Project 4: Dietary Nutrient Intake and Blood Pressure Prediction

Authors

• Yishan Lin: ylin183@jh.edu, JHED ID: ylin183
• Yimin Ding: yding75@jh.edu, JHED ID: yding75
• Yingqi Wang: ywang911@jh.edu, JHED ID: ywang911
• Ruiqing Cai: rcai9@jh.edu, JHED ID: rcai9

Overview

This project investigates the relationships between dietary nutrient intake and blood pressure levels using data from the NHANES (2021–2023) dataset. Hypertension is a significant risk factor for cardiovascular diseases, and understanding dietary influences can help guide public health strategies.

Objectives

• Statistical Analysis: Explore associations between sodium, potassium, calcium, magnesium, fiber, and blood pressure (systolic and diastolic).
• Machine Learning: Build a predictive model to identify hypertension risk based on demographic and dietary factors.

Methods

Data Preparation

• Utilized NHANES datasets for demographic, dietary, and laboratory data.
• Imputed missing values using Predictive Mean Matching (PMM) to ensure completeness.
• Stratified analysis by demographic factors to provide subgroup insights.

Statistical Models

• Regression models to identify significant nutrient-blood pressure relationships.
• Functional programming approaches (e.g., reusable functions for cleaning and processing) were implemented for reproducibility.

Machine Learning

• Random Forest models trained on demographic and dietary variables to predict:
  • Systolic Blood Pressure (SYBP)
  • Diastolic Blood Pressure (DIBP)
  • Combined Blood Pressure Category (Normal/High)
• Parallel processing was used to improve computational efficiency.

Key Results

1. Statistical Analysis:
   • Sodium intake showed a weak negative association with blood pressure.
2. Machine Learning:
   • Predictive performance for binary blood pressure outcomes:
     • Sensitivity for "Normal" classification was high (~96%).
     • Specificity for "High" classification was limited (~10–39%).
   • Random Forest models highlighted age and BMI as the most critical predictors.
3. Dashboard Insights:
   • Dynamic visualization options (e.g., violin plots, scatterplots) provided granular insights into nutrient-blood pressure interactions.

Technical Highlights

• Programming Paradigms: Functional and object-oriented approaches streamlined workflows.
• Parallel Processing: Boosted computational efficiency for machine learning.
• Visualization: An interactive dashboard enhanced data accessibility and usability.

Requirements

• R version 4.0 or higher
• Required R packages:
  • tidyverse
  • randomForest
  • caret
  • doParallel
  • mice
  • quarto
• Running the Code
  • Clone this repository.
  • Open the Final_project.qmd file to explore data preparation, statistical models, and machine learning implementations.
  • Launch the dashboard with Interactive Dashboard.qmd.
• Accessing Output
  • Summaries and visualizations are available in the Final project.html files
  • The Final Project Write up.pdf contains a comprehensive report.

License

This project is for educational purposes and uses NHANES data, which is publicly available under the CDC guidelines for non-commercial use. '