The utilization of myocardial perfusion imaging prior to coronary angiography to improve outcomes and cost-effectiveness of patients with stable coronary artery disease: a comparative cohort study in China

Abstract

Background

The incremental value of myocardial perfusion imaging (MPI) before coronary angiography (CAG) for suspected or known stable coronary artery disease (SCAD) has not been validated in China. This study investigated the necessity of MPI for these patients regarding improvements in cardiac outcomes and cost-effectiveness in China.

Methods

The initial cohort comprised 7,437 patients with suspected or known SCAD between 2018 and 2019. After excluding patients with acute coronary syndrome, previous myocardial infarction (MI) and revascularization, enrolled patients were divided into the CAG group and the MPI group according to the initial strategy (CAG or MPI). Then, two groups were matched by propensity score. The cost, revascularization, MI, and all-cause mortality of patients were followed.

Results

The MPI and CAG groups each included 130 patients after matching, with similar basic characteristics (P > 0.05). Significantly decreased cardiac events were observed in the MPI group compared to the CAG group (6 vs. 16, P < 0.05). Furthermore, the number of revascularization (5 vs. 12), MI (1 vs. 3), and all-cause mortality (0 vs. 1) in the MPI group was also lower. The Cox model showed that fewer patients in the MPI group had cardiac events (HR 0.27, 95% CI 0.10–0.71). Moreover, patients in the MPI group had fewer first-visit costs ($718 vs. $1389) and fewer downstream costs ($120 vs. $344) than those in the CAG group (P < 0.001).

Conclusions

Our findings suggest that an MPI-first strategy is associated with fewer cardiac events and lower costs compared with an initial CAG strategy.

Trial registration

Not applicable. This study is a retrospective analysis.

Background

Stable coronary artery disease (SCAD) masks these patients with chronic atherosclerotic diseases and increases their risk of adverse cardiovascular events [1, 2]. So far, invasive coronary angiography (CAG) remains frequently applied in the diagnostic pathway for stable angina, despite its high cost and associated risks [3–5]. Some patients with stable chest pain and checked by CAG actually do not develop severe obstructive coronary artery disease, suggesting that CAG may be unnecessary and potentially avoidable for many cases [6, 7]. Updated ESC and ACC/AHA guidelines recommend stress testing and non-invasive imaging for diagnosing patients with suspected or known SCAD [8]. Single-photon emission computed tomographic (SPECT) myocardial perfusion imaging (MPI) is the most common non-invasive method for accurately diagnosing myocardial ischemia and determining its location, severity, and extent. As reported, more than 10 million times of MPI are performed annually in the US and about 230,000 in Japan, thus avoiding the overuse of invasive examinations and treatments for SCAD patients [9, 10].

Therefore, a step-by-step approach through applying selective CAG after risk stratification using MPI has been proposed to improve the rationality and accuracy of percutaneous coronary intervention (PCI), and meanwhile maximize cost-effectiveness [11, 12]. However, current reports seem not completely consistent regarding improvements in cardiac outcomes and cost-effectiveness. Hung et al. showed that MPI before CAG improves the diagnosis of suspected or known SCAD, reducing revascularization and MI rates and improving survival [13]. Other two earlier studies have demonstrated that the utilization of MPI prior to CAG reduces the total cost of patients even though it has no significant effect on their clinical outcomes [14, 15]. MPI showing no or mild ischemia suggests a better prognosis and possible treatment with medication alone, without further CAG or PCI [16]. In contrast, high-risk patients can benefit from targeted revascularization [17].

This study aims to investigate differences in cardiac outcomes and costs between SCAD patients with and without initial MPI prior to CAG in China, due to inconsistencies in MPI benefits and a lack of cost-effectiveness data.

Methods

Study setting, population, and design

This retrospective study was conducted at the First Hospital of Shanxi Medical University, a tertiary academic medical center in central China with approximately 3,000 beds. The study population comprised patients with suspected or diagnosed SCAD. Inclusion criteria were as follows: (1) patients were admitted to our hospital from January 2018 to December 2019; (2) patients with chest pain following exertion or agitation, retrosternal pain, chest pain relieved by nitroglycerin or rest for > 10 min; (3) patients diagnosed with SCAD were included (n = 7437). Patients with MPI or CAG were first included (n = 6,020), and then divided into the initial MPI group (n = 2,394) and the initial CAG group (n = 3,626) according to the initial diagnostic strategy. The exclusion criteria for the initial MPI group were as follows: (1) patients with only rest-MPI (n = 1,846); (2) patients with previous myocardial infarction (MI) or acute coronary syndromes (ACS) (n = 11); (3) patients with coronary revascularization (n = 2); (4) patients with other heart diseases (n = 62); (5) patients with cancer (n = 70); (6) patients with incomplete information (n = 73). The same exclusion criteria were applied to the initial CAG group. Thus, 330 cases were in the MPI group and 300 in the CAG group, respectively. Moreover, patients who had an endpoint event within 90 days after the examination of MPI or CAG were excluded (n = 15). Next, the two groups underwent 1:1 propensity score matching (PSM) based on age, gender, and Charlson’s Comorbidity Index (CCI). In the end, 130 cases were included in both the matched MPI group and the matched CAG group, respectively (Fig. 1).

Fig. 1.

Selection process of patients included in this study. Abbreviation: matched CAG/MPI group, grouping after 1:1 propensity score matching; CAG, coronary angiography; MPI, myocardial perfusion imaging; CCI, Charlson’s comorbidity index

Data collections procedure and variables

Age, gender, family history, lifestyle (history of smoking or alcohol consumption), comorbidities (hypertension, hyperlipidemia, diabetes), CCI, and results of CAG or MPI test were recorded for both groups. CCI was calculated for each patient according to the CCI scoring criteria. All the above information was obtained from the patient’s medical record system and the case room database of the hospital.

Follow-up and endpoint

All patients with suspected or known SCAD were followed since their first visit to our hospital. Patients who experienced cardiac events such as revascularization, MI, or death were followed until the first occurrence of the cardiac event, while those who did not experience any cardiac event were followed until the end of the trial (31 December 2022). Initial diagnosis and treatment costs and downstream costs due to suspected or known SCAD were recorded for each group at the end of the follow-up period.

Cost analysis

Cost is defined as all money spent by each patient with suspected or known SCAD, consisting of the cost of each hospitalization for tests and treatment, as well as the cost of medication after discharge. It is worth mentioning that no distinction is made between health insurance expenditure and personal expenditure.

For the costs associated with the first hospital admission for suspected or known SCAD, we have collected information on the costs of SCAD-related tests and treatments of patients from the inpatient and outpatient medical records.

The calculation of downstream costs, i.e., costs related to suspected or known SCAD incurred outside the hospital, varies depending on their conditions. For those patients routinely attending follow-up appointments in our hospital, the medical record system was queried to obtain information about their relevant costs. For those patients who did not return to our hospital for follow-up but had specific medication prescriptions available, they were followed up by telephone to find out if they had complied with the medication instructions, and if they had been re-admitted to other hospitals for SCAD with the associated costs. As for patients who did not return to our hospital for follow-up and whose medication prescription could not be traced, the costs associated with SCAD were followed up by telephone.

Machine learning explainability and prognostic nomogram construction

To enhance the transparency and clinical interpretability of our prognostic model post-PSM, we implemented an Explainable AI (XAI) framework using Shapley Additive exPlanations (SHAP) [18, 19]. SHAP values were calculated to quantify the global and local contributions of each clinical feature (including group, age, gender, smoking, and CCI) to the predicted survival risk. Furthermore, SHAP dependence plots were utilized to visualize the continuous risk profiles of key covariates.

Using the feature importance rankings from our SHAP analysis, we selected the most predictive variables. We then built a simplified prognostic nomogram to predict 2-, 3-, and 5-year survival probabilities.

To evaluate the generalization and robustness of the nomogram (addressing model calibration), internal validation was performed using a bootstrapping approach with 1,000 resamples. Discriminative ability was assessed using Harrell’s concordance index (C-index), and the agreement between the nomogram-predicted and actual survival probabilities was evaluated via a 3-year calibration curve.

Data processing and analysis

Categorical variables were expressed as frequencies (percentages), and comparisons between groups were made using the chi-square test. Two independent samples t-test was used to compare continuous variables between groups. Survival curves were compared using the Kaplan-Meier method and the log-rank test. The Cox proportional risk regression model was used to estimate the hazard ratio (HR) and its 95% confidence interval (CI) for the primary endpoint between the two groups. Subsequently, a SHAP framework was applied to evaluate feature importance and model interpretability. A prognostic nomogram was constructed and internally validated via a bootstrapping approach with 1,000 resamples.

Raw data were analyzed using SPSS software (IBM SPSS Statistics 26 system, Chicago, IL, USA) and R software (version 4.5.2; R Foundation for Statistical Computing, Vienna, Austria) with the rms, kernelshap, and shapviz packages. All applicable tests were two-tailed, and P < 0.05 was considered statistically significant. Additionally, a post-hoc power analysis was conducted using the powerSurvEpi package in R to confirm sample size adequacy. Given the matched sample size of 260, an observed hazard ratio of 0.27, and an alpha level of 0.05, the statistical power was 86.67%, indicating that the study had sufficient power to detect the observed differences.

Results

The matched CAG and MPI groups each comprised 130 patients. As summarized in Table 1, no significant differences were observed in age, gender, CCI, or co-existing diseases between the two matched groups.

Table 1.

Demographic characteristics of the study population after matching

	Matched CAG group	Matched MPI group	P-value
Number	130	130
Mean age	52.7 ± 6.5	52.5 ± 12.0	0.85
Men, n (%)	83 (63.8)	75 (57.7)	0.31
CCI, n (%)			0.60
0	84 (64.6)	84 (64.6)
1	45 (34.6)	46 (35.4)
2+	1 (0.8)	0 (0.0)
Co-existing diseases
Hypertension, n (%)	49 (37.7)	50 (38.5)	0.90
Hyperlipidemia, n (%)	40 (30.8)	50 (38.5)	0.19
Diabetes, n (%)	45 (34.6)	43 (33.1)	0.79
Smoking, n (%)	34 (26.2)	33 (25.4)	0.89
Alcohol, n (%)	24 (18.5)	23 (17.7)	0.87
Results of CAG test
Negative	74 (56.9%)	-
<50%	34 (26.2%)	-
<75%	20 (15.4%)	-
≥ 75%	2 (1.5%)	-
Results of MPI test	44	130
Negative	21 (47.7%)	86 (66.2%)
Mild	9 (20.5%)	26 (20.0%)
Moderate	14 (31.8%)	18 (13.8%)
Severe	-	-

Matched CAG/MPI group = Groupings after 1:1 propensity score matching

CAG (Coronary angiography); MPI (Myocardial perfusion imaging); CCI (Charlson’s Comorbidity Index)

As shown in Table 1, nearly 56.9% of patients in the CAG group had negative CAG results (n = 74), 26.2% of patients had coronary stenosis less than 50% (n = 34), 15.4% of patients had stenosis less than 75% (n = 20), and only 1.5% had stenosis more than 75% (n = 2), suggesting more than half of patients are not really necessary to perform CAG. Of note, 33.8% of patients in the CAG group had concomitant MPI findings (n = 44). Among them, 47.7% of patients indeed had no myocardial ischemia (n = 21), and 52.3% of patients just had mild to moderate ischemia (n = 23), as indicated by MPI results. In the MPI group, all patients did not receive CAG. 66.2% of patients had no myocardial ischemia (n = 86), and 33.8% of patients were found to have mild to moderate ischemia (n = 44). In addition, no severe ischemia was observed in this group. Thus, these results together reveal the importance of MPI prior to CAG regarding the selection of diagnostic procedures for patients with SCAD.

During the follow-up period, cardiac outcomes of both groups were summarized in Table 2. More revascularization procedures were performed in the CAG group than in the MPI group (12 vs. 5, P < 0.05). The MPI group had a significantly lower rate of revascularization than the CAG group (HR 0.28, 95% CI 0.10–0.83; P = 0.02). The overall number of MI was fewer in the MPI group than that in the CAG group (1 vs. 3; HR 0.34, 95% CI 0.04–3.33; P = 0.36), even though the risk of MI was not statistically different between the two matched groups. Moreover, there was only 1 case of death in the CAG group. In summary, adverse cardiac events in the MPI group were slightly lower than those in the CAG group (4.6% vs. 12.3%). However, there was no significant difference due to the small sample size.

Table 2.

Incidence densities and hazard ratios of adverse cardiac events

	Matched CAG group	Matched MPI group	P-value *
Event, n (%)	16 (12.3)	6 (4.6)	0.59
Text results, n (%, grades)	8 (65.0, Negative)	1 (16.7, Negative)
	2 (12.5, < 50%)	5 (83.3, Moderate)
	6 (37.5, < 75%)	-
Men, n (%)	11 (8.5)	5 (3.8)	0.49
Mean Age	53.56 ± 3.25	55.5 ± 10.60	0.51
With co-existing diseases, n (%)	4 (3.1)	3 (2.3)	0.26
Follow-up, years	3.18 ± 1.19	3.42 ± 1.22	0.69
HR † (95% CI)	1 (reference)	0.27 (0.10–0.71)	0.008
Types of cardiac events
Revascularization, n (%)	12 (75.0)	5 (83.3)
HR ‡ (95% CI)	1 (reference)	0.28 (0.10–0.83)	0.02
Myocardial infarction, n (%)	3 (18.8)	1 (16.7)
HR ‡ (95% CI)	1(reference)	0.34(0.04–3.33)	0.36
All-cause death, n (%)	1 (6.3)	0
HR ‡ (95% CI)	None	None	None

^* Bold indicates statistical significance

^† HR = Hazard ratio of all cardiac events from the univariate Cox model

^‡ HR = Hazard ratio for each of three endpoints (revascularization, myocardial infarction, and all-cause death), from the univariate Cox model

Matched CAG/MPI group = Groupings after 1:1 propensity score matching

CAG (Coronary angiography); MPI (Myocardial perfusion imaging); CCI (Charlson’s Comorbidity Index)

Among patients with adverse cardiac events in the MPI group (n = 6), 5 patients (83.3%) were found to have moderate myocardial ischemia, and only 1 patient (16.7%) was found to show no significant abnormalities in myocardial perfusion. In contrast, among patients with adverse cardiac events in the CAG group (n = 16), 8 patients (65.0%) had negative CAG results, 2 patients (12.5%) had < 50% coronary stenosis, and 6 patients (37.5%) had 50%~75% coronary stenosis. No significant differences were seen in gender, age, co-existing diseases, and follow-up time for those patients between the two groups.

Although there was no statistical difference in the incidence of adverse cardiac events between the two groups, the cumulative risk of cardiac events in the MPI group was significantly lower than that in the CAG group (P < 0.001, log-rank sum test), as shown in Fig. 2a. The survival curve in the MPI group was consistently higher than that of the CAG group (Fig. 2b), and the MPI group had a 73% lower incidence risk of cardiac events than the CAG group (HR 0.27, 95% CI 0.10–0.71; P = 0.008).

Fig. 2.

Cumulative risk of cardiac events in patients receiving initial MPI and initial CAG. a. Cumulative risk curve; b. Survival curve. Abbreviation: matched CAG/MPI group, grouping after 1:1 propensity score matching; CAG, coronary angiography; MPI, myocardial perfusion imaging

In the SHAP summary plot (Fig. 3a), the initial diagnostic strategy (group) and age emerged as the two most dominant features driving the model’s risk allocation, whereas gender, smoking, and CCI demonstrated minimal baseline contributions. For example, the MPI strategy (high feature values, yellow) clustered tightly on the negative SHAP side (approximately − 0.55), signifying a lower risk profile. Conversely, the CAG strategy (low feature values, purple) shifted toward the positive SHAP side (up to + 0.7), indicating an elevated clinical risk. SHAP dependence plots also showed a clear linear, positive correlation between survival risk and older age (Fig. 3b), while the risk impact of CCI remained stable and near-zero across different scores (Fig. 3c).

Fig. 3.

Explanatory risk profiling of diagnostic strategies and clinical covariates on long-term survival. a. Global SHAP summary plot: Quantifies the individualized prognostic contributions of clinical features post- matching; CAG=0, MPI=1; b. SHAP dependence plot for Age; c. SHAP dependence plot for CCI

Based on these XAI findings, we developed a simplified prognostic nomogram incorporating only group and age (p = 2, Fig. 4a) to predict 2-, 3-, and 5-year survival probabilities. The model yielded a bootstrap-verified C-index of 0.655 (Fig. 4b). Internal validation via a 3-year calibration curve (n = 260, events d = 22, B = 1,000 bootstrap resamples) demonstrated high agreement between nomogram-predicted probabilities and actual Kaplan-Meier estimates, particularly within the high-survival spectrum (> 0.95). This internal step confirms that our simplified tool generalizes stably.

Fig. 4.

Construction and internal validation of the prognostic nomogram. a. The SHAP-derived parsimonious nomogram incorporating diagnostic strategy (group) and age to predict 2-, 3-, and 5-year survival probabilities; b. The 3-year calibration curve of the nomogram

At last, the average costs incurred by patients with suspected or known SCAD in both groups, including the first diagnostic and treatment cost and the downstream cost, were compared in Fig. 5. Firstly, the average first diagnostic and treatment cost per person in the MPI group and the CAG group was $717.72 and $1389.22, respectively. Of note, the MPI group had a 48.3% lower cost than the CAG group (P < 0.001). Secondly, the CAG group had an average downstream cost of $344.29 per patient per year, whereas the MPI group had an average downstream cost of $119.63 per patient per year, showing a 65.3% decrease in the MPI group compared to that in the CAG group (P < 0.001).

Fig. 5.

First costs and downstream costs of receiving initial MPI and initial CAG. a. Comparison of first cost; b. Comparison of downstream cost. Abbreviation: first cost, first-time diagnostic and treatment cost per person; downstream cost, downstream cost per year per person; matched CAG/MPI group, grouping after 1:1 propensity score matching; CAG, coronary angiography; MPI, myocardial perfusion imaging

Discussion

Our retrospective single-center clinical study was conducted at a general hospital in central China to compare effects of the initial MPI and CAG-guided strategies on clinical outcomes and costs in patients with suspected or known SCAD. In fact, the ideal controlled study should randomly assign patients to different diagnostic strategies, and then follow their clinical outcomes over a predetermined period [20, 21]. However, this type of study may be lengthy and expensive, and meanwhile induce some other practical problems. Therefore, we have chosen to conduct a retrospective controlled clinical study. It is worth mentioning that we apply PSM by logistic regression based on age, gender, and CCI to ensure the comparability of our study [22, 23].

We first evaluated the effect of MPI and CAG-guided strategies on clinical outcomes in patients with suspected or known SCAD. It turned out that patients in the MPI group had a lower incidence of adverse cardiac events than those in the CAG group. The relatively higher rate of revascularization in the CAG group may be related to unnecessary invasive angiography and PCI [24]. In a similar vein, one retrospective study in Taiwan have also revealed that an MPI-guided selective CAG strategy in patients with suspected SCAD was associated with lower rates of revascularization, MI, and improved survival compared with initial invasive CAG [13]. In the USA, 11.6% of PCI procedures have been considered inappropriate, with significant variations across hospitals (ranging from 0% to 55%), as revealed by an earlier report crossing 1,091 hospitals [25]. In addition, around 20% of PCI procedures are inappropriate among nearly 1 million cases of PCI each year in China [26]. This can kind of explain the higher rate of revascularization in the CAG group, as observed in our study. In conclusion, fewer cardiac events occurred in the MPI group than those in the CAG group, which may be attributed to microvascular damage caused by contrast medium or improper operation of CAG [27, 28].

While traditional Cox regression provides generalized hazard ratios, it often fails to depict how individual risk fluctuates at the bedside. By utilizing SHAP analysis, we successfully rank-ordered feature importance post-matching, demonstrating that the initial diagnostic modality (group) and age are the fundamental drivers of long-term prognosis, far outweighing traditional risks like smoking or CCI.

Interestingly, the SHAP analysis revealed a counter-intuitive phenomenon where a higher CCI score appeared to exert a marginal “protective” effect. First, looking at the numbers, the absolute SHAP values for CCI stayed very close to the zero-baseline (between − 0.02 and + 0.02). Compared to age and diagnostic group, CCI’s prognostic contribution in this matched cohort is very small. It likely represents statistical noise rather than a real clinical effect.

Second, from a clinical perspective, this finding reflects a typical “treatment paradox” and selection bias. Doctors usually monitor high-CCI patients more strictly and give them intensive, guideline-directed medical therapy before any invasive procedures. This extra care can partially reduce their long-term risk. Also, clinicians routinely avoid sending frail and highly vulnerable patients with elevated CCI scores for invasive procedures like CAG. Because of this triage process, the high-CCI individuals who actually remained in our cohort are essentially a comparatively stronger, resilient subgroup.

Driven by these XAI insights, we constructed a two-variable nomogram. The C-index of 0.655 shows moderate discrimination. We expected this, since it is inherently difficult to predict long-term survival in a matched cohort where most patients survive. However, the bootstrap calibration curve (B = 1,000) demonstrated remarkably high agreement between predicted and observed 3-year survival. This rigorous validation confirms that our model performs stably and does not overfit the limited sample (n = 260).

We then investigated the impact of MPI and CAG-guided strategies on patients’ costs, and showed that patients subjected to the initial MPI had a lower first diagnostic and treatment cost as well as a lower downstream cost than those patients with the initial CAG. Therefore, initial MPI prior to CAG offers promising cost-effectiveness for patients with suspected or known SCAD. On the one hand, many patients without myocardial ischemia based on MPI results have no further need for CAG [29], leading to some savings in unnecessary examinations. On the other hand, a considerable number of patients with normal myocardial blood supply or mild ischemia by MPI can be managed medically, thus avoiding the occurrence of coronary complications [30]. As a result, these patients have a lot of savings on CAG costs, the preoperative test costs, and hospitalization costs.

Our major finding is that the initial MPI strategy not only improves cardiac outcomes but also provides promising cost-effectiveness for patients in comparison to the initial CAG strategy, which is consistent with the previously published studies [13, 31, 32]. Of course, our study has several advantages. First of all, this study for the first time evaluates the clinical outcomes and associated costs of MPI versus CAG in SCAD patients from China. Secondly, patients with adverse cardiac events within 90 days were excluded from the follow-up analyses in this study, thus ensuring that long-term clinical outcomes were assessed. Finally, we estimated the risk of cardiac events in the two groups after propensity-score matching, which basically ruled out different clinical outcomes due to the varying severity of the disease. It is worth noting that while this study was unable to prove causality, the observed association is consistent with the notion that the initial use of MPI may avoid unnecessary revascularization of physiologically insignificant coronary stenosis [33, 34]. Nevertheless, the results from this study are still generalizable given the fact that the costs of tests across general hospitals in China vary very little despite that it is a single-center, non-prospective clinical study.

We recognize several limitations in this study. (1) Our study is primarily limited by its single-center, retrospective design and the small matched sample size. Although we implemented robust internal validation to verify the nomogram’s calibration and generalizability, external validation in independent, multi-center cohorts remains imperative to confirm these findings before widespread clinical application. (2) There was limited adjustment for potential confounding factors, such as symptom severity and pre-test probability, as specific laboratory test results and detailed clinical presentations were not fully captured in our dataset. (3) Some follow-up cost data relied on the memory of patients and their families, introducing a potential risk of recall bias. (4) Data regarding the baseline economic situation of the patients were not collected. (5) SCAD patients who underwent Coronary CT Angiography (CCTA) were not included, even though CCTA is commonly used for CAD screening in China [35].

Current clinical guidelines strongly recommend non-invasive imaging pathways, such as MPI and Coronary Computed Tomography Angiography (CCTA), prior to invasive CAG to optimize clinical decision-making [8]. Therefore, future prospective research should further investigate the long-term clinical benefits and cost-effectiveness of routinely utilizing MPI alongside CCTA before CAG. This approach would better reflect real-world clinical practice and validate the patient management pathways recommended by current guidelines. To support these future clinical endeavors, researchers should conduct multicenter studies to expand datasets and integrate them with XAI analyses. This combination can help avoid false causality caused by data sample limitations and bias [18, 19]. Given that diagnostic accuracy directly influences cost-effectiveness and downstream clinical decisions, integrating XAI into imaging interpretation is expected to become increasingly prominent. Ultimately, such methodological advancements will enhance decision-making transparency and further validate the patient management pathways recommended by current guidelines.

Conclusion

In conclusion, our findings suggest that an MPI-first strategy is associated with fewer cardiac events and lower costs compared with an initial CAG strategy.

Abbreviations

SCAD

Stable coronary artery disease

CAG

Coronary angiography

SPECT

Single-photon emission computed tomographic

MPI

Myocardial perfusion imaging

PCI

Percutaneous coronary intervention

MI

Myocardial infarction

ACS

Acute coronary syndromes

CCI

Charlson’s Comorbidity Index

HR

Hazard ratio

CI

Confidence interval

CCTA

Coronary CT Angiography

XAI

Explainable artificial intelligence

SHAP

Shapley Additive exPlanations

C-index

Harrell’s concordance index

Author contributions

All authors contributed to the study conception and design. Material preparation, data collection and analysis were performed by LY, YJ, YH, SL. The first draft of the manuscript was written by LY and all authors commented on previous versions of the manuscript. All authors read and approved the final manuscript.

Funding

This work was supported by grant No. U22A6008 from the National Natural Science Foundation of China and Shanxi Province Higher Education “Billion Project” Science and Technology Guidance Project.

Data availability

The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

This study was approved by the Ethics Committee of First Hospital of Shanxi Medical University (approval number K-128). Given the retrospective design, the requirement for informed consent was waived. To maintain anonymity, personal identifiers were not included in the data. The research was conducted in accordance with the Declaration of Helsinki.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.