This study included 1131 MI patients from the multicenter prospective REduction of rIsk for Contrast-Induced Nephropathy (REICIN) study from January 2013 to June 2016 (trial registration: ClinicalTrials.gov NCT01402232). Only adult patients (≥ 18 years of age; referred to CAG or PCI) with providing written informed consent were studied. This study was conducted in accordance with the Declaration of Helsinki and was approved by the Research Ethics Committee of Guangdong Provincial People’s Hospital, Guangdong Academy of Medical Sciences (No. GDREC2012141H).Follow-up data was monitored and recorded by trained nurses and research assistants through outpatient interviews and telephones.

Biochemistry data CK-MB was evaluated on admission, at 3-h intervals in the first 24 h, and daily in the first 3 days following admission. The peak CK-MB before CAG or PCI was chosen. Immunosuppressive method was used to determine the activity of CK-MB and was applied to each sub-center. Due to its non-normal distribution, the CK-MB variable was log-transformed. Receiver operating characteristic (ROC) curve was used to determine the cut-off point of optimal prognostic performance. Then the CK-MB was for the next analysis as a categorical variable based on the cut-off point.

Serum creatinine (Scr) concentration was measured at admission and within 24, 48 and 72 h after CAG or PCI. Other biochemical indicators were evaluated on admission. The echocardiography examination was used to evaluate the left ventricular ejection fraction (LVEF).

The primary endpoint was CI-AKI₀₃₅₀, defined as an increase in the Scr by over 0.3 mg/dL or over 50% from baseline within the first 48 h after the CAG [9]. The secondary endpoint was 3-year all-cause mortality.

Age, creatinine and ejection fraction (ACEF) score was calculated by evaluating age, Scr and LVEF [10]. MRS was calculated by evaluating the presence of hypotension, congestive heart failure(CHF), anemia, and diabetes mellitus(DM), the use of intra-aortic balloon pump(IABP), age > 75 years, the amount of contrast medium, and the basal renal function. There are two types of MRS: Mehran score A using Scr as a criterion for renal function, and Mehran score B using estimated glomerular filtration rate (eGFR) [11].

The PREdictive Value of COntrast voluMe to creatinine Clearance Ratio (PRECOMIN, trial registration:ClinicalTrials.gov NCT01400295) study [12] was a prospective single-center observational study that reviewed all consecutive patients (n = 3369) undergoing CAG and/or PCI between January 2010 and October 2012 according to the institutional protocol. The PRECOMIN study included 1312 MI patients, of which 511 samples had no data deletion. Among 511 patients, 58 (11.35%) patients fulfilled the diagnostic criteria for CI-AKI₀₃₅₀.

Continuous variables were expressed as mean (standard deviation [SD]) or medians interquartile range (IQRs), and discrete variables were expressed as frequency counts and percentages. The differences in variables among groups were evaluated by the t-test or chi-square test. The association between CK-MB and CI-AKI was tested by univariable and multivariable logistic regression. And then, CK-MB was integrated with ACEF score and MRS to compare the predictive power of before and after addition. The performances were evaluated based on discrimination and calibration. Discrimination was evaluated with the ROC curve and expressed by the C-statistic. The C-statistics were compared by the Delong test. We also compared the models using the continuous net reclassification index (NRI) and integrated discrimination and improvement (IDI). The calibration of these models was described by the Hosmer–Lemeshow test.

To evaluate the stability of the integrated models, these models were validated internally using 1000 bootstrap samples and externally validated in the PRECOMIN study data set. We calculated an optimal bootstrap-corrected C-statistic as described by Riley et al. by fitting the prediction model in each of the 1000 bootstrap samples [13]. External validation was furthermore assessed by both discrimination and calibration.

From January 2013 to June 2016, a total of 1131 consecutive MI patients who underwent CAG or PCI were included. The mean age was 60.87 ± 12.05 years, and 932 patients (82.40%) were males. 62 (5.48%) patients fulfilled the diagnostic criteria for CI-AKI₀₃₅₀. The cut-off point of log-transformed CK-MB for the best predictive value of CI-AKI was 4.7. Therefore, patients were divided into two groups based on the cut-off point: 920 (81.34%) patients with log-transformed CK-MB ≤ 4.7, and 211(18.66%) patients with log-transformed CK-MB > 4.7. All of the baseline clinical characteristics of the patients are shown in Table 1. Overall, there were 607(53.67%) patients with chronic kidney disease (CKD), 295(26.08%) patients with DM and 562 (49.69%) patients with hypertension.

Compared with patients with log-transformed CK-MB ≤ 4.7, patients with log-transformed CK-MB > 4.7 demonstrated lower LVEF (52 ± 11vs. 56 ± 11, p < 0.001) and higher random plasma glucose (8.7 ± 3.9 vs. 8.0 ± 3.9 mmol/l, p = 0.022). These patients also showed higher incidences of hypotension on the day of admission (12.8% vs. 6.4%, p = 0.003), anterior infarction (51.2% vs. 31.6%, p < 0.001) and IABP (6.6% vs. 3.0%, p = 0.022). Nonetheless, there were no significant differences in the incidences of CKD, DM and CHF between the two groups (p > 0.05) (Table 1).

Compared with patients with log-transformed CK-MB ≤ 4.7, patients with log-transformed CK-MB > 4.7 displayed a significantly greater incidence of CI-AKI (11.9% vs. 4.0%, p < 0.001, Table 1). The area under the CK-MB curve in relation to CI-AKI was 0.625 (95% confidence interval [CI]: 0.550 to 0.701), and a Hosmer–Lemeshow χ² statistic of 11.37 (p = 0.182). Univariate logistic regression analysis indicated that log-transformed CK-MB > 4.7 was significantly correlated with CI-AKI (odds ratio [OR]: 3.21, 95% CI: 1.89–5.46, p < 0.001, Table2). In multivariable logistic regression, log-transformed CK-MB > 4.7 had an independent association with CI-AKI (adjusted OR: 3.40, 95% CI: 1.93–5.98, p < 0.001, Table 2). Other variables associated with CI-AKI included age, Scr and LVEF. Table 2 depicts the performance of each covariate in logistic regression analysis.

The addition of the categorical variable log-transformed CK-MB to ACEF score, Mehran score A and Mehran score B did contribute to increase in C-statistics, which ranged from 0.680 to 0.733 (p = 0.046), 0.694 to 0.727 (p = 0.091), 0.704 to 0.734 (p = 0.102), respectively (Fig. 1, Table 3). And the CK-MB added significant discriminative value to the traditional models when evaluated by NRI and IDI (Table 3). These models embodied a good calibration for CI-AKI based on the Hosmer–Lemeshow test (Table 3). By internal bootstrap validation, the bootstrap-corrected C-statistics ranged from 0.680 to 0.731, 0.695 to 0.725, 0.702 to 0.731, respectively (Table 3). Additionally, the external validation data set had similar good performance in discrimination and calibration (Table 4, Fig. 2). The external validation shown the addition of CK-MB to ACEF score, Mehran score A and Mehran score B did contribute to increase in C-statistics, which ranged from 0.700 to 0.724 (p = 0.059), 0.716 to 0.742 (p = 0.024), 0.680 to 0.718 (p = 0.002), respectively (Table 4).

First, the definition of CI-AKI was diverse. We adopted a definition of CI-AKI₀₃₅₀ based on the increase in Scr, and used both baseline and postprocedural values, which only gave a moderately accurate evaluation of renal function. However, the definition of CI-AKI in our research was commonly cited in previous studies. Second, the log-transformed CK-MB may be complex in clinical applications. This defect will limit to generalize our results. Third, since there were 855 missing data of troponin T and 822 missing data of troponin I in our study, it is hard to detect the predictive value of troponin for CI-AKI. Fourth, the single center in PRECOMIN study is one of the centers in REICIN study, but the subjects in the two studies were enrolled at different periods.