Systematic Review of Risk Prediction Models for Heart Failure in Patients with Coronary Heart Disease

JIANG Xiaorui¹,², YAN Yuyao², WEI Jingjing¹, QIAO Lijie¹, PENG Guangcao¹, ZHU Mingjun¹*

¹ Department of Cardiovascular Diseases, the First Affiliated Hospital of Henan University of Traditional Chinese Medicine, Zhengzhou 450000, China
² Henan University of Traditional Chinese Medicine, Zhengzhou 450000, China

Corresponding author: ZHU Mingjun, Professor/Doctoral Supervisor; E-mail: zhumingjun317@163.com

Abstract

Background Heart failure (HF) is a major chronic disease that poses a serious threat to global health, with coronary heart disease (CHD) being its most common etiology. Developing risk prediction models for HF in CHD patients based on relevant risk factors can help healthcare professionals identify high-risk populations early and implement timely interventions.

Objective To systematically evaluate risk prediction models for HF in Chinese CHD patients, providing a reference for the development, selection, and dissemination of relevant predictive models.

Methods Eight databases—CNKI, VIP, Wanfang Data, SinoMed, PubMed, Cochrane Library, Web of Science, and Embase—were searched for studies on risk prediction models for HF in Chinese CHD patients from inception to October 2024. Two reviewers independently screened literature and extracted data. The Prediction Model Risk of Bias Assessment Tool (PROBAST) was used to evaluate the risk of bias and applicability of included studies.

Results A total of 27 studies reporting 64 risk prediction models were included. The area under the receiver operating characteristic curve (AUC) ranged from 0.511 to 0.989, with 63 models achieving an AUC > 0.7, indicating good overall predictive performance. However, PROBAST assessment revealed that all 27 studies had a high risk of bias and low applicability. Key predictive factors included age, left ventricular ejection fraction, diabetes history, hypertension history, NT-proBNP (N-terminal pro-B-type natriuretic peptide), and Gensini score.

Conclusion The stability and external validity of existing risk prediction models for HF in Chinese CHD patients require further validation through prospective, large-scale studies. Future model development should strictly adhere to PROBAST guidelines to ensure the design and implementation of high-quality, generalizable predictive models.

Keywords Coronary heart disease; Heart failure; Risk prediction model; Systematic review

1.1 Literature Search

Eight databases were systematically searched: CNKI, Wanfang Data, VIP, SinoMed, PubMed, Cochrane Library, Web of Science, and Embase. The search covered studies from database inception to October 2024. A combination of subject headings and free-text terms was used. Chinese search terms included "heart failure," "coronary heart disease," "myocardial infarction," "acute coronary syndrome," "risk prediction," "prediction model," and "predictive factors." English search terms included "Coronary Heart Disease," "Myocardial Infarction," "Acute Coronary Syndrome," "Heart Failure," "cardiac failure," "Prediction Model," and "Risk Prediction." Search strategies for CNKI and PubMed are detailed in Table 1 [TABLE:1].

1.2 Inclusion and Exclusion Criteria

1.2.1 Inclusion Criteria: (1) Study population: Chinese patients with acute myocardial infarction (AMI) or CHD confirmed by coronary angiography or coronary CT angiography (CCTA); (2) Study objective: development or validation of risk prediction models for HF in CHD patients; (3) Outcome: confirmed HF diagnosis; (4) Models containing at least two variables; (5) Study design: cohort, case-control, or cross-sectional studies.

1.2.2 Exclusion Criteria: (1) Studies with incomplete data or unavailable full text; (2) Conference abstracts, animal experiments, reviews, or methodological articles; (3) Studies analyzing risk factors without constructing a prediction model; (4) Studies with sample size < 100.

1.3 Literature Screening and Data Extraction

EndNote X9 was used to remove duplicates. Two reviewers independently screened titles and abstracts, then reviewed full texts. Disagreements were resolved through discussion with a third reviewer. Data were extracted using the Critical Appraisal and Data Extraction for Systematic Reviews of Prediction Modelling Studies (CHARMS) checklist and recorded in Excel.

1.4 Risk of Bias and Applicability Assessment

Two reviewers independently assessed the risk of bias and applicability using the PROBAST tool, which evaluates 20 items across four domains (participants, predictors, outcome, statistical analysis) for bias risk and three domains (participants, predictors, outcome) for applicability. Each domain was rated as "high risk," "low risk," or "unclear."

2.1 Literature Retrieval

The initial search yielded 5,180 records, with 75 additional studies identified through other sources. After removing duplicates, 4,554 records remained. Title and abstract screening excluded 4,448 records, and full-text review excluded another 79, resulting in 27 included studies [12-38]. The literature screening flowchart is shown in Figure 1 [FIGURE:1].

2.2 Basic Characteristics of Included Studies

The 27 included studies were published between 2019 and 2024, comprising 22 Chinese and 5 English articles. Study designs included 3 prospective cohort studies [17,21-22], 19 retrospective cohort studies [12-14,16,18,20-31,33,35-38], and 5 case-control studies [15-16,19,32,34]. Data sources were clinical records and electronic medical databases, with maximum follow-up of 7.3 years and outcome event rates ranging from 3.54% to 48.94%. Basic characteristics are summarized in Table 2 [TABLE:2].

Nineteen studies [13-14,17-20,22-26,28,30-31,33-36,38] reported ethics committee approval, while eight [12,15-16,21,27,29,32,37] did not. Nine studies [13-14,16,20-22,25,28,37] converted continuous variables to categorical variables, six [12,23,30,32-33,36] partially converted them, and twelve [15,17-19,24,26-27,29,31,34-35,38] maintained continuous variables. Conversion methods included clinical significance [12,16,21,33], ROC curve optimal cutoffs [13,20,22,25,28], and population distribution characteristics [32,37]. Three studies [15,26,33] reported specific missing data and deleted cases with missing values, two [18,31] used imputation, and two [23-24] used combined multiple imputation and deletion methods without reporting specific missing sample sizes. Candidate variables ranged from 14 to 72, with total sample sizes of 120 to 44,772 (Table 3 [TABLE:3]).

2.3 Model Construction

The 27 studies reported 64 prediction models. Eighteen [13,16-20,22,25-26,28-30,32-35,37-38] used logistic regression (LR), while three [12,14,21] used Cox proportional hazards regression. Variable selection methods included univariate analysis [20,24,27,30,32,36], univariate followed by multivariate analysis [12,14-15,17,19,21-23,25,28,33,37-38], LASSO analysis [16,29], LASSO followed by multivariate analysis [13,26,29,34-35], and univariate followed by LASSO [18]. All models reported discrimination performance: 13 studies [18,22-24,27,29,31-38] used AUC, six [14,16-17,20-21,28] used C-index, and eight [12-13,15,18-19,25-26,30] used both. AUC ranged from 0.511 to 0.989, and C-index from 0.720 to 0.953. Twenty-two studies [12-21,23-30,33,35-36,38] reported calibration methods, including Hosmer-Lemeshow (H-L) goodness-of-fit test and calibration curves (Table 4 [TABLE:4]).

2.4 Model Validation

Twenty-one studies [12-19,22-27,29-31,33-36] reported model validation. Thirteen [15-16,19,22-23,25-27,29-30,34-36] conducted internal validation only (using Bootstrap, cross-validation, or random split), three [12,17,33] conducted external validation only (temporal, geographic, or combined), and five [13-14,18,24,31] used both internal and external validation (Table 5 [TABLE:5]).

2.5 Model Presentation

The models included 4-13 predictors (Figure 2 [FIGURE:2]). Twenty-one studies [12-23,25-26,28,30,32-35,38] presented models as nomograms, four [24,27,36-37] as regression equations, one [29] used both, and one [31] did not specify the presentation format (Table 5 [TABLE:5]).

2.6 Risk of Bias and Applicability Assessment

2.6.1 Risk of Bias: All 27 studies were rated as high risk of bias according to PROBAST criteria. (1) Participants domain: Five studies [15-16,19,32,34] were high risk due to case-control design; one [29] was unclear due to unspecified inclusion/exclusion criteria. (2) Predictors domain: Five studies [15-16,19,32,34] were high risk because predictors were assessed with knowledge of outcomes; 16 [12-14,20,23-28,30,33,35-38] were unclear; 19 [12-14,16,18,20-31,33,35-38] did not report whether predictors were assessed blinded to outcomes. (3) Outcome domain: 17 studies [13,15,17,19-20,22-23,25-30,32-33,37-38] were high risk: 11 [15,17,19-20,22-23,25-26,29-30,32,37-38] had outcome definitions that included predictors; nine [13,15,19-20,23-24,27-28,33] had insufficient time intervals between predictor assessment and outcome determination; six [12,16,18,21,31,34] were unclear, with five [12,15,20-21,31] not reporting diagnostic criteria and six [16,18,29,31-32,34] not reporting time intervals. (4) Statistical analysis domain: 22 studies [12,14-15,17,20-28,30-38] were high risk: 14 [14-15,20-22,26-28,30,32,35-38] had events per variable (EPV) < 20; five [14,23,30,33,36] had unclear rationale for converting continuous variables; three [15,26,33] used complete case analysis; seven [20,25-26,28,30,32,36] used univariate analysis for variable selection; nine [15,22,24,28,31-32,34,36-37] did not report calibration or used only H-L test; 12 [12,17,20-21,23,27-28,31-33,37-38] lacked internal validation or used inadequate random split methods. Five studies [13,16,18-19,29] had unclear risk, and 20 [12-14,16-17,19-22,25,27-30,32,34-38] did not report missing data handling. No studies reported data complexity (Figure 3 [FIGURE:3]).

2.6.2 Applicability Assessment: Seven studies [14,17,22,30,35-36,38] had good applicability, 13 [13,15,19-20,23-28,31,33] had low applicability, and seven [12,16,18,21,29,32,34] had unclear applicability risk. (1) Participants: Five studies [25-26,31,33,37] had high applicability risk due to restricted populations (young/middle-aged [26], elderly [31,37], female only [25], or elderly hip fracture patients [33]). (2) Predictors: All studies had low applicability risk in this domain. (3) Outcome: Nine studies [13,15,19-20,23-24,27-28,33] had high applicability risk due to short intervals between predictor assessment and outcome determination; eight [12,16,18,21,29,31-32,34] were unclear, with five [12,15,20-21,31] lacking standard outcome definitions and six [16,18,29,31-32,34] not reporting time intervals.

3.1 Overall Good Predictive Performance with Logistic Regression as the Primary Method

This review included 64 risk prediction models developed since 2019, with increasing publications in recent years. One model had an AUC of 0.511, which upon investigation used all-subsets regression with only three predictors, likely resulting in omitted variable bias and poor performance. The remaining 63 models achieved AUC/C-index > 0.7, with 43 undergoing calibration, indicating good performance in predicting post-CHD HF.

Traditional modeling methods (LR, Cox regression) remain dominant, though machine learning is gaining traction due to its computational power and predictive accuracy. Six studies [15,23-24,27,31,36] compared LR with machine learning models. While XU Qian [31] found LR superior, LI et al. [24] demonstrated that machine learning models (SVC, Ada Boost, RF, DT) outperformed LR. This discrepancy may reflect differences in data characteristics, sample size, and variable selection. LR offers robustness and interpretability for smaller samples with linear relationships [39-40], whereas machine learning excels at capturing complex patterns in high-dimensional, nonlinear datasets [41]. Shapley Additive Explanations (SHAP), a game theory-based method for attributing feature contributions [42], can enhance interpretability of machine learning models like XGBoost and RF, representing a promising direction for future research.

3.2 High Risk of Bias and Low Applicability in Current Models

The high risk of bias stemmed primarily from methodological limitations in study design and statistical analysis, with heterogeneity in populations, predictor definitions, and follow-up durations. (1) Participants: Five case-control studies [15-16,19,32,34] were included, though PROBAST recommends randomized trials or prospective cohorts for risk prediction. Case-control designs cannot calculate absolute risk and have unknown source populations, limiting suitability. (2) Predictors: Retrospective designs risk information bias and selection bias. Prospective studies with standardized predictor definitions are needed to improve stability and data quality. (3) Outcome: Seven studies [15,19,23-24,27-28,33] assessed predictors and outcomes during hospitalization, yet HF can occur post-discharge. The vulnerable period extends up to 3 months after diagnosis [44], suggesting follow-up should cover this entire period. (4) Statistical analysis: Fourteen studies [14-15,20-22,26-28,30,32,35-38] had insufficient sample size (EPV < 20), violating PROBAST recommendations [45] and risking overfitting. Continuous variables should be categorized using standard definitions or nonlinear terms to minimize information loss. (5) Missing data: Complete case analysis compromises robustness; multiple imputation is recommended with transparent reporting. Only five studies [13-14,18,24,31] conducted both internal and external validation. Notably, TAN et al. [18] developed a model using data from 44,772 patients across 7 hospitals in Chongqing, demonstrating good discrimination (AUC = 0.720) and robust multi-center validation, offering valuable clinical utility.

3.3 Common Predictive Factors Despite Heterogeneity

Among 189 predictors across 27 studies, several high-frequency factors emerged: age, left ventricular ejection fraction (LVEF), diabetes history, hypertension history, NT-proBNP, and Gensini score. Age was the most frequent predictor, as aging-related cardiomyocyte apoptosis, oxidative stress, and chronic inflammation impair cardiac function [46-47]. LVEF is a well-established marker of myocardial contractility. Diabetes and hypertension are recognized cardiovascular risk factors that promote HF through endothelial injury, oxidative stress, and inflammatory pathways [48-51]. NT-proBNP reflects myocardial volume and pressure overload; the Framingham Heart Study confirmed its value in predicting incident HF [52]. The Gensini score quantifies coronary lesion severity, with higher scores indicating increased ischemic risk [53].

International HF risk models include the Framingham Heart Study [54] and Health ABC Study [55] for high-risk elderly populations, and the ARIC risk score [56], which simplified prediction using age, sex, race, and NT-proBNP. The PCP-HF study [57] developed a 10-year risk equation for the general population based on 33,010 individuals across 7 cohorts, incorporating age, blood pressure, fasting glucose, BMI, cholesterol, smoking, and QRS duration. These studies highlight racial differences in HF risk, underscoring the need for population-specific models. While providing valuable references, these models require adaptation for Chinese CHD patients to enhance clinical feasibility and accuracy.

3.4 Future Directions

Future development of HF risk prediction models for Chinese CHD patients should employ prospective, multi-center, large-scale cohort designs covering diverse regions and socioeconomic groups, with strict inclusion/exclusion criteria to reduce heterogeneity. Predictor selection should prioritize routinely available electronic health record variables, avoiding expensive or inaccessible measures. Standardized definitions for predictors and outcomes should be established through expert consensus. Model development could integrate traditional and machine learning approaches, using SHAP for feature contribution analysis and decision trees for visualization to improve interpretability. For clinical translation, models should be implemented as scoring tables, online calculators, or integrated into electronic medical records and mobile applications. Dynamic updating and further clinical validation will support personalized treatment and early intervention.

Limitations: (1) Only Chinese and English literature was searched, potentially missing relevant studies; (2) Heterogeneity in populations and outcome definitions limited generalizability, and quantitative analysis was not performed; (3) Most models were single-center, small-sample studies lacking both internal and external validation, affecting stability and external validity.

Conclusion: This systematic review included 27 studies reporting 64 risk prediction models for HF in CHD patients. While predictive performance was generally good, the risk of bias was high. The LR-based model by TAN et al. [18] and the hybrid LR-machine learning model by LI et al. [24] demonstrated stable performance and rigorous validation, offering valuable references for clinical practice. Future model development should strictly follow PROBAST guidelines, optimize predictor selection, and conduct comprehensive validation to provide reliable risk assessment and decision support.

Author Contributions: JIANG Xiaorui designed the study, performed statistical analysis, and drafted the manuscript. JIANG Xiaorui and YAN Yuyao screened literature, assessed bias and applicability, and extracted and verified data. WEI Jingjing and QIAO Lijie revised the manuscript and ensured quality control. PENG Guangcao reviewed the manuscript.

Conflict of Interest: None declared.

ORCID: JIANG Xiaorui: https://orcid.org/0009-0005-6790-1991

Funding: This work was supported by the National Natural Science Foundation of China Key Project (82020120), Henan Provincial Key R&D Special Project (231111310200), National Administration of Traditional Chinese Medicine "Hundred, Thousand, Ten Thousand" Talent Project (Qihuang Project) [2021]203, and Henan Provincial Traditional Chinese Medicine Science Research Special Project (2022JDZX012).

Citation: JIANG XR, YAN YY, WEI JJ, et al. Systematic review of risk prediction models for concurrent heart failure with coronary heart disease [J]. Chinese General Practice, 2025. DOI: 10.12114/j.issn.1007-9572.2025.0038. [Epub ahead of print]

References

[1] METRA M, TEERLINK JR. Heart failure [J]. Lancet (London, England), 2017, 390(10106): 1981-95. DOI: 10.1016/s0140-6736(17)31071-1.

[2] LIU MB, HE XY, YANG XH, et al. Interpretation of key points in the "China Cardiovascular Health and Disease Report 2023" [J]. Chinese General Practice, 2025, 28(1): 20-38. DOI: 10.12114/j.issn.1007-9572.2024.0293.

[3] WANG H, LI YY, CHAI K, et al. Mortality in patients admitted to hospital with heart failure in China: a nationwide Cardiovascular Association Database-Heart Failure Centre Registry cohort study [J]. Lancet Glob Health, 2024, 12(4): e611-622. DOI: 10.1016/S2214-109X(23)00605-8.

[4] Chinese Society of Cardiology, Chinese College of Cardiovascular Physicians, Chinese Heart Failure Association, et al. Chinese guidelines for the diagnosis and treatment of heart failure 2024 [J]. Chinese Journal of Cardiology, 2024, 52(03): 235-75. DOI: 10.3760/cma.j.cn112148-20231101-00405.

[5] KLAJDA MD, SCOTT CG, RODEHEFFER RJ, et al. Diabetes mellitus is an independent predictor for the development of heart failure: a population study [J]. Mayo Clinic Proceedings, 2020, 95(1): 124-33. DOI: 10.1016/j.mayocp.2019.07.008.

[6] BISCAGLIA S, GUIDUCCI V, ESCANED J, et al. Complete or culprit-only PCI in older patients with myocardial infarction [J]. N Engl J Med, 2023, 389(10): 889-898. DOI: 10.1056/NEJMoa2300468.

[7] GHO JMIH, POSTEMA PG, CONIJN M, et al. Heart failure following STEMI: a contemporary cohort study of incidence and prognostic factors [J]. Open Heart, 2017, 4(2): e000551. DOI: 10.1136/openhrt-2016-000551.

[8] ZHANG R, ZHENG LQ, PAN GW. Application and development of disease risk prediction models [J]. Chinese Journal of Health Statistics, 2015, 32(4): 724-726.

[9] LI Y. Study on long-term survival status, independent risk factors and prediction models for acute heart failure patients in Beijing [D]. Beijing: Peking Union Medical College, 2021. DOI: 10.27648/d.cnki.gzxhu.2021.000213.

[10] MOONS KGM, DE GROOT JAH, BOUWMEESTER W, et al. Critical appraisal and data extraction for systematic reviews of prediction modelling studies: the CHARMS checklist [J]. PLoS Med, 2014, 11(10): e1001744. DOI: 10.1371/journal.pmed.1001744.

[11] WOLFF RF, MOONS KGM, RILEY RD, et al. PROBAST: a tool to assess the risk of bias and applicability of prediction model studies [J]. Ann Intern Med, 2019, 170(1): 51-58. DOI: 10.7326/M18-1376.

[12] LI YY, ZHANG YF, XU Y, et al. Establishment and evaluation of a risk prediction model for heart failure after PCI in acute myocardial infarction patients [J]. Journal of Clinical Cardiology, 2019, 35(10): 916-922. DOI: 10.13201/j.issn.1001-1439.2019.10.012.

[13] SUN RM. Research on risk factors and establishment of a risk prediction model for early heart failure after acute myocardial infarction [D]. Lanzhou: Lanzhou University, 2020. DOI: 10.27204/d.cnki.glzhu.2020.003235.

[14] FEI F, PENG YH, XU ZY, et al. Study on a heart failure risk prediction model for acute myocardial infarction patients [J]. Chinese Journal of Cardiovascular Medicine, 2021, 26(6): 525-530. DOI: 10.3969/j.issn.1007-5410.2021.06.006.

[15] LI X. Risk factors and prediction model for heart failure after ST-segment elevation myocardial infarction [D]. Guangzhou: Guangdong Medical University, 2021.

[16] LI Y. Integrated traditional Chinese and Western medicine clinical prediction model for coronary heart disease with chronic heart failure [D]. Guangzhou: Guangzhou University of Traditional Chinese Medicine, 2021. DOI: 10.27044/d.cnki.ggzzu.2021.000408.

[17] LIU LW. Establishment and validation of a model for new-onset acute heart failure after coronary intervention [D]. Guangzhou: Southern Medical University, 2021.

[18] TAN JT, HE YX, LI ZB, et al. Establishment and validation of a non-invasive diagnostic nomogram to identify heart failure in patients with coronary heart disease [J]. Front Cardiovasc Med, 2022, 9: 875702. DOI: 10.3389/fcvm.2022.875702.

[19] CAO JY, ZHANG LX, ZHOU XJ, et al. Establishment and validation of a risk prediction model for in-hospital heart failure after PCI in coronary heart disease patients [J]. Journal of Army Medical University, 2022, 44(12): 1272-1279. DOI: 10.16016/j.2097-0927.202110047.

[20] TANG YP, LI SJ, SU ZD. Analysis of influencing factors and construction of a prediction model for heart failure in acute myocardial infarction patients [J]. South China Journal of Cardiovascular Diseases, 2022, 28(5): 438-443. DOI: 10.3969/j.issn.1007-9688.2022.05.11.

[21] WANG B, HU GH, CHEN LM, et al. Analysis of influencing factors and establishment of a prediction model for postoperative heart failure in acute myocardial infarction patients [J]. Modern Practical Medicine, 2022, 34(8): 1002-1005. DOI: 10.3969/j.issn.1671-0800.2022.08.009.

[22] ZHOU CL, JIN XY, ZHAO J, et al. Study on a nomogram prediction model for heart failure after acute myocardial infarction PCI [J]. Journal of Cardiovascular Rehabilitation Medicine, 2022, 31(5): 586-590. DOI: 10.3969/j.issn.1008-0074.2022.05.12.

[23] CHEN SY, PAN XL, MO JH, et al. Establishment and validation of a prediction nomogram for heart failure risk in patients with acute myocardial infarction during hospitalization [J]. BMC Cardiovasc Disord, 2023, 23(1): 619. DOI: 10.1186/s12872-023-03689-3.

[24] LI XW, SHANG CM, XU CY, et al. Development and comparison of machine learning-based models for predicting heart failure after acute myocardial infarction [J]. BMC Med Inform Decis Mak, 2023, 23(1): 165. DOI: 10.1186/s12911-023-02262-2.

[25] YAN QQ, YE LF, ZHANG QG, et al. Nomogram to predict the incidence of new-onset heart failure after acute coronary syndrome among women [J]. Front Cardiovasc Med, 2023, 10: 1131813. DOI: 10.3389/fcvm.2023.1131813.

[26] CHEN MQ, TANG XL. Construction and effect analysis of a heart failure prediction model for young and middle-aged patients with unstable angina after emergency PCI [J]. Chinese Journal of Cardiovascular Research, 2023, 21(9): 842-847. DOI: 10.3969/j.issn.1672-5301.2023.09.013.

[27] HU WB, LIU LJ, LIN XF, et al. Value of gradient boosting decision tree in predicting heart failure during hospitalization after PCI in acute myocardial infarction patients [J]. Journal of Practical Cardio-Cerebro-Pulmonary Vascular Disease, 2023, 31(5): 14-19. DOI: 10.12114/j.issn.1008-5971.2023.00.133.

[28] HUANG L, JIN CL, FU MN, et al. Risk factors for heart failure after percutaneous coronary intervention [J]. Journal of Chinese Physician, 2023, 25(10): 1525-1529. DOI: 10.3760/cma.j.cn431274-20221225-01350.

[29] MA MX, MA P, XU QB, et al. Construction and validation of a risk prediction model for heart failure after acute myocardial infarction [J/OL]. Tianjin Medical Journal, 2023: 1-6. (2023-09-01). http://kns.cnki.net/KCMS/detail/detail.aspx?filename=TJYZ20230831001&dbname=CJFD&dbcode=CJFQ.

[30] WEI JH, LY JX, SONG Y. Prediction study of heart failure after PCI in unstable angina patients with coronary heart disease [J]. Modern Scientific Instruments, 2023, 40(5): 52-56.

[31] XU Q. Construction and validation of a prediction model for heart failure in elderly coronary heart disease patients based on real-world data [D]. Chongqing: Chongqing Medical University, 2023.

[32] ZHONG WD. Study on high-risk populations for heart failure after myocardial infarction and correlation with traditional Chinese medicine constitution and syndrome types [D]. Changchun: Changchun University of Traditional Chinese Medicine, 2023. DOI: 10.26980/d.cnki.gcczc.2023.000436.

[33] RAN SD, WANG ZQ, FU MM, et al. Characteristics of postoperative heart failure in older hip fracture patients combined with coronary heart disease and construction of a prediction nomogram, a retrospective cohort study [J]. Clin Interv Aging, 2024, 19: 599-610. DOI: 10.2147/CIA.S450590.

[34] CHEN H, CHEN Z, CHEN X. Diagnostic prediction model for heart failure risk after PCI in acute ST-segment elevation myocardial infarction in Wuhan: a case-control study [J]. Clinical Research and Practice, 2024, 9(13): 1-5, 17. DOI: 10.19347/j.cnki.2096-1413.202413001.

[35] CHEN JW, WANG XD, WENG WC, et al. Prediction model and efficacy evaluation of heart failure after percutaneous coronary intervention in acute myocardial infarction patients [J]. Zhejiang Medical Journal, 2024, 46(18): 1938-1943. DOI: 10.12056/j.isn.1006-2785.2024.46.18.2023-2748.

[36] LIU CY, SUN J, GAO Y, et al. Construction and evaluation of a prediction model for heart failure 1 year after PCI in acute coronary syndrome patients [J]. Chinese Journal of Medicine, 2024, 59(10): 1077-1080.

[37] LIU JJ, LIU H, YU XP. Construction of a heart failure prediction model after PCI in elderly acute myocardial infarction patients [J]. Journal of Clinical Medicine in Practice, 2024, 28(15): 84-89. DOI: 10.7619/jcmp.20240209.

[38] SHI YX, WANG WG, CHENG CF, et al. Construction and importance matrix analysis of a Nomogram risk model for heart failure in patients with angina after coronary intervention [J]. Hainan Medical Journal, 2024, 35(16): 2288-2294.

[39] HELMREICH JE. Regression modeling strategies with applications to linear models, logistic and ordinal regression and survival analysis (2nd edition) [J]. J Stat Soft, 2016, 70(Book Review 2). DOI: 10.18637/jss.v070.b02.

[40] LUO W, PHUNG D, TRAN T, et al. Guidelines for developing and reporting machine learning predictive models in biomedical research: a multidisciplinary view [J]. J Med Internet Res, 2016, 18(12): e323. DOI: 10.2196/jmir.5870.

[41] OBERMEYER Z, EMANUEL EJ. Predicting the future - big data, machine learning, and clinical medicine [J]. N Engl J Med, 2016, 375(13): 1216-1219. DOI: 10.1056/NEJMp1606181.

[42] EJIYI CJ, QIN Z, UKWUOMA CC, et al. Comparative performance analysis of Boruta, SHAP, and Borutashap for disease diagnosis: a study with multiple machine learning algorithms [J]. Network, 2024: 1-38. DOI: 10.1080/0954898X.2024.2331506.

[43] Chinese College of Cardiovascular Physicians, China Cardiovascular Health Alliance, Expert Consensus Working Group on Prevention and Treatment of Heart Failure after Myocardial Infarction. 2020 Expert consensus on prevention and treatment of heart failure after myocardial infarction [J]. Chinese Circulation Journal, 2020, 35(12): 1166-80. DOI: 10.3969/j.issn.1000-3614.2020.12.002.

[44] RIZZI MA, SARASOLA AG, ARBÉ AA, et al. Factors associated with in-hospital mortality and adverse outcomes during the vulnerable post-discharge phase after the first episode of acute heart failure: results of the NOVICA-2 study [J]. Clin Res Cardiol, 2021, 110(7): 993-1005. DOI: 10.1007/s00392-021-01884-3.

[45] CHEN XP, ZHANG Y, ZHUANG YY, et al. PROBAST: a tool to assess the risk of bias and applicability of prediction model studies [J]. Chinese Journal of Evidence-Based Medicine, 2020, 20(6): 737-744. DOI: 10.7507/1672-2531.201910087.

[46] DUAN XR, GE YH, WEI YJ, et al. Manifestations, causes, and prevention of aging [J]. Journal of Zhengzhou University: Medical Sciences, 2024, 59(6): 741-745. DOI: 10.13705/j.issn.1671-6825.2024.10.070.

[47] YANG JX, PENG Y, LIN YY, et al. Risk factors and research progress of prediction models for heart failure after PCI in acute myocardial infarction patients [J]. Sichuan Medical Journal, 2024, 45(03): 305-8. DOI: 10.16252/j.cnki.issn1004-0501-2024.03.012.

[48] HOU YJ, QIN MZ. Management of comorbidities in elderly chronic heart failure [J]. Chinese Journal of Practical Internal Medicine, 2023, 43(08): 628-33. DOI: 10.19538/j.nk2023080103.

[49] LI RM, DAI GH, GAO WL, et al. Study on early warning indicators of death in chronic heart failure patients with primary hypertension [J]. Chinese Journal of Integrated Traditional and Western Medicine, 2024, 44(09): 1051-62. DOI: 10.7661/j.cjim.20240320.045.

[50] XU K, ZHANG XM, DANG J, et al. Clinical study of sacubitril/valsartan combined with dapagliflozin in elderly chronic heart failure patients with type 2 diabetes [J]. Chinese Journal of Clinical Pharmacology, 2024, 40(17): 2464-8. DOI: 10.13699/j.cnki.1001-6821.2024.17.002.

[51] RAI A, NARISAWA M, LI P, et al. Adaptive immune disorders in hypertension and heart failure: focusing on T-cell subset activation and clinical implications [J]. J Hypertens, 2020, 38(10): 1878-1889. DOI: 10.1097/HJH.0000000000002456.

[52] VASAN RS, BENJAMIN EJ, LARSON MG, et al. Plasma natriuretic peptides for community screening for left ventricular hypertrophy and systolic dysfunction: the Framingham heart study [J]. JAMA, 2002, 288(10): 1252-1259. DOI: 10.1001/jama.288.10.1252.

[53] RAMPIDIS GP, BENETOS G, BENZ DC, et al. A guide for gensini score calculation [J]. Atherosclerosis, 2019, 287: 181-183. DOI: 10.1016/j.atherosclerosis.2019.05.012.

[54] VASAN RS, PENCINA MJ, et al. General cardiovascular risk profile for use in primary care: the Framingham Heart Study [J]. Circulation, 2008, 117(6): 743-753. DOI: 10.1161/CIRCULATIONAHA.107.699579.

[55] BUTLER J, KALOGEROPOULOS A, GEORGIOPOULOU V, et al. Incident heart failure prediction in the elderly: the health ABC heart failure score [J]. Circ Heart Fail, 2008, 1(2): 125-133. DOI: 10.1161/CIRCHEARTFAILURE.108.768457.

[56] AGARWAL SK, CHAMBLESS LE, BALLANTYNE CM, et al. Prediction of incident heart failure in general practice: the Atherosclerosis Risk in Communities (ARIC) Study [J]. Circ Heart Fail, 2012, 5(4): 422-429. DOI: 10.1161/CIRCHEARTFAILURE.111.964841.

[57] KHAN SS, NING H, SHAH SJ, et al. 10-Year risk equations for incident heart failure in the general population [J]. J Am Coll Cardiol, 2019, 73(19): 2388-97. DOI: 10.1016/j.jacc.2019.02.057.

Received: January 26, 2025; Revised: March 20, 2025

Edited by: LI Weixia