Introduction

The ‘no-reflow’ phenomenon refers to the failure to achieve myocardial perfusion despite the successful restoration of epicardial coronary artery patency after percutaneous coronary intervention (PCI) [1, 2]. This complication, particularly in the setting of acute myocardial infarction (AMI), is associated with adverse clinical outcomes, including left ventricular remodeling, arrhythmia, and increased mortality. It is also relatively common in AMI, with an incidence rate exceeding 20% [3,4,5]. No-reflow has been extensively studied, and well-defined predictors and management strategies have been established [6,7,8,9,10,11,12,13].

In contrast, no-reflow in patients undergoing elective PCI for non-AMI is less common, with reported rates of approximately 1–2% [2, 3, 6, 7]. Owing to its low incidence, limited attention has been paid to its pathophysiology, predictors, and clinical relevance in this setting. As elective PCI comprises a large proportion of contemporary interventional practices, addressing this knowledge gap is clinically important.

This study aimed to determine the incidence and identify the clinical and procedural predictors of no-reflow in patients undergoing elective PCI for non-AMI using a large, single-center, prospective registry. In addition, we evaluated the utility of machine learning (ML) models for risk prediction of no-reflow.

Methods

Study Design and Population

This was a prospective, single-center cohort study conducted at a tertiary academic hospital in South Korea using data from a PCI registry (ClinicalTrials.gov Identifier: NCT02038127). The registry consecutively enrolled patients who underwent PCI between January 2013 and December 2024. Of the 10,801 patients included in the registry, 3,904 with non-AMI who underwent elective PCI were analyzed after applying the exclusion criteria, as illustrated in Fig. 1.

Fig. 1
Fig. 1
Full size image

Study Flow Diagram. AMI, acute myocardial infarction; MI, myocardial infarction; NSTEMI, Non-ST elevation myocardial infarction; PCI, percutaneous coronary intervention; SCD, sudden cardiac death; STEMI, ST-elevation myocardial infarction

The study protocol was approved by the local Institutional Review Board (approval number CR320095) and conducted in accordance with the Declaration of Helsinki, revised in 2013. Written informed consent was obtained from all the participants.

Definitions of clinical diagnosis

The definition of AMI was based on the acute myocardial injury, a rise and/or fall of cardiac troponin values with at least 1 value above the 99th percentile upper reference limit, and at least 1 of the following: presentation of chest pain; consecutive electrocardiogram changes suggesting myocardial infarction (MI); development of pathologic Q waves; imaging evidence about an ischemic etiology; identification of a coronary thrombus by angiography [14]. Hypertension was defined as a prior history of hypertension, current use of antihypertensive medication, or elevated blood pressure during the index hospitalization (systolic blood pressure ≥ 140 mmHg and/or diastolic blood pressure ≥ 90 mmHg). Diabetes mellitus was defined as a fasting plasma glucose level ≥ 126 mg/dL, a glycated hemoglobin (HbA1c) level ≥ 6.5%, or the current use of glucose-lowering medication, and.

chronic kidney disease (CKD) was defined as an estimated glomerular filtration rate (eGFR) less than 60 mL/min/1.73 m2 at admission.

Definition of no-reflow and grouping

No-reflow was defined as a transient or persistent thrombolysis in myocardial infarction flow grade of < 3 following PCI in the absence of mechanical obstruction, such as dissection, residual thrombus, or coronary spasm. Angiographic assessments were performed by an operating physician at the time of the procedure. At the patient level, no-reflow was considered to be present if it occurred in any treated vessel during multivessel PCI. At the lesion level, no-reflow was considered to be present if it occurred at any stage of treatment, including predilation, stent deployment, drug-coated balloon inflation, or adjunctive ballooning. All patients and lesions were categorized based on the presence or absence of no-reflow.

PCI and medical treatment

PCI was performed using standard techniques. The selection of vascular access, use of predilation or direct device delivery, administration of glycoprotein IIb/IIIa inhibitors (GPIs), application of intravascular imaging, and need for pressure wire assessment were left to the operator’s discretion. The techniques applied for bifurcation or chronic total occlusion lesions, as well as stent and device selection, were also determined by the treating physician. Staged PCI procedures were permitted and were not considered separate revascularization procedures.

All patients received a loading dose of aspirin (300 mg) and a P2Y12 inhibitor, including clopidogrel (600 mg), prasugrel (60 mg), or ticagrelor (180 mg). Dual antiplatelet therapy with aspirin (100 mg) and a maintenance dose of the selected P2Y12 inhibitor was recommended for at least 12 months following the procedure. Guideline-directed medical therapies, including beta-blockers, renin–angiotensin–aldosterone system inhibitors, and statins, were prescribed to all eligible patients.

Data collection

Baseline clinical, laboratory, and electrocardiographic data were collected before PCI. Echocardiographic data were obtained before or after the procedure, depending on the clinical schedule. Angiographic and procedural characteristics were assessed using diagnostic and interventional coronary angiography. For each treated lesion, trained research personnel recorded information on device selection, lesion morphology, and procedural complications. Data quality was ensured through routine source verification, electronic audit trials, and cross-validation of institutional medical records.

Statistical analysis

Continuous variables are summarized as the mean with standard deviation or median with interquartile range and were compared using Student’s t-test or the Mann–Whitney U test, as appropriate. Categorical variables are presented as counts with percentages and were compared using the chi-square or Fisher’s exact test.

Multivariable logistic regression was used to identify the independent predictors of no-reflow. Variables were selected using the least absolute shrinkage and selection operator (LASSO) method. Cluster-robust standard errors were applied using the patient identifier as a clustering variable to account for correlations among lesions in the same patient. Odds ratios (ORs) with 95% confidence intervals (CIs) are reported. The optimal penalty parameter (lambda) was determined by 10-fold cross-validation, and variables with non-zero coefficients at the lambda corresponding to the minimum cross-validated error were selected for inclusion in the subsequent multivariable logistic regression model. To assess the stability and interpretability of the multivariable model coefficients, multicollinearity among covariates included in the final logistic regression model was evaluated using variance inflation factors (VIFs). Because VIFs are not directly available for logistic regression, they were calculated from an equivalent linear regression model including the same set of covariates. All VIF values were below commonly accepted thresholds (all < 5), indicating no significant multicollinearity.

Four ML models were developed to predict no-reflow at the lesion level: logistic regression, random forest, extreme gradient boosting (XGBoost), and Light Gradient Boosting Machine (LightGBM). The dataset was randomly split into training and testing sets in an 80:20 ratio, with stratification based on the target outcome. Class imbalance was addressed using model-specific weighting strategies: class weighting for logistic regression and random forest, scale balancing for XGBoost, and internal imbalance handling for LightGBM. The synthetic minority oversampling technique (SMOTE) and Adaptive Synthetic Sampling (ADASYN) were also applied.

All analyses were performed using SPSS (version 27.0; IBM Corp., USA) and Python version 3.10 (Python Software Foundation, USA). Statistical significance was set at p < 0.05.

Results

Incidence of no-reflow

Among the 3,904 patients who underwent elective PCI for non-AMI, angiographic no-reflow occurred in 94 (2.4%). In context, no-reflow was more frequent in the excluded subgroups: 16.3% in patients with ST-elevation myocardial infarction (STEMI), 12.1% in those having recent myocardial infarction, and 7.7% in those with non–ST elevation myocardial infarction (NSTEMI; Fig. 1).

Baseline characteristics of the study population

The baseline clinical characteristics of the study population according to no-reflow status are summarized in Table 1. There were no significant differences in age, sex, cardiovascular risk factors, or laboratory parameters between groups. However, atrial fibrillation was more frequent in the no-reflow group than in the group without no-reflow (9.6% vs. 4.4%, p = 0.031). T-wave inversion on baseline electrocardiography was more common in patients with no-reflow than in those without (28.7% vs. 18.8%; p = 0.022). Other variables, including blood pressure, renal function, hemoglobin level, and left ventricular ejection fraction, showed no significant differences between the groups.

Table 1 Baseline characteristics of the study population
Full size table

Angiographic and procedural characteristics

The angiographic and procedural characteristics are summarized in Table 2. A total of 5,228 lesions were analyzed, including 96 no-reflow lesions, 51 lesions without no-reflow from patients who experienced no-reflow elsewhere, and 5,081 lesions from patients without no-reflow.

Table 2 Angiographic and procedural characteristics
Full size table

Compared to patients without no-reflow, those who experienced no-reflow had a more complex coronary anatomy. The prevalence of multivessel disease was higher in the no-reflow group (78.7% vs. 63.4%, p = 0.002), and the median Synergy Between PCI With Taxus and Cardiac Surgery (SYNTAX) score was also higher (16.5 vs. 14.0, p = 0.006). Preprocedural diameter stenosis was more severe, and the target lesion length was longer in patients with no-reflow compared with those without no-reflow.

Procedurally, the no-reflow group received longer stents (36.0 mm vs. 28.0 mm, p < 0.001), had a higher frequency of device-to-vessel ratios > 1.1 (32.3% vs. 18.6%, p < 0.003), and showed significantly lower procedural success rates (69.8% vs. 91.1%, p < 0.001). Reference vessel diameter was determined using quantitative coronary angiography (QCA). The device-to-vessel ratio was calculated as the ratio of the maximal device diameter─estimated from each devices’ compliance chart at the highest inflation pressure used during PCI─to the QCA-derived reference vessel diameter.

Despite the exclusion of patients with thrombotic lesions from the study population, thrombectomy was performed more frequently in the no-reflow group than in the group without no-reflow (12.5% vs. 0.2%, p < 0.001). The overall rate of procedural complications, excluding no-reflow, was also significantly higher in the no-reflow group.

In-hospital complications and 1-year clinical outcomes

The in-hospital complications and clinical outcomes are summarized in Table 3. The patients who experienced no-reflow had a significantly higher rate of periprocedural complications than those who did not; the incidence of cardiogenic shock was markedly elevated (26.6% vs. 1.3%, p < 0.001), and vasopressor or inotrope support was required more frequently (24.5% vs. 1.3%, p < 0.001). Atropine administration (11.7% vs. 0.8%, p < 0.001) and atrioventricular block (2.1% vs. 0.2%, p = 0.027) were more common. The use of GPIs was substantially higher in the no-reflow group (19.1% vs. 0.6%; p < 0.001). Periprocedural myocardial injury was more pronounced in the no-reflow group, with levels of creatine kinase-myocardial band elevated to ≥ 5 times the baseline (44.7% vs. 15.1%, p < 0.001) and troponin I elevated to ≥ 5 times the baseline (86.2% vs. 60.4%, p < 0.001).

Table 3 In-hospital complications and 1-year clinical outcomes
Full size table

The in-hospital mortality rate tended to be higher in the no-reflow group (1.1% vs. 0.1%, p = 0.071). However, there were no significant differences between the groups in one-year patient-oriented composite outcomes (POCO), including all-cause death, myocardial infarction, repeat revascularization, or device-oriented composite outcomes (DOCO), defined as cardiovascular death, target vessel myocardial infarction, or target lesion revascularization. The Kaplan–Meier survival curves for both endpoints are shown in Fig. 2.

Fig. 2
Fig. 2
Full size image

Kaplan–Meier Curves for One-Year Clinical Outcomes According to No-Reflow Status. Kaplan–Meier survival curves for (A) patient-oriented composite outcome (POCO) and (B) device-oriented composite outcome (DOCO) at 1 year, stratified by the presence or absence of reflow. POCO, composite of all-cause death, myocardial infarction, or repeat revascularization; DOCO, composite of cardiovascular death, target vessel myocardial infarction, or target lesion revascularization

Predictors of no-reflow

The predictors of the no-reflow phenomenon are summarized in Table 4 and illustrated in Fig. 3. In multivariable logistic regression using cluster-robust standard errors, atrial fibrillation (OR 2.20, 95% CI 1.08–4.49, p = 0.031) and male sex (OR 2.08, 95% CI 1.15–3.74, p = 0.015) were independently associated with increased risk of no-reflow. Among procedural factors, a device-to-vessel ratio > 1.1 was the strongest predictor (OR 14.42, 95% CI 3.71–56.13, p < 0.001).

Table 4 Predictors for no-reflow
Full size table
Fig. 3
Fig. 3
Full size image

Predictors of No-reflow. A multivariable logistic regression analysis with cluster-robust standard errors was performed. Significant predictors included male sex, atrial fibrillation, and a device-to-vessel ratio >1.1. Current or former smoking status was inversely associated with no-reflow. Red markers indicate statistically significant associations (p < 0.05). AF, atrial flutter; A.fib, atrial fibrillation; CI, confidence interval; OR, odds ratio; PCI, percutaneous coronary intervention

Interestingly, current or former smoking was inversely associated with no-reflow (OR 0.53, 95% CI 0.32–0.89, p = 0.017). Other variables, such as diffuse lesion morphology or T-wave inversion, showed associations in the univariate analysis but were not retained in the final model.

Models for Predicting No-reflow

The model performance for predicting no-reflow is summarized in Fig. 4. Logistic regression showed the highest sensitivity (0.47), but poor precision (0.03), with an area under the curve (AUC) of 0.64. XGBoost demonstrated a better balance of precision (0.20) and discrimination (AUC 0.59), but lower sensitivity (0.05). Random forest and LightGBM yielded lower AUCs (0.54 and 0.48, respectively) and failed to detect any no-reflow events (recall 0.00). The application of SMOTE and ADASYN did not improve the predictive performance.

Fig. 4
Fig. 4
Full size image

ROC Curves for Machine Learning Models Predicting No-reflow. Panel A shows baseline model performance (logistic regression AUC = 0.64, XGBoost = 0.59, random forest = 0.54, LightGBM = 0.48). Panels B and C illustrate the effects of the oversampling techniques, SMOTE and ADASYN, respectively, on model performance. Both approaches failed to improve the discriminative ability, with all models showing AUC values close to 0.4. ADASYN, Adaptive Synthetic Sampling; AUC, area under the curve; LightGBM, Light Gradient Boosting Machine; ROC, receiver operating characteristic; SMOTE, synthetic minority oversampling technique; XGBoost, extreme gradient boosting

Discussion

In this single-center cohort of patients undergoing elective PCI for non-AMI, angiographic no-reflow occurred in 2.4% of cases. Despite its low frequency, no-reflow is associated with procedural complications, myocardial injury, and hemodynamic instability. One-year clinical outcomes were comparable between the groups, possibly because of prompt recognition and appropriate management during the procedure. The independent predictors of no-reflow included atrial fibrillation, male sex, and device oversizing, whereas current or former smoking status was inversely associated. The ML models showed limited predictive performance, underscoring the difficulty of modeling rare procedural events using standard clinical and angiographic variables.

Previous studies have evaluated the prognostic impact of no-reflow across a broad spectrum of patients who underwent PCI. In a large, multicenter, Australian registry of over 18,000 procedures, the overall incidence of no-reflow was 4%. However, no-reflow occurred far more frequently in patients with AMI: 7.0% in STEMI and 3.9% in NSTEMI compared to 1.8% in stable angina and 2.4% in unstable angina. Notably, more than 80% of no-reflow events occurred in patients with AMI. No-reflow was associated with increased 5-year all-cause mortality, although this relationship was likely driven predominantly by the high-risk AMI population, which also had a greater burden of shock, out-of-hospital cardiac arrest, and complex lesions [15]. Similarly, a Korean multicenter AMI registry reported that no-reflow was linked to higher 3-year all-cause mortality [16]. In our current study of patients with non-AMI, no-reflow was associated with acute procedural complications though not with worse long-term outcomes. These findings suggest that the clinical context and acuity of presentation may modulate both the incidence and prognostic impact of no-reflow.

While no-reflow in AMI is typically driven by thrombus burden, plaque rupture, and prolonged ischemia [10, 17,18,19,20], patients with non-AMI experience ischemia because of progressive atherosclerotic narrowing or microvascular dysfunction [21, 22]. Consequently, the procedural environment is less inflammatory and thrombotic, and the mechanisms of no-reflow are likely to be distinct. It should also be noted that the clinical classification of stable versus unstable angina is inherently overlapping in contemporary practice, which may limit the ability to detect presentation-specific differences in the predictors of angiographic no-reflow. In this setting, device-related mechanical factors, such as oversizing or inadequate lesion preparation, may play a more central role in distal embolization and microvascular obstruction.

Our findings support this pathophysiological distinction. A device-to-vessel ratio greater than 1.1 was the strongest independent predictor of no-reflow, emphasizing the importance of procedural precision. Additionally, atrial fibrillation and male sex were associated with increased risk, whereas current or former smoking showed an inverse association.

Atrial fibrillation is a common arrhythmia among patients undergoing PCI, with a reported prevalence of 5–10% [23, 24], and is associated with increased in-hospital mortality, heart failure, and cardiogenic shock [24, 25]. Proposed mechanisms include impaired microvascular perfusion, endothelial dysfunction, and a prothrombotic state, all of which may predispose patients to microcirculatory obstruction, even in the absence of an overt thrombus [26]. Our findings extend the relevance of atrial fibrillation as a no-reflow risk factor in non-AMI settings.

Interestingly, in contrast to some previous studies reporting an increased risk in women [27, 28], female sex was independently associated with a lower risk of no-reflow. This discrepancy may reflect sex-related differences in vessel size, endothelial function, or the procedural approach. Similarly, the inverse association between smoking and no-reflow should be interpreted cautiously, as it may reflect residual confounding rather than a true biological effect, despite prior hypotheses suggesting enhanced clopidogrel metabolism as a potential explanation for the smoker’s paradox [29].

Taken together, our results suggest that no-reflow remains a clinically relevant complication, even in elective PCI for patients with non-AMI. Although a device-to-vessel ratio > 1.1 showed the strongest association with no-reflow in our analysis, this finding should be interpreted as indicative of a potential procedural risk factor rather than a precise estimate of effect size. The observed association supports the importance of careful lesion assessment and appropriate device sizing, potentially aided by intravascular imaging, to mitigate the risk of distal embolization. Importantly, the identification of potentially modifiable procedural factors may provide insights into strategies for reducing no-reflow, although causal inferences cannot be drawn from the present analysis. In addition, the inclusion of clinical and electrocardiographic markers, such as atrial fibrillation, in risk assessment models may help refine preprocedural planning and patient counseling in selected patients.

Despite the use of multiple ML algorithms and class-balancing techniques, the model performance in predicting no-reflow was suboptimal. This limitation is likely attributable to the rarity of no-reflow events, which poses a fundamental challenge for supervised learning models. Moreover, the input features were based primarily on conventional clinical and angiographic parameters, which may lack the granularity required to capture the complex pathophysiology of no-reflow. These findings suggest that traditional variables alone may be insufficient for constructing robust prediction models in this context. Future efforts should focus on incorporating more advanced data sources, such as intravascular imaging, physiological indices, or circulating biomarkers, along with larger, multi-institutional datasets to improve model generalizability and discriminative power.

Limitations

This study had several limitations. First, it was a single-center observational study, and unmeasured confounders may have influenced the results despite multivariable adjustment. Second, no-reflow was defined solely by angiographic assessment (TIMI flow grade < 3) without objective microvascular indices or independent core laboratory adjudication, introducing misclassification, particularly in non–AMI populations. However, routine assessment of microvascular perfusion in all non-AMI patients is not cost-effective in real-world practice, and angiographic evaluation therefore reflects contemporary clinical decision-making. Third, lesion characteristics were assessed using angiography rather than intravascular imaging, which may have limited the accuracy of the lesion complexity assessment. Although numerous intravascular imaging predictors, such as the presence of a necrotic core, thin-cap fibroatheroma, or attenuated plaque, have been identified in previous studies [30, 31], our study exclusively focused on angiographic findings, given the limited real-world utilization of intracoronary imaging. Fourth, detailed data on pre-procedural statin loading or statin intensity prior to PCI was not available in the present dataset. Given that high-dose statin loading has been suggested to reduce the risk of no-reflow, the potential impact of differences in statin intensity between groups could not be evaluated and cannot be excluded. Fifth, the limited performance of ML models in this study should be interpreted in the context of a low event rate, resulting in a restricted number of events per variable (EPV). In such settings, complex ML algorithms are prone to overfitting noise rather than identifying stable predictive signals, particularly when the signal-to-noise ratio is low, as is typical for rare procedural complications such as angiographic no-reflow. Despite the use of standard regularization and cross-validation techniques, ML did not provide incremental predictive value over conventional regression models. This finding does not indicate an inherent limitation of ML approaches, but rather reflects the constraints imposed by data structure and event frequency. In low-event scenarios, parsimonious, hypothesis-driven models may offer more robust and interpretable risk estimation. Finally, the unexpected associations observed for sex and smoking status should be interpreted cautiously. The inverse association with smoking likely reflects residual confounding rather than a true protective effect, as smoking status may serve as a surrogate for differences in age, comorbidity burden, medication use, or lesion phenotype that were not fully captured in this dataset. Similarly, sex-related differences in vessel characteristics and procedural strategies may influence susceptibility to angiographically defined no-reflow independent of intrinsic biological risk. Accordingly, these variables should be regarded as markers of underlying clinical and procedural profiles, and external validation in independent cohorts is warranted. Despite these limitations, our study provides novel insights into the predictors of no-reflow in patients with non-AMI undergoing elective PCI and highlights the need for procedural optimization and individualized risk assessment in this population.

Conclusions

In this single-center cohort of patients undergoing elective PCI, several clinical and procedural characteristics—including atrial fibrillation, male sex, and device oversizing—were statistically associated with the occurrence of the angiographic no-reflow phenomenon. These associations should be interpreted cautiously, given the low event rate, the angiography-based definition, potential residual confounding, and the limited discriminative performance of both logistic regression and exploratory ML analyses. Rather than indicating practice-changing risk factors, the identified associations should be viewed as descriptive and hypothesis-generating, characterizing procedural contexts in which angiographic no-reflow is more frequently observed. Consistent with this interpretation, angiographic no-reflow was not associated with adverse short-term or one-year clinical outcomes. Further prospective studies using dedicated microvascular assessment are required to elucidate the mechanisms, prognostic implications, and potential preventive strategies for no-reflow phenomenon during elective PCI.