Heatwave Forecasts and ER Visit Alerts

ISEF Category: Translational Medical Science

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

The Hook

Heat waves do more than make people uncomfortable. They can push emergency rooms toward overload. If you can predict that spike a few days early, cities can warn people and move resources faster. That turns weather data into a health tool.

What Is It?

This project asks a simple question with a hard answer: can weather forecasts help predict emergency department visits? You start with NOAA heat data, then compare it with health records from sources like MIMIC-IV or public state ED datasets. The goal is to see whether a heat event, plus a short time lag, lines up with more ER visits.

Think of it like a bridge between two systems. One side is the atmosphere, which gives you temperature, humidity, and heat index. The other side is human health, which shows up as visit counts. Your job is to test whether the bridge holds up well enough to support a 72-hour alert prototype.

Why This Is a Good Topic

This is a strong science fair topic because it is measurable, timely, and tied to a real public health need. You can test whether heat forecasts predict ER traffic, compare different lag times, and see which weather features matter most. A student can learn data cleaning, time-series analysis, and model evaluation without needing to run a wet lab.

Research Questions

• How does a 72-hour NOAA heat forecast relate to changes in emergency department visits in the same region?
• What is the effect of humidity-adjusted heat index versus air temperature alone on predicted ED visit spikes?
• Does adding a one-day or two-day lag improve the accuracy of a heat-health alert model?
• To what extent do weekday patterns change the link between heat events and ED visits?
• Which weather variable, such as maximum temperature, heat index, or overnight low, best predicts ED visit surges?
• How does model performance change when you compare one city to several cities with different climates?

Basic Materials

• Laptop or desktop computer with internet access.
• Spreadsheet software such as Google Sheets or Excel.
• Python installed with pandas, matplotlib, scipy, and scikit-learn.
• NOAA weather and heat data from public archives.
• Public ED visit dataset, such as MIMIC-IV if you have access, or a state open-data portal.
• Notes document for tracking variable definitions and data sources.
• Digital calendar or project planner for logging analysis steps.

Advanced Materials

• Laptop or desktop computer with a Python or R environment.
• Access to MIMIC-IV, subject to required training and permissions.
• Public health or state syndromic surveillance dataset, if available.
• NOAA climate data archive and local forecast products.
• GIS software such as QGIS for mapping alert zones.
• Statistical package for regression, time-series models, or generalized additive models.
• Version control system such as Git for tracking code changes.

Software & Tools

• Python: Cleans weather and ED data, builds lag features, and fits prediction models.
• R: Runs time-series and regression analyses with strong plotting tools.
• Google Sheets: Helps you inspect small tables and check your variable definitions.
• QGIS: Maps heat alerts to neighborhoods or service areas.
• PubMed: Finds review articles on heat exposure, emergency care, and climate-health forecasting.

Experiment Steps

1. Define the exact health outcome you will predict, such as daily ED visit counts or heat-related chief complaints.
2. Choose the weather variables you will test, then decide which forecast horizon matters most for your alert model.
3. Match weather and health data on date, location, and lag so the two datasets speak the same language.
4. Build a baseline model first, then compare it with a heat-aware model so you can measure added value.
5. Set up controls for season, day of week, and holidays so heat does not get credit for unrelated patterns.
6. Plan how you will judge alert quality, using metrics that reward early warning without flooding users with false alarms.

Common Pitfalls

• Using raw temperature alone, which can miss the health impact of humidity and overnight cooling.
• Mismatching time zones or date cutoffs, which shifts weather exposure away from the correct ED visit day.
• Treating all ED visits as heat-related, which hides the signal inside unrelated cases.
• Ignoring season and weekday patterns, which can make a model look better than it really is.
• Building an alert rule that is too sensitive, which creates false alarms every time the forecast warms up.

What Makes This Competitive

A competitive project will do more than plot heat against visits. You can strengthen it by testing several lag structures, comparing multiple cities, and measuring whether your model beats a simple baseline. Strong entries also separate signal from confounders like season, holidays, and day of week. If you add a clear alert threshold and explain how health agencies could act on it, your project feels much more real.

Project Variations

• Focus on pediatric ED visits instead of all-age visits to test whether children respond differently to heat.
• Compare coastal and inland cities to see whether humidity changes the heat-health link.
• Swap ED visits for ambulance calls or urgent care visits if your local dataset tracks those outcomes more cleanly.

Learn More

• NOAA National Weather Service and climate data portals: Search NOAA for local heat alerts, forecasts, and historical weather archives.
• MIMIC-IV: Search PhysioNet for the dataset and its documentation on ICU and hospital records.
• US CDC Heat and Health Tracker: Find background on heat-related illness, risk factors, and public health response.
• PubMed: Search for review articles on heat exposure, emergency department visits, and climate-health prediction models.
• MIT OpenCourseWare, Intro to Probability and Statistics: Use it for free refreshers on regression, model testing, and uncertainty.