Determinants of Developing Cardiovascular Disease Risk with Emphasis on Type-2 Diabetes and Predictive Modeling Utilizing Machine Learning Algorithms

doi:10.21203/rs.3.rs-4724144/v1

Download PDF

Research Article

Determinants of Developing Cardiovascular Disease Risk with Emphasis on Type-2 Diabetes and Predictive Modeling Utilizing Machine Learning Algorithms

https://doi.org/10.21203/rs.3.rs-4724144/v1

This work is licensed under a CC BY 4.0 License

You are reading this latest preprint version

Background

This research aims to enhance our comprehensive understanding of the influence of type-2 diabetes on the development of Cardiovascular diseases (CVD) risk, its underlying determinants, and to construct precise predictive models capable of accurately assessing CVD risk within the context of Bangladesh.

Methods

This study combined data from the 2011 and 2017-18 Bangladesh Demographic and Health Surveys, focusing on individuals with hypertension. CVD development followed WHO guidelines. Eight machine learning algorithms (Support Vector Machine, Logistic Regression, Decision Tree, Random Forest, Naïve Bayes, K-Nearest Neighbor, Light GBM, and XGBoost) were analyzed and compared using six evaluation metrics to assess model performance.

Results

The study reveals that individuals aged 35–54 years, 55–69 years, and ≥ 70 years face higher CVD risk with adjusted odds ratios (AOR) of 2.140, 3.015, and 3.963, respectively, compared to those aged 18–34 years. 'Rich' respondents show increased CVD risk (AOR = 1.370, p < 0.01) compared to 'poor' individuals. Also, 'normal weight' (AOR = 1.489, p < 0.01) and 'overweight/obese' (AOR = 1.871, p < 0.01) individuals exhibit higher CVD risk than 'underweight' individuals. The predictive models achieve impressive performance, with 75.21% accuracy and an 80.79% AUC, with Random Forest (RF) excelling in specificity at 76.96%.

Conclusion

This research holds practical implications for targeted interventions based on identified significant factors, utilizing ML models for early detection and risk assessment, enhancing awareness and education, addressing urbanization-related lifestyle changes, improving healthcare infrastructure in rural areas, and implementing workplace interventions to mitigate stress and promote physical activity.

cardiovascular diseases (CVD)

diabetes

risk factors

machine learning

Bangladesh

Cardiovascular diseases (CVD) pose a significant global health challenge, ranking among the most pressing concerns worldwide. With its status as the primary cause of mortality on a global scale, CVD encompasses a range of heart and blood vessel conditions [1, 3, 4]. This group of diseases stands as the leading cause of death in middle-aged and elderly populations, accounting for one-third of all fatalities [5]. Tragically, low-income and middle-income countries experience a disproportionately high burden of cardiovascular disease-related deaths [6]. Moreover, the economic impact weighs heavily on patients and their families, imposing a considerable financial strain [7, 8]. On a broader societal scale, cardiovascular diseases exact a significant economic cost [9], with estimates indicating a staggering global burden of $3.7 trillion between 2010 and 2015 [4].

CVD accounts for over 31% of global deaths [15]. In Asian countries, heart diseases are responsible for 22% of total mortality [16]. In Bangladesh, a staggering increase in cardiovascular disease-related deaths, with the number rising from 11 per 10,000 people in 1986 to 411 per 10,000 people in 2006 [12]. In 2018 alone, 0.256 million people lost their lives to cardiovascular disease in Bangladesh. Notably, cardiovascular disease stands as the leading cause of death in middle- and low-income nations, contributing to over 75% of all fatalities [14, 19]. According to a study published in January 2017, cardiovascular disease (CVD) has become the leading cause of death worldwide, climbing to the top of the list of the top 10 causes of death over the previous 15 years. In 2015 alone, it claimed 15 million lives [17]. The World Health Organization (WHO) estimated that 17.5 million individuals died from cardiovascular disease (CVD) in 2005 [1]. By 2015, CVD had claimed the lives of approximately 17.9 million individuals [14, 18]. Each year, 12 million people succumb to cardiovascular disease, making it the primary cause of mortality in both developing and developed nations. The WHO further predicts a 24.5% increase in deaths by 2030 [10, 13]. Among all CVDs, heart attacks and strokes account for 85% of deaths [6].

Multiple factors, either directly or indirectly, contribute to cardiovascular diseases [3, 12, 13]. High blood pressure, body mass index (BMI), smoking, diabetes, cholesterol levels, age, gender, stress, and family history all serve as risk factors for these disorders [3, 5, 12, 13]. Early detection and prediction play crucial roles in combating this disease, with counseling and medication forming the foundation of treatment when diagnosed early enough [2, 3]. Therefore, emphasis on the pre-detection of cardiovascular disease becomes paramount [10, 14].

Understanding the complex interplay between various risk factors and their effects on CVD is crucial for effective risk assessment and prevention strategies. Previous research has identified several established risk factors, such as hypertension, diabetes, obesity, and family history. However, there is still a need to explore research with advanced techniques to identify factors that may contribute to the development of CVD. Besides these factors it is highly necessary to identify a predictive model that has the ability of early prediction of the disease’s status. Machine learning algorithms offer a promising approach for predicting CVD risk with improved accuracy and precision. These algorithms can handle large and complex datasets, identify non-linear relationships, and discover patterns that may not be apparent through conventional statistical methods. Therefore, this manuscript aims to investigate the factors associated with the high risk of developing cardiovascular disease and explore the feasibility of predicting CVD using machine learning algorithms. Through the exploration of a wide array of variables and the utilization of advanced computational techniques, this research aims to enhance our comprehensive understanding of the influence of type-2 diabetes on the development of CVD risk, its underlying determinants, and to construct precise predictive models capable of accurately assessing CVD risk within the context of Bangladesh.

Sources of Data

Our research draws upon data extracted from two distinct surveys: the 2011 Bangladesh Demographic and Health Survey (BDHS) and the 2017-18 BDHS. In the case of the 2011 BDHS, we meticulously compiled information from a cohort of 1,154 individuals who had been diagnosed with hypertension, ensuring that their data records were complete and devoid of any missing values. Similarly, for the 2017-18 BDHS, we gathered data from a larger group of 2,973 individuals who had hypertension, again ensuring that there were no gaps or missing data points. By combining and merging the datasets from both survey years, we were able to amass a comprehensive dataset comprising a total of 4,127 respondents who had been diagnosed with hypertension. This unified data set forms the core of our research and serves as the basis for our analyses and findings. The data is freely available in the following link:

https://dhsprogram.com/data/available-datasets.cfm

Sample Design and Sampling Frame

Starting in 1984, the DHS Program has provided technical support to conduct more than 300 demographic and health surveys across more than 90 countries. These surveys employ a meticulously designed sampling methodology. The process begins with a stratified multistage cluster design overseen by the DHS. In the initial phase, probability proportional to size (PPS) is applied to choose the principal sample units (PSUs) within each stratum. Subsequently, in the second phase, an exhaustive inventory of households is compiled for each selected cluster. Within these chosen clusters, households are then systematically sampled with equal probability to ensure a representative and robust data collection process.

Variable Description

Dependent Variable

In this research, our dependent variable of interest is the "Development of Cardiovascular Diseases Risk." To construct this variable, we have derived it from a combination of factors related to hypertensive patients and diabetes. We have substantially reorganized the data originally presented in the 1999 WHO/ISH recommendations to delineate three primary risk groups for significant cardiovascular events expected to occur within the next decade among individuals with hypertension [5]. Initially, we focus on hypertensive individuals with systolic blood pressure (SBP) ≥ 140 or diastolic blood pressure (DBP) ≥ 90, categorizing them into three distinct stages: Stage 1 (SBP 140–159 or DBP 90–99), Stage 2 (SBP 160–179 or DBP 100–109), and Stage 3 (SBP ˃ 180 or DBP ˃ 110). In addition, we consider the type-2 diabetes (T2D) status of these individuals, where those with a fasting plasma blood glucose level ≥ 7.0 mmol/L are classified as having diabetes, while those with levels below this threshold are categorized as not having diabetes. Based on the presence or absence of diabetes, we further classify hypertensive patients into three categories: low risk, medium risk, and high risk. This composite variable, referred to as the "Development of Cardiovascular Diseases Risk," encompasses these three distinct risk categories and serves as the primary focus of our study (see, Table 1).

Table 1

Measuring variable with Blood pressure and Diabetes
Blood pressure (mmHg)
Type-2 Diabetes Status	Stage 1	Stage 2	Stage 3
Type-2 Diabetes Status		SBP 140–159 or DBP 90–99	SBP 160–179 or DBP 100–109	SBP ˃180 or DBP ˃110
Absent	Low	Medium	High
Present	Medium	High	High

For our analysis, we pinpointed individuals at a higher risk of developing cardiovascular diseases (CVD). This high-risk group includes those with a previous history of diabetes and/or those falling into the higher blood pressure categories (Stage 2 or Stage 3). In other words, individuals meeting these criteria are considered to be at high risk for CVD). That is

$$\:High\:risk\:of\:developing\:CVD=\left\{\begin{array}{c}1;Yes;if\:T2D\:present\:and/or\:fall\:into\:the\:Stage\:2\:or\:3\:\:\:\\\:0;Otherwise\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\end{array}\right.$$

Independent Variables

The independent variables in our study encompass a set of 10 attributes: division, gender, employment status, age, education, wealth index, residential location, marital status, household size, and BMI. Notably, the 2011 BDHS featured seven divisions: Barisal, Chittagong, Dhaka, Khulna, Rajshahi, Rangpur, and Sylhet. In contrast, the 2017-18 BDHS data included eight divisions, with the addition of Mymensingh. It is worth noting that Dhaka and Mymensingh divisions were merged on September 14, 2015, resulting in seven categories for our analysis. We also categorized the wealth index into five groups: lowest, poorer, medium, richer, and wealthiest. To simplify our analysis, we combined the poorest and poorer categories into "poor," designated the medium category as "middle," and retained the wealthiest category as "wealthy."

Data pre-processing

Data cleaning

Upon combining the data from the 2011 BDHS and the 2017-18 BDHS, we initially had a dataset comprising 21,959 respondents. However, not all observations contained complete information. Specifically, out of these, 914 samples provided comprehensive data pertaining to the development of cardiovascular diseases and other relevant features. Following a rigorous process of data cleaning to address missing values, we retained a final dataset containing 914 complete observations.

Data balancing

Imbalance in class distribution has been a persistent challenge for researchers in various machine learning applications, adversely impacting accuracy [3]. When the number of samples varies significantly across different classes, it poses a significant issue. To address this, a common approach involves balancing the underrepresented class in the dataset [20]. In this study, we employed the Synthetic Minority Oversampling Technique (SMOTE) to rectify this class imbalance issue. Following the balancing of the dataset, all subsequent analyses were conducted using this balanced dataset, wherein various machine learning algorithms were applied for the classification of cardiovascular disease development.

Statistical Analysis and machine learning models

After filtering out missing, ineligible, and non-responsive cases for the questions, we scrutinized the completeness of the extracted data. Initially, we conducted univariate analysis and assessed the bivariate associations between independent and dependent variables using the chi-square test. Subsequently, employing multivariable binary logistic regressions, we explored the impact of other factors. To address the complexity of the sample design, we applied sample weights provided with the data to all our analyses. All statistical analyses were conducted using MS-Excel and SPSS Windows version 25.0 on the remaining dataset. The outcomes are presented in tabular form.

Following this, we employed a variety of machine learning models, including Support Vector Machine (SVM), Decision Tree, Random Forest, Naïve Bayes, K-Nearest Neighbor, Light Gradient Boosting Machine (LightGBM), and Extreme Gradient Boosting Machine (XGBoost), to identify the most accurate model for predicting CVD risk. The performance of these machine learning models was evaluated using various metrics such as Accuracy, Precision, Sensitivity, Specificity, F1 Score, and Area Under the Curve (AUC) values.

Table 2 provides a comprehensive overview of the data from BDHS-2011, BDHS-2017-18, and the combined dataset, offering insights into the socio-demographic characteristics of the respondents. As depicted in Table 2, noteworthy findings emerged. In the BDHS-2011 data, the Khulna division exhibited the highest representation, accounting for 20.5% of the sample, while the Sylhet division had the lowest frequency at 9.7%. Among the respondents, males constituted the majority, comprising 53.2% of the total. A significant proportion of respondents, approximately 47.8%, had no formal education or were in the preschool category. Furthermore, a substantial portion, 57.2%, were not actively employed. The highest frequency of respondents, at 38.7%, was observed among those with 1–4 household members. In terms of age distribution, the 55–69 years age group stood out with the highest frequency, encompassing 40.0% of the respondents. Those categorized as 'rich' based on wealth index accounted for the largest share, representing 53.7% of the dataset. Additionally, rural areas were predominant, with 62.0% of the respondents residing there. The highest frequency, at 71.3%, was recorded among respondents who were currently married. In relation to body weight, the category of "normal-weighted" respondents held the highest frequency at 58.0%, while the category of respondents with a low risk of developing cardiovascular diseases had the highest frequency at 77.7%.

Table 2

Descriptive statistics of the BDSH 2011, BDHS 2017-18 & Combined data
Variables	Categories	BDHS-2011	BDHS 2017-18	Combined data
Variables	Categories	Total (%)	Total (%)	Total (%)
Division	Barisal	113 (9.8)	351 (11.8)	464 (11.2)
	Chittagong	139 (12.0)	410 (13.8)	549 (13.3)
	Dhaka	199 (17.2)	595 (20.0)	794 (19.2)
	Khulna	236 (20.5)	452 (15.2)	688 (16.7)
	Rajshahi	150 (13.0)	410 (13.8)	560 (13.6)
	Rangpur	205 (17.8)	452 (15.2)	657 (15.9)
	Sylhet	112 (9.7)	303 (10.2)	415 (10.1)
Sex of household member	Male	614 (53.2)	1272 (42.8)	1886 (45.7)
Sex of household member	Female	540 (46.8)	1701 (57.2)	2241 (54.3)
Highest education level attained	No education	552 (47.8)	966 (32.5)	1518 (36.8)
	Primary	276 (23.9)	866 (29.1)	1142 (27.7)
	Secondary	195 (16.9)	725 (24.4)	920 (22.3)
	Higher	131 (11.4)	416 (14.0)	547 (13.3)
Work Status	No	660 (57.2)	1246 (41.9)	1906 (46.2)
Work Status	Yes	494 (42.8)	1727 (58.1)	2221 (53.8)
Number of household member	1–4	447 (38.7)	1298 (43.7)	1745 (42.3)
	5–6	395 (34.2)	973 (32.7)	1368 (33.1)
	≥ 7	312 (27.0)	702 (23.6)	1014 (24.6)
Age	18–34 years	Null	618 (20.8)	618 (15.0)
	35–54 years	418 (36.2)	1209 (40.7)	1627 (39.4)
	55–69 years	462 (40.0)	754 (25.4)	1216 (29.5)
	≥ 70 years	274 (23.7)	392 (13.2)	666 (16.1)
Wealth index	Poor	326 (28.2)	1010 (34.0)	1336 (32.4)
	Middle	208 (18.0)	581 (19.5)	789 (19.1)
	Rich	620 (53.7)	1382 (46.5)	2002 (48.5)
Place of residence	Rural	716 (62.0)	1881 (63.3)	2597 (62.9)
Place of residence	Urban	438 (38.0)	1092 (36.7)	1530 (37.1)
Marital status	Never married/ divorced/ separated/ widow	331 (28.7)	661 (22.2)	992 (24.0)
Marital status	Currently married	823 (71.3)	2312 (77.8)	3135 (76.0)
Body mass index (BMI)	Underweight	275 (23.8)	348 (11.7)	623 (15.1)
	Normal	669 (58.0)	1545 (52.0)	2214 (53.6)
	Overweight/ Obese	210 (18.2)	1080 (36.3)	1290 (31.3)
High risk of developing CVD	No	897 (77.7)	2285 (76.9)	3182 (77.1)
High risk of developing CVD	Yes	257 (22.3)	688 (23.1)	945 (22.9)

Table 3

Bivariate association in combined dataset
Variables	High risk of developing CVD		$\:{\varvec{\chi\:}}^{2}$	P-value
	No (%)	Yes (%)
Division			13.622	< 0.05
Barisal	364(8.8)	100(2.4)
Chittagong	405(9.8)	144(3.5)
Dhaka	600(14.5)	194(4.7)
Khulna	522(12.6)	166(4.0)
Rajshahi	432(10.5)	128(3.1)
Rangpur	537(13.0)	120(2.9)
Sylhet	322(7.8)	93(2.3)
Sex of household member			0.777	0.378
Male	1466(35.5)	420(10.2)
Female	1716(41.6)	525(12.7)
Highest education level attained			2.037	0.565
No education, preschool	1166(28.3)	352(8.5)
Primary	897(21.7)	245(5.9)
Secondary	700(17.0)	220(5.3)
Higher	419(10.2)	128(3.1)
Work status			8.212	< 0.01
No	1431(34.7)	475(11.5)
Yes	1751(42.4)	470(11.4)
Number of household members			1.416	0.493
1–4	1352(32.8)	393(9.5)
5–6	1040(25.2)	328(7.9)
≥ 7	790(19.1)	224(5.4)
Age			62.839	< 0.01
18–34 years	542(13.1)	76(1.8)
35–54 years	1270(30.8)	357(8.7)
55–69 years	900(21.8)	316(7.7)
≥ 70 years	470(11.4)	196(4.7)
Wealth index			36.153	< 0.01
Poor	1091(26.4)	245(5.9)
Middle	627(15.2)	162(3.9)
Rich	1464(35.5)	538(13)
Place of residence			11.734	< 0.01
Rural	2047(49.6)	550(13.3)
Urban	1135(27.5)	395(9.6)
Marital status			2.135	0.144
Never married/ divorced/ separated/ widow	748(18.1)	244(5.9)
Currently married	2434(59.0)	701(17.0)
Body mass index (BMI)			19.913	<0.01
Underweight	515(12.5)	108(2.6)
Normal	1717(41.6)	497(12.0)
Overweight/ Obese	950(23.0)	340(8.2)

In the BDHS 2017-18 dataset, as presented in Table 2, the Dhaka division exhibited the highest representation, constituting 20.0% of the sample, whereas Sylhet had the lowest frequency at 10.2%. Female respondents accounted for the majority, comprising 57.2% of the dataset. The largest frequency among respondents was observed in the category with no formal education or preschool background, accounting for 32.5%. A significant proportion of respondents, approximately 58.1%, were currently employed. In terms of household size, those with 1–4 household members had the highest frequency at 43.7%. The age group ranging from 35 to 54 years represented the largest share, with 40.7% of respondents falling into this category. Additionally, a notable 46.5% were classified under the 'rich' category based on their wealth index. Rural areas were predominant, with 63.3% of respondents residing there. Furthermore, the highest frequency, at 71.8%, was recorded among respondents who were currently married. Concerning body weight, 'normal-weighted' respondents held the highest frequency at 52.0%, while 'underweight' respondents had the lowest frequency at 11.7%. Among respondents, those with the lowest risk of developing cardiovascular diseases exhibited the highest frequency, encompassing 76.9% of the sample.

In the combined dataset, as illustrated in Table 2, the Sylhet division displayed the lowest frequency at 10.1%, whereas the Dhaka division exhibited the highest representation at 19.2%. Females constituted the majority of respondents, accounting for 54.3% of the dataset. Among respondents, the highest frequency was observed in the category reporting no formal education or preschool experience, standing at 36.8%. A significant proportion, approximately 53.8%, were currently employed. In terms of household size, the largest frequency was found among those with 1–4 household members, representing 42.3% of the sample. The age group spanning 35–54 years had the highest frequency, with 39.4% of respondents falling into this category. Furthermore, 48.5% of the population fell under the high wealth index classification. Rural areas were predominant, with 62.9% of respondents residing there. Additionally, the highest frequency, at 76.0%, was reported among respondents who were currently married. In relation to body weight, respondents categorized as 'normal-weighted' had the highest frequency, accounting for 53.6%. Conversely, those with a low risk of developing cardiovascular disease had the highest frequency, which was 77.1%.

Table 4 presents the results of binary logistic regression analysis conducted exclusively with the combined dataset. The table reveals that all age categories yield statistically significant results $\:(p\le\:0.05).$ Specifically, respondents aged 35–54 years, 55–69 years, and those above 70 years are 2.140 times, 3.015 times, and 3.963 times more likely, respectively, to exhibit a high risk of developing cardiovascular diseases compared to respondents aged 18–34 years. Regarding the wealth index, affluent respondents are 1.370 times more likely $\:(AOR=1.370,\:p\le\:0.01)$ to face a high risk of developing cardiovascular diseases in comparison to their less affluent counterparts. Additionally, the likelihood of having a high risk for developing cardiovascular diseases is 1.489 times higher for respondents with a normal weight and 1.871 times higher for those categorized as overweight or obese, as opposed to respondents classified as underweight. Notably, the variable of BMI exhibits statistically significant results across all categories $\:(p\le\:0.05).$

Table 4

Logistic regression model showing factors affecting the development of cardiovascular diseases in combined data
Variable	Categories	AOR	P value	95% C.I for OR
Variable	Categories	AOR	P value	Lower	Upper
Division	Barisal (ref)
	Chittagong	1.235	0.170	0.914	1.670
	Dhaka	1.129	0.401	0.851	1.497
	Khulna	1.060	0.694	0.793	1.417
	Rajshahi	1.099	0.541	0.811	1.490
	Rangpur	0.876	0.395	0.646	1.188
	Sylhet	1.098	0.580	0.789	1.527
Sex of household member	Male (ref)
Sex of household member	Female	1.143	0.184	0.938	1.392
Highest education level attained	No education, preschool (ref)
	Primary	.992	0.934	0.811	1.213
	Secondary	1.180	0.161	0.936	1.487
	Higher	1.139	0.380	0.852	1.522
Work status	No (ref)
Work status	Yes	1.002	0.982	0.827	1.215
Number of household members	1–4 (ref)
	5–6	1.030	0.741	0.865	1.226
	≥ 7	.856	0.122	0.702	1.043
Age	18–34 years (ref)
	35–54 years	2.140	< 0.01	1.619	2.829
	55–69 years	3.015	< 0.01	2.247	4.047
	≥ 70 years	3.963	< 0.01	2.838	5.535
Wealth index	Poor (ref)
	Middle	1.074	0.542	0.854	1.351
	Rich	1.370	< 0.01	1.111	1.690
Place of residence	Rural (ref)
Place of residence	Urban	1.111	0.215	0.941	1.312
Marital status	Never married/ divorced/ separated/ widow (ref)
Marital status	Currently married	0.962	0.715	0.783	1.183
Body mass index (BMI)	Underweight (ref)
	Normal	1.489	< 0.01	1.170	1.896
	Overweight/ Obese	1.871	< 0.01	1.429	2.451
Note: Ref: Reference category

Table 5 provides a comprehensive evaluation of selected algorithms used to assess classification performance. The findings in Table 5 indicate that the Random Forest (RF) model consistently excels, achieving the highest accuracy rate at 75.21%, precision at 75.38%, sensitivity at 73.08%, F1 score at 75.19%, and an AUC value of 80.79%. In terms of specificity, the SVM model stands out, boasting the highest value at 77.75%. Consequently, when considering the overall classification results, the Random Forest model emerges as the most effective algorithm for predicting the onset of cardiovascular disease.

Table 5

Classification performance measure of the algorithms and comparison
Algorithms	Accuracy	Precision	Sensitivity	Specificity	F1 score	AUC
SVM	73.38%	73.43%	68.68%	77.75%	73.31%	75.65%
DT	72.31%	72.42%	69.66%	74.78%	72.28%	74.37%
RF	75.21%	75.38%	73.08%	76.96%	75.19%	80.79%
LightGBM	70.58%	70.66%	68.64%	72.59%	70.58%	74.74%
XGBoost	74.35%	74.35%	71.81%	76.62%	74.33%	80.28%
LR	60.02%	60.03%	59.56%	60.46%	60.02%	64.74%
KNN	71.01%	71.39%	59.43%	60.59%	70.82%	79.16%
NB	60.34%	60.44%	56.53%	62.04%	60.29%	63.46%

10-fold cross validation of the accuracy of the classifiers

Table 6 presents the average classification accuracy resulting from 10-fold cross-validation for all machine learning models. The table's findings reveal that the Random Forest (RF) model achieved the highest average accuracy score, reaching 71%. In contrast, both SVM and XGBoost models obtained the same accuracy score, which stood at 69%. Additionally, the Decision Tree (DT) and K-Nearest Neighbor (KNN) models demonstrated identical average accuracy rates of 68%. Moreover, LightGBM, Logistic Regression (LR), and Naïve Bayes (NB) models exhibited average accuracy scores of 67%, 59%, and 58%, respectively. Notably, among the selected models, Naïve Bayes (NB) yielded the lowest average accuracy.

Table 6

10-fold cross validation scores
Fold	ML classifier
Fold	SVM	DT	RF	LightGBM	XGBoost	LR	KNN	NB
Fold-01	0.74	0.68	0.70	0.67	0.70	0.60	0.69	0.54
Fold-02	0.68	0.67	0.69	0.69	0.67	0.57	0.69	0.59
Fold-03	0.72	0.70	0.75	0.67	0.72	0.60	0.70	0.57
Fold-04	0.61	0.63	0.66	0.68	0.71	0.64	0.67	0.61
Fold-05	0.65	0.67	0.70	0.64	0.68	0.57	0.68	0.56
Fold-06	0.71	0.69	0.74	0.65	0.68	0.57	0.69	0.61
Fold-07	0.71	0.68	0.72	0.68	0.67	0.56	0.62	0.57
Fold-08	0.68	0.69	0.73	0.64	0.67	0.63	0.68	0.57
Fold-09	0.69	0.68	0.71	0.66	0.74	0.58	0.69	0.59
Fold-10	0.66	0.67	0.71	0.68	0.71	0.58	0.66	0.59
Average	0.69	0.68	0.71	0.67	0.69	0.59	0.68	0.58

Figure 1 visually depicts the classification performance of all classifiers (SVM, DT, RF, LightGBM, XGBoost, LR, KNN, NB) through receiver operating curves (ROC) and their corresponding area under the curve (AUC) values. The ROC curve serves as a competency measurement plot for any classifier, representing the trade-off between true positive rate (sensitivity) and false positive rate (1-specificity) for object classification. Different points on the curve correspond to various decision thresholds used to classify objects as positive or negative, revealing the optimal balance between sensitivity and false positive rate. The results from Fig. 1 indicate that the Random Forest (RF) classifier achieved the highest AUC value at 80.79%, outperforming all other classifier models. The XGBoost model secured the second-highest AUC value at 80.28%. Notably, all other classifiers achieved AUC values exceeding 70%, with the exceptions being Logistic Regression (LR) and the Naïve Bayes classifier, both of which scored below 65%.

Cardiovascular disease, a highly heritable trait, causes major deaths worldwide. Though the prevalence of CVD has been found high worldwide, the awareness rate has been found very low [37]. CVD is concerned as highly heritable trait, but the micronutrients intake, age, socio-economic condition, and environmental toxic metal condition can also cause severe risk of CVD [38, 39, 40]. Throughout our study, we have tried to identify the risk factors and predict CVD using different machine learning models. The hypertensive patients from BDHS-2011 and BDHS-2017-18 datasets have been used throughout the study for analysis.

Several recent studies have used ML algorithms to predict the cardiovascular diseases which indicate the reliability and the feasibility of this method in this case [41, 42, 43]. Chandralekha & Shenbagavadivu compared supervised and unsupervised ML models and found that Decision Tree has more classification accuracy, and precision with 73%, and 91% respectively. Another study conducted by Arunachalam found Support Vector Machine and Multilayer Perceptron with the highest accuracy score (91.7%). It also identified chest pain type, thalassemia, age, depression, cholesterol, gender, blood pressure as the most effective factors for CVD.

In this study, some statistical analysis such as frequency distribution and chi-square test were conducted to identify the patterns and also the significant factors. A slight increment of CVD was found from BDHS 2011 to BDHS 2017-18 with the prevalence of 22.3% and 23.1% respectively. Chi-square analysis determined division, work status, age, wealth index, place of residence and bmi as significant factors. Besides statistical analysis, 8 different ML classifier models were used to predict CVD. Among these, Random Forest was identified with the highest accuracy, precision, sensitivity, and F1 score with 78%, 78%, 74%, and 78% respectively. The features division, age, highest education level, bmi, wealth index, place of residence and work status has been identified as most important.

The highest prevalence of CVD had been occurred in Dhaka that might be the result of rapid urbanization, dietary changes, increased consumption of tobacco, limited physical activity, low level of awareness, and also the poor detection and control rate [44]. This study has also shown that bmi is also working as a significant factor for CVD because obesity irritates plaque in the arteries and predisposes, releases substances in the blood that make plaque rapture, and also develops atrial fibrillation, increases triglyceride levels which triggers heart attacks, plaque rupturing, and stevens notes [45]. Moreover, age is also an important feature for CVD, since it has been linked to obesity, persistent inflammation, and oxidative stress which may increase the risk of heart diseases [46]. Our study has also been found that the prevalence of CVD is higher in rural areas that may happen because of the low level of awareness among people and also the inadequate health qualities [47]. Another important risk factor determined by our study is wealth index which has also been found positively correlated with CVD. This may happen because of the accessibility of high-calories food from well-off families and also related with less physical activities [48].

Working status has also been found positively correlated with CVD that means that less physical activity as well as intaking high-calories food and also stress may increase the risk of CVD [49]. Moreover, the ML models determined education level as the most important significant factors for CVD. This explains the fact that low education may lead to low awareness and knowledge of healthy lifestyle, and also the risk of CVD [50].

The findings of this study provide valuable insights and practical implications for addressing cardiovascular disease (CVD). The statistical analysis identified several significant factors associated with CVD, including division, work status, age, wealth index, place of residence, and BMI. These factors can help healthcare professionals and policymakers prioritize interventions and allocate resources effectively.

The study also employed machine learning (ML) models, with Random Forest achieving the highest accuracy, precision, sensitivity, and F1 score for predicting CVD. This suggests that ML models can be utilized as a reliable tool for early detection and risk assessment of CVD. The identified important features, such as division, age, highest education level, BMI, wealth index, place of residence, and work status, can guide the development of targeted interventions. For example, focusing on urban areas like Dhaka, where a higher prevalence of CVD was observed, interventions can address factors like rapid urbanization, dietary changes, increased tobacco consumption, limited physical activity, low awareness, and inadequate detection and control rates.

Promoting awareness and education about healthy lifestyles, especially among individuals with lower education levels, can help mitigate the risk of CVD. Targeted interventions in rural areas, aiming to improve health infrastructure and increase awareness, can contribute to reducing the burden of CVD in those communities. Addressing the correlation between wealth index and CVD requires strategies to promote healthy eating habits and physical activity among all socioeconomic groups. Workplace interventions focusing on reducing stress and promoting physical activity can also contribute to preventing CVD.

Limitations

Since there was a significant gap between the two BDHS datasets that were combined, this may have influenced the results. Respondents related to the topic were very limited, for which the sample size is very small. Fasting plasma glucose (FPG) readings are used to monitor diabetes in BDHS, but they do not constitute a clinical diagnosis of the disease because, according to the WHO, "FPG alone cannot be used to diagnose diabetes, as it fails to diagnose around 30% of cases of previously undiagnosed diabetes. However, there is still room for improvement in the method that is currently being used.

In literature, numerous researches disclose many classification techniques to identify the better diagnosis for cardiovascular diseases but the performance of the classifier is still inconsistent and none of the research was done based on Bangladesh. So, the aim of this study is to improve the literature with suggested classification techniques that yield a classifier with better accuracy for predicting the development of cardiovascular diseases in Bangladesh. Eight machine learning algorithms were used to predict the development of cardiovascular diseases and compared. For the measures of the classification performance, six types of evaluating measure were used such as accuracy, precision, sensitivity, specificity, F1 score and AUC value. I demonstrated my analysis by showing an accuracy at 78% and area under curve (AUC) at 84% through Random Forest classifier. For specificity, KNN outperformed other algorithms. This concept has the potential to revolutionize the medical industry. By using this technique, it may be possible to identify heart disease-at-risk patients quickly, potentially reducing the rising death rate. Future advancements in machine learning algorithms will lead to a rise in the prevalence of this type of diagnosis. The model might be improved and modified if more patient data is used. Adapting this approach to other types of datasets will be intriguing in the future, as it could provide a time and money saving option for cardiovascular patients and doctors alike.

Conflict of Interest:

Any authors have no conflict of interest.

Ethical approval:

This study used a secondary data collected by NIPORT, Bangladesh and MEASURE DHS. All procedures performed in this study involving human participants were in accordance with the ethical standards of the national research committee and with the 1964 Helsinki Declaration and its later amendments or comparable ethical standards. As the data is freely available in the website, ethical review and approval was not required for the study on human participants in accordance with the local legislation and institutional requirements.

Funding:

No fund has been received

Author Contribution

S.D., R.H and A.T. wrote the main manuscript text and prepared figures as well as Tables. All authors reviewed the manuscript.

Data Availability

Data is freely available in the public domain with the following link: https://dhsprogram.com/data/dataset_admin/login_main.cfm?CFID=300458265&CFTOKEN=26bd09600ba5696-DE5E82D0-A6C1-5C66-C216C2CD9B9E82D9

Jin Z, Oresko J, Huang S, Cheng AC. HeartToGo: a personalized medicine technology for cardiovascular disease prevention and detection. 2009 IEEE/NIH Life Science Systems and Applications Workshop. IEEE; 2009, April. pp. 80–3.
Sabab SA, Munshi MAR, Pritom AI. (2016, December). Cardiovascular disease prognosis using effective classification and feature selection technique. In 2016 International Conference on Medical Engineering, Health Informatics and Technology (MediTec) (pp. 1–6). IEEE.
Rahim A, Rasheed Y, Azam F, Anwar MW, Rahim MA, Muzaffar AW. An integrated machine learning framework for effective prediction of cardiovascular diseases. IEEE Access. 2021;9:106575–88.
Maiga J, Hungilo GG. (2019, October). Comparison of machine learning models in prediction of cardiovascular disease using health record data. In 2019 International Conference on Informatics, Multimedia, Cyber and Information System (ICIMCIS) (pp. 45–48). IEEE.
World Health Organization, & International Society of Hypertension Writing Group. (2003). 2003 World Health Organization (WHO)/International Society of Hypertension (ISH) statement on management of hypertension. Journal of hypertension, 21(11), 1983–1992.
Kaptoge, S., Pennells, L., De Bacquer, D., Cooney, M. T., Kavousi, M., Stevens, G.,… Di Angelantonio, E. (2019). World Health Organization cardiovascular disease risk charts: revised models to estimate risk in 21 global regions. The Lancet Global Health,7(10), e1332-e1345.
Keeley EC, Boura JA, Grines CL. Primary angioplasty versus intravenous thrombolytic therapy for acute myocardial infarction: a quantitative review of 23 randomised trials. Lancet. 2003;361(9351):13–20.
World Health Organization. Prevention of cardiovascular disease: guidelines for assessment and management of total cardiovascular risk. World Health Organization; 2007.
Romiti S, Vinciguerra M, Saade W, Cortajarena A, I., Greco E. (2020). Artificial intelligence (AI) and cardiovascular diseases: an unexpected alliance. Cardiology Research and Practice, 2020.
Sharma G, Rani G, Dhaka VS. (2020, November). A Review on Machine Learning Techniques for Prediction of Cardiovascular Diseases. In 2020 Sixth International Conference on Parallel, Distributed and Grid Computing (PDGC) (pp. 237–242). IEEE.
Yadav SS, Jadhav SM, Nagrale S, Patil N. (2020, March). Application of machine learning for the detection of heart disease. In 2020 2nd International Conference on Innovative Mechanisms for Industry Applications (ICIMIA) (pp. 165–172). IEEE.
Rahman M, Nakamura K, Seino K, Kizuki M. Sociodemographic factors and the risk of developing cardiovascular disease in Bangladesh. Am J Prev Med. 2015;48(4):456–61.
Louridi N, Amar M, Ouahidi E. B. (2019, October). Identification of cardiovascular diseases using machine learning. In 2019 7th mediterranean congress of telecommunications (CMT) (pp. 1–6). IEEE.
Nashif S, Raihan MR, Islam MR, Imam MH. Heart disease detection by using machine learning algorithms and a real-time cardiovascular health monitoring system. World J Eng Technol. 2018;6(4):854–73.
Li P, Hu Y, Liu ZP. Prediction of cardiovascular diseases by integrating multi-modal features with machine learning methods. Biomed Signal Process Control. 2021;66:102474.
Lakshmanarao A, Swathi Y, Sundareswar PSS. Machine learning techniques for heart disease prediction. Forest. 2019;95(99):97.
Alić B, Gurbeta L, Badnjević A. (2017, June). Machine learning techniques for classification of diabetes and cardiovascular diseases. In 2017 6th mediterranean conference on embedded computing (MECO) (pp. 1–4). IEEE.
Ruan, Y., Guo, Y., Zheng, Y., Huang, Z., Sun, S., Kowal, P., … Wu, F. (2018). Cardiovascular disease (CVD) and associated risk factors among older adults in six low-and middle-income countries: results from SAGE Wave 1. BMC public health, 18(1), 1–13.
Sajid MR, Muhammad N, Zakaria R, Shahbaz A, Bukhari SAC, Kadry S, Suresh A. Nonclinical features in predictive modeling of cardiovascular diseases: A machine learning approach. Interdisciplinary Sciences: Comput Life Sci. 2021;13(2):201–11.
Baccouche A, Garcia-Zapirain B, Olea C, C., Elmaghraby A. Ensemble deep learning models for heart disease classification: A case study from Mexico. Information. 2020;11(4):207.
Ramalingam VV, Dandapath A, Raja MK. Heart disease prediction using machine learning techniques: a survey. Int J Eng Technol. 2018;7(28):684–7.
Katarya R, Meena SK. Machine learning techniques for heart disease prediction: a comparative study and analysis. Health Technol. 2021;11(1):87–97.
Al’Aref, S. J., Anchouche, K., Singh, G., Slomka, P. J., Kolli, K. K., Kumar, A.,… Min, J. K. (2019). Clinical applications of machine learning in cardiovascular disease and its relevance to cardiac imaging. European heart journal, 40(24), 1975–1986.
Sharma V, Yadav S, Gupta M. (2020, December). Heart disease prediction using machine learning techniques. In 2020 2nd International Conference on Advances in Computing, Communication Control and Networking (ICACCCN) (pp. 177–181). IEEE.
Aljanabi M, Qutqut MH, Hijjawi M. Machine learning classification techniques for heart disease prediction: a review. Int J Eng Technol. 2018;7(4):5373–9.
Breiman L. Random forests. Mach Learn. 2001;45(1):5–32.
Jiang, Y., Zhang, X., Ma, R., Wang, X., Liu, J., Keerman, M., … Guo, H. (2021). Cardiovascular disease prediction by machine learning algorithms based on cytokines in Kazakhs of China. Clinical epidemiology, 13, 417.
Rubini PE, Subasini CA, Katharine AV, Kumaresan V, Kumar SG, Nithya TM. A cardiovascular disease prediction using machine learning algorithms. Annals of the Romanian Society for Cell Biology; 2021. pp. 904–12.
Krishnan S, Geetha S. (2019, April). Prediction of Heart Disease Using Machine Learning Algorithms. In 2019 1st international conference on innovations in information and communication technology (ICIICT) (pp. 1–5). IEEE.
Islam S, Jahan N, Khatun ME. (2020, March). Cardiovascular disease forecast using machine learning paradigms. In 2020 Fourth International Conference on Computing Methodologies and Communication (ICCMC) (pp. 487–490). IEEE.
Kim JO, Jeong YS, Kim JH, Lee JW, Park D, Kim HS. Machine learning-based cardiovascular disease prediction model: A cohort study on the Korean National Health Insurance Service Health Screening Database. Diagnostics. 2021;11(6):943.
Khatun F, Rahman R, ., Das S, Talukdar A, IDENTIFICATION OF SIGNIFICANT FACTORS FOR BIRTH MODE IN BANGLADESH: EVIDENCE FROM CROSS-SECTIONAL STUDY. Khulna Univ Stud. 2022;664–75. https://doi.org/10.53808/KUS.2022.ICSTEM4IR.0121-se.
Jinjri WM, Keikhosrokiani P, Abdullah NL. (2021, July). Machine Learning Algorithms for The Classification of Cardiovascular Disease-A Comparative Study. In 2021 International Conference on Information Technology (ICIT) (pp. 132–138). IEEE.
Krishnani D, Kumari A, Dewangan A, Singh A, Naik NS. (2019, October). Prediction of coronary heart disease using supervised machine learning algorithms. In TENCON 2019–2019 IEEE Region 10 Conference (TENCON) (pp. 367–372). IEEE.
Alalawi HH, Manal SA. Detection of Cardiovascular Disease using Machine Learning Classification Models. International Journal of Engineering Research & Technology (IJERT) ISSN; 2021. 2278 – 0181.
Hossain SME, Chetty G. (2011, June). Next generation identity verification based on face-gait biometrics. In Proceedings of the International Conference on Biomedical Engineering and Technology (Vol. 11, pp. 142–148).
Ene-Iordache, B., Perico, N., Bikbov, B., Carminati, S., Remuzzi, A., Perna, A., Islam,N., Bravo, R. F., Aleckovic-Halilovic, M., Zou, H., Zhang, L., Gouda, Z., Tchokhonelidze,I., Abraham, G., Mahdavi-Mazdeh, M., Gallieni, M., Codreanu, I., Togtokh, A., Sharma,S. K., … Remuzzi, G. (2016). Chronic kidney disease and cardiovascular risk in six regions of the world (ISN-KDDC): A cross-sectional study. The Lancet Global Health,4(5), e307–e319. https://doi.org/10.1016/S2214-109X(16)00071-1.
Dehghan, M., Mente, A., Zhang, X., Swaminathan, S., Li, W., Mohan, V., Iqbal, R.,Kumar, R., Wentzel-Viljoen, E., Rosengren, A., Amma, L. I., Avezum, A., Chifamba,J., Diaz, R., Khatib, R., Lear, S., Lopez-Jaramillo, P., Liu, X., Gupta, R., … Mapanga,R. (2017). Associations of fats and carbohydrate intake with cardiovascular disease and mortality in 18 countries from five continents (PURE): A prospective cohort study.The Lancet, 390(10107), 2050–2062. https://doi.org/10.1016/S0140-6736(17)32252-3.
Tada H, Fujino N, Hayashi K, Kawashiri M, Takamura M. Human genetics and its impact on cardiovascular disease. J Cardiol. 2022;79(2):233–9.
Chowdhury R, Ramond A, O’Keeffe LM, Shahzad S, Kunutsor SK, Muka T, Gregson J, Willeit P, Warnakula S, Khan H. (2018). Environmental toxic metal contaminants and risk of cardiovascular disease: Systematic review and meta-analysis. BMJ, 362.
Chandralekha M, Shenbagavadivu N. Performance analysis of various machine learning techniques to predict cardiovascular disease: An emprical study. Appl Math Inf Sci. 2018;12(1):217–26.
Arunachalam S. Cardiovascular disease prediction model using machine learning algorithms. Int J Res Appl Sci Eng Technol. 2020;8:1006–19.
Sharma D, Gotlieb N, Farkouh ME, Patel K, Xu W, Bhat M. (2022). Machine Learning Approach to Classify Cardiovascular Disease in Patients With Nonalcoholic Fatty Liver Disease in the UK Biobank Cohort. J Am Heart Association, 11(1), e022576.
Al Kibria GM, Burrowes V, Choudhury A, Sharmeen A, Swasey K. Sex differences in prevalence and associated factors of prehypertension and hypertension among Bangladeshi adults. Int J Cardiol Hypertens. 2019;1:100006. https://doi.org/10.1016/j.ijchy.2019.100006.
Cercato C, Fonseca FA. Cardiovascular risk and obesity. Diabetol Metab Syndr. 2019;11(1):74. https://doi.org/10.1186/s13098-019-0468-0.
Rodgers JL, Jones J, Bolleddu SI, Vanthenapalli S, Rodgers LE, Shah K, Karia K, Panguluri SK. Cardiovascular Risks Associated with Gender and Aging. J Cardiovasc Dev Disease. 2019;6(2):19. https://doi.org/10.3390/jcdd6020019.
Thompson SC, Nedkoff L, Katzenellenbogen J, Hussain MA, Sanfilippo F. Challenges in Managing Acute Cardiovascular Diseases and Follow Up Care in Rural Areas: A Narrative Review. Int J Environ Res Public Health. 2019;16(24). https://doi.org/10.3390/ijerph16245126. Article 24.
Clark AM, DesMeules M, Luo W, Duncan AS, Wielgosz A. Socioeconomic status and cardiovascular disease: Risks and implications for care. Nat Reviews Cardiol. 2009;6(11). Article 11. https://doi.org/10.1038/nrcardio.2009.163.
Erem C, Hacihasanoglu A, Kocak M, Deger O, Topbas M. Prevalence of prehypertension and hypertension and associated risk factors among Turkish adults: Trabzon Hypertension Study. J Public Health. 2009;31(1):47–58. https://doi.org/10.1093/pubmed/fdn078.
Petrelli A, Sebastiani G, Napoli AD, Macciotta A, Filippo PD, Strippoli E, Mirisola C, d’Errico A. Education inequalities in cardiovascular and coronary heart disease in Italy and the role of behavioral and biological risk factors. Nutr Metabolism Cardiovasc Dis. 2022;32(4):918–28. https://doi.org/10.1016/j.numecd.2021.10.022.
ROC. curve.

No competing interests reported.

Download PDF

Reviews received at journal
05 Sep, 2024
Reviewers agreed at journal
30 Aug, 2024
Reviewers agreed at journal
30 Aug, 2024
Reviewers agreed at journal
02 Aug, 2024
Reviewers agreed at journal
29 Jul, 2024
Reviewers invited by journal
29 Jul, 2024
Editor assigned by journal
21 Jul, 2024
Submission checks completed at journal
19 Jul, 2024
First submitted to journal
11 Jul, 2024

You are reading this latest preprint version

Determinants of Developing Cardiovascular Disease Risk with Emphasis on Type-2 Diabetes and Predictive Modeling Utilizing Machine Learning Algorithms

Status:

Version 1

Abstract

Background

Methods

Results

Conclusion

Figures

Introduction

Methods

Sources of Data

Sample Design and Sampling Frame

Variable Description

Dependent Variable

Independent Variables

Data pre-processing

Data cleaning

Data balancing

Statistical Analysis and machine learning models

Results

10-fold cross validation of the accuracy of the classifiers

Discussion

Limitations

Conclusion

Declarations

Conflict of Interest:

Ethical approval:

Funding:

Author Contribution

Data Availability

References

Additional Declarations

Status:

Version 1