Unveiling Coronary Heart Disease Prediction through Machine Learning Techniques: Insights from the Suita Population-Based Cohort Study

doi:10.21203/rs.3.rs-4120297/v1

Download PDF

Article

Unveiling Coronary Heart Disease Prediction through Machine Learning Techniques: Insights from the Suita Population-Based Cohort Study

https://doi.org/10.21203/rs.3.rs-4120297/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this latest preprint version

We leveraged machine learning (ML) techniques, namely logistic regression (LR), random forest (RF), support vector machine (SVM), extreme gradient boosting (XGBoost), and LightGBM to predict coronary heart disease (CHD) and identify the key risk factors involved. Based on the Suita study, 7672 men and women aged 30 to 84 years without cardiovascular disease were recruited from 1989 to 1999, in Suita City, Osaka, Japan. Over an average period of 15 years, participants were diligently monitored until the onset of their initial cardiovascular event or relocation. CHD diagnoses encompassed primary heart attacks, sudden death, or coronary artery disease with bypass surgery or intervention.

RF achieved the highest AUC (95% CI) of 0.79 (0.70–0.87), outperforming LR, SVM, XGBoost, and LightGBM. Shapley Additive Explanations (SHAP) on the best model identified the top CHD predictors. Notably, systolic blood pressure, non-HDL-c, glucose levels, age, metabolic syndrome, HDL-c, estimated glomerular filtration rate, hypertension, elbow joint thickness, and diastolic blood pressure were key contributors. Remarkably, elbow joint thickness was identified as a previously unrecognized risk factor associated with CHD. These findings indicated that ML methods accurately predict incident CHD risk. Additionally, ML has identified new incident CHD risk variables.

Health sciences/Biomarkers

Health sciences/Cardiology

Health sciences/Diseases

Health sciences/Medical research

Health sciences/Risk factors

Coronary Heart Disease (CHD)

Machine Learning (ML)

Logistic Regression (LR)

Random Forest (RF)

Support Vector Machine (SVM)

eXtreme Gradient Boost (XGBoost)

Light Gradient Boosted Machine (LightGBM)

Shapley Addictive ExPlanations (SHAP)

Coronary heart disease (CHD) is a prevalent and life-threatening cardiovascular condition characterized by narrowing or blockage of the coronary arteries [1, 2]. Accurate prediction and timely risk assessment are crucial for effective preventive measures and personalized interventions [3]. While conventional risk assessment models have been used, there is growing recognition of the potential of machine learning in enhancing CHD prediction [4, 5].

Machine learning algorithms have proven their ability to analyze complex data and identify intricate patterns and relationships that are not easily detected by traditional statistical methods [6, 7]. By integrating diverse data sources, such as demographics, medical history, lifestyle habits and diagnostic findings, these algorithms can predict the likelihood of developing CHD. This approach offers comprehensive risk evaluation, adaptability to new data, and the potential to uncover novel risk factors and disease mechanisms [8, 9]. Several studies have demonstrated the effectiveness of machine learning models in deriving quantitative markers for coronary artery disease and predicting the presence of heart disease. For example, a study developed and validated a coronary artery disease-predictive machine learning model using electronic health records and assessed its probabilities as in silico scores for coronary artery disease in participants in two longitudinal biobank cohorts [10]. Another study applied an ensemble ML model for coronary disease prediction, using ML classifiers to predict heart disease [11]. These findings highlight the potential of machine learning in driving innovation and improving the accuracy of CHD diagnosis and prediction [12].

However, challenges exist in utilizing machine learning for CHD prediction, including data quality, feature selection, model interpretability, and generalizability. These issues must be carefully addressed to ensure the reliability and robustness of the predictive models. Rigorous validation, regulatory compliance, and effective communication strategies are essential for its successful integration into clinical practice.

This study aimed to address the role of machine learning techniques in predicting incident CHD and identifying novel risk factors. This study sought to deepen our understanding of the factors contributing to CHD development by analyzing a comprehensive dataset. These findings will enhance risk assessment, enabling the development of personalized interventions and preventive strategies. Moreover, this study provides insights into the strengths and limitations of machine learning models in CHD prediction, guiding their integration into clinical workflows.

2.1. Study Participants

The Suita Study, a prospective population-based cohort study, was conducted in Suita City, Osaka, Japan. Previous publications have provided extensive information regarding the methodology and selection criteria of the study [13, 14]. Briefly, from 1989 to 1999, a total of 7672 men and women aged 30 to 84 years who did not have a previous history of cardiovascular disease were recruited for the study. Participants were selected from the population registry of the municipality and were followed up every two years for an average of 15 years until their first occurrence of stroke, myocardial infarction (MI), death, or relocation. Opt-out procedures were implemented for those who preferred not to participate in the artificial intelligence research. Informed consent was obtained from all participants. The study was conducted in accordance with the guidelines and regulations outlined in the Declaration of Helsinki, with approval from the Institutional Review Board at the National Cerebral and Cardiovascular Center. (No. R21024-2).

2.2. Data collection

A comprehensive and prospective data collection process was implemented, encompassing various aspects such as demographics, medical history, medical imaging, laboratory data, lifestyle habits, and outcomes. Detailed information regarding the data collection procedure can be found elsewhere [13, 14]. These evaluations served as the baseline examination for the current investigation.

2.3. Risk Factors and Anthropomorphic Measurements

Each participant's blood pressure (BP) was measured using a mercury column sphygmomanometer, an appropriately sized cuff, and a standardized protocol to ensure accuracy and precision [14]. Before the initial blood pressure reading, the participants were instructed to rest for at least 5 minutes to establish a stable baseline. Blood pressure readings were obtained by averaging the second and third measurements, which were performed at intervals of more than one minute to allow for adequate observation and recording. Hypertension was defined as a systolic blood pressure ≥ 140 mmHg, diastolic blood pressure ≥ 90 mmHg, or the use of antihypertensive medications. Body mass index (BMI) was calculated as weight (kg) divided by the square of height (m²).

At baseline, routine blood tests were performed, including serum total and high-density lipoprotein cholesterol and glucose levels. Non-high-density lipoprotein cholesterol levels were calculated by subtracting high-density lipoprotein cholesterol (HDL-c) from total cholesterol. Diabetes mellitus was defined as a fasting plasma glucose (FPG) ≥ 126 mg/dL, a non-FPG ≥ 200 mg/dL, or the use of diabetes mellitus medication. Metabolic syndrome was defined as a combination of abdominal obesity, impaired fasting glucose, atherogenic dyslipidemia, and elevated blood pressure according to the original Japanese criteria for metabolic syndrome, which were a waist circumference ≥ 85 cm in men and ≥ 90 cm in women and/or a BMI ≥ 25.0 kg/m², an essential component plus ≥ 2 (definite MetS) of the following [15, 16]: (1) systolic blood pressure ≥ 130 mm Hg and/or diastolic blood pressure ≥ 85 mm Hg or medication use; (2) triglyceride level ≥ 150 mg/dL and/or HDL cholesterol level < 40 mg/dL; and (3) fasting glucose level ≥ 100 mg/dL and/or medication use. The estimated glomerular filtration rate (eGFR) (mL/min/1.73 m²) was calculated according to the original Modification of Diet in Renal Disease (MDRD) equation modified by the Japanese coefficient (0.881) as follows: 0.881 × 186 × serum creatinine^− 1.154 × age^− 0.203 × (0.742 if female) [17].

Smoking status and drinking statuses were categorized as current, quit, or never. A questionnaire was used to ask participants about their past and present history of CHD.

2.4. Confirmation of CHD

Medical records were carefully reviewed by hospital doctors or researchers who were blinded to the baseline data to provide an unbiased approach to the analysis. Myocardial infarctions were classified as definite or probable according to the criteria established by the MONICA project [18].

Every two years, each participant’s health was evaluated at the National Cerebral and Cardiovascular Center in Osaka, Japan, to detect the occurrence of CHD. Yearly questionnaires were also completed by all participants by mail or telephone. CHD surveillance was completed by systematically searching for death certificates [13, 14]. CHD was diagnosed based on specific criteria such as primary heart attack, sudden death within 24 hours of acute illness onset, and coronary artery disease with bypass surgery or intervention.

2.5. Statistical Methods

The data are presented as percentages, means (standard deviations), or medians (IQRs) depending on the variable characteristics. Differences in patient characteristics according to incident CHD status were evaluated using t tests, Mann‒Whitney U tests, or chi‒squared tests, as appropriate. Our study was conducted using Python 3.7.11 (https://www.python.org/, accessed on 1 June 2023) along with compatible open-source packages for data analysis and ML model building. To train the model, we utilized Scikit-Learn version 1.0.2 (https://scikit-learn.org, accessed on 5 April 2022).

2.6. Data preprocessing and variable selection

Our original dataset consisted of 7672 participants and 169 variables.

We imputed the missing continuous variables using random forest and replaced the missing categorical variables with the mode using one-hot encoding. Normalization was performed on the data using a standardization model: var' = (var - mean(var))/SD, where var is the original variable and SD is its standard deviation. To develop a data-driven model, we extracted all possible variables from the raw dataset. We removed multicollinearity by considering the clinical meaning of variables, variance inflation factor (VIF), and correlation coefficients between variables [19]. A VIF greater than 10 is generally considered to indicate very high multicollinearity and can lead to unreliable results and poor estimated coefficients [19]. We used a cutoff VIF value of less than 10 and eliminated redundant and unnecessary variables from the dataset. This resulted in a dataset with 7133 participants and 53 variables. All variables are presented in Supplementary Table S1, Supplementary Figure S1, and Figure S2.

2.7. Model development

In this study, there were 302 incident coronary heart disease (CHD) patients, representing 4.2% of the population. However, the imbalance between the groups with and without incident CHD considerably affected the supervised ML models.

To enhance the performance of the supervised model for CHD incidence, the dataset resulting from the preceding stage of data preprocessing and variable selection was downsampled to a group free of CHD and randomly split into training (80%) and testing (20%) datasets.

For incident CHD prediction, we applied five supervised models. We used a combination of cross-validation (5-fold cross-validation of training and testing with a randomized data split) and a hyperparameter optimization framework. We trained five classification algorithms: logistic regression (LR), random forest (RF), support vector machine (SVM), extreme gradient boost (XG Boost), and light gradient boosting machine (LightGBM).

Each model had unique strengths that made it the best model for predicting CHD in our cohort.

The application of logistic regression (LR) in cardiovascular disease studies supports the statement that LR has been widely used for risk prediction in this domain. This research paper provides evidence of the extensive use of LR in modeling the probability of events such as heart attacks based on risk factors such as cholesterol level, blood pressure, and age [20]. LR has been a staple in medical research for risk prediction, particularly in the context of cardiovascular diseases. It has been employed to model the probability of various events and to explore the effects of relationships between multiple predictors [20]. LR is a simple approach to prediction that provides baseline accuracy scores for comparisons with other nonparametric machine learning models. RF is an ensemble learning method that builds multiple decision trees and combines their predictions. It is robust and effective for complex data. In cardiovascular research, RF has been applied to predict the risk of heart disease by incorporating various risk factors [21, 22]. Its ability to handle nonlinear relationships and interactions makes it valuable for capturing the complexity of cardiovascular risk [23]. SVMs have been applied in cardiovascular disease studies for risk prediction, demonstrating their effectiveness in handling high-dimensional data and identifying complex patterns to predict outcomes [24, 25]. SVM classifies data by separating classes with a boundary [26, 27], whereas XGBoost is a powerful tool for classification and regression [27]. XGBoost has been utilized to predict outcomes such as heart failure and cardiovascular events based on patient data and risk factors, demonstrating its efficiency and accuracy in this domain [28, 29]. A literature review and research articles indicate that LightGBM, a distributed gradient boosting framework known for its speed and efficiency, has been increasingly applied in healthcare and cardiovascular research. Its ability to handle large datasets efficiently makes it suitable for predicting cardiovascular events [30, 31].

We evaluated the efficacy of these models using several performance metrics, such as accuracy, area under the receiver operating characteristic curve (AUC), recall, precision, and F1-score. These performance metrics are clearly explained in Supplementary Table S2.

2.8. Extraction of important variables for CHD risk

Once we identified the best performance model from the aforementioned machine learning models, we utilized the Shapley additive explanations (SHAP) method to determine the predictors that had the greatest impact on the prediction model [19, 22, 32, 33]. SHAP is a unified framework for predictive interpretation that expresses the importance of variables by comparing situation predictions with baseline values when there are specific values for a given feature [32, 34].

Explanations of the SHAP bar plot and SHAP heat plot are provided in Supplementary files Figure S1 and Figure S2.

The flowchart in Fig. 1 represents the development of a model for predicting CHD.

The study identified 302 cases of incident coronary heart disease (CHD) from the total sample, constituting 4.2% of the population. Individuals who experienced incident CHD were generally older, with a median age of 63 years [56, 70.7], and were predominantly male (66.6%). This group exhibited a greater prevalence of risk factors such as high blood pressure, elevated non-HDL cholesterol (non-HDL-c) levels, and increased elbow joint thickness. Conversely, lower estimated glomerular filtration rate (eGFR) and HDL cholesterol (HDL-c) levels were observed. Furthermore, these participants had a greater likelihood of having comorbidities, including hypertension, metabolic syndrome, and diabetes, indicating a significant presence of these conditions among individuals with incident CHD, as outlined in Table 1.

Table 1

Characteristics of study participants with and without coronary heart disease incidence (healthy Japanese, aged 30–84 years, Suita study at baseline).
	Overall	Coronary heart disease, n (%)
	7133	No 6831 (95.8)	Yes 302 (4.2)	p-value
Age, Years	55 [44, 65]	55 [43, 64]	63 [56, 70.7]	< 0.001
Male, n (%)	3284 (46.0)	3083 (45.1)	201 (66.6)	< 0.001
BMI, kg/m2	22.52 (3.11)	22.49 (3.10)	23.30 (3.19)	< 0.001
SBP, mmHg	123 [110, 138]	123 [110, 137]	138 [125, 152]	< 0.001
DBP, mmHg	77 [70, 85]	77 [70, 85]	83 [74, 89]	< 0.001
eGFR, mL/min/1.73 m2	90.24 [74.02, 104.76]	90.56 [74.77, 104.90]	79.40 [67.48, 92.62]	< 0.001
Hypertension, n (%)	2165 (30.4)	1996 (29.2)	169 (56.0)	< 0.001
MetS, n (%)	1755 (24.6)	1598 (23.4)	157 (52.0)	< 0.001
Diabetes, n (%)	860 (12.1)	781 (11.4)	79 (26.2)	< 0.001
TG, mg/dL	99 [71, 144]	98 [70.00, 143]	122.5 [89.25, 174]	< 0.001
non-HDL-c, mg/dL	153.08 (37.26)	152.22 (36.90)	172.77 (40.04)	< 0.001
HDL-c, mg/dL	54.6 (14.3)	54.9 (14.3)	48.0 (13.1)	< 0.001
Glucose, mg/dL	95 [90, 102]	95 [89, 101]	100 [93, 110]	< 0.001
Fructosamine, µmol/L	251 [237, 266]	251 [237, 265]	257 [242, 276.25]	< 0.001
WBC, /mm³	5.56 (1.53)	5.54 (1.51)	5.98 (1.76)	< 0.001
RBC, 10³/mm³	4.54 (0.44)	4.53 (0.44)	4.61 (0.45)	0.004
Hemoglobin, g/dL	13.90 (1.56)	13.88 (1.56)	14.30 (1.49)	< 0.001
Smoking status, n (%)				< 0.001
Current	2072 (29.5)	1968 (29.3)	104 (35.1)
Quit	1127 (16.1)	1050 (15.6)	77 (26.0)
Never	3819 (54.4)	3704 (55.1)	115 (38.9)
Elbow, mm	6.30 [5.80, 6.70]	6.30 [5.80, 6.70]	6.50 [6.10, 6.90]	< 0.001
Calcium, mg/dL	9.35 (0.46)	9.35 (0.46)	9.33 (0.42)	0.586
Coronary heart disease (CHD) was diagnosed by a first-ever acute myocardial infarction, sudden cardiac death within 24 hours of illness, or coronary artery disease followed by bypass or angioplasty.
Values are presented as mean (standard deviation) for continuous variables with approximately normally distribution or by median [interquartile range] with skewed distribution and n (%) for categorical variables. Differences in characteristics were evaluated by using the unpaired Student’s t-test, Wilcoxon rank sums test, or Chi squared test.

Abbreviations: BMI, body mass index; SBP, systolic blood pressure; DBP, diastolic blood pressure; TG, triglycerides; non-HDL-c, non-high-density lipoprotein cholesterol; eGFR, estimated glomerular filtration rate; MetS, metabolic syndrome.

The performance metrics of the five machine learning models employed in our CHD prediction study provide valuable insights into their effectiveness, as shown in Table 2. Logistic regression (LR) yielded an accuracy and AUC of 0.68 and 0.76 (95% CI: 0.68–0.85), respectively, showing its ability to achieve a good balance between sensitivity, precision, and F1 score, all at 0.68. The random forest (RF) model demonstrated the highest AUC at 0.79, indicating superior discriminatory power, while maintaining competitive accuracy, sensitivity, precision, and F1 score at 0.72, 0.72, 0.73, and 0.72, respectively. The support vector machine (SVM) offered reliable performance, with an AUC of 0.75 and consistent values across other metrics of 0.70. XGBoost and LightGBM exhibited similar AUC values of 0.74, highlighting their competitive predictive abilities, although with slightly lower accuracy, sensitivity, and precision. In summary, RF performed the best in terms of both discrimination and classification accuracy, while SVM, XGBoost, and LightGBM offered viable alternatives because of their unique advantages in CHD prediction.

Table 2: Performance of different machine learning approaches

	Accuracy	AUC	Recall	Precision	F1-score
LR	0.68 (0.59-0.75)	0.76 (0.68-0.85)	0.68 (0.61-0.75)	0.68 (0.61-0.76)	0.68 (0.53-0.82)
RF	0.72 (0.63-0.79)	0.79 (0.70-0.87)	0.72 (0.65-0.79)	0.73 (0.65-0.80)	0.72 (0.57-0.86)
SVM	0.70 (0.62-0.78)	0.75 (0.66-0.84)	0.70 (0.63-0.77)	0.70 (0.63-0.78)	0.70 (0.55-0.85)
XG Boost	0.66 (0.57-0.74)	0.74 (0.65-0.83)	0.66 (0.59-0.73)	0.66 (0.59-0.74)	0.66 (0.51-0.81)
Light GBM	0.69 (0.60-0.76)	0.74 (0.65-0.83)	0.69 (0.61-0.76)	0.69 (0.61-0.76)	0.68 (0.54-0.83)

Abbreviation: AUC, Area Under the Curve; LR, Logistic Regression; RF, Random Forest; SVM, Support Vector Machine; XG Boost, eXtreme Gradient Boost; Light GBM, Light Gradient Boosted Machine

Further analysis using SHAP values is depicted in Fig. 2. The RF was employed to calculate the SHAP values, enabling the identification of the top ten variables that contributed to incident CHD. Notably, factors related to blood pressure, such as systolic blood pressure (SBP), hypertension, and diastolic blood pressure (DBP), played a significant role in predicting incident CHD, with systolic blood pressure emerging as the most influential risk factor. Age also emerged as an important risk factor, as it reflects the cumulative exposure to various risk factors over time. Variables such as systolic blood pressure, non-HDL cholesterol (non-HDL-c), and glucose levels were more important than age in predicting CHD occurrence. Interestingly, an inverse association was observed between HDL cholesterol (HDL-c) and the estimated glomerular filtration rate (eGFR) and incident CHD. Furthermore, a previously unidentified risk factor, referred to as elbow joint thickness, was found to be associated with CHD.

The present study investigated the prediction of incident coronary heart disease (CHD) using machine learning techniques and explored the key risk factors associated with CHD occurrence.

This study revealed that comorbidities such as hypertension, metabolic syndrome, and diabetes were more prevalent among individuals with incident CHD, consistent with previous research [35, 36]. The presence of these comorbidities suggests the need for comprehensive management and preventive strategies targeting multiple risk factors in individuals at risk of CHD.

Optimal machine learning model

In the context of CHD prediction, our study evaluated the performance of five distinct machine learning models for identifying individuals at risk. All five models had good discrimination between the classes of CHD, resulting in a high AUC. LR, a conventional method, demonstrated its utility by accurately classifying binary outcomes, assuming a linear relationship between input variables and CHD incidence, and it serves as a straightforward prediction method that provides accuracy scores for comparison with other nonparametric machine learning models [37]. Nevertheless, LR is not well suited for the analysis of nonlinear and high-dimensional datasets. SVM, with its capacity to manage both linear and nonlinear relationships via a variety of kernels, is a robust alternative [24–26]. Moreover, tree-based ensemble models such as RF, XGBoost, and LightGBM have demonstrated their ability to capture complex data patterns [23, 28, 30, 31, 38]. In summary, RF stood out as the top performer, achieving the highest AUC and other metric values, underscoring its effectiveness in CHD prediction. This finding aligns with similar studies [18, 27].

The predictive modeling results demonstrated that the random forest (RF) model outperformed other supervised models in accurately predicting incident CHD, which is in line with the findings of similar studies [39, 40].

Key risk factors for CHD

Regarding the key risk factors identified in this study, blood pressure variables, particularly systolic blood pressure, were found to play a significant role in predicting incident CHD. This finding aligns with the well-established association between elevated blood pressure and the development of CHD [41, 42]. Non-HDL cholesterol and glucose levels also emerged as important predictors of incident CHD, highlighting the significance of dyslipidemia and impaired glucose metabolism in CHD risk [43–45].

Age, a traditional risk factor, retained its significance in predicting incident CHD, reflecting the cumulative exposure to various risk factors over time [46, 47]. Metabolic syndrome, characterized by a cluster of metabolic abnormalities, has emerged as a noteworthy risk factor for incident CHD. This finding highlights the importance of considering the collective impact of metabolic factors in assessing CHD risk [36, 48]. An inverse relationship was observed between incident CHD and HDL-c and eGFR. This paradox showed the expected protective effects of higher HDL cholesterol levels [49] and optimal renal function in reducing CHD risk [50].

Comparison of Risk Factors in Our Study with Framingham and Suita Scores and Other Studies:

The primary focus of the Framingham and Suita scores revolves around predicting CHD. The Framingham risk score incorporates six coronary risk factors: age, sex, smoking habits, blood pressure, total cholesterol, and HDL cholesterol [51]. In the context of predicting CHD, the Suita score, which is tailored for the Japanese population, outperformed the original Framingham risk score, encompassing similar factors but introducing an assessment of chronic kidney disease (CKD) stage for enhanced accuracy [52]. Our study, leveraging machine learning techniques, shares similarities with the risk factors included in both the Framingham and Suita scores. However, noteworthy disparities arise due to the distinct methodologies employed. Unlike the predefined variables in traditional scores, our models autonomously selected risk factors based on their predictive capabilities. This approach introduces an element of objectivity, allowing for the identification of novel risk factors and potentially refining CHD risk assessment models beyond the constraints of preestablished variables.

Comparing these results with those of previous studies, the identified variables align with established risk factors for incident CHD, including age, blood pressure, lipid profile, and metabolic abnormalities [2, 36, 48]. However, the specific ranking of these variables based on SHAP values provides additional insights into their relative importance in predicting CHD incidence.

The present work emphasizes the considerable capacity of machine learning techniques to effectively capture complex relationships and identify previously unrecognized risk factors in the domain of CHD risk evaluation. Machine learning approaches offer a great framework for investigating and understanding the complex nature of CHD. The discovery of an association between elbow joint thickness and CHD serves as a compelling example of how machine learning can effectively uncover previously overlooked risk factors. However, further research is needed to confirm this association. Therefore, this underscores the advantages of utilizing machine learning algorithms to uncover hidden insights inside large datasets, ultimately assisting researchers and doctors in making informed decisions based on data. In contrast, conventional analytical methods are constrained by predefined rules and models established by experts, potentially limiting their ability to provide novel insights that extend beyond explicit definitions.

Limitations

The study's generalizability is also open to examination. The focus on a specific population might restrict the applicability of findings to diverse populations with varying genetic, environmental, and lifestyle factors. In addition, the complexity of machine learning models can make interpretation difficult. Although these models exhibit enhanced predictive accuracy, the lack of transparency of their internal workings might obscure a clear understanding of the underlying mechanisms involved. This "black-box" nature might hinder the translation of results into actionable insights. Therefore, it would be advantageous to undertake additional research to externally validate the findings of our study and assess the efficacy of machine learning algorithms in clinical practice.

In conclusion, this study emphasizes the predictive value of machine learning models for accurately identifying individuals at risk of incident CHD. Additionally, the use of machine learning techniques has led to the identification and exploration of novel risk factors associated with incident CHD. The incorporation of these findings into clinical practice has the potential to enhance CHD risk assessment and improve patient outcomes.

Author contributions

Study concept and design: TV, YK, MA; data analysis and interpretation: TV, MI, MY; drafting of the manuscript: TV; resources: YK; data curation: YK, MA; supervision: YK, MA; reviewing and editing: TV, RD, AMM, JTT, AA, MT, YMN, TI. All authors critically revised and approved the final version of the manuscript.

Data Availability: The dataset examined in this study is not available to the public due to the inclusion of individuals' personal information, but is available from the corresponding author an reasonable request.

Conflicts of interest: YMN reports a study grant from Bayer outside the submitted work. All authors declare no conflicts of interest.

Acknowledgments: This article was supported by the Japan Science and Technology Agency (JST) COI-NEXT Grant number JPMJPF2018 to M.A.

Global burden of 369 diseases and injuries in 204 countries and territories, 1990-2019: a systematic analysis for the Global Burden of Disease Study 2019. Lancet 2020. 396 (10258): p. 1204-1222. DOI: 10.1016/s0140-6736(20)30925-9.
Roth GA, Mensah GA, Johnson CO, Addolorato G, Ammirati E, Baddour LM, et al. Global Burden of Cardiovascular Diseases and Risk Factors, 1990-2019: Update From the GBD 2019 Study. J Am Coll Cardiol 2020. 76 (25): p. 2982-3021. DOI: 10.1016/j.jacc.2020.11.010.
Lim HY, Burrell LM, Brook R, Nandurkar HH, Donnan G, and Ho P. The Need for Individualized Risk Assessment in Cardiovascular Disease. J Pers Med 2022. 12 (7). DOI: 10.3390/jpm12071140.
Matheson MB, Kato Y, Baba S, Cox C, Lima JAC, and Ambale-Venkatesh B. Cardiovascular Risk Prediction Using Machine Learning in a Large Japanese Cohort. Circ Rep 2022. 4 (12): p. 595-603. DOI: 10.1253/circrep.CR-22-0101.
Weng SF, Reps J, Kai J, Garibaldi JM, and Qureshi N. Can machine-learning improve cardiovascular risk prediction using routine clinical data? PLoS One 2017. 12 (4): p. e0174944. DOI: 10.1371/journal.pone.0174944.
Jiang T, Gradus JL, and Rosellini AJ. Supervised Machine Learning: A Brief Primer. Behav Ther 2020. 51 (5): p. 675-687. DOI: 10.1016/j.beth.2020.05.002.
Linardatos P, Papastefanopoulos V, and Kotsiantis S. Explainable AI: A Review of Machine Learning Interpretability Methods. Entropy (Basel) 2020. 23 (1). DOI: 10.3390/e23010018.
Gonsalves A, Thabtah FA, Mohammad RMA, and Singh G. Prediction of Coronary Heart Disease using Machine Learning: An Experimental Analysis. Proceedings of the 2019 3rd International Conference on Deep Learning Technologies 2019.
Kumar Y, Koul A, Singla R, and Ijaz MF. Artificial intelligence in disease diagnosis: a systematic literature review, synthesizing framework and future research agenda. J Ambient Intell Humaniz Comput 2023. 14 (7): p. 8459-8486. DOI: 10.1007/s12652-021-03612-z.
Forrest IS, Petrazzini BO, Duffy Á, Park JK, Marquez-Luna C, Jordan DM, et al. Machine learning-based marker for coronary artery disease: derivation and validation in two longitudinal cohorts. Lancet 2023. 401 (10372): p. 215-225. DOI: 10.1016/s0140-6736(22)02079-7.
Bani Hani SH and Ahmad MM. Machine-learning Algorithms for Ischemic Heart Disease Prediction: A Systematic Review. Curr Cardiol Rev 2023. 19 (1): p. e090622205797. DOI: 10.2174/1573403x18666220609123053.
Alizadehsani R, Abdar M, Roshanzamir M, Khosravi A, Kebria PM, Khozeimeh F, et al. Machine learning-based coronary artery disease diagnosis: A comprehensive review. Comput Biol Med 2019. 111: p. 103346. DOI: 10.1016/j.compbiomed.2019.103346.
Kokubo Y, Watanabe M, Higashiyama A, Nakao YM, Kobayashi T, Watanabe T, et al. Interaction of Blood Pressure and Body Mass Index With Risk of Incident Atrial Fibrillation in a Japanese Urban Cohort: The Suita Study. Am J Hypertens 2015. 28 (11): p. 1355-1361. DOI: 10.1093/ajh/hpv038.
Kokubo Y, Kamide K, Okamura T, Watanabe M, Higashiyama A, Kawanishi K, et al. Impact of high-normal blood pressure on the risk of cardiovascular disease in a Japanese urban cohort: the Suita study. Hypertension 2008. 52 (4): p. 652-659. DOI: 10.1161/hypertensionaha.108.118273.
Nakao YM, Miyamoto Y, Ueshima K, Nakao K, Nakai M, Nishimura K, et al. Effectiveness of nationwide screening and lifestyle intervention for abdominal obesity and cardiometabolic risks in Japan: The metabolic syndrome and comprehensive lifestyle intervention study on nationwide database in Japan (MetS ACTION-J study). PLoS One 2018. 13 (1): p. e0190862. DOI: 10.1371/journal.pone.0190862.
Iso H, Cui R, Takamoto I, Kiyama M, Saito I, Okamura T, et al. Risk Classification for Metabolic Syndrome and the Incidence of Cardiovascular Disease in Japan With Low Prevalence of Obesity: A Pooled Analysis of 10 Prospective Cohort Studies. J Am Heart Assoc 2021. 10 (23): p. e020760. DOI: 10.1161/jaha.121.020760.
Imai E, Horio M, Nitta K, Yamagata K, Iseki K, Hara S, et al. Estimation of glomerular filtration rate by the MDRD study equation modified for Japanese patients with chronic kidney disease. Clin Exp Nephrol 2007. 11 (1): p. 41-50. DOI: 10.1007/s10157-006-0453-4.
Tunstall-Pedoe H, Kuulasmaa K, Amouyel P, Arveiler D, Rajakangas AM, and Pajak A. Myocardial infarction and coronary deaths in the World Health Organization MONICA Project. Registration procedures, event rates, and case-fatality rates in 38 populations from 21 countries in four continents. Circulation 1994. 90 (1): p. 583-612. DOI: 10.1161/01.cir.90.1.583.
Martin-Morales A, Yamamoto M, Inoue M, Vu T, Dawadi R, and Araki M. Predicting Cardiovascular Disease Mortality: Leveraging Machine Learning for Comprehensive Assessment of Health and Nutrition Variables. Nutrients 2023. 15 (18). DOI: 10.3390/nu15183937.
Bewick V, Cheek L, and Ball J. Statistics review 14: Logistic regression. Crit Care 2005. 9 (1): p. 112-118. DOI: 10.1186/cc3045.
Pal M and Parija SR. Prediction of Heart Diseases using Random Forest. Journal of Physics: Conference Series 2021. 1817.
Su X, Xu Y, Tan Z, Wang X, Yang P, Su Y, et al. Prediction for cardiovascular diseases based on laboratory data: An analysis of random forest model. J Clin Lab Anal 2020. 34 (9): p. e23421. DOI: 10.1002/jcla.23421.
Liaw A and Wiener MC. Classification and Regression by randomForest. 2007.
Son YJ, Kim HG, Kim EH, Choi S, and Lee SK. Application of support vector machine for prediction of medication adherence in heart failure patients. Healthc Inform Res 2010. 16 (4): p. 253-259. DOI: 10.4258/hir.2010.16.4.253.
Unnikrishnan P, Kumar DK, Poosapadi Arjunan S, Kumar H, Mitchell P, and Kawasaki R. Development of Health Parameter Model for Risk Prediction of CVD Using SVM. Comput Math Methods Med 2016. 2016: p. 3016245. DOI: 10.1155/2016/3016245.
Hamaguchi T, Saito T, Suzuki M, Ishioka T, Tomisawa Y, Nakaya N, et al. Support Vector Machine-Based Classifier for the Assessment of Finger Movement of Stroke Patients Undergoing Rehabilitation. Journal of Medical and Biological Engineering 2020. 40 (1): p. 91-100. DOI: 10.1007/s40846-019-00491-w.
Sandhya Y. Prediction of Heart Diseases using Support Vector Machine. 2020.
Rajliwall N, Davey R, and Chetty G. Cardiovascular Risk Prediction Based on XGBoost. 2018. 246-252.
Peng M, Hou F, Cheng Z, Shen T, Liu K, Zhao C, et al. Prediction of cardiovascular disease risk based on major contributing features. Scientific Reports 2023. 13 (1): p. 4778. DOI: 10.1038/s41598-023-31870-8.
Yang H, Chen Z, Yang H, and Tian M. Predicting Coronary Heart Disease Using an Improved LightGBM Model: Performance Analysis and Comparison. IEEE Access 2023. PP: p. 1-1. DOI: 10.1109/ACCESS.2023.3253885.
Chowdary BV, Kedarnath J, Vyshnavi RG, Lavakush V, and Shashidhar C. An Effective and Efficient Heart Disease Prediction Model Using Distributed High Performance Light GBM. 2021 Fifth International Conference on I-SMAC (IoT in Social, Mobile, Analytics and Cloud) (I-SMAC) 2021: p. 662-667.
Bloch L and Friedrich CM. Data analysis with Shapley values for automatic subject selection in Alzheimer's disease data sets using interpretable machine learning. Alzheimers Res Ther 2021. 13 (1): p. 155. DOI: 10.1186/s13195-021-00879-4.
Su PY, Wei YC, Luo H, Liu CH, Huang WY, Chen KF, et al. Machine Learning Models for Predicting Influential Factors of Early Outcomes in Acute Ischemic Stroke: Registry-Based Study. JMIR Med Inform 2022. 10 (3): p. e32508. DOI: 10.2196/32508.
Yu Z, Ji H, Xiao J, Wei P, Song L, Tang T, et al. Predicting Adverse Drug Events in Chinese Pediatric Inpatients With the Associated Risk Factors: A Machine Learning Study. Front Pharmacol 2021. 12: p. 659099. DOI: 10.3389/fphar.2021.659099.
Reaven GM. Relationships among insulin resistance, type 2 diabetes, essential hypertension, and cardiovascular disease: similarities and differences. J Clin Hypertens (Greenwich) 2011. 13 (4): p. 238-243. DOI: 10.1111/j.1751-7176.2011.00439.x.
Grundy SM, Stone NJ, Bailey AL, Beam C, Birtcher KK, Blumenthal RS, et al. 2018 AHA/ACC/AACVPR/AAPA/ABC/ACPM/ADA/AGS/APhA/ASPC/NLA/PCNA Guideline on the Management of Blood Cholesterol: A Report of the American College of Cardiology/American Heart Association Task Force on Clinical Practice Guidelines. Circulation 2019. 139 (25): p. e1082-e1143. DOI: 10.1161/cir.0000000000000625.
Landwehr N, Hall M, and Frank E. Logistic Model Trees. Machine Learning 2005. 59: p. 161-205. DOI: 10.1007/s10994-005-0466-3.
Budholiya K, Shrivastava S, and Sharma V. An optimized XGBoost based diagnostic system for effective prediction of heart disease. Journal of King Saud University - Computer and Information Sciences 2020. 34. DOI: 10.1016/j.jksuci.2020.10.013.
Chang V, Ganatra MA, Hall K, Golightly L, and Xu Q. An assessment of machine learning models and algorithms for early prediction and diagnosis of diabetes using health indicators. Healthcare Analytics 2022. 2: p. 100118. DOI: 10.1016/j.health.2022.100118.
Hassan CAU, Iqbal J, Irfan R, Hussain S, Algarni AD, Bukhari SSH, et al. Effectively Predicting the Presence of Coronary Heart Disease Using Machine Learning Classifiers. Sensors (Basel) 2022. 22 (19). DOI: 10.3390/s22197227.
Whelton PK, Carey RM, Aronow WS, Casey DE, Jr., Collins KJ, Dennison Himmelfarb C, et al. 2017 ACC/AHA/AAPA/ABC/ACPM/AGS/APhA/ASH/ASPC/NMA/PCNA Guideline for the Prevention, Detection, Evaluation, and Management of High Blood Pressure in Adults: Executive Summary: A Report of the American College of Cardiology/American Heart Association Task Force on Clinical Practice Guidelines. Hypertension 2018. 71 (6): p. 1269-1324. DOI: 10.1161/hyp.0000000000000066.
Feinstein MJ, Hsue PY, Benjamin LA, Bloomfield GS, Currier JS, Freiberg MS, et al. Characteristics, Prevention, and Management of Cardiovascular Disease in People Living With HIV: A Scientific Statement From the American Heart Association. Circulation 2019. 140 (2): p. e98-e124. DOI: 10.1161/cir.0000000000000695.
Saito I, Yamagishi K, Kokubo Y, Yatsuya H, Iso H, Sawada N, et al. Non-High-Density Lipoprotein Cholesterol and Risk of Stroke Subtypes and Coronary Heart Disease: The Japan Public Health Center-Based Prospective (JPHC) Study. J Atheroscler Thromb 2020. 27 (4): p. 363-374. DOI: 10.5551/jat.50385.
Guo LL, Chen YQ, Lin QZ, Tian F, Xiang QY, Zhu LY, et al. Non-HDL-C Is More Stable Than LDL-C in Assessing the Percent Attainment of Non-fasting Lipid for Coronary Heart Disease Patients. Front Cardiovasc Med 2021. 8: p. 649181. DOI: 10.3389/fcvm.2021.649181.
Martín-Timón I, Sevillano-Collantes C, Segura-Galindo A, and Del Cañizo-Gómez FJ. Type 2 diabetes and cardiovascular disease: Have all risk factors the same strength? World J Diabetes 2014. 5 (4): p. 444-470. DOI: 10.4239/wjd.v5.i4.444.
Ambale-Venkatesh B, Yang X, Wu CO, Liu K, Hundley WG, McClelland R, et al. Cardiovascular Event Prediction by Machine Learning: The Multi-Ethnic Study of Atherosclerosis. Circ Res 2017. 121 (9): p. 1092-1101. DOI: 10.1161/circresaha.117.311312.
McClelland RL, Jorgensen NW, Budoff M, Blaha MJ, Post WS, Kronmal RA, et al. 10-Year Coronary Heart Disease Risk Prediction Using Coronary Artery Calcium and Traditional Risk Factors: Derivation in the MESA (Multi-Ethnic Study of Atherosclerosis) With Validation in the HNR (Heinz Nixdorf Recall) Study and the DHS (Dallas Heart Study). J Am Coll Cardiol 2015. 66 (15): p. 1643-1653. DOI: 10.1016/j.jacc.2015.08.035.
Powell-Wiley TM, Poirier P, Burke LE, Després JP, Gordon-Larsen P, Lavie CJ, et al. Obesity and Cardiovascular Disease: A Scientific Statement From the American Heart Association. Circulation 2021. 143 (21): p. e984-e1010. DOI: 10.1161/cir.0000000000000973.
Muscella A, Stefàno E, and Marsigliante S. The effects of exercise training on lipid metabolism and coronary heart disease. Am J Physiol Heart Circ Physiol 2020. 319 (1): p. H76-h88. DOI: 10.1152/ajpheart.00708.2019.
Charoen P, Nitsch D, Engmann J, Shah T, White J, Zabaneh D, et al. Mendelian Randomisation study of the influence of eGFR on coronary heart disease. Sci Rep 2016. 6: p. 28514. DOI: 10.1038/srep28514.
Jahangiry L, Farhangi MA, and Rezaei F. Framingham risk score for estimation of 10-years of cardiovascular diseases risk in patients with metabolic syndrome. J Health Popul Nutr 2017. 36 (1): p. 36. DOI: 10.1186/s41043-017-0114-0.
Nishimura K, Okamura T, Watanabe M, Nakai M, Takegami M, Higashiyama A, et al. Predicting coronary heart disease using risk factor categories for a Japanese urban population, and comparison with the framingham risk score: the suita study. J Atheroscler Thromb 2014. 21 (8): p. 784-798. DOI: 10.5551/jat.19356.

No competing interests reported.

20240318forSReSupplCHDML.docx

Download PDF

Version 1

posted

You are reading this latest preprint version

Unveiling Coronary Heart Disease Prediction through Machine Learning Techniques: Insights from the Suita Population-Based Cohort Study

Status:

Version 1

Abstract

Figures

1. Introduction

2. Methods

2.1. Study Participants

2.2. Data collection

2.3. Risk Factors and Anthropomorphic Measurements

2.4. Confirmation of CHD

2.5. Statistical Methods

2.6. Data preprocessing and variable selection

2.7. Model development

2.8. Extraction of important variables for CHD risk

3. Results

4. Discussion

Declarations

References

Additional Declarations

Supplementary Files

Status:

Version 1