Predictive Horizons: Unveiling Cardiovascular Insights With Conditional Inference Trees

OBJECTIVES: Cardiovascular diseases (CVDs) pose a significant global burden. Early detection and accurate prediction are essential for prevention and effective treatment. This study aims to identify risk factors for early heart disease prediction using machine learning techniques.

METHODS: This study applied the Nearmiss algorithm to address the issue of low prevalence (8.6%) by conducting under-sampling on units with higher prevalence indicator values through the KNN algorithm, aiming to enhance variability among the retained units. Subsequently, the conditional tree hyperparameters were fine-tuned during model construction. The Bonferroni correction was applied for testing the global null hypothesis and subsequent partial hypotheses, facilitating the construction of the Conditional Inference Tree (CIT). The analysis utilized a CDC dataset as part of the Behavioral Risk Factor Surveillance System (BRFSS), conducted across the United States and territories

RESULTS: In the study, a 76% accuracy rate for conditional inference trees in the early diagnosis of heart diseases was established. The sensitivity and specificity of the model were found to be 80% and 72%, respectively. The analysis revealed that individuals with walking difficulty in the 40–60 age group with multimorbidity and poor general health have a significant probability (~95%) of developing heart disease. On the other hand, individuals with good overall health who do not smoke or drink and do not have multiple medical conditions have a ~5% risk of developing heart disease.

CONCLUSIONS: The conditional inference tree is a useful tool for the identification of patients who are at risk for heart disease. The tool can be used to identify risk factors that can be targeted in tailored care plans and early intervention, which ultimately help lower the risk of cardiovascular disease and improve patient outcomes.