A predictive nomogram for intramyocardial hemorrhage in patients with ST-segment elevation myocardial infarction undergoing primary percutaneous coronary intervention based on coronary angiography-derived index of microcirculatory resistance

Abstract

Background

Intramyocardial hemorrhage (IMH) after ST-segment elevation myocardial infarction (STEMI) reperfusion increases the risk of major adverse cardiovascular events (MACE). We aimed to integrate a coronary angiography-derived index of microcirculatory resistance (caIMR) with clinical features in a nomogram model for predicting IMH in patients with STEMI following primary percutaneous coronary intervention (PCI).

Methods

A retrospective study included 309 STEMI patients admitted at Xuzhou Medical University for primary PCI from 2022 to 2023 in training and validation cohorts. Their caIMR was calculated from coronary angiography images and IMH was assessed by cardiac Magnetic Resonance (CMR). A nomogram was constructed through logistic regression analyses. Predictive accuracy, calibration, and clinical usefulness were validated by an area under the curve (AUC) of the receiver operating characteristic (ROC) curve, calibration curves, and decision curve analysis (DCA), respectively. All statistical tests were two-sided.

Results

A total of 247 patients were assigned to the training cohort and 62 to the validation cohort randomly. A nomogram was established using two independent predictors of IMH derived from multivariable analysis: caIMR and left ventricular ejection fraction (LVEF). The AUC for the nomogram was 0.844 (95% confidence interval (CI): 0.793–0.896) in the training cohort and 0.834 (95% CI: 0.724–0.944) in the validation cohort. The predicted and actual estimates were significantly correlated in both cohorts, indicating good calibration of the nomogram. The nomogram was clinically useful for the prediction of IMH, within a 10–88% and 10–74% threshold probability in the training and validation cohorts respectively, exhibited by DCA. Furthermore, the nomogram showed superior discriminative ability for MACE prediction compared to CMR-confirmed IMH (AUC 0.726 vs. 0.584, P < 0.001).

Conclusions

This novel caIMR-based nomogram resulted in high accuracy of prediction for IMH and MACE in STEMI patients undergoing primary PCI. It may become a convenient tool to predict IMH immediately after the PCI procedure, and help with prognostic judgment and early clinical intervention.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12872-025-05057-0.

Background

Great progress has been made in percutaneous coronary intervention (PCI) techniques, and yet the mortality of ST-segment elevation myocardial infarction (STEMI) has continued its upward trend [1]. One of the reasonable explanations is that although the flow of epicardial blood vessels was restored after primary PCI, a significant proportion of patients experienced myocardial perfusion injury and cardiac microvascular dysfunction (CMD), for instance, microvascular obstruction (MVO) and intramyocardial hemorrhage (IMH), which can be sustained or even strengthened after the flow restoration [2, 3] MVO and IMH identified by cardiac magnetic resonance (CMR) were found closely related to the poor prognosis of STEMI, resulting in recurrent major adverse cardiovascular events (MACE), especially remodeling and heart failure (HF) [4, 5] Early determination of CMD contributed to early intervention to improve prognosis. However, CMR is difficult to be performed immediately or early after primary PCI in STEMI patients. Besides, CMR strictly requires patients to coordinate their breathing movements for a long time to achieve good image quality. The complex process, contraindications, and high costs may make it difficult for some STEMI patients to complete CMR, thus hindering early intervention.

The index of microcirculatory resistance (IMR) is another important physiological indicator that can quantitatively assess the severity of CMD [6, 7] For STEMI patients undergoing primary PCI, the elevated IMR of the criminal vessel has been found an independent predictor for MVO, IMH, and the risk of subsequent cardiac mortality or HF [8, 9] Most study results supported that an IMR normal range was less than 25, and an IMR value greater than 40 U was strongly associated with poor prognosis [6, 8–10]. The predictive ability of IMR in STEMI patients after primary PCI was found to correlate well with CMR results [11]. The traditional IMR measurement requires the pressure guidewire as well as vasodilator injection for maximum congestive condition, which increases the complexity of PCI procedure and thus makes IMR not commonly used in clinical practice [12]. IMR derived by the angiographic images (caIMR) has been proven to have a diagnostic accuracy and to be highly correlated with IMR, without the steps mentioned above [13]. It was therefore proposed as an alternative to traditional pressure guidewire-based IMR for prognostic evaluation [13, 14] There have been few studies quantifying the predictive value of caIMR on IMH assessed by CMR. An accurate but easy score is warranted to predict the early occurrence of IMH so that timely treatment can be given for ultimate prognosis improvement. Hence, this study aims to develop and validate a nomogram model incorporating caIMR for the prediction of the IMH risk after primary PCI in STEMI patients.

Methods

Study design

The retrospective study was conducted among STEMI patients who completed the CMR procedure within a week after successful primary PCI from January 2022 and January 2023 at the affiliated hospital of Xuzhou Medical University, China. The study was approved by the ethics committee of this center and the informed consents from patients were waived (Approval No. XYFY2024-KL042-01). Patients were grouped randomly with a 4:1 ratio, in which 247 patients were assigned to the training cohort, and 62 patients were assigned to the validation cohort. The sample size met the principle of 10 Events Per Variable (EPV).

Study population

Patients were included if (1) STEMI was diagnosed according to guidelines; [15] (2) patients were treated with successful primary PCI; (3) the quality of CMR and angiography images met the requirement of reconstruction.

Patients were excluded if (1) necessary clinical data, digital subtraction angiography (DSA), and CMR images were missing; (2) DSA images were unqualified or incomplete; (3) CMR sequences were unqualified or incomplete; (3) patients had contraindications for caIMR detection. The flowchart of patient selection is presented in Fig. 1.

Fig. 1.

The flowchart of patient selection

Data collection

The demographics (age, gender), body mass index (BMI), heart rate, blood pressure, smoking status, medical (hypertension, diabetes) and drug history (renin-angiotensin-aldosterone system inhibitors, beta-blocker), Killip level of cardiac function, detailed information of PCI procedure (infarction-related artery (IRA), multi-vessel disease, stent size, intra-aortic balloon pump (IABP), thrombus aspiration, caIMR) were extracted from medical and image records. Left ventricular ejection fraction (LVEF) and the existence of IMH were obtained from CMR. Laboratory data included general biochemical indicators (C-reactive protein (CRP), low-density lipoprotein cholesterol (LDL-C), serum creatinine (SCr), creatine kinase-MB isoenzyme (CKMB), and cardiac Troponin T (cTnT). If multiple test results of laboratory indicators existed, the peak levels during the acute phase of STEMI were collected. Clinical outcomes included (1) MACE within a 12-month follow-up period, defined as a composite of cardiac death, repeat revascularization (e.g., target vessel re-intervention), recurrent myocardial infarction, and readmission due to heart failure; (2) in-hospital HF, defined as the development of clinical signs and symptoms of HF during the index hospitalization, including new or worsening dyspnea, pulmonary congestion on imaging, rales, elevated jugular venous pressure, or the need for intravenous diuretics, inotropes, or mechanical circulatory support.

Primary PCI procedure

Primary PCI procedures and medical treatment of STEMI were performed by experienced senior cardiologists according to the latest guidelines [15, 16]. Successful PCI was defined as successful stent implantation with residual stenosis < 20% and final Thrombolysis in Myocardial Infarction (TIMI) flow grade 3.

CaIMR acquisition and calculation

In the diastole, a higher flow rate can be achieved, resulting in lower microcirculation resistance. Intracoronary injection of contrast agent can induce a certain degree of hyperemia during angiography when we can obtain a caIMR value from the angiography images.

Immediately after primary PCI, nitroglycerin was injected, and then coronary angiography of at least two postures (the angle between the included two postures ≥ 30 °) was performed (contrast agent injection at a speed of 4 ml/s and stable for more than 3 cardiac cycles; 15 frames/s recording rate of image). Based on the aforementioned angiographic images, a three-dimensional model of the coronary artery was generated by the fully automatic caIMR system, as well as a 3-dimensional mesh reconstruction of the coronary artery generated along the vessel path from the entrance to the distal segment of the target vessel. CaIMR was estimated using computational pressure-flow dynamics with a specialized software system. Detailed calculation formulas were referred to in the previous published study. All images and metrics are analyzed and measured at offline workstations [17].– [18] Normal reference range: caIMR < = 40 U (Figure S1).

CMR procedure

Patients with STEMI underwent CMR imaging no more than 1 week after primary PCI. The protocol, sequence, and setting parameters of CMR scan refer to the 2017 consensus statement by the Society for CMR (SCMR) endorsed by the European Association for Cardiovascular Imaging (EACVI) and 2020 SCMR position paper [19, 20].

CMR was performed on 3.0-T systems (Ingenia, Philips Healthcare, Best, Netherlands). A digital stream (dS) phased-array surface coil is used in combination with an integrated dS posterior spinal matrix coil. All images were acquired by electrocardiography-gated and breath-holding. A gadolinium contrast agent was administered (0.15 mmol/kg), and then, a stack of short-axis images of the left ventricle was acquired using a cine sequence. After 15 min, late gadolinium enhancement (LGE) images were acquired in the same image positions. Imaging parameters were a 350 × 350 mm field of view, 2.6/1.3ms repetition time/echo time (TR/TE), and 8 mm slice thickness. In the T2*mapping sequence, spin echo images of T2 and T2* were collected by holding their breath before the administration of the contrast agent. The high-signal edema area and low-signal core IMH enveloped by the high-signal area were quantified in the short-axis direction of myocardial infarction.

All CMR images were analyzed by 2 radiologists with CMR experience for more than 5 years using CVI 42 software (Circle Cardiovascular Imaging Inc., Canada).

The delayed enhancement area (infarction area) was defined as the area where the signal strength was 5 standard deviations greater than normal distal myocardium at the same layer. MVO area was defined as the area with low signal or without signal within the high signal infarcted core area on the LGE image, quantified as the percentage of MVO volume to total left ventricular volume. The IMH region was mapped on the T2* image, namely an area where the mean signal strength was 2 standard deviations lower than the distant myocardium. A drop of T2* value below 20 was considered the presence of IMH. A suggestion of IMH existence was defined as the T2* value fell below 20ms. The IMH volume and T2* value of the IMH area were measured. Besides assigning patients to the training and validation cohorts, Patients were also grouped according to the existence of IMH.

Statistical analysis

R software (version 4.3.1) was used for all statistical analysis. A two-sided p-value of < 0.05 was considered statistically significant in all comparisons. Variables including cTnT, CKMB, and CRP were first transformed to their natural logarithm (ln) values for comparability. Mean (standard deviation, SD) or median (inter-quartile range, IQR) was used for continuous variables according to normality tested by Shapiro-Wilk. Student t-test was used for normally distributed quantitative data and Wilcoxon rank sum test for non-normally distributed quantitative data. Categorical variables were expressed as a percentage, using chi-square for analysis. In the training cohort, potential predictors for IMH were first screened using univariate logistic regression with a significance threshold of P < 0.05. They were considered candidate variables for the multivariate model, expressed as odds ratio (OR) and 95% confidence intervals (CI). Prior to multivariate modeling, we assessed multicollinearity among these candidate variables using variance inflation factor (VIF), defining a VIF > 5 as indicative of significant collinearity. Subsequently, the multivariate logistic regression model was constructed using a stepwise backward elimination approach based on the Akaike information criterion (AIC). The independent predictors identified in this final multivariate model were then used to develop a nomogram.

The predictive performance of the nomogram was evaluated using three methods. First, the discriminatory ability was verified by the area under the curve (AUC) of the receiver operating characteristic (ROC) curve. Second, nomogram performance was determined by calibration curves and the Hosmer-Lemeshow test. Third, decision curve analysis (DCA) was used to assess the net benefit of the nomogram for different threshold probabilities. Additionally, the nomogram model was internally validated using Bootstrap with 1,000 resamples in the training cohort to enhance the robustness. The predictive performance of the nomogram model was also verified in the validation cohort. Additional analyses assessed the nomogram’s association with clinical outcomes (in-hospital HF/MACE) and compared its predictive accuracy with CMR-detected IMH alone.

Results

Clinical features and patient characteristics

From January 2022 to January 2023, there were a total of 420 STEMI patients successfully treated with primary PCI and who underwent CMR within the following 1 week, in which 111 patients were excluded because they did not fulfill the inclusion or exclusion criteria. Finally, 309 patients were grouped randomly with a 4:1 ratio into the training and validation cohorts.

Table 1 showed the comparison of baseline characteristics between the two cohorts, without significant differences in the aspects of demographics, medical and drug history, PCI procedure, and laboratory results. Of the 247 patients comprising the training cohort, the mean age was 56.8 ± 12.1 years, and 85.4% were male. These patients with a caIMR > 40 U accounted for 21.9%. Examples of images that met caIMR and CMR analysis criteria were shown in Fig. 2.

Table 1.

Comparison of baseline characteristics between the training and validation cohorts

	Training cohort N = 247	Validation cohort N = 62	Statistical value	P value
Age, years, Mean (SD)	56.8 (12.10)	54.9 (12.80)	t = 1.06	0.291
Male, n (%)	211 (85.40)	55 (88.70)	X2 = 0.45	0.504
Heart rate, bpm, Mean (SD)	82.10 (53.10)	81.80 (15.90)	t = 0.06	0.950
SBP, mmHg, Median (IQR)	128 [121;140]	128 [120;138]	Z = −0.02	0.984
DBP, mmHg, Mean (SD)	76.80 (7.13)	77.00 (7.93)	t= −0.16	0.872
Hypertension, n (%)	120 (48.60)	30 (48.40)	X2 < 0.001	0.978
Smoking history, n (%)	108 (43.70)	34 (54.80)	X2 = 2.46	0.116
Killip level, n (%)			X2 < 0.001	1.000
I	239 (96.76)	60 (96.77)
II	8 (3.24)	2 (3.23)
Medications
ACEI/ARB, n (%)	191 (77.30)	44 (71.00)	X2 = 1.10	0.294
Beta-blocker, n (%)	23 (9.31)	6 (9.68)	X2 < 0.001	0.930
PCI procedure
IRA, n (%)			X2 = 0.29	0.864
LAD	129 (52.20)	30 (48.40)
LCX	37 (15.00)	10 (16.10)
RCA	81 (32.80)	22 (35.50)
Mutivessel Disease, n (%)	140 (56.70)	29 (46.80)	X2 = 1.96	0.161
Stent diameter, mm, Median (IQR)	3.00 [2.75;3.50]	3.00 [2.50;3.50]	Z = −0.86	0.389
Stent total length, mm, Median (IQR)	29.00 [21.00;36.00]	29.00 [21.50;36.00]	Z = −0.09	0.931
IABP, n (%)	3 (1.21)	0 (0.00)	X2 < 0.001	1.000
Thrombus Aspiration, n (%)	29 (11.70)	9 (14.50)	X2 = 0.354	0.552
caIMR>40 U, n (%)	54 (21.90)	16 (25.80)	X2 = 0.44	0.507
caIMR value, Median (IQR)	23.60 [17.70;37.10]	25.50 [18.30;38.00]	Z = −0.65	0.515
LVEF, %, Median (IQR)	45.00 [38.00;54.00]	47.50 [35.20;55.80]	Z = −0.19	0.852
Laboratory data
*Peak cTnT, ng/mL, Median (IQR)	7.97 [6.77;8.63]	8.07 [7.17;8.53]	Z = −0.58	0.565
*CKMB, ng/mL, Median (IQR)	4.65 [3.20;5.62]	4.67 [3.53;5.41]	Z = −0.19	0.847
*CRP, mg/L, Mean (SD)	2.57 (1.34)	2.37 (1.44)	t = 0.98	0.328
LDL-C, mmol/L, Median (IQR)	2.68 [2.17;3.20]	2.60 [2.03;3.38]	Z = −0.19	0.847
SCr, µmol/L, Median (IQR)	70.00 [62.00;78.00]	70.00 [60.20;80.80]	Z = −0.59	0.556

Fig. 2.

Comparison of the characteristics of CMR imaging in two STEMI patients without (A to C) and with (D to F) IMH, respectively. Figure A to C corresponsing to the meaured caIMR value 27.6 of the IRA (left anterior descending artery (LAD), showed the same infarct zone using the T2-weighted, T2*-weighted and LGE sequence, respectively. The red arrows pointed to the high signal infarct zone. Figure D to F corresponsing to the meaured caIMR value 48.6 of the IRA (LAD), showed the same IMH area within the infarct zone using the T2-weighted, T2*-weighted and LGE sequence, respectively. Base on the high signal infarct zone circled in green line, the IMH area was circled in red line

*Variables were firstly transformed to their natural logarithm (ln) values for comparability. Abbreviations: SD, standard deviation; IQR, inter-quartile range; SBP, systolic blood pressure; DBP, diastolic blood pressure; ACEI/ARB, angiotensin-converting enzyme inhibitors/angiotensin receptor blockers; IRA, infarction related artery; LAD, left anterior descending artery; LCX, left circumflex artery; RCA, right coronary artery; IABP, intra-aortic balloon pump; caIMR, coronary-angiography-derived index of microcirculatory resistance; LVEF, left ventricular ejection fraction; cTnT, cardiac Troponin T; CKMB, creatine kinase-MB; CRP, C-reactive protein; LDL-C, low density lipoprotein cholesterol; Scr, serum creatinine.

Overall IMH incidence in the training and validation cohorts

From CMR results, IMH occurred in 72 and 17 patients in the training and validation cohorts, respectively. Baseline characteristics were further compared between IMH (n = 89) and non-IMH (n = 220) groups (Table 2). IRA of the two groups differed significantly. In the aspect of laboratory results, the levels of peak ln (cTnT), ln (CKMB), and ln (CRP) seemed much higher in the IMH group than in non-IMH groups. Worse LVEF was also observed among Patients in the IMH group. Patients with a caIMR > 40 U accounted for 57.3% and 8.64% of IMH and non-IMH groups, respectively (P < 0.001).

Table 2.

Comparison of baseline characteristics in two groups based on whether the IMH existed

	Non-IMH N = 220	IMH N = 89	Statistical value	P value
Age, years, Median (IQR)	57 [47;66]	58 [49;66]	Z = −030	0.766
Male, n (%)	190 (86.40)	76 (85.40)	X2 = 0.05	0.823
Heart rate, bpm, Mean (SD)	79 [69;86]	80 [73;91]	Z = −1.74	0.081
SBP, mmHg, Median (IQR)	129 [121;142]	126 [120;136]	Z = −1.26	0.208
DBP, mmHg, Median (IQR)	77 [71;81]	79 [71;83]	Z = −1.38	0.169
Hypertension, n (%)	105 (47.70)	45 (50.60)	X2 = 0.20	0.652
Smoking history, n (%)	104 (47.30)	38 (42.70)	X2 = 0.53	0.465
Killip level, n (%)			X2 = 1.78	0.182
I	211 (95.91)	88 (98.88)
II	9 (4.09)	1 (1.12)
Medications
ACEI/ARB, n (%)	164 (74.50)	71 (79.80)	X2 = 0.95	0.329
Beta-blocker, n (%)	200 (90.90)	80 (89.90)	X2 = 0.08	0.780
PCI procedure
IRA, n (%)			X2 = 10.20	0.006
LAD	101 (45.90)	58 (65.20)
LCX	35 (15.90)	12 (13.50)
RCA	84 (38.20)	19 (21.30)
Mutivessel Disease, n (%)	117 (53.20)	52 (58.40)	X2 = 0.70	0.402
Stent diameter, mm, Median (IQR)	3.00 [2.75;3.50]	3.00 [2.50;3.50]	Z = −1.27	0.203
Stent total length, mm, Median (IQR)	29.00 [21.00;36.00]	29.00 [23.00;35.00]	Z =−0.004	0.997
IABP, n (%)	1 (0.45)	2 (2.25)	X2 = 5.00	0.200
caIMR>40 U, n (%)	19 (8.64)	51 (57.30)	X2 = 85.65	<0.001
LVEF, %, Median (IQR)	49.00 [40.00;55.20]	39.00 [33.00;47.00]	Z = −5.52	<0.001
Laboratory data
*Peak cTnT, ng/mL, Median (IQR)	7.67 [6.54;8.42]	8.51 [7.76;8.90]	Z = −5.04	<0.001
*CKMB, ng/mL, Median (IQR)	4.48 [3.18;5.47]	5.13 [3.96;5.70]	Z = −2.76	0.006
*CRP, mg/L, Median (IQR)	2.48 [1.51;3.32]	2.87 [1.97;3.71]	Z = −2.02	0.044
LDL-C, mmol/L, Median (IQR)	2.61 [2.15;3.25]	2.73 [2.14;3.15]	Z = −0.37	0.710
SCr, µmol/L, Median (IQR)	70.00 [62.00;79.00]	70.00 [60.00;77.00]	Z = −1.03	0.304

*Variables were firstly transformed to their natural logarithm (ln) values for comparability. Abbreviations: SD, standard deviation; IQR, inter-quartile range; SBP, systolic blood pressure; DBP, diastolic blood pressure; CEI/ARB, angiotensin-converting enzyme inhibitors/angiotensin receptor blockers; IRA, infarction related artery; LAD, left anterior descending artery; LCX, left circumflex artery; RCA, right coronary artery; IABP, intra-aortic balloon pump; caIMR, coronary-angiography-derived index of microcirculatory resistance; LVEF, left ventricular ejection fraction; cTnT, cardiac Troponin T; CKMB, creatine kinase-MB; CRP, C-reactive protein; LDL-C, low density lipoprotein cholesterol; Scr, serum creatinine.

Potential predictors of IMH through univariate and multivariate logistic regression analyses

Table 3 presented the results of univariate and multivariate logistic regression analyses. In the training cohort, univariate analysis identified six variables significantly associated with IMH occurrence (P < 0.05): IRA, ln (cTnT), ln (CRP), caIMR value, caIMR > 40 U, and LVEF. These six candidate variables were entered into a multivariate logistic regression model. The final model selection was rigorously performed using a backward stepwise elimination procedure guided by AIC, aiming for the most parsimonious model with optimal fit. Variables were sequentially removed based on their statistical contribution to the model fit (greatest reduction in AIC). Following this statistically driven selection process, only caIMR > 40 U (OR: 17.27, 95% CI: 8.17–39.09, P < 0.001) and LVEF (OR: 0.93, 95% CI: 0.90–0.96, P < 0.001) were retained as independent predictors.

Table 3.

Logistic regression analysis for determining independent predictors of IMH

Characteristics	Univariate Logistic Regression			Multivariate Logistic Regression
	OR	95% CI	P	OR	95%CI	P
Age	1.00	0.97–1.02	0.772
Heart rate	1.00	0.99–1.01	0.866
SBP	0.98	0.96-1.00	0.065
DBP	1.01	0.97–1.05	0.622
Hypertension	0.96	0.55–1.67	0.889
Killip	0.34	0.04–2.86	0.324
Mutivessel Disease	1.36	0.77–2.39	0.287
Thrombus Aspiration	0.77	0.31–1.88	0.56
IRA(LAD vs. LCX)	0.58	0.25–1.33	0.20
IRA(LAD vs. RCA)	0.44	0.23–0.86	0.015
Beta-blocker	0.60	0.25–1.45	0.252
ACEI/ARB	1.01	0.52–1.95	0.974
*cTnT	1.44	1.16–1.78	0.001
*CKMB	1.16	0.97–1.39	0.103
*CRP	1.27	1.03–1.58	0.028
LDL-C	1.03	0.97–1.09	0.332
SCr	1.00	0.98–1.02	0.711
caIMR > 40 U	14.93	7.27–30.67	< 0.001	17.27	8.17–39.08	< 0.001
LVEF	0.94	0.92–0.97	< 0.001	0.93	0.89–0.96	< 0.001

*Variables were firstly transformed to their natural logarithm (ln) values for comparability. Abbreviations: SBP, systolic blood pressure; DBP, diastolic blood pressure; IRA, infarction related artery; LAD, left anterior descending; LCX, left circumflex; RCA, right coronary artery; ACEI/ARB, angiotensin-converting enzyme inhibitors/angiotensin receptor blockers; cTnT, cardica troponin T; CKMB, creatine kinase-MB; CRP, C-reactive protein; LDL-C, low-density lipoprotein cholesterol; SCr, serum creatinine; LVEF, left ventricular ejection fraction; caIMR, coronary angiography-derived index of microcirculatory resistance.

Construction of the nomogram prediction of IMH

Based on the backward step-down selection process with the Akaike information criterion and the independent risk factors, namely caIMR and LVEF were finally integrated to construct the prediction nomogram for IMH in STEMI patients treated with primary PCI (Fig. 3). The formula of the model to calculate the risk of IMH after primary PCI in STEMI patients is , where p represents the probability of IMH, and caIMR >40 U is a binary indicator (1 if caIMR > 40 U, 0 otherwise).

Fig. 3.

The nomogram model for predicting the risk of IMH after primary PCI in STEMI patients. LVEF, left ventricular ejection fraction; caIMR, coronary angiography-derived index of microcirculatory resistance

To enhance the clinical applicability of the nomogram, we incorporated two representative case examples in Figure S2. Clinicians can utilize this prediction tool in four specific steps: (1) Identify the corresponding value on the “LVEF” axis, (2) Locate the patient’s value on the “caIMR > 40 U” axis (Yes = 1, No = 0), (3) Sum the derived points, and (4) Project the total points to obtain the predicted IMH risk probability. The entire process can be completed within 30 s.

Predictive accuracy of the nomogram for IMH

We calculated the AUC values to evaluate the discriminative ability of the prediction nomogram to identify patients at risk for IMH. ROC curve analysis indicated that an AUC value for the nomogram to predict IMH was 0.844 (95%CI, 0.793 to 0.896) in the training cohort, and the corresponding value was 0.834 (95%CI, 0.724 to 0.944) in the validation cohort (Fig. 4A and B), which suggested good discriminative ability of the nomogram for IMH.

Fig. 4.

Comparing different risk models on the predictive accuracy of IMH using the ROC curves in the training A and the validation cohort B, respectively. AUC indicates an area under the curve. LVEF, left ventricular ejection fraction; caIMR, coronary angiography-derived index of microcirculatory resistance

Figure 5 presented the calibration assessment of the nomogram model. The calibration curves of the nomogram in the training and validation cohorts were plotted: the x- and the y-axis represented the predicted risk and the observed actual risk of IMH occurrence respectively, and the black dotted line represented the ideal reference line. The calibration curve of the model was close to the reference line suggesting that the predicted probability was in good agreement with the actual probability. The Apparent line (blue) meant the fit between the predicted and the actual risk of IMH, while the Bias-corrected line (red) referred to the result using bootstrap sampling 1000 times in the data constructing the model and thus corrected the overfitting. The chi-square of the training and validation cohorts obtained by the Hosmer-Lemeshow goodness of fit test is 4.79 (P = 0.69) and 9.99 (P = 0.19), respectively, indicating insignificant deviation between the probabilities predicted by the model and the observed probabilities.

Fig. 5.

The calibration curves of the nomogram in the training (A) and validation (B) cohorts

The DCA was also performed to illustrate the clinical usefulness of the nomogram model. The decision curves were shown in Fig. 6, which suggested that in the training cohort, the predicted threshold probability of IMH occurrence in STEMI patients with primary PCI was 0.10–0.88 and the highest net benefit was 0.23. Besides, the nomogram had a good efficiency in predicting IMH when the threshold probability was 0.10 to 0.74 in the validation cohort, with the highest net benefit was 0.22. For example, if the individual threshold probability of a patient is 60% (i.e. if the probability of IMH is 60% and will choose early treatment for prognosis improvement), the net benefit of using the nomogram to decide whether to receive treatment is 0.08, which is higher than the net benefit of all treatment, no treatment or only using caIMR or LVEF.

Fig. 6.

Assessing the nomogram model using the decision curves analysis in the training (A) and validation (B) cohorts, respectively

Association between nomogram-derived IMH risk score and outcomes

To establish the clinical relevance of IMH risk prediction, we analyzed associations between the nomogram-derived IMH risk score and concrete outcomes: in-hospital heart failure (HF) and MACE. Multivariate logistic regression adjusted for baseline characteristics identified both CMR-detected IMH and the nomogram risk score as independent predictors of in-hospital HF (IMH: OR 2.75, 95% CI 1.53–4.95, P < 0.001; nomogram: OR 11.41, 95% CI 3.82–34.08, P < 0.001) and MACE (IMH: OR 1.99, 95% CI 1.01–3.87, P = 0.043; nomogram: OR 12.79, 95% CI 4.39–38.05, P < 0.001), with detailed results provided in Table S1–S4.

Crucially, the nomogram demonstrated superior prognostic discrimination versus isolated IMH assessment for both in-hospital HF (AUC 0.748, 95% CI 0.692–0.804 vs. AUC 0.633, 95% CI 0.580–0.687, P < 0.001) and MACE (AUC 0.726, 0.641–0.812 vs. AUC 0.584, 95% CI 0.505–0.663, P < 0.001). This clinically meaningful improvement in risk stratification provides a foundation for targeted interventions in high-risk STEMI patients post-PCI (Figure S3-S4).

Discussion

We developed and validated the first nomogram combining caIMR (> 40 U threshold) and LVEF to predict IMH risk immediately post-primary PCI. The model demonstrated excellent discrimination (AUC 0.84 − 0.83) and calibration, outperforming individual parameters. This addresses a critical gap in early CMD detection where advanced imaging like CMR is not accessible.

Despite successful primary PCI in STEMI, up to 65% of patients fail to achieve adequate myocardial reperfusion due to CMD, with IMH representing its most severe form [21–23]. Current diagnostic methods (CMR, invasive IMR) face practical limitations in acute settings, creating an urgent need for rapid risk stratification tools. In our study, multivariable analysis identified both caIMR and LVEF as significant IMH predictors, with caIMR demonstrating particularly strong association. Consequently, in the constructed nomogram, caIMR carried greater weight than LVEF, reinforcing its primacy in acute risk stratification. While a more complex model incorporating additional variables such as infarct size and troponin T showed slightly improved AUC (0.84–0.86), it also resulted in increased AIC and minimal net reclassification improvement. These findings support the selection of a parsimonious model that offers robust predictive performance without sacrificing simplicity.

Notably, while both continuous caIMR values and the categorical threshold (caIMR > 40 U) were evaluated, only the dichotomized variable (caIMR > 40 U) independently predicted IMH risk with statistical significance. This suggests a non-linear relationship between absolute caIMR values and IMH probability. The > 40 U threshold—validated in our prior work [17, 18]—proved clinically robust for identifying patients at heightened risk of MVO and adverse outcomes. Although categorizing continuous variables can reduce statistical power, the dichotomized form aligns with known pathophysiologic thresholds and enhances clinical interpretability and rapid decision-making [24, 25, 29, 30]. Regarding LVEF, extensive evidence establishes its prognostic importance in STEMI [26]. IMH contributes to adverse cardiac remodeling, which results from the loss of coronary microvascular integrity and the subsequent myocardial iron accumulation due to erythrocyte extravasation and hemoglobin degradation and occurs in about half of STEMI patients [27, 28] The inclusion of LVEF further strengthens the model’s applicability.

Beyond its predictive capacity for IMH, this nomogram also demonstrated superior performa nce in forecasting downstream clinical outcomes, including MACE and in-hospital HF, compared to CMR-detected IMH. These findings highlight its practical value as an alternative tool for early risk stratification, particularly in acute settings. The model also addresses the narrow therapeutic window for mitigating reperfusion injury and CMD. Wider adoption could potentially help address disparities in STEMI outcomes highlighted in the Report on Cardiovascular Health and Disease in China 2023 [31].

The nomogram’s major strength lies in its simplicity, requiring only two routinely avail able parameters (caIMR and LVEF) for immediate post-primary PCI risk assessment. Three practical implementation pathways exist, including (1) real-time decision support: embedding the nomogram formula into Electronic Health Record systems for automated risk reporting; (2) mobile application: developing smartphone apps allowing physicians to input parameters and instantly display risk stratification (e.g., Figure S2 cases: LVEF 54%/caIMR < = 40 → 8.3% low-risk; LVEF 31%/caIMR > 40 → 88.6% high-risk); (3) risk-directed management: initiating early interventions (e.g., intensified anti-thrombotic regimens, targeted anti-inflammatory therapy, or short-term mechanical circulatory support) for high-risk patients.

The study had some limitations. In this study, we employed a caIMR threshold of > 40 U to define significant coronary microvascular dysfunction. This threshold was selected primarily based on clinical consensus established in prior studies [16–18]. Although restricted cubic spline analysis suggested a non-linear increase in risk beginning around 37–38 U, the choice of 40 U aimed to align with existing standards for clinical applicability and interpretability. Sensitivity analyses confirmed that this threshold effectively identified high-risk patients and that the main findings were robust to this choice. Second, the nomogram was based on data from a single center, so the lack of immediate external validation necessitates caution in generalizing these results. Generalizability may be limited by heterogeneity in caIMR measurement techniques, differences in adjuvant therapy regimens, and varied definitions of IMH on CMR. We are about to initiate collaboration with the IMPROVE-AMI study (NCT06683131) for external validation. A third limitation is because of the retrospective characteristics of this study; it is difficult to prove this hypothesis with existing data. So more prospective studies are needed to further verify the predictive value of the nomogram. Fourth, the relatively low proportion of female participants (13.9%) may limit the generalizability of the model across sexes. Given that sex-related differences in microvascular dysfunction have been previously reported, further validation in more gender-balanced cohorts is warranted. Finally, it remains to be determined whether this nomogram applies to non-STEMI or other high-risk patients to predict the risk of IMH.

Conclusions

In summary, we developed and validated a nomogram containing caIMR to accurately predict the risk of IMH after primary PCI in STEMI patients. The predictive accuracy of the nomogram proposed is significantly better than that of using either caIMR or LVEF alone. This clinically practical tool enables early IMH risk stratification and prognostic assessment, particularly for MACE and in-hospital HF. However, external validation across varying clinical settings remains necessary to confirm the model’s generalizability beyond our study population.

Abbreviations

PCI

Percutaneous Coronary Intervention

STEMI

ST-segment Elevation Myocardial Infarction

CMD

Cardiac Microvascular Dysfunction

MVO

MicroVascular Obstruction

IMH

Intramyocardial Hemorrhage

CMR

Cardiac Magnetic Resonance

MACE

Major Adverse Cardiovascular Events

HF

Heart Failure

IMR

Index of Microcirculatory Resistance

caIMR

Coronary Angiography-derived Index of Microcirculatory Resistance

DSA

Digital Subtraction Angiography

BMI

Body Mass Index

IRA

Infarction Related Artery

IABP

Intra-Aortic Balloon Pump

LVEF

Left Ventricular Ejection Fraction

CRP

C-Reactive Protein

LDL-C

Low-Density Lipoprotein Cholesterol

SCr

Serum Creatinine

CKMB

Creatine Kinase-MB

cTnT

cardiac Troponin T

TIMI

Thrombolysis in Myocardial Infarction

LGE

Late Gadolinium Enhancement

SD

Standard Deviation

IQR

Inter-Quartile Range

OR

Odds Ratio

CI

Confidence Interval

AUC

Area Under the Curve

ROC

Receiver Operating Characteristic

DCA

Decision Curve Analysis

Author contributions

Y.D. and Y.L. obtained and analyzed all patients’ data. Y.Y. and S.H. were responsible for the analysis of CMR images. Y.W. and Y.D. were major contributors in writing the manuscript. Q.C. was responsible for revising the manuscript. Y.Z. was mainly in charge of the study design and manuscript writing instruction. All authors read, reviewed and approved the final manuscript.

Funding

This study was supported by the Key Research and Development Plan (Social Development) of the Science and Technology Project of Xuzhou City, Jiangsu Province (Grant No. KC23274), and the Affiliated Hospital of Xuzhou Medical University (Grant No. 2022ZL.01).

Data availability

All data analyzed in this study can be accessed from the corresponding authors (xyfyduanyang0125@163.com) upon reasonable request.

Declarations

Ethics approval and consent to participate

This work was supported by grants from the National Natural Science Foundation of China (81873486), the Science and Technology Development Program of Jiangsu Province– Clinical Frontier Technology (BE2022754), Clinical Medicine Expert Team (Class A) of Jinji Lake Health Talents Program of Suzhou Industrial Park (SZYQTD202102), Suzhou Key Discipline for Medicine (SZXK202129), Demonstration of Scientific and Technological Innovation Project (SKY2021002), Suzhou Dedicated Project on Diagnosis and Treatment Technology of Major Diseases (LCZX202132), Research on Collaborative Innovation of Medical Engineering Combination (SZM2021014, SZM2022003), Suzhou Key Laboratory of Diagnosis and Treatment of Panvascular Diseases (SZS2023021), and Gusu Talent Program (GSWS2022119). The funders had no roles in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.