A retrospective analysis was conducted on consecutive patients from January 2016 to December 2017 at Fujian Medical University Union Hospital. The study focused on patients who were 65 years or older and underwent PCI, with a 5-year clinical follow-up period. The exclusion criteria for the study were: (1) patients who were under 65 years old, (2) acute myocardial infarction (AMI) within 7 days [14], (3) lack of follow-up data, (4) situations where quantitative flow ratio (QFR) calculation could not be performed, including an interrogated lesion involving a myocardial bridge or bypass graft; severe overlap in the stenosed segment or severe tortuosity of any interrogated vessel; and poor angiographic image quality.

The QFR computations and QCA analyses were performed by two independent investigators blinded to the clinical data using the AngioPlus system (Pulse Medical Imaging Technology Shanghai, China) according to standard operating procedures. The three-dimensional (3D) reconstructions of three main vessels were performed based on the automated contouring of two angiographic projections captured at 15 frames/s and at least 25° apart. After 3D reconstruction, QFRs were computed using contrast flow velocity models [15]. In this study, QFR was computed in each participant at the time of pre-PCI and post-PCI. Post-PCI QFR, defined as including at least one lesion treated with PCI, was retrospectively computed in all eligible vessels [16]. The QCA information of all vessels consisted of blood flow resistance (BFR), percent diameter stenosis (DS%) and the percent of area stenosis (AS%). Furthermore, the QFR and QCA data for patients with multiple lesions are presented as the mean of those values.

An electronic medical record system was utilized to retrospectively gather relevant clinical data of the patients at the time of their first hospitalization. This allowed the researchers to access and analyze the necessary information for their analysis. Laboratory indices during the initial hospitalization such as low-density lipoprotein cholesterol (LDL-C), creatinine, N-terminal pro brain natriuretic peptide (NT-proBNP), and high-sensitivity C-reactive protein (hs-CRP) were calculated using standard laboratory techniques. Echocardiography was employed to determine left ventricular ejection fraction (LVEF) and E/E’. E/E’ is the ratio of peak mitral early filling velocity (E) to early diastolic mitral annular velocity (E’), an indicator of diastolic cardiac function. All patients received standard pharmacological treatment according to clinical guidelines, and the information on medications used, including statins, antiplatelet agents, angiotensin-converting-enzyme inhibitors (ACEIs), and angiotensin-receptor blockers (ARBs), was reported. Information of other atherosclerosis-related diseases, such as the atherosclerosis of the carotid arteries and strokes, and in patients with atrial fibrillation and their use of anticoagulants, was also obtained.

Target vessel revascularization (TVR) was defined as any subsequent PCI or surgical bypass involving any segment of the target vessel, including the target lesion and non–target lesions that underwent revascularization. Target vessel and non-target lesion revascularization was defined as any repeat percutaneous intervention or surgical bypass of the target vessel for pre-existing disease, disease progression or other reasons unrelated to the target lesion. Additionally, the definition of target vessel was the entire major intervened coronary vessel, including side branches [17]. We analyzed patients with TVR as our endpoint. Furthermore, TVR was performed through detecting the significant stenosis by angiography based on whether patients suffered from chest pain or other symptoms. And those who required repeat revascularization were also included in the TVR groups.

Continuous variables were presented as mean $\pm$ standard deviation (SD) or median (interquartile range, IQR) and compared using appropriate statistical tests such as the 2-sample Student’s t-test, Welch’s t-test, or Mann-Whitney U test. Categorical variables were presented as numbers and percentages and compared using the chi-squared test or Fisher’s exact test.

To develop the overall survival nomogram, a multivariable Cox regression model was chosen in a training set. Initially, univariable Cox regression was performed, and variables with a p-value 0.10 were selected as candidate variables. The nomogram’s performance was assessed through discrimination and calibration in both the training set and the validation set. The receiver operating characteristic curve (ROC) was used to measure the area under the curve (AUC). The optimal threshold was determined using the Youden index derived from the ROC curve. Model performance was further evaluated through survival analysis using Kaplan–Meier curves and landmark analysis. Calibration plots were used to compare the actual Kaplan–Meier survival estimates with the predicted probability of freedom from 2-year and 5-year TVR, and C-index was used to assess the performance of the nomogram.

In this study, the death is a competing risk data, which affect the incidence of TVR during follow-up. Fine-Gray model was performed to analysis this phenomenon. We categorized the population into three groups: the deceased individuals as the competitive risk subgroup, those experiencing TVR as the risk group, and the rest of the population as the control group. Subsequently, we performed Fine-Gray analysis over a 5-year follow-up, plotted the curves, and meticulously observed the outcomes.

All statistical analyses were performed using SPSS 25.0 (IBM Inc., New York, NY, USA) and R version 4.2.2 (R Foundation for Statistical Computing, Vienna, Austria).

From January 2016 to December 2017, the angiographic data of 3328 vessels from 1570 patients were screened. A total of 448 vessels in 195 patients were excluded due to a failure in computing QFR, 1419 vessels in 684 patients were excluded due to unsuitable clinical conditions, resulting in 1461 vessels from 691 patients being included in the final analysis. The enrolled patients were randomly stratified into a training set and a validation set at a 7:3 ratio. Finally, 483 patients with 1014 lesions were divided into a training set, and 208 patients with 447 lesions were divided into a validation set (Fig. 1).

Fig. 1.

Study flow chart. PCI, percutaneous coronary intervention; QFR, quantitative flow ratio.

Tables 1,2 provided baseline characteristics and angiographic data of the training set and validation set. There were no statistically significant differences in the general clinical and angiographic data between both sets. During the follow-up period of 2 and 5 years, the mean of the overall observation period was 56 months. Additionally, 61 patients died during following up, and the remaining population had an average follow-up period of 58.3 months. A total of 28 patients during 2 years and 48 patients within 5 years experienced TVR. The rates of TVR were 4.05% and 6.95%, respectively. In the training set, the TVR rates after 2 years and 5 years were 4.55% and 7.66% respectively. Similarly, in the validation set, the TVR rates after 2 years and 5 years were 2.88% and 5.29% respectively. No significant differences were observed between the two groups in terms of TVR rates (all p $>$ 0.05).

Appropriate variables from patients in the training set were selected to develop the nomogram model. The clinical characteristics and angiographic data of the training set are shown in Tables 3,4. Patients in the training set were further stratified into a control group and a TVR group. Patients in the TVR group experienced higher rates of DM, previous MI, and previous PCI compared to the control group (56.76% vs. 37.44%, p = 0.021; 32.43% vs. 5.83%, p 0.001; 40.54% vs. 9.64%, p 0.001). There were no significant differences in the QCA analysis in the pre-PCI (all p $>$ 0.05). The conditions of the vessels after PCI were similar but the QFR of the TVR group was less than the control group (0.93 $\pm$ 0.08 vs. 0.96 $\pm$ 0.06, p = 0.001).

According to the above baseline characteristics, angiographic data and risk factors of cardiovascular hazard events, 13 variables were selected into the univariate Cox regression analysis (Table 5). 5 candidate variables were found to satisfy the threshold of p 0.10. The multivariate Cox regression analysis indicated that DM, previous MI, previous PCI, and post-PCI QFR were significant independent predictors of the rate of TVR in the training set (p 0.05). We eventually created a nomogram for TVR prediction by using these factors (Fig. 2). Each predictor corresponded to a specific point by drawing a straight line upwards to the axis point. Scores for each variable were added and located on the “Total Points” axis. Finally, a vertical line was drawn straight down from the plotted total axis point to the 2-year or 5-year TVR probability axis to determine the probability of TVR. The Nomogram score was calculated using the following formula:

Nomogram score = [200 + (–200 $\times$ post-PCI QFR)] + (51.41004 $\times$ previous PCI) + (46.5815 $\times$ previous MI) + (18.14027 $\times$ DM)

Fig. 2.

Nomogram for predicting 2-year and 5-year TVR. Points were composed of diabetes mellitus, post-PCI QFR, previous MI and previous PCI. The score for each value was assigned by drawing a line upward to the points line, and the sum of the four scores was plotted on the total points line. Finally, the probability line was used to determine the probability of freedom from 2-year and 5-year TVR. MI, myocardial infarction; QFR, quantitative flow ratio; TVR, target vessel revascularization; PCI, percutaneous coronary intervention.

The calibration plots of predictions from the nomogram model in the training set, validation set and total population are displayed in Fig. 3. The C-index was used to assess the performance of the nomogram model. The results revealed that the C-index were 0.771 and 0.736 in the training set during the 2-year and 5-year periods (Fig. 3A,B), 0.846 and 0.801 in the validation set (Fig. 3C,D), 0.774 and 0.758 in the total population (Fig. 3E,F), respectively.

Fig. 3.

The calibration plots for predicting 2-year and 5-year TVR probability. (A) 2-year TVR in the training set. (B) 5-year TVR in the training set. (C) 2-year TVR in the validation set. (D) 5-year TVR in the validation set. (E) 2-year TVR in the whole population. (F) 5-year TVR in the whole population. TVR, target vessel revascularization.

The ROC analysis indicates that the nomogram had an excellent performance for predicting TVR in the training, validation and whole population sets (Fig. 4). The nomogram yielded an AUC of 0.742 and 0.789 for predicting 2-year to 5-year TVR risk in the training set (Fig. 4A,B) and 0.837 and 0.783 in the validation set (Fig. 4C,D). Additionally, the AUC of the nomogram in the total population for predicting 2-year and 5-year TVR were 0.764 and 0.786, respectively (Fig. 4E,F).

Fig. 4.

The ROC curves for comparisons of different variable. (A) The ROC curves to predict 2-year TVR in training set. (B) The ROC curves to predict 5-year TVR in training set. (C) The ROC curves to predict 2-year TVR in validation set. (D) The ROC curves to predict 5-year TVR in validation set. (E) The ROC curves to predict 2-year TVR in whole population. (F) The ROC curves to predict 5-year TVR in the whole population. ROC, receiver operating characteristic curve; AUC, area under curve; MI, myocardial infarction; QFR, quantitative flow ratio; LVEF, left ventricular ejection fraction; TVR, target vessel revascularization; PCI, percutaneous coronary intervention.

The distribution of the nomogram score in three sets was shown in Supplementary Fig. 1. Based on the 2-year and 5-year ROC analysis, a prognostic score cut-off point of 20.07 was determined. The patients were then divided into two groups: Group A (score $\le$20.07) and Group B (score $>$20.07). Kaplan–Meier curves were recorded and a landmark analysis was performed and implemented at 1 year (Fig. 5A,B). The results showed there was no significant difference between the two groups in the incidence of TVR within the first year (log-rank p = 0.406) and the curve appeared to be well separated after one year, implying reasonable discrimination (log-rank p 0.001).

Fig. 5.

The Kaplan–Meier curves and landmark analysis to predict TVR. (A) The Kaplan–Meier curves to predict 2-year TVR and landmark analysis at one year. (B) The Kaplan–Meier curves to predict 5-year TVR and landmark analysis at one year. Group A: score $\le$20.07; Group B; score $>$20.07. The cut-off points of 2-year and 5-year ROC are 20.07. ROC, receiver operating characteristic curve; TVR, target vessel revascularization.

During the follow-up period, we encountered competing risks between the occurrence of death and our primary endpoint, TVR. To further investigate and analyze this situation, we employed the Fine-Gray model, as depicted in Supplementary Fig. 2. The results revealed no significant difference in the percentage of death between the groups with scores $\le$20.07 and scores $>$20.07. The incidence of TVR remained higher in the groups with scores $>$20.07 compared to those with scores $\le$20.07 even after accounting for competing risks (p 0.001).

Our study has several limitations. First, the present study is a retrospective, single-center analysis, with small a sample size. Therefore, the findings need to be confirmed by further prospective multicenter cohort studies. Data from other centers are also required to access the current model in more external validation sets. Second, not all screened patients were included in the final analysis, which inevitably introduces selection bias. Finally, model performance is not extremely perfect, the prediction of this nomogram is properly performed in lower- risk patients in most cases, and there is room for improvement.