Development of a convenient and effective hypertension risk prediction model and exploration of the relationship between Serum Ferritin and Hypertension Risk: a study based on NHANES 2017—March 2020

Shuang Guo, Jiuxin Ge, Shan-Na Liu, Jiayu Zhou, Chang Li, Hanjie Chen, Li Chen, Yuqiang Shen, Qingli ZhouFrontiers in Cardiovascular Medicine 2023 · DOI

This study, published in Frontiers in Cardiovascular Medicine, presents a high-performance machine learning (ML) "fusion" model that predicts hypertension risk using just 10 low-cost clinical features. Beyond the engineering feat, the researchers uncover a significant, dose-dependent relationship between serum ferritin—a common marker for iron storage—and the risk of high blood pressure.

TL;DR

Researchers developed a "fusion" machine learning model that achieves an impressive 0.966 AUC in predicting hypertension risk using only 10 easily accessible clinical features. A standout finding is the strong correlation between serum ferritin levels and hypertension, suggesting this common biomarker could be a vital tool for early clinical screening.

What Problem Does This Paper Address?

Hypertension is a "silent killer" and a leading cause of cardiovascular disease worldwide. While standard office blood pressure measurements are the norm, they are prone to errors like "white coat hypertension" (stress-induced high readings) or "masked hypertension" (normal readings in-office despite high pressure elsewhere).

Existing machine learning models for hypertension often rely on a massive number of variables or focus strictly on traditional risk factors like age and BMI. This paper aims to bridge the gap by creating a model that is both convenient (using few, low-cost features) and effective (high accuracy), while specifically investigating the role of serum ferritin (SF), a biomarker often overlooked in cardiovascular risk assessments.

Methods

The researchers employed a multi-stage pipeline to move from raw data to a refined prediction tool:

  1. Feature Selection: Starting with 120 features, they used LASSO regression and correlation analysis to remove redundant variables. They then applied Recursive Feature Elimination (RFE) to find the "sweet spot"—the minimum number of features that maintain maximum accuracy.
  2. Data Balancing: To handle the imbalance between hypertensive and non-hypertensive subjects, they used SMOTE (Synthetic Minority Over-sampling Technique) to generate synthetic examples of the minority class, ensuring the model didn't become biased.
  3. Model Stacking: Instead of picking one "best" algorithm, they used a Stacking Strategy. They integrated the top four performing tree-based learners (Random Forest, XGBoost, LightGBM, and Extra Trees) into a single "fusion model."
  4. Statistical Association: They performed a univariate analysis and multiple logistic regression to specifically quantify how serum ferritin levels (divided into quartiles Q1–Q4) influenced hypertension risk.

Datasets

The study utilized data from the NHANES (National Health and Nutrition Examination Survey) 2017–March 2020.

  • Initial size: 15,561 participants.
  • Final cohort: 8,073 people (after excluding minors, pregnant women, cancer patients, and those with significant missing data).
  • Target Classes: Normal blood pressure, pre-hypertension, and hypertension.

Key Insights and Ideas

  • The Power of Ten: The model proved that you don't need 120 variables for high-accuracy screening. The final 10 features included: Age, BMI, waist circumference, SBP, DBP, Creatinine, UACR (urine albumin-creatinine ratio), Serum Ferritin, HbA1C, and a doctor's recommendation to reduce salt.
  • Serum Ferritin as a "Dark Horse" Predictor: SF is typically used to check for anemia or iron overload. However, this study ranked it as a top-tier predictor for hypertension. The authors suggest this link may be due to SF's role in oxidative stress and inflammation, which can damage blood vessel walls.
  • Tree-Based Models Rule: The study found that tree-based algorithms (like Random Forest and XGBoost) significantly outperformed traditional methods like Logistic Regression or K-Nearest Neighbors for this specific dataset, likely due to their ability to capture non-linear relationships between physiological markers.

Key Results

  • Model Performance: The fusion model achieved an AUC of 0.966, with accuracy, precision, and recall all hovering around 0.87. This significantly outperformed individual base models.
  • The Ferritin Gradient: There was a clear "dose-response" relationship. Compared to the lowest level group (Q1), the odds ratio (OR) for hypertension increased steadily through the quartiles:
  • Q2: 1.396
    • Q3: 1.499
    • Q4 (Highest): 1.645
  • Clinical Utility: The 10 features identified are "low-cost and easily accessible," making the model a candidate for automated diagnostic aids in resource-limited hospital environments.

Limitations and Future Work

  • Cross-Sectional Limitation: Because the NHANES data is a "snapshot" in time, the study can show a correlation but cannot definitively prove that high serum ferritin causes hypertension.
  • Geographic Bias: The data is limited to the US population. The authors acknowledge the need to validate these findings in other cohorts, such as Chinese populations, to ensure the model's universality.
  • Temporal Dynamics: Future research is needed to explore the temporal relationship—specifically, whether rising ferritin levels can predict the future onset of hypertension years before it appears on a pressure cuff.

Summary updated

Abstract

The hypertension risk prediction model developed in this study is efficient in predicting hypertension with only 10 low-cost and easily accessible features, which is cost-effective in assisting clinical diagnosis. We also found a trend correlation between serum ferritin levels and the risk of hypertension.
BibTeX
@inproceedings{guo2023development,
  author = {Shuang Guo and Jiuxin Ge and Shan-Na Liu and Jiayu Zhou and Chang Li and Hanjie Chen and Li Chen and Yuqiang Shen and Qingli Zhou},
  title = {{Development of a convenient and effective hypertension risk prediction model and exploration of the relationship between Serum Ferritin and Hypertension Risk: a study based on NHANES 2017—March 2020}},
  booktitle = {Frontiers in Cardiovascular Medicine},
  year = {2023},
  doi = {10.3389/fcvm.2023.1224795},
}
Figure