0.599).

The regression model revealed the risk of developing CS was positively correlated with pre-procedural HR and negatively correlated with pre-procedural SBP. Furthermore, the scatter plot showed the relationship between logit probability of shock and pre-procedural SBP was nonlinear and a inverse proportional function was well fitted (P < 0.001,adjust R square = 0.9904)(Fig. 1a). The relationship between logit probability of shock and pre-procedural HR was also nonlinear and a logistic function was fitted (P < 0.001, adjust R square = 0.9973)(Fig. 1b). SI was formulated as the ratio of pre-procedural HR to pre-procedural SBP. The logit probability of shock was positively linear correlated with SI (P < 0.001, adjust R square = 0.9549). The relationship between probability of shock and SI was located at the rapid rise part of a sigmoid curve and the logistic function was well fitted (P < 0.001, adjust R square = 0.9898) (Fig. 1c and d). Regression analysis demonstrated that SI was a significant independent predictor (Hosmer-Lemeshow χ² = 8.12, P = 0.42)(Table 4). However, there was no interplay between SI and Killips classification no matter in CS cohort, non-CS cohort or global cohort. That meant Killips classification was not able to further promote the discriminability of SI.

With the calibration of ROC analysis, the AUC of SI was significantly higher than that of pre-procedural SBP(0.8113 vs 0.7582, P = 0.04) and pre-procedural HR(0.8113 vs 0.7111, P < 0.001) respectively(Fig. 2a). The AUC of SI was similar to the AUC of combination of SBP, HR and Killips classification(0.8133 vs 0.8137, P = 0.97) (Fig. 2b). Moreover, the AUC of SI in anterior MI patients was not significantly different from that in non-anterior MI patients(0.8332 vs 0.7944, P = 0.46)(Fig. 2c).

The diagnostic parameters were calculated at different cutoff of SI (Table 5). Of note, SI had a high specificity and low sensitivity because of the relatively low incidence of developing CS during pPCI. In other words, it was suitable for exclusion diagnosis. When the cutoff was between 0.90~ 0.95, the accuracy were all above 85% and the negative predictive value were higher than 90%. If the cutoff was set at 0.93, the negative predictive value was as high as 95%.

In the validation series, total 29 patients had undergone CS during primary PCI. Logistic regression analysis demonstrated that the odd ratio of SI (per 0.1 changes) for predicting this event was 1.94(95% CI: 1.54–2.46). Hosmer-Lemeshow test revealed a excellent model fit(χ² = 8.57, P = 0.38). The ROC curve of SI was depicted in Fig. 3. The AUC of this curve was 0.8245(95% CI: 0.7441–0.9048). The incidence of developing CS at different breakpoints was shown in Fig. 4. If the cutoff value was set at 0.93, the sensitivity, specificity, positive predictive value and negative predictive value was 48.3%, 94.4%, 53.8% and 93.2% respectively. The accuracy and Youden index was 88.9% and 42.7% respectively.

Firstly, the negative predictive value of SI had been proven approximate 95% by analysis and validation series. However, the difference of positive predictive value between analysis series and validation series was markedly due to a bit small sample size of validation series. Secondly, left main artery(LM) as IRA is a strong predictor of developing CS, which has been reported in previous study [22]. In our center, all the patients with LM as IRA had accepted prophylactic IABP support, which exclude these patients from current study. Therefore, we could not investigate the possible interaction of LM with SI. Thirdly, SI norms changes by age as aforementioned. We did not analyze the predictive value by age stratification due to age not being significant covariate by regression analysis. Maybe the age range in the data was not large enough or the analysis series sample was not large enough for SI to discriminate the diagnostic capability in different age groups. The last but not least, diabetes usually worsens the prognosis of the patients with acute myocardial infarction accompanied with multivessel disease [27, 28]. It may affect the anti-apoptotic properties of atherosclerotic plaques and reduce the mobilization of stem cells to repair the damaged myocardial tissue [29, 30]. However, in the current study, diabetes did not play an important role in developing CS during primary PCI. It may be considered that diabetes influences the long-term prognosis rather than instant consequence of acute myocardial infarction. Moreover, incretin, a novel antidiabetic drug, has been identified a protective effect on the cardiovascular events. It could improve the cardiovascular prognosis of diabetic patients by pleiotropic effect [27, 28]. In our patient cohort, we did not have the detail percentage of the patients who had accepted incretin therapy, which preclude us to evaluate whether incretin therapy could protect the patients against developing CS during primary PCI.