This single-center, retrospective observational study was conducted at the Beijing Friendship Hospital from January 2015 to December 2023. We included patients diagnosed with STEMI who underwent primary percutaneous coronary intervention (PCI). Patients were categorized into two groups based on the presence or absence of CCC observed during the index PCI procedure. The CCC group comprised patients with angiographically visible CCC (Rentrop grade $\ge$1), while the non-CCC group included patients without angiographic evidence of CCC (Rentrop grade 0). A 1:3 matched control group was created, matching patients in the non-CCC group to those in the CCC group by the time of admission time. The study protocol was showed in Fig. 1 and it was approved by the Institutional Review Board of the Beijing Friendship Hospital (Approval No. 2018-P2-030-01), these patients were informed during their hospitalization that their medical data might be used for medical research, and their informed consent was obtained.

Fig. 1.

Flowchart of this retrospective case-control study. CCC, coronary collateral circulation; LASSO, least absolute shrinkage and selection operator; ROC, receiver operating characteristic.

The inclusion criteria were as follows: (1) patients aged $\ge$18 years; (2) patients diagnosed with acute STEMI, diagnosed following guidelines set by the Chinese Society of Cardiology; (3) patients eligible for PCI, having no contraindications, and who underwent either primary PCI or percutaneous transluminal coronary angioplasty (PTCA) within 12 hours after STEMI occurred; (4) patients whose complete angiographic data, including Rentrop collateral grade, were available; and (5) patients whose complete clinical and laboratory data, encompassing cardiovascular risk factors, medical history, symptoms, and biochemical markers, were available.

The exclusion criteria were as follows: patients with (1) a history of myocardial infarction or previous revascularization with pre-existing collaterals; (2) severe mechanical complications, acute left heart failure, sudden cardiac death, or cardiogenic shock, to avoid potential difficulties for the assessment of CCC; (3) severe valvular or congenital heart diseases, or other structural heart diseases which may affect normal cardiovascular function; (4) malignancy, advanced renal disease, severe infection, severe liver injury, or other severe comorbidities (5) incomplete coronary angiography or clinical data, and (6) patients who did not provide informed consent.

Baseline demographic, clinical, laboratory, and angiographic data were collected from the medical records of the patients. This data included age, gender, cardiovascular risk factors (such as hypertension, diabetes, dyslipidemia, and smoking status), history of prior myocardial infarction, culprit vessel and Rentrop collateral grade. Laboratory parameters, including complete blood count, lipid profile, and cardiac biomarkers, were also recorded.

CCC was assessed by two experienced interventional cardiologists who were blinded to the clinical data of the patients. The degree of CCC was graded using the Rentrop classification system, which ranges from “zero” (no visible collaterals) to “three” (complete filling of the epicardial vessel distal to the occlusion) [6].

Continuous variables were presented as either mean $\pm$ standard deviation, or median (interquartile range), depending on their distribution. Categorical variables were presented as frequencies and percentages. Differences between the CCC and non-CCC groups were analyzed using the student’s t-test, Mann-Whitney U test, or chi-square test, as appropriate.

We performed least absolute shrinkage and selection operator (LASSO) regression analysis on the collected variables to identify the most relevant predictors of CCC. Variables selected by LASSO regression analysis were then subjected to univariable and multivariable logistic regression analyses to determine the independent predictors of CCC. Based on the results of the multivariable analysis, a nomogram was constructed to visually predict the probability of CCC.

The performance of the nomogram was evaluated using receiver operating characteristic (ROC) curve analysis, calibration plots, and decision curve analysis (DCA). The area under the ROC curve (AUC) was calculated to assess the discriminative ability of the nomogram. Calibration plots assessed the agreement between predicted probabilities and observed outcomes, while DCA quantified the net benefits of the nomogram at various threshold probabilities to determine its clinical usefulness.

All statistical analyses were conducted using R software (version 4.4.0, The R Foundation for Statistical Computing, Vienna, Austria). A two-sided p-value of 0.05 was considered statistically significant.

Table 1 shows the baseline characteristics of the 668 enrolled patients, categorized into two groups: 501 patients in the non-CCC group and 167 in the CCC group. The table highlights several key findings between the non-CCC and CCC groups. Notably, the CCC group has a significantly higher percentage of patients with a history of CHD (26.95% vs. 13.97%) and higher fibrinogen levels (median 3.00 g/L vs. 2.84 g/L), with p-values of 0.000 and 0.002, respectively. Additionally, the CCC group has a slightly higher average body mass index (BMI) (25.86 kg/m² vs. 25.13 kg/m²) and lower osmolality (median 287.50 mOsm/kg vs. 289.30 mOsm/kg), with p-values of 0.021 for both. These findings suggest that CHD history, fibrinogen levels, BMI, and osmolality may be associated with the development of CCC. For additional relevant patient information, please refer to the Supplementary Materials.

Table 2 showed, in the CCC group, a substantial proportion of patients had good collateral blood flow (78.44%), while none in the non-CCC group did. The CCC group had higher rates of left anterior descending (LAD), left circumflex (LCX), and right coronary artery (RCA) closures, with p-values of 0.009, 0.001, and 0.001, respectively. The Gensini score, which reflects the severity of coronary artery disease, was significantly higher in the CCC group (median 116.00) compared to the non-CCC group (median 83.00), with a p-value 0.001. Additionally, admission and peak N-terminal pro B-type natriuretic peptide (NT-proBNP) levels, indicating more severe heart stress, with p-values of 0.008 and 0.022, respectively. The use of intra-aortic balloon pump (IABP) was significantly more frequent in the CCC group (6.59% vs. 1.80%), and their hospital stay was longer (median 9 days vs. 8 days), both with p-values of 0.004 and 0.001, respectively. There were no significant differences in major adverse cardiac events (MACE), cardiogenic death, recurrent myocardial infarction, cerebral infarction, or cerebral hemorrhage between the groups. For further patient details, please refer to the Supplementary Materials.

LASSO regression analysis (Fig. 2) was used to identify potential predictors of CCC. Using an optimal lambda value, 30 significant predictors were selected from a total of 88 items (As showed in Tables 1,2, and the Supplementary Materials). The selected variables included: age, history of CHD, prior myocardial infarction, diabetes, duration of diabetes, use of beta-blockers on admission, family history of early-onset CHD, family history of ischemic stroke, family history of hemorrhagic stroke, NT-proBNP at admission, red blood cells, mean corpuscular hemoglobin, alanine aminotransferase (ALT), globulin, lactic acid, low-density lipoprotein cholesterol, high-sensitivity C-reactive protein, potassium, chloride, carbon dioxide, osmolality, prothrombin time, levels of fibrinogen, thyroxine, left main closure, LAD closure, LCX closure, RCA closure, RCA lesion, and Gensini score. These variables were identified as potential predictors of CCC and were further analyzed to evaluate their relevance in predicting CCC development.

Fig. 2.

LASSO regression analysis selecting related item for CCC. Note: (1) Left: Coefficient profile plot. The coefficient profile plot illustrates the change in the magnitude of each coefficient as the penalty parameter (lambda) increases. As lambda increases, more coefficients are reduced towards zero in a simpler model with fewer predictors. (2) Right: Cross-validation plot from LASSO regression analysis. The cross-validation plot shows the performance of the model, such as mean squared error, for various lambda values. This plot is used to identify the optimal lambda value that minimizes out-of-sample prediction error, achieving a balance between model complexity and predictive accuracy. (3) Using an optimal lambda_min value of 0.01472, 30 variables were selected as significant predictors from 88 items. The selected variables and their coefficients are as follows: age (–9.0062 $\times$ 10^-03), history of CHD (5.4443 $\times$ 10^-01), prior myocardial infarction (3.8011 $\times$ 10^-01), diabetes (–1.1056 $\times$ 10^-01), duration of diabetes (3.4746 $\times$ 10^-03), beta-blockers use on admission (3.3271 $\times$ 10^-01), family history of early-onset CHD (–2.5897 $\times$ 10^-02), family history of ischemic stroke (–2.3142 $\times$ 10^-01), family history of hemorrhagic stroke (1.7612 $\times$ 10^-01), NT-proBNP at admission (2.8717 $\times$ 10^-06), red blood cells (7.3242 $\times$ 10^-02), mean corpuscular hemoglobin (–2.5730 $\times$ 10^-02), ALT (–2.0129 $\times$ 10^-04), globulin (6.3066 $\times$ 10^-03), lactic acid (–1.0391 $\times$ 10^-01), low-density lipoprotein cholesterol (1.0075 $\times$ 10^-02), high-sensitivity C-reactive protein (1.1336 $\times$ 10^-03), potassium (7.1077 $\times$ 10^-02), chloride (–6.0830 $\times$ 10^-02), carbon dioxide (1.0784 $\times$ 10^-02), osmolality (–5.1357 $\times$ 10^-03), prothrombin time (–1.1948 $\times$ 10^-02), levels of fibrinogen (1.0555 $\times$ 10^-01), thyroxine (1.0420 $\times$ 10^-02), LM closure (–9.7422 $\times$ 10^-01), LAD closure (6.5542 $\times$ 10^-01), LCX closure (7.4271 $\times$ 10^-01), RCA closure (1.7897 $\times$ 10⁺⁰⁰), RCA lesion (2.6392 $\times$ 10^-01), and Gensini score (–1.19458 $\times$ 10^-02). These variables were identified as potential predictors of CCC and were further analyzed to assess their relevance in predicting CCC.

Table 3 summarizes the results of both univariable and multivariable logistic regression analyses used to identify predictors of CCC. In the multivariable model, several factors were independently associated with CCC: a history of CHD (OR = 2.129, 95% CI: 1.262–3.590, p = 0.005), higher fibrinogen levels (OR = 1.375, 95% CI: 1.119–1.689, p = 0.002) and Gensini scores (OR = 1.012, 95% CI: 1.005–1.018, p = 0.001), lower osmolality (OR = 0.970, 95% CI: 0.947–0.993, p = 0.011), LAD closure (OR = 3.368, 95% CI: 1.889–6.003, p 0.001), LCX closure (OR = 3.434, 95% CI: 1.746–6.752, p 0.001), and RCA closure (OR = 11.156, 95% CI: 6.488–19.182, p 0.001).

Fig. 3 presents a nomogram, a graphical statistical tool designed to estimate the probability of CCC in patients with STEMI. This nomogram incorporates several variables, including history of CHD, osmolality, levels of fibrinogen, and LAD, LCX, and RCA closures, as well as the Gensini score. Each variable is assigned a specific point value based on its measurement, which is then totaled to generate an overall score. This total score is used to determine the predicted probability of CCC, as indicated on the nomogram’s linear predictor scale at the bottom.

Fig. 3.

Nomogram for predicting CCC in patients with STEMI. This image is a nomogram designed to predict the probability of an outcome based on several clinical variables. To use the nomogram, first identify the variables, which include coronary heart disease history (binary: 0 or 1), osmolality (ranging from 250 to 320), fibrinogen (0 to 9), LAD closure (binary: 0 or 1), LCX closure (binary: 0 or 1), RCA closure (binary: 0 or 1), and Gensini score (ranging from 20 to 280). For each variable, locate its value on the corresponding scale and draw a vertical line up to the “Points” scale at the top, recording the number of points for each variable. Next, sum the points from all variables to get a total score, which can range from 0 to 350. Use the “Total Points” scale to find the corresponding value on the “Linear Predictor” scale, which ranges from –3 to 4. Finally, convert the linear predictor value to a predicted probability using the “Predicted Probability” scale, which ranges from 0.05 to 0.95. This nomogram provides a visual and quantitative method to assess the likelihood of an outcome by integrating multiple clinical factors into a single predictive model.

The nomogram (Fig. 4) was validated using ROC curve analysis. The optimal cutoff value for the nomogram was identified as 153.05, which resulted in a sensitivity of 0.749 and a specificity of 0.236. This indicates that the nomogram correctly identified 74.9% of patients with CCC but only 23.6% of patients without CCC. The positive predictive value (PPV) was 0.514, indicating that 51.4% of patients scoring above the cutoff actually had CCC. Conversely, the negative predictive value (NPV) was 0.901, showing that 90.1% of patients scoring below the cutoff did not have CCC. The Youden index, which measures the ability of a nomogram to discriminate between patients with and without CCC, was –0.016. The AUC was 0.827 (95% CI, 0.791–0.864), indicating good overall predictive accuracy of the nomogram.

Fig. 4.

ROC curve analysis for nomogram validation. ROC, receiver operating characteristic; AUC, the area under the ROC curve.

The performance of the nomogram was evaluated using a calibration curve (Fig. 5), which compares the predicted probabilities from the nomogram with the actual observed frequencies. The test produced a Hosmer-Lemeshow statistic of 0.050 with 10 degrees of freedom, resulting in a p-value of 1.000. It indicates no statistically significant difference between the predicted probabilities and the observed frequencies. Consequently, the calibration curve demonstrates excellent agreement, suggesting that the nomogram is well-calibrated and provides accurate predictions of the presence of CCC across various risk levels.

Fig. 5.

Calibration curve analysis for nomogram validation.

The presents a clinical DCA (Fig. 6), a tool used to evaluate the clinical utility of the nomogram model for predicting the presence of CCC. The graph shows the trend of net benefit of the model changes as the threshold probability for defining high risk increases. The red line, representing the nomogram, shows a decreasing net benefit as the high-risk threshold increases. This indicates that as the criteria for intervention become more stringent, the ability of the model to provide a net benefit diminishes. For comparison, the gray line, labeled “all”, shows the net benefit if all patients were classified as high risk, while the black line labeled “none” represents the net benefit if no patients were classified as high risk. These benchmark lines help assess the performance of the nomogram against these extreme scenarios.

Fig. 6.

DCA for nomogram validation. DCA, decision curve analysis.

The observational design of this study limits its ability to establish causality between the identified predictors and CCC development. To mitigate potential biases, we ensured the accuracy of historical data by cross-referencing multiple sources and using standardized data collection procedures. However, the sample size may not be large enough to fully capture the diversity of the patient population, potentially limiting the generalizability of the findings. Additionally, the relatively short follow-up period may not adequately reflect long-term outcomes or the evolution of CCC. Variability in laboratory parameters due to differences in testing methods and patient conditions at the time of admission may affect the reliability of the associations identified. Conducted at a single institution, the findings may reflect local practices that do not represent broader populations. These limitations highlight the need for further research to validate and expand upon these findings, ideally incorporating multicenter data and longer follow-up periods to enhance their robustness and applicability.