Construction and validation of a diagnostic model for obstructive coronary artery disease in patients with suspected unstable angina pectoris based on resting ¹⁸F-fluorodeoxyglucose positron emission tomography myocardial ischemia memory imaging

Abstract

Background

Identifying obstructive coronary artery disease (OCAD) via non-invasive imaging modalities in patients with suspected unstable angina (UA) holds substantial clinical significance. This study aimed to develop and validate a diagnostic model for OCAD in patients with suspected UA, by leveraging resting ¹⁸F-fluorodeoxyglucose (¹⁸F-FDG) positron emission tomography (PET) myocardial ischemia memory imaging combined with clinical indicators.

Methods

We retrospectively analyzed 162 patients with a Global Registry of Acute Coronary Events (GRACE) score ≤140 who presented with chest pain or chest tightness and were clinically suspected of having UA. After collecting clinical indicators, predictive factors were screened using logistic regression. A diagnostic model was constructed using binary logistic regression based on the predictive factors, with internal validation via 1,000 bootstrap resamples. The discriminative ability, calibration, and clinical net benefit of the established model were evaluated by the receiver operating characteristic (ROC) curve, calibration curve, and decision curve analysis (DCA). Furthermore, the enhancement value of the final model over the basic model was quantified using net reclassification improvement (NRI) and integrated discrimination improvement (IDI).

Results

Of the 162 enrolled patients with suspected UA, 89 (54.9%) were diagnosed with OCAD. Six predictors were incorporated into the diagnostic model, including hyperlipidemia, diabetes, typical angina pectoris, ¹⁸F-FDG PET results, serum creatinine, and left ventricular ejection fraction (LVEF). The area under the curve (AUC) of the model was 0.89 [95% confidence interval (CI): 0.84–0.94], with a sensitivity of 0.84 and a specificity of 0.81; the Brier score was 0.1319. The Hosmer-Lemeshow goodness-of-fit test revealed good model fit (χ²=6.15, P=0.63). After internal validation via the bootstrap method, the optimism-corrected AUC was 0.87 (95% CI: 0.82–0.93). Calibration curve and DCA demonstrated that the model exhibited satisfactory calibration and promising clinical utility.

Conclusions

The OCAD diagnostic model for suspected UA patients, based on resting ¹⁸F-FDG PET and clinical indicators, demonstrated excellent diagnostic performance.

Introduction

Despite significant breakthroughs in the treatment of cardiovascular diseases, such diseases remain one of the leading causes of death worldwide (1). As a critical, life-threatening subtype of cardiovascular diseases, acute coronary syndrome (ACS) is triggered by the rupture or erosion of unstable coronary atherosclerotic plaques and is characterized by acute myocardial ischemia. It is often associated with high mortality rates and poor long-term prognosis (2). With the widespread application of high-sensitivity troponin assays and electrocardiography (ECG), ST-segment elevation myocardial infarction (STEMI) and non-ST-segment elevation myocardial infarction (NSTEMI) can be rapidly identified. However, some patients with unstable angina (UA) lack typical angina symptoms or characteristic ECG changes, rendering diagnosis challenging in certain cases (3,4). Therefore, the early identification of obstructive coronary artery disease (OCAD) in patients with suspected UA is of great clinical significance.

Currently, the main diagnostic test for OCAD is coronary angiography. However, this invasive procedure may induce unnecessary complications in patients. Previous studies have shown that among the patients who underwent coronary angiography, only 41% were clearly diagnosed as having OCAD (5), suggesting that a large proportion of coronary angiography examinations are unnecessary. Therefore, exploring non-invasive, accurate diagnostic methods has become an urgent clinical need. In recent years, diagnostic models developed based on non-invasive imaging examinations combined with clinical data have shown promising application prospects. Common imaging modalities include coronary computed tomography angiography (CCTA), echocardiography, and myocardial perfusion imaging (MPI) (6-8). Nevertheless, these examinations are not suitable for all patients due to factors such as calcified plaque interference, renal insufficiency, contrast agent allergy, and internal metal implants. Additionally, conventional imaging examinations primarily focus on anatomical assessment; if further evaluation of myocardial ischemia is required, a stress test has to be performed. However, stress tests may pose potential risks to patients with suspected UA. Against this backdrop, resting ¹⁸F-fluorodeoxyglucose positron emission tomography (¹⁸F-FDG PET) myocardial ischemia memory imaging can address the aforementioned limitations. Its core mechanism is as follows: during myocardial ischemia, the energy supply mode of myocardial cells shifts from fatty acid aerobic oxidation to glucose anaerobic metabolism. Even after blood flow is restored, abnormalities at the myocardial metabolic level persist for a period—a phenomenon termed “ischemic memory” (9,10). Based on this characteristic, resting ¹⁸F-FDG PET myocardial ischemia memory imaging has become an important tool for evaluating myocardial ischemia (11). Previous studies have demonstrated that this technique has certain value in identifying OCAD in patients with suspected UA (12), but its diagnostic efficacy still has room for improvement. Therefore, this study aims to develop an OCAD diagnostic model by integrating resting ¹⁸F-FDG PET myocardial ischemia memory imaging with clinical indicators, with the goal of providing a more accurate non-invasive diagnostic protocol for clinical practice. We present this article in accordance with the STARD reporting checklist (available at https://qims.amegroups.com/article/view/10.21037/qims-2026-1-0411/rc).

Methods

Study subjects

This was a retrospective study that consecutively collected data from patients with low-to-moderate-risk suspected UA [Global Registry of Acute Coronary Events (GRACE) score ≤140] who presented with chest pain or chest tightness at The Third Affiliated Hospital of Soochow University from September 2020 to April 2024. Inclusion criteria: (I) patients with angina symptoms such as chest tightness and chest pain, highly suspected of UA by cardiologists. The suspicion was based on a reduced threshold, increased severity, or higher frequency of angina attacks, and the presence of coronary artery disease (CAD) risk factors: family history of CAD, elderly patients, male, smoking, hypertension, diabetes and hyperlipidemia (13); (II) initial negative myocardial biomarkers; (III) normal or non-diagnostic ECG; (IV) GRACE score ≤140; (V) the interval between the most recent chest pain episode and resting ¹⁸F-FDG PET myocardial metabolic imaging was ≤7 days. Exclusion criteria: (I) clinically confirmed STEMI, NSTEMI, and high-risk UA (GRACE score >140); (II) severe cardiomyopathy, valvular heart disease, or congenital heart disease; (III) hemodynamically unstable or severe heart failure; (IV) serious infections or metabolic disorders; (V) patients with malignant tumors or undergoing radiotherapy and chemotherapy; (VI) patients with incomplete medical records lacking critical clinical information (critical information includes clinical data essential for definitive diagnosis and all 14 clinical indicators required for subsequent statistical analyses); (VII) pregnant or breastfeeding women.

This study was approved by the Ethics Review Committee of The Third Affiliated Hospital of Soochow University (approval No. 2020033) and was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. All patients provided written informed consent. The study flowchart is shown in Figure 1.

Figure 1.

Research flowchart. ECG, electrocardiography; FDG, fluorodeoxyglucose; GRACE, Global Registry of Acute Coronary Events; NSTEMI, non-ST-segment elevation myocardial infarction; PET, positron emission tomography; STEMI, ST-segment elevation myocardial infarction; UA, unstable angina.

Definition of clinical indicators

Patient data were retrospectively collected from the hospital’s electronic medical records. Based on the risk factors associated with CAD identified in previous literature and combined with the actual availability of hospital data, we ultimately collected 13 clinical variables, excluding the results of resting ¹⁸F-FDG PET myocardial ischemia memory imaging. These include demographic and clinical risk stratification data (age, gender, presence of typical angina pectoris), cardiovascular disease risk factors (hypertension, smoking history, hyperlipidemia, diabetes), clinical laboratory indicators (aspartate aminotransferase, serum creatinine, uric acid), cardiac electrophysiological and structural functional indicators [QRS duration, left ventricular ejection fraction (LVEF), left atrial diameter]. Typical angina pectoris is defined as oppressive dull pain in the sternum or precordium, which can radiate to the shoulders and arms, induced by emotional excitement or physical activity, relieved by rest or nitroglycerin administration, lasting for several minutes to tens of minutes (14). The GRACE score was calculated based on the patient’s age, heart rate, systolic blood pressure, serum creatinine, Killip class, cardiac arrest, ST segment deviation, and changes in cardiac markers (15). Hypertensive patients were defined as those whose blood pressure was measured three times on different days, with systolic blood pressure ≥140 mmHg and/or diastolic blood pressure ≥90 mmHg, or those taking antihypertensive drugs. Diabetic patients were defined as those diagnosed with diabetes based on laboratory indicators such as glycated hemoglobin and blood glucose, or those receiving hypoglycemic treatment. Patients with hyperlipidemia were defined as those diagnosed by physicians based on laboratory indices including triglycerides (TGs), low-density lipoprotein (LDL), high-density lipoprotein (HDL), total cholesterol, and other related parameters, or those undergoing lipid-lowering therapy. Left atrial diameter was measured by an echocardiographer with at least 5 years of experience using a color doppler echocardiography system, and LVEF was determined using the biplane Simpson method.

Resting ¹⁸F-FDG PET myocardial ischemia memory imaging

All patients underwent ¹⁸F-FDG PET myocardial ischemia memory imaging after a 12-hour fast. The radioactive tracer used for the examination was provided by Nanjing Jiangyuan Andeke Company, with an injection dose of 111–185 MBq. After tracer injection, patients were required to rest for 1 hour, followed by examination using a Biograph mCT 64s PET/CT scanner (Siemens AG, Erlangen, Germany). During the examination, patients were placed in a supine position and instructed to maintain stable breathing. The scanning procedure was as follows: first, a 64-slice spiral computed tomography (CT) scan was performed for localization and attenuation correction, with scanning parameters set to a tube current of 35 mA and a tube voltage of 120 kV; Subsequently, gated PET myocardial imaging was conducted. The imaging parameters were: energy peak of 511 keV, matrix size of 128×128, magnification factor of 2.0, and acquisition time of 10 minutes. Iterative reconstruction (four iterations, eight subsets) was used for image reconstruction, and finally, short-axis, horizontal long-axis, and vertical long-axis images of the left ventricle were obtained. The results of the ¹⁸F-FDG PET myocardial ischemia memory imaging were analyzed using visual semi-quantitative methods. A 3-point scoring system (0–2 points) was used for myocardial segmental metabolism: 0 points indicated no significant resting ¹⁸F-FDG uptake, 1 point indicated mild uptake, and 2 points indicated obvious uptake. Myocardial ¹⁸F-FDG uptake was classified into the following four patterns: pattern 1 was the “no uptake” pattern, representing a score of 0 in all myocardial segments (Figure 2A); pattern 2 was the “uniform uptake” pattern, representing a score of 1 or 2 in all myocardial segments (Figure 2B); pattern 3 was the “focal uptake” pattern, representing a score of 1 or 2 in some myocardial segments while a score of 0 in others (Figure 2C); pattern 4 was the “diffuse with focal uptake” pattern, representing a score of 2 in some myocardial segments while a score of 1 in others (Figure 2D). Patterns 1 and 2 were defined as negative (no significant myocardial ischemia); patterns 3 and 4 were defined as positive (presence of myocardial ischemia). However, focal uptake or diffuse uptake with focal involvement in the left ventricular basal segments and papillary muscles was defined as normal (16,17). The results of ¹⁸F-FDG PET myocardial ischemia memory imaging were classified as positive or negative. When pattern 3 or pattern 4 was observed, it was considered positive; otherwise, it was defined as negative. All images were independently evaluated by two nuclear medicine physicians. Both physicians had at least 5 years of work experience and were not exposed to the patients’ clinical background information during the evaluation process. The agreement rate of initial diagnoses between the two nuclear medicine physicians was 98.77%, with a Kappa coefficient of 0.975. In cases of disagreement between the two physicians, a third physician reviewed the images again and made the final diagnosis.

Figure 2.

Four uptake patterns of ¹⁸F-FDG PET myocardial ischemia memory imaging. (A) The “no uptake” pattern. (B) The “uniform uptake” pattern. (C) The “focal uptake” pattern. (D) The “diffuse with focal uptake” pattern. FDG, fluorodeoxyglucose; PET, positron emission tomography.

Coronary angiography

Coronary angiography was completed within 7 days after myocardial ischemia memory imaging. The results of coronary angiography were evaluated by two cardiologists with more than 10 years of clinical experience, and the cardiologists were completely unaware of the results of the patients’ myocardial metabolic imaging. When the opinions of the two experts differ, a third expert will conduct a re-diagnosis and reach a conclusion. OCAD was defined as follows: luminal stenosis ≥70% in the left anterior descending artery, left circumflex artery, right coronary artery, or their major branches; or luminal stenosis ≥50% in the left main coronary artery. Patients not meeting the above criteria were classified as having non-OCAD (18). With coronary angiography as the diagnostic gold standard, we divided patients with suspected UA into OCAD group and non-OCAD group.

Statistical analysis

Normality testing was first performed on continuous variables. For variables that followed a normal distribution, data were presented as mean ± standard deviation, and intergroup comparisons were carried out using the independent samples t-test. In contrast, variables deviating from a normal distribution were described as median (interquartile range), with intergroup differences analyzed via the Mann-Whitney U test. Categorical variables were expressed as constituent ratios or rates, and statistical analysis was conducted using the Chi-squared test or corrected Chi-squared test as appropriate. Univariate and multivariate logistic regression analyses were sequentially applied to screen potential predictive variables, and a diagnostic model was established by incorporating the finally selected predictors into binary logistic regression. The model’s receiver operating characteristic (ROC) curve was plotted, and its area under the curve (AUC), sensitivity and specificity were computed. The model’s performance in terms of accuracy, calibration, and clinical applicability was assessed using the Brier score, calibration curve, and DCA. Internal validation of the model was performed using 1,000 bootstrap resamples to correct for model optimism. Additionally, net reclassification improvement (NRI) and integrated discrimination improvement (IDI) were calculated to quantify the incremental diagnostic value of the comprehensive model for OCAD diagnosis—relative to the model utilizing only clinical indicators and the model relying solely on resting ¹⁸F-FDG PET myocardial ischemia memory imaging results. All P values were two-tailed, and a P value <0.05 was considered statistically significant. SPSS 26.0 and R (Version 4.5.1) were employed for performing statistical analyses.

Results

Baseline characteristics

This study enrolled a total of 162 patients presenting with suspected UA. The mean age of the entire cohort was 65.2±8.5 years, and 102 patients (63.0%) were men. Among these, 89 patients (54.9%) were diagnosed with OCAD, with a mean age of 65.4±8.7 years and 65 patients (73.0%) being men; 73 patients (45.1%) had non-OCAD, with a mean age of 65.0±8.3 years and 37 patients (50.7%) being men. Statistically significant differences were observed between the OCAD and non-OCAD groups in terms of gender, presence of typical angina pectoris, hyperlipidemia, diabetes, results of resting ¹⁸F-FDG PET myocardial ischemia memory imaging, QRS duration, serum creatinine level, and LVEF. Furthermore, although age, hypertension, and smoking history showed no significant differences between the two groups, age, prevalence of hypertension, and smoking rate in the OCAD group were all higher than those in the non-OCAD group. All patient baseline characteristics are presented in Table 1.

Table 1. Baseline characteristics of groups according to obstructive coronary artery disease.

Variables	All (n=162)	Non-OCAD (n=73)	OCAD (n=89)	P
Male	102 (63.0)	37 (50.7)	65 (73.0)	0.003*
Age (years)	65.2±8.5	65.0±8.3	65.4±8.7	0.789
Typical angina	81 (50.0)	22 (30.1)	59 (66.3)	<0.001*
Hypertension	107 (66.1)	43 (58.9)	64 (71.9)	0.082
Hyperlipidemia	104 (64.2)	39 (53.4)	65 (73.0)	0.010*
Diabetes	52 (32.1)	17 (23.3)	35 (39.3)	0.030*
Smoking history	49 (30.3)	21 (28.8)	28 (31.5)	0.710
18F-FDG PET	70 (43.2)	9 (12.3)	61 (68.5)	<0.001*
QRS (ms)	101 [96, 109]	99 [93, 107]	102 [98, 110]	0.018*
AST (U/L)	19.9 [16.8, 22.8]	20.2 [16.8, 23.4]	19.6 [16.8, 22.3]	0.710
Creatinine (μmol/L)	70.0±15.3	65.8±13.8	73.4±15.7	0.001*
Uric acid (μmol/L)	338.1±79.2	339.4±86.2	337.1±73.4	0.859
LVEF (%)	62 [60, 65]	63 [61, 65]	62 [58, 64]	<0.001*
LA (mm)	37 [35, 41]	37 [34, 39]	38 [35, 42]	0.063

Data are presented as median [interquartile range], mean ± standard deviation, or n (%). *, P<0.05. AST, aspartate aminotransferase; FDG, fluorodeoxyglucose; LA, left atrium; LVEF, left ventricular ejection fraction; OCAD, obstructive coronary artery disease; PET, positron emission tomography; QRS, QRS complex.

Selection of predictive factors

A total of 14 variables were subjected to multicollinearity testing, and variables with a variance inflation factor (VIF) >5 were excluded (19). Since the VIF values of all 14 candidate predictors were less than five, all variables were included in the univariate logistic regression analysis (Table 2). Variables with a P value <0.05 in the univariate analysis were further enrolled into the multivariate binary logistic regression, and a bidirectional stepwise regression method was adopted to screen out independent variables. The model achieved the best fit when six predictors were finally included. These six factors (hyperlipidemia, diabetes, typical angina, ¹⁸F-FDG PET myocardial ischemia memory imaging, serum creatinine, and LVEF) were incorporated into the diagnostic model as the final predictors.

Table 2. Univariate logistic regression analysis of the influencing factors of obstructive coronary artery disease.

Variables	β	SE	Z	P	OR (95% CI)
Age	0.01	0.02	0.27	0.788	1.01 (0.97–1.04)
QRS duration	0.03	0.01	2.26	0.024*	1.03 (1.01–1.05)
AST	−0.02	0.03	−0.79	0.429	0.98 (0.93–1.03)
Serum creatinine	0.04	0.01	3.09	0.002*	1.04 (1.01–1.06)
Uric acid	0.00	0.00	−0.18	0.858	1.00 (1.00–1.00)
LVEF	−0.17	0.05	−3.57	<0.001*	0.84 (0.77–0.93)
Left atrial diameter	0.03	0.03	1.21	0.226	1.04 (0.98–1.10)
Male
0					1.00 (reference)
1	0.97	0.33	2.90	0.004*	2.64 (1.37–5.08)
Typical angina
0					1.00 (reference)
1	1.52	0.34	4.47	<0.001*	4.56 (2.34–8.87)
Hypertension
0					1.00 (reference)
1	0.58	0.33	1.73	0.083	1.79 (0.93–3.44)
Hyperlipidemia
0					1.00 (reference)
1	0.86	0.33	2.57	0.010*	2.36 (1.22–4.55)
Diabetes
0					1.00 (reference)
1	0.76	0.35	2.16	0.031*	2.14 (1.07–4.25)
Smoking history
0					1.00 (reference)
1	0.13	0.34	0.37	0.710	1.14 (0.58–2.23)
18F-FDG PET
Negative					1.00 (reference)
Positive	2.74	0.42	6.48	<0.001*	15.49 (6.76–35.49)

*, P<0.05. 0: no. 1: yes. AST, aspartate aminotransferase; CI, confidence interval; FDG, fluorodeoxyglucose; LVEF, left ventricular ejection fraction; OR, odds ratio; PET, positron emission tomography; SE, standard error.

Construction of the diagnostic model

An OCAD diagnostic model was constructed in 162 participants using multivariate binary logistic regression based on the above six risk factors. The results of the regression analysis are presented in Table 3, and a nomogram was used for visualization (Figure 3). The nomogram in this study included six variables: typical angina pectoris, hyperlipidemia, diabetes, ¹⁸F-FDG PET myocardial ischemia memory imaging, serum creatinine, and LVEF. The result of each variable corresponded to a point value at the top of the nomogram. The total point was obtained by summing the points of all variables, and the probability corresponding to the total point represented the probability of OCAD.

Table 3. Multivariate logistic regression analysis of the influencing factors of obstructive coronary artery disease.

Variables	β	SE	Z	P	OR (95% CI)
Serum creatinine	0.03	0.02	2.01	0.044*	1.03 (1.01–1.06)
LVEF	−0.16	0.07	2.32	0.020*	0.85 (0.74–0.98)
Typical angina
0					1.00 (reference)
1	0.95	0.44	2.18	0.030*	2.59 (1.10–6.10)
Hyperlipidemia
0					1.00 (reference)
1	1.09	0.47	2.30	0.021*	2.98 (1.18–7.56)
Diabetes
0					1.00 (reference)
1	1.13	0.51	2.23	0.026*	3.09 (1.14–8.36)
18F-FDG PET
Negative					1.00 (reference)
Positive	2.67	0.50	5.35	<0.001*	14.47 (5.44–38.52)

*, P<0.05. 0: no. 1: yes. CI, confidence interval; FDG, fluorodeoxyglucose; LVEF, left ventricular ejection fraction; OR, odds ratio; PET, positron emission tomography; SE, standard error.

Figure 3.

A nomogram model for predicting the risk of OCAD in individuals with suspected unstable angina. Ischemia memory imaging: ¹⁸F-FDG PET myocardial ischemia memory imaging. FDG, fluorodeoxyglucose; LVEF, left ventricular ejection fraction; OCAD, obstructive coronary artery disease; PET, positron emission tomography.

Evaluation and validation of the diagnostic model

Figure 4 shows the apparent ROC curve of the diagnostic model. The AUC for diagnosing OCAD was 0.89 (95% CI: 0.84–0.94). At the maximum Youden index of 0.65, the sensitivity was 0.84 and the specificity was 0.81. Figure 5 presents the calibration curve of the model. The apparent curve closely approximated the ideal curve, and the bias-corrected curve also showed close agreement with the ideal curve. The Brier score was 0.1319, suggesting favorable overall accuracy of the predicted probabilities; the Hosmer-Lemeshow goodness-of-fit test did not indicate poor model fit (χ²=6.15, P=0.63). Combined with the calibration curve, these results demonstrate that the model has good calibration. Internal validation of the model was performed using bootstrap resampling with 1,000 iterations. The optimism-corrected ROC curve is shown in Figure 6. The mean optimism was 0.018, yielding a corrected AUC of 0.87 (95% CI: 0.82–0.93), which suggests a small optimism bias and stable diagnostic performance.

Figure 4.

ROC curve of obstructive coronary artery disease diagnostic model. AUC, area under the curve; CI, confidence interval; ROC, receiver operating characteristic.

Figure 5.

Calibration curve of diagnostic model. The horizontal axis represents the predicted probability derived from the model, and the vertical axis denotes the actual observed frequency. The red solid line indicates the apparent curve of the model, while the black dashed line stands for the calibration curve after 1,000 bootstrap resamplings. The grey diagonal line is the ideal reference curve. The red shaded area refers to the 95% confidence interval of the prediction curve, and the grey histogram shows the distribution density of study samples across different predicted probabilities. CI, confidence interval; MAE, mean absolute error.

Figure 6.

ROC curve corrected by 1,000 bootstrap resamplings. Solid red line: apparent ROC curve. Dotted blue line: optimism-corrected ROC curve. Shaded area: corrected 95% CI. AUC, area under the curve; CI, confidence interval; ROC, receiver operating characteristic.

Clinical application value of the diagnostic model

DCA was performed to evaluate the clinical utility of the model (Figure 7). The red curve represents the net benefit of selective treatment based on the CAD diagnostic model; the blue oblique line represents the net benefit of treating all patients, and the green horizontal dashed line represents the net benefit of treating none. The results showed that when the threshold probability for CAD treatment ranged from 17% to 99%, the net benefit of selective treatment for patients with suspected UA guided by the model was consistently higher than that of either the treat-all or treat-none strategy. This suggests that the model can assist clinical decision-making by helping avoid unnecessary overtreatment in low-risk patients while ensuring timely intervention for high-risk individuals, indicating favorable clinical utility for initial CAD risk assessment in patients with suspected UA.

Figure 7.

Decision curve analysis of the diagnostic model. The red solid line represents the net benefit of using this diagnostic model; the blue dashed line represents the net benefit of treating all patients; the green dashed line represents the net benefit of treating no patients.

Enhancement value of the diagnostic model

To evaluate the enhancement value of this comprehensive model in OCAD diagnostic efficacy, we constructed Model 1 using the same logistic regression method with the five clinical indicators: hyperlipidemia, diabetes, typical angina pectoris, serum creatinine, and LVEF. Model 2 was constructed using ¹⁸F-FDG PET myocardial ischemia memory imaging alone, and the comprehensive model was defined as Model 3. The AUC of Model 1 was 0.81 (95% CI: 0.74–0.88); Model 2 was 0.78 (95% CI: 0.72–0.84); Model 3 was 0.89 (95% CI: 0.84–0.94). Figure 8 shows the ROC curves of the three models, and Model 3 exhibited the highest discriminatory ability for diagnosing OCAD. When comparing Model 3 with Model 1, the NRI was 0.17 (95% CI: 0.01–0.31) and the IDI was 0.18 (95% CI: 0.12–0.23). When comparing Model 3 with Model 2, the NRI was 0.06 (95% CI: −0.05 to 0.16) and the IDI was 0.15 (95% CI: 0.10–0.20).

Figure 8.

The ROC curves of the three diagnostic models, where the blue line represents Model 1, the orange line represents Model 2, and the green line represents Model 3. Model 1: only incorporating 5 clinical indicators. Model 2: only incorporating ¹⁸F-FDG PET myocardial ischemia memory imaging. Model 3: incorporating all 6 indicators. AUC, area under the curve; FDG, fluorodeoxyglucose; PET, positron emission tomography; ROC, receiver operating characteristic.

Discussion

According to clinical practice guidelines, only high-risk patients with ACS require early coronary angiography (20). For intermediate- and low-risk patients, however, premature coronary angiography not only fails to reduce mortality but may also increase the risk of procedural complications and bleeding (21,22). Therefore, for intermediate- and low-risk populations with suspected UA, non-invasive imaging examinations should be prioritized for initial assessment, and the need for further coronary angiography should be determined based on the assessment results. Nevertheless, simple non-invasive imaging examinations often only provide information on the local anatomical structure of the coronary arteries, resulting in certain limitations in diagnostic efficacy. In a large-scale study conducted in the United States, nearly 280,000 patients with suspected CAD were diagnosed with myocardial ischemia via stress testing and subsequently underwent coronary angiography; however, only 38% of these patients were confirmed to have OCAD (5). This indicates that a large number of patients diagnosed with “myocardial ischemia” may only have mild-to-moderate coronary artery stenosis. If treatment strategies are developed solely based on the degree of coronary artery stenosis, some patients may not benefit. Currently, some scholars have proposed renaming “CAD” to “myocardial ischemia syndrome”—a viewpoint that profoundly reflects the core value of “myocardial ischemia” in the diagnosis and treatment decision-making of patients with CAD (23).

In the field of non-invasive evaluation of myocardial ischemia, stress MPI has achieved remarkable progress (24). The CAD prediction model constructed by combining MPI with clinical indicators exhibits excellent diagnostic performance, with an AUC of up to 0.92 (25). However, for patients with suspected UA, stress testing carries potential risks of inducing or exacerbating myocardial ischemia and even triggering malignant arrhythmias, which limits its clinical application to a certain extent. Similarly, other stress modalities such as stress echocardiography, stress cardiac magnetic resonance imaging, and treadmill exercise ECG face the same safety concerns in patients with UA. Although resting MPI is useful for diagnosing patients with suspected UA the diagnostic efficacy of standalone resting MPI is inferior to that of resting ¹⁸F-FDG PET myocardial ischemia memory imaging. Previous studies have demonstrated that resting MPI combined with clinical indicators yields an AUC of 0.91 for the diagnosis of CAD (8). Nevertheless, this approach only reflects the current myocardial perfusion status and cannot identify previous episodes of myocardial ischemia (12). As a non-invasive examination technique that requires no stress stimulation and can accurately identify previous myocardial ischemia events, resting ¹⁸F-FDG PET myocardial ischemia memory imaging is gradually demonstrating unique advantages in the field of CAD diagnosis and treatment. Based on this, this study constructed a diagnostic model for OCAD by taking the results of resting ¹⁸F-FDG PET myocardial ischemia memory imaging as the core and combining them with clinical indicators. The aim is to provide a more accurate and safer new assessment method for patients with suspected UA and offer reliable references for clinical diagnosis and treatment decisions.

In this study, one imaging indicator and five independent risk factors for OCAD were screened from 14 clinical variables, including ¹⁸F-FDG PET myocardial ischemia memory imaging, typical angina pectoris, hyperlipidemia, diabetes, serum creatinine, and LVEF. All these factors were closely associated with CAD. Firstly, the coronary arteries of CAD patients are narrowed to varying degrees due to the formation of atherosclerotic plaques. When myocardial oxygen consumption increases abruptly or coronary blood supply decreases suddenly, the stenotic vessels cannot provide sufficient blood for the myocardium, leading to myocardial ischemia. At this time, a large number of metabolites accumulate in myocardial cells, stimulate cardiac nerve endings, and eventually induce angina pectoris (26,27). In previous studies, the detection rate of obstructive coronary atherosclerosis in patients with angina pectoris was significantly higher than that in people without angina pectoris, and angina pectoris has been confirmed as a major independent predictor of cardiovascular death or non-fatal myocardial infarction in patients with chronic coronary syndrome (28,29). Hyperlipidemia is a recognized risk factor for CAD and an independent indicator for predicting cardiovascular events. Data have shown that more than 25% of CAD risk loci are significantly associated with lipid traits at the genome-wide level, suggesting that the lipid metabolism pathway is a core component of CAD genetic susceptibility (30,31). From the pathophysiological perspective, LDL and TG deposit in the subendothelium of blood vessels and are phagocytosed by activated macrophages to form foam cells, which further trigger chronic inflammatory reactions and eventually promote atherosclerotic plaques (32). With the elevation of LDL, very low-density lipoprotein (VLDL) and TG levels, the risk of CAD increases accordingly, while the increase of HDL can reduce the risk of CAD (33,34). In addition, elevated lipid levels are associated with an increased risk of mortality in patients diagnosed with CAD, and achieving lipid control targets can reduce the risk of cardiovascular events (35). Diabetes is also an important risk factor for CAD patients. Under the synergistic effect of long-term hyperglycemia and insulin resistance, diabetic patients continuously induce oxidative stress and chronic low-grade inflammation, promote the formation of advanced glycation end products, LDL oxidation, endothelial dysfunction and macrophage foam cell formation, which accelerate the formation of coronary atherosclerotic plaques step by step and eventually lead to CAD (36). Meta-analyses have shown that the prevalence of CAD in diabetic patients is about twice that in non-diabetic patients, and the risk of major adverse cardiovascular events is significantly higher in CAD patients with diabetes (37). As the final product of muscle metabolism, serum creatinine is mainly excreted through renal filtration, and its serum concentration is the most commonly used indicator for clinical evaluation of renal function. Within a certain range, serum creatinine level is closely related to the risk of CAD and the degree of coronary stenosis. In patients diagnosed or highly suspected of CAD, decreased renal function is significantly associated with an increased risk of adverse cardiovascular events (38,39). The potential mechanism may be that when renal function declines, pathological conditions such as imbalanced body volume load, electrolyte disturbance and retention of inflammatory mediators are activated. These factors can further accelerate the occurrence and progression of atherosclerosis by promoting vascular endothelial injury and lipid deposition. LVEF is a core indicator for the quantitative evaluation of cardiac pumping function. CAD leads to insufficient myocardial blood supply, and myocardial cells are in anoxic state, which easily causes cell degeneration and necrosis, resulting in impaired myocardial contractility and eventually reduced LVEF. Clinical studies have confirmed that the degree of coronary stenosis is significantly negatively correlated with LVEF (40). In CAD patients, the lower the LVEF level, the higher the risk of all-cause mortality and major adverse cardiovascular events (41). In summary, the five clinical indicators screened in this study have established clear and fully pathophysiologically supported associations with the occurrence, development and adverse prognosis of CAD from the key dimensions of typical symptoms, core risk factors and cardiac function of CAD, providing sufficient logical and clinical evidence for them to become the core predictors of the diagnostic model.

In addition, to evaluate the differences between the comprehensive model and the single clinical indicator model or single imaging model, we compared Model 3 with Model 1 and Model 2, respectively. The results showed that the AUC of Model 3 was 0.89, higher than that of Model 1 and Model 2, indicating that the comprehensive model had the highest discriminatory ability in diagnosing OCAD. The DeLong test was used to evaluate the differences among the three models. No significant difference was found between Model 1 and Model 2, while Model 3 showed significant differences compared with both Model 1 and Model 2. This suggests that although ¹⁸F-FDG PET myocardial ischemia memory imaging alone provides functional information, it shows no statistical advantage over traditional clinical indicators, and ischemia memory imaging cannot simply replace clinical risk assessment. The combination of clinical indicators and imaging features allows evaluation from different pathological dimensions of CAD, demonstrating a synergistic complementary effect. In the model improvement assessment, for Model 3 vs. Model 1: NRI =0.17 (95% CI: 0.01–0.31), IDI =0.18 (95% CI: 0.12–0.23); for Model 3 vs. Model 2: NRI =0.06 (95% CI: −0.05 to 0.16), IDI =0.15 (95% CI: 0.10–0.20). Compared with Model 1, Model 3 indicated that the net proportion of correctly reclassified patients exceeded that of incorrectly reclassified patients by 17 percentage points, and an IDI of 0.18. Compared with Model 2, Model 3 yielded an NRI of 0.06 and an IDI of 0.15. Notably, the NRI for Model 3 versus Model 2 was not statistically significant. This may be because the clinical indicator model already had favorable diagnostic performance, and the addition of functional information from ¹⁸F-FDG PET myocardial ischemia memory imaging mainly enabled precise re-stratification of patients with uncertain risk based on clinical indicators, thus providing clear NRI. In contrast, relying solely on imaging features without the epidemiological background of CAD resulted in relatively low predicted probabilities. Although the addition of clinical indicators ultimately improved overall discriminative ability, the upward and downward adjustments during reclassification may have partially offset each other, leading to a lack of statistical significance. From a pathological mechanism perspective, clinical indicators reflect long-term chronic risk exposure of CAD, forming the necessary background for CAD pathogenesis.¹⁸F-FDG PET myocardial ischemia memory imaging mainly captures functional markers of acute ischemic events, representing signs of disease activity. Interpretation of imaging findings usually requires integration of the patient’s clinical background to achieve greater accuracy, which is consistent with real-world clinical practice and emphasizes the irreplaceable fundamental role of traditional risk factors in the era of precision medicine.

The model constructed in this study can provide a novel, safe and efficient evaluation method for patients with suspected UA pectoris. However, this study has certain limitations: first, the sample size included is relatively limited and only from a single center, which may restrict the extrapolation and application of the research results to a wider population. Second, this study only conducted internal validation to confirm the stability of the model, and the argumentative power of internal validation is weaker than that of external validation. In the future, it is necessary to expand the sample size and conduct external validation with multi-center data to further verify the clinical applicability of the model. In addition, during the period from the onset of chest pain to the completion of ischemic memory imaging, patients may use various drugs to relieve myocardial ischemia, which may cause confounding interference with the results of ischemic memory imaging. Subsequent studies will strictly control relevant confounding variables, reduce the impact of drugs and other factors on imaging results, and further improve and optimize the diagnostic model.

Conclusions

A diagnostic model for OCAD was constructed by combining the results of resting ¹⁸F-FDG PET myocardial ischemia memory imaging with five clinical indicators—hyperlipidemia, diabetes, typical angina pectoris, serum creatinine level, and LVEF. This model demonstrates favorable diagnostic efficacy in patients with suspected UA and is anticipated to serve as a safe, efficient approach for evaluating this patient population.

Supplementary

The article’s supplementary files as

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. This study was approved by the Ethics Review Committee of The Third Affiliated Hospital of Soochow University (approval No. 2020033). Written informed consent was taken from all individual participants.

Data Sharing Statement

Available at https://qims.amegroups.com/article/view/10.21037/qims-2026-1-0411/dss

qims-16-08-638-dss.pdf ^{(114.4KB, pdf)}

DOI: 10.21037/qims-2026-1-0411