A Nomogram for Predicting In-Stent Restenosis Risk in Patients Undergoing Percutaneous Coronary Intervention: A Population-Based Analysis

Introduction

Coronary heart disease (CHD), with high morbidity and high mortality rate, is still a serious public health concern around the world. PCI is fast becoming a key instrument in revascularization for patients with CHD, as well as an important technology in the management of CHD patients.¹ Although the clinical application of coronary stents brought about a dramatic improvement in patients’ clinical and procedural outcomes, the mid-and long-term outcome of stent implantation remains significantly hampered by the risk of developing ISR with a prevalence rate of 3–20% over time^1,2 Predictive models have the advantage of formally combining risk factors to allow more accurate risk estimation. And it is essential to establish a model to predict ISR in patients with CAD and drug-eluting stents (DESs) implantation.

The risk factors for ISR after PCI were systematically summarized. The preoperative factors comprised the following: the morphological characteristics of the diseased vessel, the location of the lesion, the degree of stenosis, and part of blood biochemical indicators which are associated with inflammatory responses and lipid metabolism.^3–8 Among these, the levels of monocyte and LDL-C are considered to be critical factors related to inflammatory response and to lipid metabolism,^9–11 respectively.

Although a few previous studies have analyzed potential predictors related to the high incidence rate of ISR and established a relevant nomogram for ISR in patients undergoing PCI, there are still limitations to the predictive model. As a starting point, new factors for inflammatory response and lipid metabolism have emerged in recent years. Such as the neutrophil/lymphocyte ratio^12,13, which reflects the body’s levels of oxidative stress and inflammation, as well as residual cholesterol, which is a more accurate indicator of the body’s lipid metabolism than LDL cholesterol.^14,15 Secondly, the majority of prediction models lacked a quantitative predictor of coronary lesions before PCI, such as the GS score system, a technique based on the artery morphology, coronary anatomy, and severity of stenosis in lesions.

Currently, a new preoperative model based on preoperative blood biochemical parameters for PCI, a technique for assessment for the severity of CAD, and procedural characteristics are scarce to evaluate the probability of ISR. The aim of this study was to analyzed post-PCI ISR patients in preoperative blood biochemical parameters for PCI, GS scores,¹⁶ and procedural characteristics. This work will generate fresh insight into developing a preoperative risk factor nomogram that may help clinicians discern high-risk ISR patients, optimize treatment strategies.

Materials and Methods

According to the Declaration of Helsinki, this study was approved by the Ethics Committee of the Central Hospital of Enshi Autonomous Prefecture. Due to the retrospective nature of this study, patient consent was waived for the evaluation of their medical information. This study was an analysis of an observational cohort study conducted from January 2017 to December 2020 at Enshi Central Hospital, China. A total of 1015 CAD patients undergoing PCI with DES were enrolled. All patients took statins and anti-platelet aggregation drugs regularly after surgery, and all received 6–24 months of follow-up coronary angiography. Patients were excluded if they (1) had a history of coronary artery bypass grafting, heart failure (cardiac function class more than 4), (2) active or acute inflammatory diseases, (3) had a liver failure or renal failure, (4) had evidence of active infection, such as fever, cough, or diarrhea, and (5) were missing clinical and angiographic data (Figure 1). The main outcome measure was ISR, which is defined as ≥50% luminal narrowing at follow-up angiography.

Figure 1 The study design and the selection procession of CAD patients. Abbreviation: receiver operator characteristic.

Demographic information, biochemical parameters, clinical, and angiographic characteristics were collected. Demographic information and clinical characteristics included age, gender, chronic obstructive pulmonary disease, diabetes, stroke, smoking, and patient medication history (ACEI, diuretic). Biochemical parameters included platelet parameters of platelet distribution width (PDW), leukocytes, monocytes, mean platelet volume (MPV), glucose (GLU), neutrophils (N), lymphocyte, monocyte, hemoglobin (Hb), platelet, procalcitonin (PCT), total cholesterol (TC), triglycerides (TG), high-density lipoprotein cholesterol (HDL-C), low-density lipoprotein cholesterol (LDL-C), the width of red blood cell volume distribution (RDW), alanine transaminase (ALT), creatinine (Cr) and left ventricular systolic function (EF%). The specific data concerning the angiography information included stent numbers, GS score, lesion location (left main coronary artery, left circumflex artery, left anterior descending branch, right coronary artery and others. RC = TC- LDL-C- HDL-C. The neutrophil/lymphocyte ratio = the neutrophil count /the lymphocyte count. The GS score was calculated using the scoring schema defined by Gensini et al.¹⁶

R software version 3.6.3 was used for statistical analysis, and two-tailed analysis with P. The Student’s t-test was used to detect differences between continuous variables with a normal distribution. The chi-square test or Fisher’s exact test was used to compare categorical variables. In the cases of skewed distribution, data were expressed as IQR and compared using the Mann–Whitney U-test. LASSO-penalized regression analysis, which is capable of estimating parameters in high-dimensional regression, was used to select ISR predictors with the R package Glmnet. The Hosmer–Lemeshow 2 statistic, calibration curve, and 1000-fold bootstrap were used to test the prediction model.

We assessed the nomogram model performance in terms of discrimination, calibration plots, and the Hosmer–Lemeshow 2 statistic. The discrimination of the model has been validated through Area under the ROC, which implies the better accuracy of the nomogram. The diagnostic value of models whose AUC is between 0.7 and 0.8 and Hosmer–Lemeshow 2 > 0.05 is acceptable A very perfect agreement was observed in the calibration plot of our nomogram. Concerning its clinical usefulness, we performed decision curve analysis (DCA) to assess if clinical decisions taken based on this model would improve patient outcomes.

Results

According to the inclusion and exclusion criteria, 1015 patients were screened, and 477 patients with complete follow-up data were selected. A total of 477 patients was enrolled and divided into the ISR group (94) and the non-ISR group (383), according to the main outcome indicator (ISR). The baseline characteristics of the patients are shown in Table 1.

Table 1 The Baseline Characteristics of the Patients

Of CAD patients included in our study, 94 subjects (19.3%) had ISR. Most of the baseline characteristics between the two groups were similar, such as gender, underlying disease (hypertension, diabetes, stroke), medication history (diuretic, ACEI), smoking, lesion location, ALT, Hb, PDW, RDW, MPV, PLT, TG, TC, TG, HDL-C, monocyte, neutrophil/lymphocyte ratio and Cr. However, as Tables 1 shown, underlying disease (COPD), GS score, LDL-C, and RC showed significant differences between the groups (all P < 0.05).

The feature selection of the risk prediction model for ISR in patients after PCI is based on the 477 patients in the cohort (parameter selection is shown in Figure 2). 35 features were reduced to 5 potential predictors, and non-zero coefficients were used in the LASSO regression model. Those factors include: history of COPD, GS score of vascular assessments before PCI, monocyte, RC, and neutrophil/lymphocyte ratio. The Norman diagram based on the regression coefficient is shown in Figure 3.

Figure 2 Risk factors selecting using LASSO model. Notes: (Left panel) Optimal parameter (lambda) selection for the LASSO model was cross validated five the minimum criterion. Partial likelihood deviation (binomial deviation) curves versus log(lambda). Dotted vertical lines are drawn at the best values of 1SE (1-SE criterion) using the minimum criterion and the maximum criterion. (Right) LASSO coefficient profiles for 35 characteristics. The coefficient profiles were produced from log (lambda) sequences.The vertical lines are drawn on the value selected using five fold cross validation, where the best lambda resulted in non-zero coefficients for five features.

Figure 3 Nomogram to predict the probability of ISR in patients with stent implantation. Abbreviations: COPD, chronic obstructive pulmonary disease; GS scores, Gensini score; RC, remnant cholesterol. Notes: The nomogram included a medical history of COPD, GS score of vascular assessment before PCI, RC, monocyte, neutrophil/lymphocyte ratio.

The binary multivariate logistic regression comprised features selected from the lasso-penalized regression analysis (Table 2). The 5 independent risk factors for ISR were COPD, GS score, monocyte, RC, and neutrophil/lymphocyte ratio. The collinearity diagnostic test indicated that there was no significant collinearity between the independent variables in the regression model, and the variance inflation factors (VIFs) were 1.012, 1.006, 1.024, 1.037, and 1.008, respectively (all VIFs<10).

Table 2 Multivariable Logistic Regression Analysis of Predictors of ISR

Internal validation was performed with 1000 repeats, and the results were consistent. In addition, the AUC of the prediction model was 0.841 (Figure 4). A Hosmer–Lemeshow goodness-of-fit test was performed to evaluate this prediction model, yielding P = 0.609, and a calibration curve was also provided in Figure 5, confirming no divergence between anticipated and observed probability. Figure 6 shows the decision curve analysis for the ISR nomogram. The results showed that the nomogram might be used to forecast the likelihood of ISR in patients having PCI with high accuracy and a broader range of threshold probabilities, and it could have clinical implications.

Figure 4 ROC curves for validating the discrimination power of nomogram. Abbreviation: ROC, receiver operator characteristic.

Figure 5 Calibration plots of the nomogram for the probability of PCI patients with ISR. Notes: The x-axis represents prediction probability. The y-axis represents the actual probability. The diagonal dotted line represents a perfect prediction by an ideal model. The solid line represents the performance of the nomogram, of which a closer fit to the diagonal dotted line represents a better prediction.

Figure 6 Decision curve analysis for the ISR prediction nomogram. Notes: The ISR risk nomogram is represented by the dotted line. The thin solid line denotes the expectation that all patients are affected by ISR. The thick solid line denotes the assumption that there are no patients with ISR. When the threshold probability is >4.5, decision curve analysis demonstrates that adopting this ISR prediction nomogram has a net benefit.

Discussion

PCI is the mainstay of revascularization in patients with coronary artery disease, and in-stent restenosis remains a problem that greatly affects the long-term prognosis of post-PCI patients. The safety of stent implantation has significantly increased in recent years as a result of technological developments. ISR, on the other hand, remains one of the most significant issues. The ISR rate reached 19.7% (94/477) in our research, which was consistent with previous studies (3–20%).³ Life-threatening consequences can arise should the ISR continue to deteriorate without prompt recognition and treatment. Therefore, early identification of risk factors for ISR is crucial in preventing major postoperative complications. Herein, we firstly developed a nomogram utilizing the five preoperative predictors from the multivariate analysis: a medical history of COPD, monocytes, GS score, RC, and neutrophil/lymphocyte ratio.

Previous studies^5,6 summarized the predictors of ISR as follows: medication history (eg, clopidogrel), prior PCI, stent characteristics, and some indicators concerning inflammatory responses and lipid metabolism (eg, TC, LDL-C, CRP, monocyte).¹¹

Similar to the previous studies,¹¹ we also found that monocytes can be one of the independent risk factors for ISR. Nan et al reported that activated monocytes release large amounts of pro-inflammatory cytokines, and then cause vasoconstriction and non-specific recruitment, proliferation, and activation of other cells, including vascular smooth muscle cells in the vascular wall, which may account for the results in our study.

Another systemic factor noted in our study to be closely related to ISR after PCI is preoperative neutrophil/lymphocyte ratio, the index associated with inflammatory response and neointimal proliferation,^4,5,17,18 which was consistent with the previous research that neutrophil/lymphocyte is a strong inflammatory marker^19–22 and closely associated with Cardiovascular disease. In contrast to previous reports, we also discovered that RC had a higher predictive value for ISR than LDL-C. According to relevant literature,^7,23,24 an increase in fasting RC level increases the degree of coronary atherosclerosis stenosis and RC may be a better indicator of lipid metabolism in the body than LDL cholesterol.

Interestingly, we also found that patients with chronic obstructive pulmonary disease may have a higher risk of developing ISR. According to relevant literature,^9,25–28 this could be due to the fact that COPD and CVD have similar risk factors. Low-grade systemic inflammation is one of the primary processes that may be responsible for the systemic impacts on distant illnesses and the increased rate of comorbidity, particularly cardiovascular comorbidity, in COPD patients.²⁸ Regretfully, no relevant mechanistic investigations have been conducted to explain why a history of COPD can predict the occurrence of ISR, which will be the subject of our next research.

The most interesting finding was that GS score is an independent predictor of ISR. It is known that the morphological characteristics of the diseased vessel, the location of the lesion, and the degree of stenosis can make PCI procedures more difficult and increase the risk of preoperative vascular injury, leading to ISR. These factors have been shown to be independent risk factors for ISR,² but no quantitative assessments of diseased vessels have been considered in any published models and no studies have examined whether the GS score, which combines these characteristics, is a strong predictor of ISR. And we improved that GS score is a strong predictor of ISR in CAD patients for the first time.

And in previous studies,^8,29,30 the level of the prognostic utility of the ISR prediction model was still not entirely satisfactory, with a C-statistic below 0.7. We, therefore, developed a predictive model based on the GS score, a proxy for coronary lesion factors,^31,32 and other predictors for patients undergoing PCI in the Enshi region.

Nevertheless, the research still has several limitations. First of all, its validity is limited by the small sample size and the low number of events although our sample size had met the required the minimum sample size of building the model 0.80 is 140. Furthermore, this was a single-Center study with no external validation. Despite an internal validation, the prediction model’s generalization may be compromised. Lastly, the study had a retrospective design with an inadequate level of evidence.

Conclusion

Above all, in the Enshi population, we created a new prediction model based on the history of COPD, GS score of vascular assessment before PCI, monocyte, RC, and neutrophil/lymphocyte ratio, to help clinicians discern high-risk ISR patients, optimize treatment strategy, thus improve the prognosis of these patients. Furthermore, we visualized the prognostic risk factors in LASSO regression by using nomogram, evaluated the accuracy and clinical practicability of the predictive model by using DCA and Calibration curve. And it is found that the model is satisfactory in terms of goodness of fit, clinical usefulness, and accuracy.