Research · In progress · Not yet peer reviewed
When sheltering in place fails.
Before a hurricane, a hospital has to choose: evacuate, or shelter in place. This study asks what happens when a hospital shelters in place and then has to evacuate anyway.
Pythonpandasnumpystatsmodels
01 · At a glance
The question, the data, and the terms.
A study of hospitals in Atlantic and Gulf hurricanes from 1990 to 2025, built from public sources only.
- The question
- What happens when sheltering in place fails, and the hospital has to evacuate anyway?
- Who it’s for
- Researchers and hospital emergency preparedness leaders.
- Authors
- Tyler Wallace, PhD (University of Georgia) and Curt Harris, PhD (Augusta University).
- Data
- Public sources only.
- NOAA National Hurricane Center: HURDAT2 and ATCF forecasts
- CMS Provider of Services and HCRIS
- NOAA SLOSH surge maps and PRISM rainfall
- USGS 3DEP elevation and the FEMA National Flood Hazard Layer
- The Census geocoder and gazetteer
- Public news, state, and hospital reports, for the registry of decisions
- Status
- Research in progress. Not peer reviewed.
- License
- No license is stated in the repository.
02 · What it is
A national cohort, and a registry of decisions.
The study has two parts, plus a replication package that re-runs every number it reports.
- A
The cohort
80,189 hospital–storm pairs: 3,241 hospitals and 193 storms, across 21 states and DC, from 1990 to 2025.
- B
The registry
261 documented hospital decisions across 31 storms. AI tools “did much of the gathering and coding,” and every record links to its sources.
- C
The replication package
It re-runs every reported number, and should end with “PASSED: zero discrepancies.”
03 · Early findings
Preliminary, and likely to change.
The findings so far are early. They have not been peer reviewed, and no person has yet checked the registry against its sources.
They are listed here so emergency preparedness leaders can see the direction of the work. Please don’t cite them yet.
PreliminaryWhat the registry shows so far
- 60 hospitals sheltered in place and then had to evacuate anyway, “almost always within two days of landfall.”
- “Power or water failure forced evacuation as often as flooding or building damage.”
04 · Limits
What isn’t settled yet.
These limits come from the project’s own documentation.
- The work is not peer reviewed, and the numbers may change.
- No person has checked the registry against its sources. A 55-record sample check is under way.
- The registry holds only the decisions someone wrote about. It is not a census of every hospital’s decision.
- “Associations between hospital siting and failure are weak and imprecise. No conclusion here rests on them.”
- Some codebook counts are out of date. Raw downloads are left out of the repository for size.
Follow the work as it’s checked.
The code, the data, and the replication package are on GitHub. Please don’t cite the numbers yet.