Wearable Signals for Hypertension Risk Prediction

ISEF Category: Translational Medical Science

Subcategory: Disease Prevention  ·  Difficulty: Advanced  ·  Setup: University Lab  ·  Time: Full Year

The Hook

Your watch may spot a blood pressure problem before a clinic visit does. Step count and sleep regularity can act like a daily health fingerprint. If those patterns shift, your body may be heading toward trouble long before a diagnosis. That makes this a strong idea for a data-driven science fair project.

What Is It?

This project asks a simple question with a serious payoff, can daily movement and sleep patterns predict who later develops hypertension? Hypertension means high blood pressure, and it often builds quietly. A student can think of this like weather forecasting, where tiny changes in pressure, wind, and humidity help predict a storm before it arrives. Here, your signals are steps, sleep timing, and sleep regularity.

The key idea is to turn wearable data into a risk score. Step count shows how active someone is. A sleep-regularity index measures how consistent sleep timing stays from night to night. You can test whether people with lower activity and more irregular sleep are more likely to show incident hypertension later in the dataset. The work blends public health, data science, and prediction.

Why This Is a Good Topic

This topic works well because you can define clear variables, build a prediction model, and measure performance with real outcomes. It connects to a real health problem that affects millions of people. You can learn data cleaning, feature engineering, model validation, and basic epidemiology. You do not need a wet lab to ask a serious question.

Research Questions

• How does average daily step count predict incident hypertension in the All of Us public tier?
• What is the effect of sleep regularity on the odds of developing hypertension later?
• Does combining step count and sleep regularity improve prediction more than either variable alone?
• To what extent does adding age, sex, and BMI improve a wearable-based hypertension risk score?
• Which sleep regularity measure, bedtime consistency or wake-time consistency, best predicts incident hypertension?
• How does prediction performance change when you compare weekly averages with monthly averages of wearable data?

Basic Materials

• Computer with internet access.
• Access to the All of Us Research Hub public tier.
• Spreadsheet software such as Google Sheets or Excel.
• Free statistical software such as R or Python.
• Data dictionary for wearable and outcome variables.
• Notebook for tracking variable definitions and analysis choices.
• Headphones or a quiet workspace for long data-cleaning sessions.

Advanced Materials

• Computer with internet access.
• Access to the All of Us Research Hub public tier or a similar secure environment.
• Python with pandas, NumPy, scikit-learn, seaborn, and statsmodels.
• R with tidyverse, survival, and pROC packages.
• Jupyter Notebook or RStudio.
• PubMed access for background literature.
• NIH and CDC hypertension data summaries for context.
• Version control tool such as Git for analysis tracking.

Software & Tools

• Python: Cleans wearable data, builds features, and fits prediction models.
• R: Runs regression, survival analysis, and model checks.
• Jupyter Notebook: Keeps code, notes, and plots in one place.
• ImageJ: Not needed for this project unless you compare visual chart exports, so skip it if you want a cleaner workflow.
• Google Sheets: Helps you inspect small data tables and plan variables before coding.

Experiment Steps

1. Define the outcome you will predict, then decide exactly what counts as incident hypertension in the dataset.
2. Choose the wearable features you will test first, such as step count, sleep regularity, and simple combinations of both.
3. Build a clean analysis table with one row per person and one row per time window.
4. Plan a baseline model, then compare it with a wearable-based model and a fuller model with demographic controls.
5. Decide how you will score prediction quality, such as AUC, sensitivity, specificity, or calibration.
6. Pre-register your main comparison on paper or in a private document so you do not change the goal after seeing results.

Common Pitfalls

• Using inconsistent sleep definitions, which makes your regularity metric meaningless across records.
• Mixing people with very different amounts of wearable wear time, which can create fake patterns.
• Predicting current hypertension instead of incident hypertension, which breaks the early-warning idea.
• Forgetting to separate training and test data, which makes the model look better than it really is.
• Treating missing wearable days as healthy behavior, which can bias the risk score.

What Makes This Competitive

A strong version of this project goes beyond a simple yes-or-no model. You can compare several ways to define sleep regularity, test whether one wearable signal adds real predictive value, and check whether the model stays accurate across age or sex groups. A competitive entry also explains why the signal matters biologically and clinically. Good validation and clear error analysis will matter more than a flashy model.

Project Variations

• Use self-tracked Fitbit export data from volunteers and compare whether the same step and sleep signals predict blood pressure trends in a smaller cohort.
• Replace incident hypertension with prehypertension or elevated blood pressure and test whether the earlier threshold changes model performance.
• Compare step count and sleep regularity against another wearable feature, such as resting heart rate or sedentary time, to see which adds the most predictive value.

Learn More

• All of Us Research Program: Search the public data documentation and training materials on the NIH All of Us site.
• PubMed: Search for review articles on sleep regularity, physical activity, and hypertension risk.
• NIH Office of Dietary Supplements and NHLBI resources: Use the public hypertension background pages and fact sheets.
• CDC Hypertension Facts: Find national statistics and plain-language background on blood pressure risk.
• American Journal of Hypertension: Search for studies on wearable data, activity patterns, and incident hypertension.
• MIT OpenCourseWare: Search for free lectures on statistics, data analysis, and machine learning basics.