Genome-Wide Association of New-Onset Hypertension According to Renin Concentration: The Korean Genome and Epidemiology Cohort Study

Abstract

The renin-angiotensin system (RAS) is a crucial regulator of vascular resistance and blood volume in the body. This study aimed to examine the genetic predisposition of the plasma renin concentration influencing future hypertension incidence. Based on the Korean Genome and Epidemiology Cohort dataset, 5211 normotensive individuals at enrollment were observed over 12 years, categorized into the low-renin and high-renin groups. We conducted genome-wide association studies for the total, low-renin, and high-renin groups. Among the significant SNPs, the lead SNPs of each locus were focused on for further interpretation. The effect of genotypes was determined by logistic regression analysis between controls and new-onset hypertension, after adjusting for potential confounding variables. During a mean follow-up period of 7.6 years, 1704 participants (32.7%) developed hypertension. The low-renin group showed more incidence rates of new-onset hypertension (35.3%) than the high-renin group (26.5%). Among 153 SNPs in renin-related gene regions, two SNPs (rs11726091 and rs8137145) showed an association in the high-renin group, four SNPs (rs17038966, rs145286444, rs2118663, and rs12336898) in the low-renin group, and three SNPs (rs1938859, rs7968218, and rs117246401) in the total population. Most significantly, the low-renin SNP rs12336898 in the SPTAN1 gene, closely related to vascular wall remodeling, was associated with the development of hypertension (p-value = 1.3 × 10−6). We found the candidate genetic polymorphisms according to blood renin concentration. Our results might be a valuable indicator for hypertension risk prediction and preventive measure, considering renin concentration with genetic susceptibility.

Shapely additive values can effectively visualize pertinent covariates in machine learning when predicting hypertension

“…Prior studies for risk prediction have utilized traditional parametric methods such as linear regression, cox proportional hazards analysis, and logistic regression 28,31,32,39,40 . In traditional parametric methods such as linear regression, each covariate can be interpreted clearly (e.g., for each one increase in x, we observe two increases in y) 6,25,41 .…”

“…One solution that has been executed are machine‐learning techniques 5,23,24,26 . These methods often have better accuracy, potentially due to their ability to leverage non‐linear and interactive relationships when performing prediction 28,30,42,43 . However, due to the complexity of the non‐parametric algorithms that are common in machine‐learning methods, it is impossible for a human to analyze each tree and execute an explanation of how the machine‐learning method works 7,28,30,35,42–45 .…”

“…15,19,[35][36][37] To provide context over the antihypertensive use amongst the Prior studies for risk prediction have utilized traditional parametric methods such as linear regression, cox proportional hazards analysis, and logistic regression. 28,31,32,39,40 In traditional parametric methods such as linear regression, each covariate can be interpreted clearly (e.g., for each one increase in x, we observe two increases in y). 6,25,41 However, each of these parametric methods relies on a large number of assumptions (normally distributed errors and linearity for linear regression, proportional hazards for cox proportional hazards) and are often destabilized due to multicollinearity.…”

“…However, the increased predictive accuracy of ML models comes at the cost of interpretability 1,14‐16,19,20 . Without methods that explain how machine learning algorithms reach their predictions, clinicians will not be able to identify if models are reliable and generalizable or just replicating the biases within the training datasets 5,26‐28 . Provision of explanations about how model predictions are researched and providing accurate summary statistics for model accuracy metrics (e.g., AUROC, Sensitivity, Specificity, F1, Balanced Accuracy) will increase the transparency of machine learning methods and increase confidence when using their predictions 18,24,25,29 .…”

“… 1 , 14 , 15 , 16 , 19 , 20 Without methods that explain how machine learning algorithms reach their predictions, clinicians will not be able to identify if models are reliable and generalizable or just replicating the biases within the training datasets. 5 , 26 , 27 , 28 Provision of explanations about how model predictions are researched and providing accurate summary statistics for model accuracy metrics (e.g., AUROC, Sensitivity, Specificity, F1, Balanced Accuracy) will increase the transparency of machine learning methods and increase confidence when using their predictions. 18 , 24 , 25 , 29 Visualization of the relationship between the independent covariates and the outcome can be done through SHapely Additive exPlanations (SHAP) to enhance model interpretability.…”