Background
Coronary heart disease severely threatens the health of individuals worldwide with a rising incidence and a leading cause of mortality. Revascularization with percutaneous coronary intervention (PCI) is a well-established and effective therapeutic strategy for advanced coronary heart disease, particularly following the introduction of drug-eluting stents (DES) [
1]. While the incidence of in-stent restenosis (ISR) and target lesion revascularization is substantially reduced by DES compared with the bare-metal stents (BMS), they are not eliminated. ISR after DES implantation with an incidence of 3–20% remains a pervasive clinical problem which should not be neglected [
2‐
4]. Hence, tools for the identification of individual patients at higher risk for ISR with stent implantation are especially needed.
Although several prior studies have analyzed potential predictive factors associated with a high incidence of ISR based on patient and procedure-related factors, there are still some limitations that confine their clinical application. Prediction model for ISR is yet to be fully developed and validated [
5‐
7]. Of note, most studies have only focused on comprehensively identifying the predictors of ISR or developing prediction models without an individual risk prediction tool, while the simplicity and ease of use for the clinicians and patients were not well considered [
7‐
9]. Thus, a more solid, well-validated and easy-to-use clinical ISR prediction model for all patients with stent implantation, especially for clinical decision in primary prevention, is urgently needed for accurate prognostication of future events.
The nomogram-based predicting model has been widely implemented in clinical studies. Featured by the advantage of visualization, a well-developed nomogram based on statistical regression models is a cogent tool to make clinical decision for clinicians and to assess straightforward the probability of disease for individual patients without complex formula, thus could benefit both doctors and patients.
Based on these premises, this study aimed to identify factors correlated to the risk of ISR for individual patients undergoing PCI, using data from an observational single-center registry study. These factors were used to develop and validate a nomogram-based clinical prediction model. This model could help clinicians discern high-risk ISR patients, optimize treatment strategy, and improve prognosis of these patients.
Discussion
Stent implantation is an effective therapy for coronary artery disease. However, ISR has always been one of the most common complications, even in the era of DES. Thus, establishing a predication model of ISR for risk-tailored screening and preventive measure implementation may be pivotal to improve clinical outcomes of patients undergoing PCI. In the present study, five clinical and angiographic characteristics including the history of prior PCI, glycemia, stents in LAD, the type of stent, and absence of clopidogrel were found to independently predict ISR in DES recipients. Moreover, the nomogram prediction model based on these independent factors was constructed and validated, which could provide clinicians with an easy-to-use clinical tool for individualized assessments of patients with high-risk of ISR. Notably, visually and prospectively informing patients of the benefits of risk factor control may improve the patient's understanding of treatment and compliance of therapies, which has great significance for reducing the risk of ISR after stent implantation.
Patients undergoing PCI are exposed to an increased risk of ischemic and bleeding events. Risk models to predict the risk of bleeding events, ischemic events and mortality after PCI have been widely reported. The PRAISE score was a machine learning-based model for the prediction of all-cause death, myocardial infarction, and major bleeding in patients after an acute coronary syndrome [
17]. The leading predictors of both myocardial infarction and major bleeding in this model was hemoglobin level, age, left ventricular ejection fraction (LVEF), and estimated glomerular filtration rate (EGFR). The ARC-HBR score is a criterion used for identifying PCI patients with high risk of bleeding events. Risk factors in this criterion includes age, oral anticoagulation, chronic kidney disease, hemoglobin, prior bleeding, thrombocytopenia, chronic bleeding diathesis, liver cirrhosis with portal hypertension, use of NSAIDs or Steroids, active malignancy, prior stroke, and surgery or trauma within 30 days [
18]. The PRECISE-DAPT score was also a model to predict bleeding risks for patients after coronary stenting [
19], which was based on five factors: integrating age, hemoglobin, white-blood-cell count, creatinine clearance, and prior bleeding. Age, hemoglobin and renal function are the common risk factors in these models. However, these factors were not found in our models for the prediction of ISR in DES, revealing that ISR was associated with different mechanisms, compared with bleeding events, myocardial infarction or death.
Although the exact mechanism of ISR in DES is unclear and probably multifactorial, it is currently accepted that factors including biological, mechanical, and technical issues can facilitate the adverse neointimal hyperplasia and contribute to ISR after stent implantation [
2]. Several studies have attempted to identify the independent predictors of ISR. Stolker et al. [
20] developed a risk model for predicting restenosis of DES from the EVENT registry and identified age < 60, prior PCI, unprotected left main PCI, saphenous vein graft PCI, minimum stent diameter ≤ 2.5 mm, and total stent length ≥ 40 mm as the predictors of ISR. In another study evaluating the incidence and predictors of target vessel revascularization among 27,107 patients undergoing implantation of BMS or DES, significant predictors of restenosis included prior PCI, emergency or salvage PCI, prior coronary artery bypass grafting (CABG), peripheral vascular disease, diabetes mellitus (DM), and angiographic characteristics [
8]. Lately, Zheng et al. [
21] analyzed 944 stented lesions from 394 patients with 2nd-generation DES implantation. Factors including DM, previous PCI, postprocedural diameter stenosis and CRP levels were found to independently predict target lesion revascularization. A large patient data pooled analysis from 6 prospective and randomized trials, which included 10,072 patients undergoing DES implantation, suggested that vessel diameter, DM, prior CABG, and prior PCI were patient- and lesion-related predictors of target lesion failure [
22].
Individual predictors likely vary between different studies on account of difference in the complexity of patients and candidate variables. However, the overlap in predictive factors, such as prior PCI, prior CABG, and DM, are strongly interlinked with accelerated ISR and repeated target lesion revascularization. Similar to those found in previous studies identifying predictors of ISR, our study also indicates that patient populations with prior PCI and history of DM are prone to ISR.
A history of PCI, which was a consistent and independent predictor of ISR, is closely related to the primary risk factors of atherosclerosis and represents the overall risk of severe coronary lesions requiring further intervention. It is also reported that repeated revascularization is more likely to occur for culprit lesions at a site of previous restenosis [
9]. As for DM, patients have a higher risk of developing ISR due to the higher inflammatory response, endothelial dysfunction, platelet hyperreactivity and more aggressive neointimal hyperplasia accompanied by elevated plasma glucose levels [
23,
24]. Generally, DM is associated with complicated coronary artery disease characterized by multivessel lesions and diffuse lesions in small vessels, which requires multiple stents with small diameter during PCI. Thus, it can explain at least partly why variables like the total length and minimum diameter of stents were not included in the prediction model in this study. In addition, a noteworthy finding of our study is that uncontrolled glycemia in patients with DM has more predictive value for ISR rather than DM itself. Hyperglycemia itself is reported to be the major cause of diabetic angiopathy [
25]. High glucose could increase the expression of monocyte chemoattractant protein-1 (MCP1) and vascular cell adhesion molecule-1 (VCAM-1), thus enhance the monocyte-endothelial cell interaction and promote the atherogenic process and endothelial dysfunction [
25]. Endothelial dysfunction is associated with smooth muscle cell proliferation after vascular injury, e.g., restenosis after PCI [
26]. Moreover, the formations of advanced glycation end-products (AGE) and AGE-modified low-density lipoproteins (AGE-LDL), which are associated with hyperglycemia, can directly affect the cells of the vascular wall, through mechanisms including the upregulation of MCP1 in both vascular smooth muscle cells and endothelial cells [
27,
28]. In addition, Marfella et al. [
29] have found that hyperglycemia could increase oxidative damage and may reduce the regenerative potential of ischemic myocardium, while tight glycemic control showed protective effects.
Although several factors associated with ISR in our study are concordant with previous findings, some key predictors including stents in LAD, type of stent, and absence of clopidogrel have not been reported consistently in literature. Most studies suggested that coronary artery intervention restenosis was more frequent for lesions in the LAD than other native coronary arteries, confirming that the LAD may be another potential risk factor for ISR [
30‐
32]. However, different opinions have also been proposed. To the contrary, lesions located in the LAD were reported to have a decreased restenosis rate [
33,
34]. In fact, in the present study the results observed after stent implantation for LAD lesions were very similar to those observed in most studies and we believe that stents located in the LAD were associated with an elevated incidence of ISR. For lesions in LAD and other complicated lesions, intravascular ultrasound (IVUS), optical coherence tomography (OCT) and other coronary imaging techniques are recommended for optimizing the treatment strategy. In addition, our findings figured out that sirolimus-eluting stent (SES) was associated with a lower risk of ISR than paclitaxel-eluting stent (PES). Sirolimus and its analogs have a cytostatic effect on coronary artery endothelial cells, while paclitaxel has a cytotoxic effect. Several studies have also indicated that use of SES has a less late luminal loss [
7,
35,
36] and a lower rate of late stent thrombosis [
37], as compared with use of PES, suggesting a better performance of SES in reducing restenosis. Finally, drugs and polymers of DES can inhibit the excessive neointimal hyperplasia. However, it delays the repair of endothelial cells. Therefore, antiplatelet drugs are still the cornerstone in the treatment of coronary heart disease, especially after PCI. Gianluca et al. [
38] investigated the clinical outcome of patients undergoing PCI for ISR with short (6 months) or long (24 months) dual antiplatelet therapy (DAPT). The main findings of this study were that patients receiving revascularization for ISR may benefit from long-term administration of aspirin plus clopidogrel. Similarly, our study showed that the absence of clopidogrel increased the risk of ISR after PCI, suggesting the benefit of appropriately prolonged DAPT duration for patients with high risk of ISR after DES implantation.
Predictor identification and risk assessment are essential and important to an effective medical decision making for preventing restenosis. Multifactorial intervention has shown remarkable benefits on the risks of major cardiovascular events and mortality in patients with diabetic kidney disease [
39]. It is expectable that treatments targeting related risk factors will help to lower the incidence of ISR in patients receiving PCI. However, the levels of prognostic utility of prediction models of ISR in prior studies remained less than totally satisfying with c-statistic below 0.7 [
8,
9,
20,
40]. In the present study, the best c-statistic derived from the nomogram model in the development set was 0.706 and was confirmed to be 0.662 in the validation set as well, suggesting that the distinct predictors improved the overall discrimination of the models. Moreover, calibration plots and decision curve analysis for the nomogram-based predication model were also performed well, making our findings more convincing and providing broad applicability in clinical practice.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.