Loading dataset

Description

Every way Americans hurt themselves badly enough to visit the ER in 2025, in the hospital's own words. Each emergency-room visit in CPSC's NEISS sample carries the coder's one-line story of what happened, the product involved, the injury, and tags from the Clef model: three emoji, how it happened, what the patient was doing, and a score for how strange it is. Sort by weird_score to find the strangest visits, or filter by mechanism, product or holiday.

The table holds 410,201 visits from about 100 sampled US hospitals. NEISS is a probability sample, so sum weight to estimate national counts. The Clef tags are model output from the narrative alone, and their scores are not calibrated probabilities. The highest weird_score rows are mostly objects stuck in the body, and they include psychiatric self-harm cases, so read the narrative before you share one.

Coverage

  • Window: treatment dates 1 January to 31 December 2025.
  • Grain: one row per sampled ER visit, keyed by case_number.
  • Cadence: a fixed snapshot. CPSC publishes each year as a new file the following spring.

Columns

  • case_number: CPSC case number.
  • treatment_date: date of the ER visit.
  • narrative: the coder's free-text account, in CPSC's capitals and abbreviations (YOM is year-old male, DX is diagnosis).
  • emoji_1 to emoji_3, emoji_1_p to emoji_3_p, emojis: the best-fitting of forty emoji and Clef's 0 to 1 score for each. A slot is filled only at 0.5 or more.
  • mechanism, activity, holiday: Clef's tags for how it happened, what the patient was doing, and any holiday the narrative names.
  • weird_score, stunt_score: Clef's 0 to 1 scores that the injury is unusual or absurd, and that a stunt, dare or prank caused it.
  • age_years, age_months: patient age. Under-twos are 0 years with their age in months.
  • sex, race, race_other, hispanic: patient demographics as CPSC codes them.
  • body_part, diagnosis, diagnosis_other, body_part_2, diagnosis_2, diagnosis_2_other: the main and second injury.
  • disposition: released, admitted, transferred, held, left or died.
  • location: where it happened, such as home, street or sports venue.
  • fire_involvement: whether a fire was involved and whether the fire department attended.
  • product_1 to product_3, product_1_code to product_3_code: the consumer products involved, in CPSC's category wording and codes.
  • alcohol_involved, drug_involved: CPSC's flags.
  • stratum, psu, weight: the NEISS sample design. weight is the number of national ER visits each row stands for.
  • tag_model: the Clef checkpoint that tagged the row.

Missing values

Empty cells mean the field was not recorded: no second body part, no second product, an unknown age. Emoji slots are empty when fewer than three emoji scored 0.5. Every visit has Clef tags. holiday is loose: other_holiday also catches vacations, and a few Thanksgiving and New Year tags rest on weak cues.

Suitable for

  • Finding strange, funny or alarming injury stories by product, holiday or mechanism.
  • Estimating national injury counts for a product with the sample weights.
  • Training and testing text classifiers on short clinical narratives.

Source and rights

The National Electronic Injury Surveillance System, run by the US Consumer Product Safety Commission. A US government work in the public domain. The Clef tags were added by Mostly Right.

Tables

  • Table healthAs published 15 October 2026
    injuriesPublished
    Table overviewPublished rows410,201Columns42 rows42 columnsUpdated

Sources

CPSC.govCPSC works to save lives and keep families safe by reducing the unreasonable risk of injuries associated with consumer products and fulfilling its vision to be the recognized global leader in consumer product safety.2 endpoints

CPSC works to save lives and keep families safe by reducing the unreasonable risk of injuries associated with consumer products and fulfilling its vision to be the recognized global leader in consumer product safety.

Usage rights
Public domain asserted.
Collection
3 requests across 2 endpoints

Collected endpoints

CPSC NEISS 2025 annual file: one row per sampled ER visit for a consumer-product injury./cgibin/NEISSQuery/Data/Archived%20Data/2025/neiss2025.tsv2 requests
CPSC NEISS code formats: the label for each product, body part, diagnosis and other coded value./cgibin/NEISSQuery/Data/Info%20Docs/neiss_fmt.txt1 request
ClefServed by Clef (asia-south2), Clef (asia-southeast1), Clef (us-central1) and Clef (europe-west4)1 endpoint
Usage rights
Public domain asserted.
Collection
1 request across 1 endpoint

Collected endpoints

One Clef answer per NEISS narrative: forty emoji yes/no scores, three choice tags and two yes/no flags.clef_injuries1 collectionAll 410,201 pages fetched

Details

Contents
1 table ยท 410,201 rows ยท 42 columns
Updated
Published
Version
v1
License
Not stated
Visibility
Public
Publisher
Mostly Right

Activity

Views153+153 in the last 30 days
Likes0+0 in the last 30 days
Uses2+2 in the last 30 days