Development and validation of a simple risk score model for predicting 3-year major adverse cardiovascular events in patients with coronary artery disease: a retrospective cohort study Authors: Mengxian Song, Liangcheng Xu, Delu Yin, Chengbo Sun Published: 2025 ABSTRACT Abstract AbstractObjectivesTo develop and validate a simple risk score for predicting major adverse cardiovascular events (MACE) in coronary artery disease (CAD) patients using routinely available clinical variables.DesignThis was a cohort study with retrospective analysis of prospectively collected data.SettingThis study was conducted at a tertiary care centre in China.ParticipantsThis cohort study included 7182 CAD patients, randomly divided into training dataset and testing dataset in a ratio of 3:1.Primary and secondary outcome measuresThe primary outcome was a composite of MACE (cardiovascular death, non-fatal MI, stroke and revascularisation). A Cox regression model was developed on a training set to identify independent predictors. Variables were assigned points based on their β-coefficients to construct a risk score. The model was validated on a testing set. Discrimination was assessed using the concordance index (C-index) and area under the receiver operating characteristic (ROC) curve (AUC). Risk groups were defined according to the total score.ResultsOver a median follow-up of 27.3 months, 487 (6.8%) MACE events occurred. Six independent predictors were identified and included in the score: age ≥65 years (three points), NT-proBNP ≥125 pg/mL (four points), HbA1c ≥7% (3 points), elevated serum creatinine (>106 µmol/L for male or >97 µmol/L for female, 4 points), low-density lipoprotein-cholesterol (LDL-C) ≥1.8 mmol/L (two points), and cardiac troponin T (cTnT) ≥0.15 ng/mL (four points). The score stratified patients into low- (0–4 points), middle- (5–9 points), and high-risk (10–14 points) groups. In the testing set, the middle- and high-risk groups had significantly increased MACE risk compared with the low-risk group (HR 1.54, 95% CI 1.03 to 2.29; HR 2.70, 95% CI 1.78 to 4.08, respectively). The model showed consistent discrimination in both training (C-index = 0.726, AUC = 0.728) and testing sets (C-index = 0.702, AUC = 0.705).ConclusionA simple risk score comprising six clinical variables effectively stratified CAD patients into distinct MACE risk categories. This tool may aid in clinical decision-making and resource prioritisation in secondary prevention, pending external validation.Keywords: Coronary heart disease, CARDIOLOGY, Cardiovascular Disease STRENGTHS AND LIMITATIONS OF THIS STUDY The prediction model is highly practical, derived from six routinely available clinical variables, facilitating potential adoption in diverse clinical settings. The prediction model is highly practical, derived from six routinely available clinical variables, facilitating potential adoption in diverse clinical settings. This study benefits from a large, prospective cohort, with a sample size far exceeding the minimum requirement for prediction model development, ensuring robust estimates. This study benefits from a large, prospective cohort, with a sample size far exceeding the minimum requirement for prediction model development, ensuring robust estimates. The model demonstrated consistent and acceptable discrimination and calibration in both the derivation and validation sets, indicating minimal overfitting. The model demonstrated consistent and acceptable discrimination and calibration in both the derivation and validation sets, indicating minimal overfitting. This was a single-centre study, and external validation in independent, multicentre populations is required to confirm generalisability. This was a single-centre study, and external validation in independent, multicentre populations is required to confirm generalisability. The pursuit of simplicity involved dichotomising continuous predictors and excluding other potential risk factors, which may limit the ceiling performance of the model. The pursuit of simplicity involved dichotomising continuous predictors and excluding other potential risk factors, which may limit the ceiling performance of the model. Introduction Coronary artery disease (CAD) remains a leading cause of global morbidity and mortality.1 Patients with established CAD are at a significantly elevated risk for subsequent major adverse cardiovascular events (MACE), including cardiovascular death, non-fatal myocardial infarction and stroke.2 The accurate identification of patients at the highest risk is fundamental to optimising secondary prevention strategies, intensifying therapies and improving long-term outcomes.3 Several risk prediction models for cardiovascular events exist, such as the GRACE score4 5 for acute coronary syndromes and the SMART risk score6 7 for secondary prevention. While valuable, these models often require specific or numerous variables, some of which may not be routinely available in all primary or secondary care settings, potentially limiting their widespread applicability for rapid, routine risk assessment.8 Concurrently, the evolving field of cardiovascular prognostics has seen the emergence of machine-learning-based models (eg, ensemble methods, neural networks) trained on high-dimensional data, which have shown promise in improving predictive accuracy for MACE.9 10 However, these sophisticated models often face limitations regarding interpretability, clinical integration and susceptibility to calibration drift over time or across different populations, underscoring the persistent need for simple, robust and transparent tools. Therefore, a pragmatic gap exists for a simple, integer-based risk score, derived from universally available clinical data, specifically tailored for the stable CAD population to facilitate routine risk stratification and guide management intensity in diverse clinical environments. There is a clinical need for a simple, pragmatic risk tool that utilises readily available clinical and laboratory data to stratify MACE risk in the broader CAD population. Such a tool could facilitate timely clinical decisions by general cardiologists and primary care physicians, enabling them to prioritise high-risk patients for more frequent follow-up, aggressive risk factor modification or advanced therapeutic interventions. Therefore, this study aimed to develop and validate a simple integer-based risk score to predict future MACE in patients with CAD, using non-invasive and commonly measured clinical variables. Methods Study design and participants This study was a retrospective analysis of prospectively collected data, conducted at the First People’s Hospital of Lianyungang. We included adult patients (aged ≥18 years) with a confirmed diagnosis of CAD (defined as a history of myocardial infarction, coronary revascularisation or angiographic stenosis >50%). The exclusion criteria included severe non-cardiovascular life-limiting diseases (active, metastatic or haematologic malignancies undergoing active therapy; end-stage liver disease; end-stage renal disease on chronic dialysis; severe haematologic disorder; severe neurodegenerative diseases or major physical disabilities precluding follow-up; advanced, untreated systemic autoimmune diseases or active severe infections), incomplete baseline data and loss to follow-up. A total of 7182 eligible patients were enrolled. To develop and validate the prediction model, the entire cohort was randomly split into two independent sets using a computer-generated random number sequence. To ensure reproducibility, a fixed random seed was set prior to the split. The training dataset (n = 5027, approximately 70%) was used for model development and internal calibration. The testing dataset (n = 2155, approximately 30%) was reserved solely for the final evaluation of model discrimination and calibration, providing an unbiased performance estimate. Baseline data were collected at the beginning of follow-up. Ethics statement This study was conducted in accordance with the Declaration of Helsinki and was approved by the Medical Ethics Committee of the First People’s Hospital of Lianyungang (approval number: KY-20250718001–01). The requirement for written informed consent was waived due to the retrospective nature of the study, as determined by the approving ethics committee. Sample size The sample size for model development was estimated a priori to minimise overfitting. Based on recommendations by Riley et al11 for developing clinical prediction models with time-to-event outcomes, the calculation targeted a shrinkage factor of 0.9. Assuming 12 candidate predictor parameters and a conservative anticipated Cox-Snell R2 (R2cs) of 0.05, the minimum required sample size was calculated using the formula: n = P/[ (S - 1) × ln(1 - R2cs / S) ], where P is the number of parameters and S is the shrinkage factor. This yielded a minimum of 2098 participants. The final cohort of 7182 patients far exceeded this requirement, ensuring sufficient statistical power for robust model development and minimising the potential for overfitting. Variables measured Baseline data were collected through electronic health records and included demographic characteristics, clinical measurements, laboratory parameters and medicine treatment. All variables were selected based on clinical relevance, routine availability in primary care settings and established associations with cardiovascular outcomes. Outcome The primary outcome was the first occurrence of a composite MACE endpoint, which included cardiovascular death, non-fatal myocardial infarction and unplanned coronary revascularisation. Cardiovascular death was defined as death attributable to ischaemic heart disease, heart failure or sudden cardiac death. Non-fatal myocardial infarction was diagnosed according to the Fourth Universal Definition of Myocardial Infarction. Revascularisation included percutaneous coronary intervention or coronary artery bypass grafting performed for clinical indications during follow-up. Patient and public involvement No patients or members of the public were involved in the design, conduct, reporting or dissemination plans of this research. The research question and outcome measures were developed based on clinical relevance and gaps in the existing literature, without direct input from patients or the public. Due to the retrospective analysis design of this study using clinical data, involvement in recruitment or conduct was not applicable. As such, the burden of participation and plans for disseminating results were not assessed in partnership with patients or public communities. Statistical analysis All baseline characteristics were considered in the model development process. These variables are routinely collected as part of the standard clinical assessment for CAD patients at our tertiary center. Consequently, these variables were recorded for all enrolled patients as part of standard clinical care. After verifying data completeness through cross-referencing with source documents, we confirmed there were no missing values for any of these candidate predictors in the final analytic cohort. Therefore, a complete-case analysis was performed, and no data imputation was required or applied. In the training set, univariate and multivariate Cox proportional hazards regression analyses were performed to identify independent predictors of MACE. Variables with a P value <0.05 in the multivariate model were retained. A risk score was created by multiplying β-coefficient by 10 and rounding to the nearest integer. The predictive accuracy of the Cox proportional hazards model was assessed by examining its calibration. We generated a calibration curve using a bootstrap resampling method with 1000 repetitions to obtain relatively unbiased estimates. The plot compares the predicted probability of survival against the actual observed survival probability. A perfectly calibrated model would lie along the 45-degree reference line. The total risk score for each patient was calculated by summing the points for each predictor present. Patients were categorised into three risk groups (low, middle, high) based on their total score. Model performance was evaluated by assessing discrimination (C-index and time-dependent AUC) and calibration (Hosmer-Lemeshow goodness-of-fit test) in both the training and testing datasets. All analyses were performed using R (version 4.5.1) and RStudio (version 2025.09.0+387) with two-sided statistical significance set at p < 0.05. Results Baseline characteristics From November 2022 to May 2025, a total of 7182 patients with established coronary artery disease were enrolled in this prospective cohort (figure 1). The median follow-up duration was 27.3 months. Over this period, 487 patients experienced the primary composite endpoint of MACE, corresponding to a cumulative incidence of 6.8%. The training set (n = 5027) and testing set (n = 2155) were well-balanced in terms of all baseline demographic, clinical and laboratory characteristics, confirming the success of the random split (table 1). Flowchart of the study. Table 1Baseline characteristicsOverallTraining datasetTesting datasetP valuen718250272155CharacteristicsAge (y)65.00 (58.00, 70.00)65.00 (58.00, 70.00)65.00 (57.00, 71.00)0.434Male5784 (80.5)4056 (80.7)1728 (80.2)0.648BMI (kg/m2)24.91 (23.03, 27.04)24.91 (23.03, 27.04)24.86 (23.03, 27.04)0.940Current smoking3426 (47.7)2392 (47.6)1034 (48.0)0.776Hypertension4754 (66.2)3298 (65.6)1456 (67.6)0.114Diabetes2741 (38.2)1906 (37.9)835 (38.7)0.523Chronic kidney disease798 (11.1)548 (10.9)250 (11.6)0.410Stroke history497 (6.9)342 (6.8)155 (7.2)0.586Hyperlipidaemia1154 (16.1)833 (16.6)321 (14.9)0.083Old myocardial infarction929 (12.9)650 (12.9)279 (12.9)1Laboratory examinationcTnT ≥ 0.15 ng/mL2261 (31.5)1584 (31.5)677 (31.4)0.959Haemoglobin (g/L)137.00 (127.00, 146.00)137.00 (127.00, 146.00)137.00 (126.00, 146.00)0.714Platelet count (× 109 /L)201.00 (168.00, 238.00)201.00 (168.00, 237.00)202.00 (168.00, 239.00)0.752MPV (fL)10.60 (10.00, 11.20)10.50 (10.00, 11.20)10.60 (10.00, 11.20)0.438PCT (%)0.21 (0.18, 0.25)0.21 (0.18, 0.25)0.21 (0.18, 0.25)0.864PDW (%)12.20 (11.00, 13.70)12.20 (11.00, 13.70)12.20 (11.00, 13.80)0.828NT-proBNP (ng/L)111.00 (47.60, 298.00)110.00 (47.15, 293.90)114.90 (49.80, 309.50)0.104HbA1c (%)6.01 (5.61, 6.86)6.01 (5.61, 6.86)6.01 (5.60, 6.86)0.884D-dimer (mg/L)0.27 (0.19, 0.45)0.27 (0.19, 0.45)0.27 (0.19, 0.46)0.779ALT (U/L)20.30 (14.40, 30.00)20.10 (14.40, 30.00)20.60 (14.60, 30.00)0.375Serum creatinine (μmol/L)80.00 (69.00, 93.00)80.00 (69.00, 93.00)80.00 (70.00, 93.00)0.630eGFR (mL/min/1.73 m2)85.24 (71.58, 93.78)85.35 (71.66, 93.88)84.96 (71.45, 93.55)0.589TC (mmol/L)3.53 (2.98, 4.27)3.52 (2.98, 4.27)3.56 (2.99, 4.28)0.792TG (mmol/L)1.51 (1.06, 2.20)1.50 (1.07, 2.20)1.53 (1.06, 2.21)0.533LDL-C (mmol/L)1.69 (1.25, 2.32)1.69 (1.25, 2.33)1.69 (1.26, 2.31)0.946DrugsStatin6655 (92.7)4660 (92.7)1995 (92.6)0.892β-blocker4776 (66.5)3342 (66.5)1434 (66.5)0.981ACEI/ARB3744 (52.1)2601 (51.7)1143 (53.0)0.325PPI2616 (36.4)1802 (35.8)814 (37.8)0.127Aspirin/indobufen6016 (83.8)4200 (83.5)1816 (84.3)0.469Clopidogrel/ticagrelor5624 (78.3)3951 (78.6)1673 (77.6)0.381Values presented are n (%) or median(inter-quartile range).ACEI, angiotensin-converting enzyme inhibitor; ARB, angiotensin receptor blocker; BMI, body mass index; cTnT, cardiac troponin T; eGFR, estimate glomerular filtration rate; HbA1c, glycosylated haemoglobin A1c; LDL-C, low-density lipoprotein-cholesterol; MPV, mean platelet volume; NT-proBNP, N-terminal pro-B-type natriuretic peptide; PCT, platelet crit; PDW, platelet distribution width; P-LCR, platelet larger cell ratio; PPI, proton pump inhibitor; TC, total cholesterol; TG, triglyceride. Baseline characteristics Values presented are n (%) or median(inter-quartile range). ACEI, angiotensin-converting enzyme inhibitor; ARB, angiotensin receptor blocker; BMI, body mass index; cTnT, cardiac troponin T; eGFR, estimate glomerular filtration rate; HbA1c, glycosylated haemoglobin A1c; LDL-C, low-density lipoprotein-cholesterol; MPV, mean platelet volume; NT-proBNP, N-terminal pro-B-type natriuretic peptide; PCT, platelet crit; PDW, platelet distribution width; P-LCR, platelet larger cell ratio; PPI, proton pump inhibitor; TC, total cholesterol; TG, triglyceride. Predictors of MACE and risk score development In the training set, univariable Cox regression identified several candidate variables associated with MACE. Subsequent multivariable Cox proportional hazards analysis, after adjusting for potential confounders, revealed six independent predictors of 3 year MACE. These predictors, along with their HRs, β-coefficients and assigned points based on the scoring system, were as shown in table 2. The final integer-based risk score, constructed by summing the points for each predictor present, yielded a theoretical range from 0 to 20 points. A practical nomogram was developed to facilitate easy calculation of individual patient risk (figure 2). Table 2Risk score assignment according to multivariable Cox regression modelVariablesLevelHR (95% CI)P valueβ coefficientScoresAge ≥65 yearsNo–––0Yes1.42 (1.14 to 1.77)0.00210.34763NT-proBNP ≥125 pg/mLNo–––0Yes1.55 (1.23 to 1.94)0.00020.43514HbA1c ≥7 g/LNo–––0Yes1.29 (1.02 to 1.63)0.03280.25523Scr >106 µmol/L for male or >97 µmol/L for femaleNo–––0Yes1.49 (1.12 to 1.99)0.00690.39914LDL-C ≥1.8 mmol/LNo–––0Yes1.27 (1.02 to 1.56)0.03640.23052cTnT ≥0.15 ng/mLNo–––0Yes1.52 (1.22 to 1.90)0.00050.39494CI, confidence interval; cTnT, cardiac troponin T; HbA1c, glycosylated haemoglobin A1c;; HR, hazard ratio; LDL-C, low-density lipoprotein-cholesterol; NT-proBNP, N-terminal pro-B-type natriuretic peptide; Scr, serum creatinine. Risk score assignment according to multivariable Cox regression model CI, confidence interval; cTnT, cardiac troponin T; HbA1c, glycosylated haemoglobin A1c;; HR, hazard ratio; LDL-C, low-density lipoprotein-cholesterol; NT-proBNP, N-terminal pro-B-type natriuretic peptide; Scr, serum creatinine. Nomogram for predicting 3-year major adverse cardiovascular events (MACE) risk in patients with coronary artery disease. Instructions for using the nomogram: First, draw a straight upward line to the 'Points’ axis to determine the points contributed by each variable. For example, age ≥65 years contributes three points, while age <65 years contributes 0 points. Second, repeat this process for all six variables: NT-proBNP ≥125 pg/mL (four points), HbA1c ≥7% (3 points), LDL ≥1.8 mmol/L (two points), cTnT ≥0.15 ng/mL (four points) and elevated creatinine (four points). Third, sum all points from the six domains to obtain the total points. Finally, draw a straight downward line from the 'Total Points’ axis to the 'Estimated risk of MACE at 3 years’ axis to obtain the predicted risk. cTnT, cardiac troponin T; HbA1c, glycosylated haemoglobin A1c; LDL-C, low-density lipoprotein-cholesterol; NT-proBNP, N-terminal pro-B-type natriuretic peptide; Scr, serum creatinine. Risk stratification and model performance Based on the distribution of scores and clinical judgement, patients were stratified into three distinct risk categories for MACE, low risk (0–4 points), middle risk (5–9 points) and high risk (10–20 points). Kaplan-Meier survival curves demonstrated excellent separation among these three groups in both the training and testing datasets, with log-rank tests confirming highly significant differences (figure 3). Kaplan-Meier curves of MACE in CAD patients categorised by score model at 3 years. (A) Kaplan-Meier curves of the training dataset. (B) Kaplan-Meier curves of the testing dataset. CAD, coronary artery disease; MACE, major adverse cardiovascular events. In the testing set, the model’s predictive power was robustly validated. Compared with the low-risk group, patients in the middle-risk group had a significantly increased hazard for MACE (HR: 1.54, 95% CI: 1.03 to 2.29). The high-risk group exhibited an even more pronounced risk, with a HR of 2.70 (95% CI: 1.78 to 4.08) relative to the low-risk group (table 3). This represents a near threefold gradient in risk from the lowest to the highest category. Table 3MACE risk by risk score stratification in the training and testing datasetsScore rangeRisk levelTraining datasetTesting datasetEvents rate (%)Hr (95% CI)P valueEvents rate (%)HR (95% CI)P value0~4Low4.28––4.70––5~9Middle6.801.70 (1.30 to 2.22)0.00016.911.54 (1.03 to 2.29)0.033810~14High12.243.30 (2.52 to 4.33)< 0.000110.962.70 (1.78 to 4.08)<0.0001CI, confidence interval; HR, hazard ratio; MACE, major adverse cardiovascular events. MACE risk by risk score stratification in the training and testing datasets CI, confidence interval; HR, hazard ratio; MACE, major adverse cardiovascular events. The model demonstrated consistent and acceptable discrimination across both cohorts. In the training set, the Concordance index (C-index) was 0.726, and the area under the receiver operating characteristic curve (AUC) for predicting 3-year MACE was 0.728 (95% CI: 0.696 to 0.759). The performance was well-preserved in the testing set, with a C-index of 0.702 and an AUC of 0.705 (95% CI: 0.675 to 0.745), indicating minimal overfitting and good generalisability. Furthermore, the calibration curve (online supplemental Figure S1) showed strong agreement between predicted and observed 3-year event-free survival probabilities, with the plotted line closely adhering to the ideal 45-degree reference line, confirming the model’s accuracy in estimating absolute risk. Discussion In this large prospective cohort study of 7182 CAD patients, we developed and validated a simple integer-based risk score for predicting 3-year MACE. The score, comprising six readily available clinical variables, effectively stratified patients into three distinct risk categories with a nearly threefold gradient in risk between the highest and lowest groups in the validation cohort. This risk stratification tool demonstrates several important clinical and methodological advantages that warrant detailed discussion. The variables selected in our final model represent fundamental pathophysiological pathways in cardiovascular disease progression. Age serves as a composite cause of macrovascular and microvascular pathologies and cumulative vascular burden.12 NT-proBNP, a well-established marker of myocardial wall stress, reflects underlying ventricular dysfunction and neurohormonal activation, which are strong predictors of adverse outcomes in CAD populations.13 Similarly, cardiac troponin T (cTnT) elevation serves as a critical biomarker for acute coronary syndrome and is associated with clinical prognosis.14 Even minimal elevations of cTnT below the diagnostic threshold for acute myocardial infarction indicate ongoing subclinical myocardial injury and have been consistently associated with increased cardiovascular risk in stable CAD patients.15 The inclusion of HbA1c as a dichotomised variable (≥7%) acknowledges the continuous relationship between glycaemic control and cardiovascular risk, while maintaining clinical practicality. This approach aligns with current guideline recommendations that emphasise the importance of glycaemic control in secondary prevention.16 Renal dysfunction, as reflected by elevated creatinine levels using sex-specific thresholds, incorporates the well-established cardiovascular-renal axis into risk assessment. Chronic kidney disease not only shares common risk factors with CAD but also exacerbates cardiovascular risk through multiple mechanisms including fluid overload, electrolyte disturbances and accelerated atherosclerosis.17 Notably, while contemporary guidelines recommend a more stringent LDL-C goal of <1.4 mmol/L for secondary prevention of chronic and acute coronary syndromes, and even suggest consideration of a goal <1.0 mmol/L in cases of recurrent events,3 18 our selection of the 1.8 mmol/L threshold was statistically driven which was identified through Cox regression model as the value that provided the maximal separation in MACE risk between the two groups within our cohort. This finding reinforces the ‘lower is better’ principle for lipid management in high-risk CAD patients and suggests that even modest elevations within the ‘controlled’ range may confer residual risk.19 20 The persistence of this association underscores the importance of aggressive lipid-lowering strategies in secondary prevention. Our model demonstrated consistent discriminative ability across both training (C-index = 0.726) and testing (C-index = 0.702) datasets. While this level of discrimination may appear modest compared with some complex prediction models, it is important to contextualise these values within the framework of practical clinical utility. For population-level risk stratification and resource allocation decisions—the primary intended use of this tool—a C-index of 0.70–0.75 provides meaningful risk differentiation.21 22 Furthermore, the minimal performance degradation between development and validation phases (ΔC-index = 0.024) indicates excellent model stability and minimal overfitting, a common challenge in clinical prediction rule development. The choice to dichotomise continuous variables, while resulting in some loss of statistical information, enhances the tool’s clinical applicability. Busy clinicians often prefer straightforward decision thresholds over complex continuous algorithms. This approach aligns with successful previous risk scores such as the CHA2DS2-VASc for atrial fibrillation and the HAS-BLED for bleeding risk, which similarly employ categorical variables for practical implementation.23 24 Our simple six-variable score compares favourably with existing risk prediction tools in several aspects. Compared with the GRACE score, which requires acute phase parameters and complex calculations,4 our model utilises stable, routinely available measures versus the SMART score, which incorporates numerous variables including specific medication details;6 our tool offers greater simplicity while maintaining reasonable predictive performance. This balance between simplicity and accuracy is particularly valuable in resource-limited settings or for rapid bedside assessment. The contemporary landscape of cardiovascular risk prediction is increasingly enriched by sophisticated methodologies, including machine learning algorithms and neural networks, which can model complex, non-linear interactions between a multitude of variables and often achieve superior discriminative performance in research settings. While these advanced techniques represent an important frontier for discovery, their utility in routine clinical practice can be constrained by the ‘black-box’ nature of their predictions, the requirement for extensive and often non-routine data inputs and the need for specialised computational infrastructure for model deployment and updating. In contrast, the integer-based risk score presented here prioritises interpretability, accessibility and immediate clinical actionability. Its strength lies in its composition from six routinely measured variables, enabling calculation at the bedside or during a clinic visit without digital aids. This simplicity facilitates seamless integration into existing clinical workflows, promotes transparent risk communication between clinicians and patients and ensures equitable applicability across diverse healthcare settings, including those with limited resources. Therefore, while the pursuit of more complex models remains vital for refining our pathophysiological understanding, we posit that there is a sustained and critical role for simple, validated tools like ours in bridging the gap between evidence-based risk stratification and real-world clinical implementation at the point of care. The practical implications of our risk score are substantial. In primary care settings, where time constraints often limit comprehensive risk assessment, this tool could facilitate rapid identification of high-risk patients requiring more intensive management. The clear stratification into three risk categories (low, middle, high) provides actionable clinical thresholds that align with current stepped-care approaches to cardiovascular prevention. In practice, patients classified as low risk (0–4 points) may continue with guideline-directed medical therapy and annual follow-up, as their estimated 3-year MACE risk remains modest. For middle-risk patients (5–9 points), our score could prompt a more proactive management approach, including closer monitoring (eg, biannual visits), escalation of lipid-lowering therapy to achieve lower LDL-C targets and more structured lifestyle or glycaemic control programmes. Most critically, high-risk patients (10–20 points), comprising those with multiple risk determinants, should trigger a comprehensive, multidisciplinary evaluation. This may include consideration of advanced imaging (eg, coronary computed tomography angiography or perfusion imaging) to further characterise plaque burden and ischaemia, intensified pharmacotherapy (eg, combination lipid-lowering, higher-dose antiplatelet regimens where indicated) and earlier referral to cardiac rehabilitation. Importantly, this simple score could also serve as a tool for shared decision-making, helping clinicians visually communicate individualised risk to patients and thereby improve adherence to prescribed preventive strategies. Future implementation studies should evaluate whether the systematic use of this risk score in outpatient cardiology or primary care settings leads to meaningful improvements in treatment escalation, patient outcomes and healthcare resource allocation. Limitations Our study has several limitations. (a) It was a single-centre study, and external validation in diverse populations is needed to confirm generalisability. (b) While we intentionally prioritised simplicity, the exclusion of potentially informative variables such as imaging parameters, functional capacity measures or novel biomarkers inevitably limits the model’s ceiling performance. (c) The categorisation of continuous variables, while creating a user-friendly tool, may lead to some loss of statistical information. (d) The performance of this model in patients with advanced non-cardiac illnesses is unknown and that its use should be restricted to patients where cardiovascular prognosis is a primary management focus. (e) We were unable to account for important post-discharge factors such as medication adherence and lifestyle modifications, which may substantially modulate MACE risk. (f) Despite adjusting for established clinical variables, residual confounding may persist due to unmeasured factors, including inflammatory biomarkers (eg, high-sensitivity C-reactive protein) and detailed imaging characteristics (eg, coronary plaque burden or complexity. (g) While the model aims to predict 3-year MACE, the median follow-up of 27.3 months may not fully capture later events, potentially affecting the accuracy of long-term risk estimation. Future iterations could explore hybrid approaches that maintain simplicity while incorporating selected high-yield additional parameters. Conclusion We have developed a simple, clinically practical risk score that effectively stratifies CAD patients into low-, middle- and high risk for future 3-year MACE. This tool may assist clinicians in tailoring the intensity of follow-up and therapeutic interventions based on an individual’s calculated risk, thereby potentially improving resource allocation and patient outcomes in the management of CAD. Supplementary material Acknowledgements We are deeply grateful to the participants and their relatives for their dedication and support. Footnotes Funding: This work was supported by a grant to DY from the Health and Family Planning Commission of Jiangsu Province (ZDB2020029) and to MS from Kangda College Scientific Research Development Fund Project (KD2025KYJJ122). Prepublication history and additional supplemental material for this paper are available online. To view these files, please visit the journal online (https://doi.org/10.1136/bmjopen-2025-113037). Provenance and peer review: Not commissioned; externally peer reviewed. Patient consent for publication: Not applicable. Ethics approval: This study involves human participants and was approved by the Medical Ethics Committee of the First People’s Hospital of Lianyungang (KY-20250718001-01). Participants gave informed consent to participate in the study before taking part. Data availability free text: The data and code that support the findings of this study are available from the corresponding author, Chengbo Sun, upon reasonable request. However, these data are not publicly available due to restrictions imposed by the ethics committee and institutional policies regarding patient confidentiality. Patient and public involvement: Patients and/or the public were not involved in the design, conduct, reporting or dissemination plans of this research. Data availability statement Data are available upon reasonable request. References Supplementary Materials Data Availability Statement Data are available upon reasonable request.