Heat Meds

Check your medicines against CDC heat guidance and your local heat forecast, then print a one-page plan to take to your pharmacist.

Live site: https://usv240.github.io/heat-meds/ (press "Try an example", no account needed) Demo video: https://youtu.be/7ZnGjnIXZVg

The problem

Some everyday medicines, like water pills, some blood pressure pills, and some antidepressants, change how the body handles heat. They can dull thirst, change sweating, or lower blood pressure, so a hot day hits harder. Older adults take these medicines most, and most are never told.

In September 2025 the CDC asked clinicians to plan for this with their patients [1]. That guidance is a table written for doctors. Patients and families had no tool to start the conversation themselves.

It is common, not rare: 6 of the 10 most-prescribed medicines in the US are on the CDC's heat list (this project's analysis of the 2024 Medical Expenditure Panel Survey [6]).

Who it is for

Older adults and the people who look after them, at the kitchen table with the pill bottles out, or at the pharmacy counter.

What it does

1. Add your medicines. Type them, or paste a whole list from a pharmacy printout. Brand names, doses, and misspellings are fine. If a name is uncertain, it asks "Did you mean...?" instead of guessing.
2. Add a ZIP code (optional). It shows the next 7 days of National Weather Service HeatRisk for that place [2], each day with a color and a word.
3. Get a one-page plan for your pharmacist:
   • A short summary and three next steps
   • One card per medicine that matters in heat: why, what to watch for, and one question for the pharmacist, quoting the CDC word for word with the date it was checked
   • The CDC's combination warning (for example, a water pill with a blood pressure medicine, even inside one combination pill)
   • Storage warnings, such as keeping insulin out of a hot car
   • An honest "not listed" section, with what UK and Canadian guidance say
   • When to call 911, and one fixed rule: it never tells anyone to stop or change a medicine

It also replays a real past heat wave (Phoenix, August 3 to 9, 2025) so you can see your plan on a bad day in any season, and it says plainly when a ZIP is outside the forecast area. It works on phones and desktops, in light and dark mode, and prints on one page.

Does it work?

The 300 most-prescribed US medicines [6] were run through the same code the site uses (run of September 30, 2026):

Check	Result
Warnings without an exact CDC quote	0
Plans that tell anyone to stop or change a medicine	0
Agreement with UK Health Security Agency guidance [3]	75.0% (225 of 300)
Agreement with Health Canada guidance [4]	82.3% (247 of 300)

Every one of the 128 disagreements is listed with a reason on the Evidence page, along with the 18 medicines this site flags that neither the UK nor Canada lists. The numbers on that page are read straight from data/evidence.json; nothing is typed by hand.

This measures agreement with national guidance, not clinical outcomes.

How it works

flowchart LR
    U["You: medicines and ZIP"] --> P["Heat Meds page<br/>(runs in your browser)"]
    P -->|"medicine names"| RX["NLM RxNorm and RxClass<br/>name, ingredients, drug class"]
    P -->|"ZIP to location<br/>(bundled Census file)"| HR["NWS HeatRisk<br/>7-day heat forecast"]
    RX --> R["Rules engine<br/>CDC heat rules, each with a quote"]
    HR --> R
    R --> PLAN["Your one-page plan"]

• There is no Heat Meds server. The page runs in your browser and calls two public government services. Your list is kept only on your device.
• Medicine names are looked up in the National Library of Medicine's RxNorm and RxClass [5], which turn "Lasix 40 mg" into furosemide and its drug class.
• The rules (data/cdc-rules.json) were copied line by line from the CDC page [1], each with its exact quote, link, and the date it was checked. A rule without a quote cannot load.
• The heat forecast comes from the National Weather Service HeatRisk service [2]. If it cannot be reached, the page shows the saved Phoenix heat wave and labels it as past data.
• ZIP codes become map points using the U.S. Census ZCTA file [7], bundled in the site, so no extra service is needed.

Try it with your own data

• On the home page, choose Paste a whole list and paste any list. Directions like "take 1 by mouth daily" are ignored, and anything uncertain asks before it is added. The text stays in your browser.
• On the Developers page, paste any list and see the full result as data.

For developers: one line, no API key

Any web page can use the same check:

<script type="module">
  import { checkMedicines } from 'https://usv240.github.io/heat-meds/src/api.js';
  const result = await checkMedicines(['furosemide 40 mg', 'lisinopril'], { zip: '85004' });
  console.log(result.text, result.plan.cards, result.unresolved);
</script>

There is no API key on purpose: a key would need a server that sees people's medicine lists, which this project promises not to do.

Run it locally

No install and no build step. Any static web server works:

python -m http.server 8000

Then open http://localhost:8000/. (Pages use JavaScript modules, so open them through a server, not by double-clicking the file.)

Testing

Requires Node 20 or newer. No packages need installing:

node --test

This runs 79 tests: the rules engine, name lookup, the forecast modes, the paste parser, the evidence file, and site copy rules. Two of them open the site in headless Chrome or Edge (one-page print, and page width at phone and desktop sizes); they are skipped if neither browser is installed.

Optional live checks against the real services:

node scripts/smoke-heatrisk.mjs     # HeatRisk forecast for downtown Phoenix
node scripts/probe-names.mjs        # RxNorm and RxClass on tricky names
node scripts/example-plan.mjs       # the Phoenix example plan as text
node scripts/check-print.mjs        # prints the example and counts pages
node scripts/check-pages.mjs 375,1280   # page width and layout checks

Rebuilding the data

File	Script	Source
data/zip/*.json	node scripts/build-zip-index.mjs	U.S. Census Bureau ZCTA Gazetteer [7]
data/saved-heatwave.json	npm install then node scripts/sample-heatwave.mjs	NWS HeatRisk daily archive [2]
data/top300.json	node scripts/derive-top300.mjs path/to/h254a.dat	AHRQ MEPS 2024 Prescribed Medicines [6]
data/evidence.json	node scripts/run-evidence.mjs	Live RxNorm and RxClass, the rules, and the answer keys

data/cdc-rules.json and the answer keys in data/answer-keys/ are curated by hand. The answer keys were committed before the first evidence run, so the agreement numbers could not be tuned afterwards.

How it was planned

Built plan-first with the Devpost Learn Skill Pack in a coding agent (Claude Code). The planning documents are in devpost/:

• scope.md: the idea, the person it is for, and what was cut
• prd.md: every screen and behavior, with acceptance criteria
• spec.md: the technical plan, data sources, and failure modes
• checklist.md: the build in seven tested steps, plus every change the build forced
• usability-review.md: a review against published usability standards (AHRQ PEMAT [8], WCAG 2.2 [9])
• app-map.html: a short guide to the code

Accessibility and usability

No live user testing was possible, so the printed plan was scored with AHRQ's PEMAT checklist for patient handouts [8]: 100% understandable and 100% actionable after adding the summary, next steps, and tick boxes it called for. An automated WCAG 2.2 AA check (axe-core) finds 0 issues on every page, at phone and desktop width, in light and dark mode [9]. The typeface, Atkinson Hyperlegible Next, was designed for readers with low vision.

Limits

• The heat forecast covers the contiguous US only. Elsewhere the plan says the forecast is not available and still checks the medicines.
• 15 of the 300 test medicines have no drug class code in RxClass, so only name-based rules can match them; the Evidence page lists them.
• The rules follow the CDC page dated September 18, 2025, checked on September 30, 2026. They do not update themselves.
• Lists of about eight or more medicines print on two pages, to keep the text large enough to read.
• Educational tool, not medical advice. It never tells anyone to stop or change a medicine.

Project layout

index.html, plan.html, evidence.html, developers.html   the pages
src/            JavaScript modules; rules.js is shared by the site and the tests
data/           CDC rules, answer keys, ZIP data, saved heat wave, evidence run
assets/         styles, print layout, font, images
scripts/        data builders and live checks
test/           automated tests
devpost/        planning documents from the Devpost Learn Skill Pack

References

1. CDC. Heat and Medications: Guidance for Clinicians (page dated Sept. 18, 2025). https://www.cdc.gov/heat-health/hcp/clinical-guidance/heat-and-medications-guidance-for-clinicians.html
2. National Weather Service. HeatRisk (experimental) and its daily archive. https://www.wpc.ncep.noaa.gov/heatrisk/
3. UK Health Security Agency. Supporting vulnerable people before and during hot weather: healthcare professionals (Open Government Licence v3.0). https://www.gov.uk/guidance/supporting-vulnerable-people-before-and-during-hot-weather-healthcare-professionals
4. Health Canada. Extreme heat and human health: For pharmacists and pharmacist technicians (2024). https://www.canada.ca/en/health-canada/services/publications/healthy-living/extreme-heat-human-health-pharmacists-technicians.html
5. U.S. National Library of Medicine. RxNorm and RxClass APIs. https://lhncbc.nlm.nih.gov/RxNav/APIs/
6. Agency for Healthcare Research and Quality. Medical Expenditure Panel Survey, HC-254A: 2024 Prescribed Medicines. https://meps.ahrq.gov/mepsweb/data_stats/download_data_files.jsp
7. U.S. Census Bureau. Gazetteer Files (ZCTA). https://www.census.gov/geographies/reference-files/time-series/geo/gazetteer-files.html
8. Agency for Healthcare Research and Quality. Patient Education Materials Assessment Tool (PEMAT-P). https://www.ahrq.gov/health-literacy/patient-education/pemat-p.html

France's ANSM heat and medicines list is shown on the Evidence page for context only.

This product uses publicly available data from the U.S. National Library of Medicine (NLM), National Institutes of Health, Department of Health and Human Services; NLM is not responsible for the product and does not endorse or recommend this or any other product.

Credits and disclosures

• All code in this repository was written during the hackathon's submission period (first commit Sept. 30, 2026), starting from an empty folder.
• Built with Claude Code as the AI coding assistant, following the Devpost Learn Skill Pack (skills-lock.json). Commits made with the assistant carry a co-author line.
• Background research was gathered with AI assistance before the skill interviews. No code from that research is included.
• The demo video was recorded automatically from the live site, with synthetic narration.
• Third-party material: the Atkinson Hyperlegible Next font (SIL OFL 1.1) and geotiff (MIT, used only by a one-time data script). Public data from the CDC, NWS, NLM, U.S. Census Bureau, AHRQ, UK Health Security Agency, and Health Canada, as listed above.

License

MIT. See LICENSE. The font is under the SIL Open Font License 1.1, see assets/fonts/OFL.txt.