Sleep Apnea and Hypertension Causality Study

ISEF Category: Translational Medical Science

Subcategory: Other  ·  Difficulty: Advanced  ·  Setup: University Lab  ·  Time: Full Year

The Hook

Sleep apnea and high blood pressure often show up together, but that does not prove one causes the other. Your project can test the direction of that link with real population data. That means you are not just counting patients, you are asking which problem may come first.

What Is It?

This project asks a causal question, not just a correlation question. Correlation means two things move together. Causation means one helps produce the other. Those are very different claims, and in medicine, the difference matters.

You can think of sleep apnea and hypertension like two roommates who both leave dirty dishes. Seeing dishes in the sink does not tell you who made them. A causal discovery method tries to sort out which variables may sit upstream and which may sit downstream by looking at patterns across many related measurements. PC and FCI are two algorithms that build a network of possible cause-and-effect links from data, while also trying to account for hidden causes.

Mendelian randomization adds a second angle. It uses inherited genetic variants as natural proxies for an exposure, such as risk for sleep apnea, to test whether that exposure may influence blood pressure. If both methods point the same way, your argument gets stronger. If they disagree, that disagreement becomes part of the story.

Why This Is a Good Topic

This is a strong science fair topic because it asks a real medical question with public data and published methods. You can test a clear outcome, compare two causal inference strategies, and learn how researchers handle confounding, bias, and hidden variables. The project also connects to a real health problem, since both sleep apnea and hypertension affect many people and often get missed early.

Research Questions

• How does including age, sex, BMI, and smoking status change the causal graph between sleep apnea and hypertension in NHANES data?
• What is the effect of removing obesity-related variables on the direction of the edge between sleep apnea and blood pressure outcomes?
• Does the PC algorithm and the FCI algorithm agree on whether sleep apnea is upstream of hypertension?
• To what extent do different blood pressure definitions change the inferred causal direction between sleep apnea and hypertension?
• Which genetic instruments for sleep apnea produce the strongest Mendelian-randomization signal for blood pressure outcomes?
• How does stratifying by sex or age group change the causal discovery results for sleep apnea and hypertension?

Basic Materials

• A computer with enough memory to run statistical software.
• A clean NHANES data set with sleep, blood pressure, BMI, age, sex, and smoking variables.
• A spreadsheet program for data screening and variable tracking.
• A statistics environment such as R or Python.
• A code notebook for documenting preprocessing choices.
• A public genetic summary-data source for Mendelian-randomization inputs.

Advanced Materials

• A workstation or server with enough memory for repeated causal-discovery runs.
• Access to NHANES raw and derived data files.
• Access to genome-wide association study summary statistics for sleep apnea and blood pressure traits.
• A curated list of candidate instrumental variables for Mendelian randomization.
• A statistical package that supports PC, FCI, and sensitivity analyses.
• A version-controlled code repository for reproducible analysis.

Software & Tools

• R: Runs causal discovery, regression, and Mendelian-randomization workflows with free packages.
• Python: Helps with data cleaning, plotting, and reproducible analysis notebooks.
• Jupyter Notebook: Keeps code, notes, and outputs in one place for a clear methods section.
• RStudio: Gives you a friendly interface for running R scripts and checking results.
• ImageJ: Not needed here, so skip it and focus on data analysis tools instead.

Experiment Steps

1. Define your causal question and decide which direction you want to test first.
2. Choose a small set of variables that could plausibly connect sleep apnea and hypertension.
3. Plan a data-cleaning rule set that keeps your analysis consistent across all runs.
4. Build a causal graph workflow and decide how you will compare PC, FCI, and Mendelian-randomization results.
5. Set up sensitivity checks that test whether your answer changes when you swap covariates or subgroup definitions.
6. Decide how you will judge agreement, disagreement, and uncertainty across methods.

Common Pitfalls

• Using too many correlated variables, which can make the causal graph unstable and hard to read.
• Treating self-reported sleep apnea and measured hypertension as if they have the same error profile.
• Ignoring sample weights in NHANES, which can distort population-level conclusions.
• Choosing genetic instruments that are too weak, which can make Mendelian-randomization estimates unreliable.
• Reading a directed edge as proof of causation, even when hidden confounding may still be present.

What Makes This Competitive

A stronger version of this project goes beyond one method and one answer. You compare algorithms, test subgroup effects, and report when the methods agree or conflict. You also explain why a result might change after removing confounders or switching outcome definitions. That kind of careful triangulation shows real scientific judgment.

Project Variations

• Focus only on adult women, and test whether the causal direction changes after stratifying by sex.
• Replace hypertension with systolic blood pressure, diastolic blood pressure, or pulse pressure as separate outcomes.
• Compare sleep apnea severity with blood pressure control among people already taking antihypertensive medication.

Learn More

• NHANES, CDC: Search the NHANES data portal for health interview, examination, and laboratory files.
• PubMed: Search review articles on sleep apnea, hypertension, causal discovery, and Mendelian randomization.
• NIH Office of Dietary Supplements: Use background pages to practice reading biomedical evidence and study design language.
• Mendelian Randomization review articles: Search PubMed for tutorials that explain instruments, pleiotropy, and sensitivity tests.
• MIT OpenCourseWare: Search for free coursework on statistics, probability, and data analysis to support your methods.