Predictive value of anthropometric indices for incident of dyslipidemia: a large population-based study

Abstract

Introduction: Dyslipidemia as a modifiable risk factor for chronic non-communicable diseases has become a worldwide concern. We aim to explore different anthropometric measures as predictors of dyslipidemia using various machine learning methods.

Method: From the baseline of the Mashhad Stroke and Heart Atherosclerotic Disorder (MASHAD) study, a total of 9,640 participants were included in the analysis. Among them, 1,388 participants did not have dyslipidemia, while 8,252 participants had dyslipidemia. Various anthropometric indices were examined, including waist-to-height ratio (WHtR), body roundness index (BRI), abdominal volume index (AVI), weight-adjusted waist index (WWI), lipid accumulation product (LAP), visceral adiposity index (VAI), conicity index (C-index), body surface area (BSA), body adiposity index (BAI), and waist-to-hip ratio (WHR). The association between these indices and dyslipidemia was assessed using logistic regression (LR), decision tree (DT), random forest (RF), neural networks (NN), K-nearest neighbors (KNN), and eXtreme Gradient Boosting (XGBoost) models.

Results: Based on our LR model, we found that several factors included, BAI, BSA, age, and WHR were significant. For example, for each unit increase in WHR, the odds of dyslipidemia increase by 9 time (OR = 90.29, 95%CI (4.09,21.08)). Additionally, our DT model indicated that BMI was the most influential predictor, followed by age and WHR. The LR model outperforms other models with the highest accuracy (0.89) and AUC-ROC score (0.89), showing strong ability to classify dyslipidemia cases. Feature importance analysis reveals variables like "BSA" contribute differently across models, with XGBoost relying more on it than LR. LR's balanced performance makes it the best choice.

Conclusion: The findings from machine learning models were in agreement, highlighting the significance of BMI, WHR, BSA, and BAI as key anthropometric indices for predicting dyslipidemia. These indices consistently emerged as strong predictors underscoring their importance in assessing the risk of dyslipidemia.

Keywords: Adiposity; Anthropometry; Dyslipidemia; Machine learning.

Fig. 1

Flow chart of this study

Fig. 2

Decision tree for dyslipidemia

Fig. 3

Variables in order of importance in RF

Fig. 4

Partial dependence on two importance variables bases on RF

Fig. 5

Metrics of Train Instances; comparison between several machines learning male’s models

Fig. 6

Confusion matrices on both train and test data. A) XGBoost, B) K-Nearest Neighbor, C) Logistic Regression, D) Random Forest, E) Neural Network