Machine Learning in Cardiovascular Disease

Table of Contents:

CVDs are diseases that affect the heart or blood vessels and include hypertension, coronary heart disease, heart failure, atrial fibrillation, and stroke. CVD is one of the leading causes of mortality or morbidity and accounts for one-third of the deaths worldwide (Eckel et al., 2014; Alwan, 2011). About 102 million (41.5%) US adults currently have at least one cardiovascular condition, and —131 million (45%) are projected to suffer from CVD by 2035-CVD’s estimated annual 2014-2015 cost was 555 billion dollars in the United States, and it is projected to double by 2035 to 1.1 trillion dollars.

Age-adjusted cardiovascular mortalities have reduced significantly over the past 50 years. Despite its significant incidence decline, CVD remains to be the leading cause of death (McClellan et al., 2019). Contributing factors for recent flattening cardiovascular mortality rate reduction include (McClellan et al, 2019)

1. Patients fail to make risk factors modifications.
2. Patients do not adhere to recommended treatments.
3. Failure to diagnose.
4. Failure to use evidence-based medication regimens.

Additionally, significant cardiovascular mortality disparities are present across different groups by race, sex, and ethnicity. These may result from patient difficulties to comply with CVD risk factor modification treatments, socioeconomic variables, and behavioral factors (McClellan et al, 2019). Al may provide solutions to alleviate some of these CVD treatment challenges by personalizing treatment, targeting high-risk patients, and potentially decreasing cardiovascular outcome disparities. Current cardiology ML studies are categorized in the following four groups:

1. Cardiovascular risk prediction
2. Phenotypic and prognostic studies
3. ECG and diagnostic accuracy improvement
4. Imaging and diagnostic accuracy improvement.

Cardiovascular Risk Prediction

Failure to diagnose and prevent CVD remains problematic. A significant proportion of apparently healthy subjects without prior symptoms continue to die suddenly (Naghavi et al., 2006). Traditional CVD risk prediction models such as the Framingham risk score assesses only at the population level with discriminatory accuracy at the individual level remaining suboptimal (75%)-As a result, individual patients experience overtreatment or under treatment with related morbidity and economic costs (Wilson Peter et al., 1998; Detrano et al., 2008).

Traditional CVD risk prediction models are limited in predictive performance because they adopt restrictive modeling methods with limited numbers of variables. Identifying asymptomatic patients who are at high risk of CVD and starting on intense preventive treatment is a major ongoing challenge, particularly with traditional CVD risk models (Franco et al., 2011). Comprehensive tools with improved accuracy are needed to address these challenges. ML models may provide solutions to CVD risk prediction by identifying current and new CVD risk factors, evaluating patient variables comprehensively, and potentially discovering new CVD therapeutic strategies (Table 2.4).

Identification of Novel Cardiovascular Disease Phenotypes

ML models have helped to classify novel heart disease genotypes and phenotypes by identifying therapeutically homogeneous patient subgroups within heterogeneous diseases such as heart failure preserved ejection fraction (HFpEF), pulmonary hypertension, and coronary artery disease (Krittanawong et al., 2017). As a result, these models may assist patients and healthcare providers with prognostic and therapeutic decisions, active and informed decision-making about invasive procedures, and efficient resource utilization (Table 2.5).