2025 NEISS Magnet Injury Data Observations

2026-10-05

–

11 min read

– MagnetSafety.org –

Shihan

1. Executive Summary

The National Electronic Injury Surveillance System (NEISS), operated by the U.S. Consumer Product Safety Commission (CPSC), samples emergency departments nationwide to capture injuries associated with consumer products. Because every case is statistically weighted, NEISS is the only product-injury database that produces nationally representative estimates.

According to the 2025 NEISS dataset, emergency departments treated an estimated 153,000 foreign-body ingestions, up from 138,000 in 2024 and 114,000 in the 2022 data. Over the last four years, magnet-related incidents have made up a steady 5.5–5.8% of all ingestions. Ingestions of spherical rare-earth magnets (SREMs) fell to their lowest level since 2017, about 0.46% of total ingestions in 2025, or 8% of all magnet cases.

2. Data Overview

Foreign-body ingestion visits to emergency departments keep accelerating. Between 2022 and 2025, estimated cases rose from 114k to 153k, an absolute gain of about 38k cases and a compound annual growth rate (CAGR) of about 10.1%. For comparison, 2003–2022 averaged a CAGR of about 2.8%, with cases rising from 68k to 114k. Over the full 2003–2025 span annual ingestions increased by about 126%, while the U.S. population grew by less than a fifth. The 2025 estimate carries a 95% confidence interval of ±32k cases (CV 0.10). Relative to the estimate, that is ±20.9%, the tightest margin since 2013.

Magnet ingestion incidents continue their upward trend, which accelerated in the second half of the series. After a dip around 2016, incidents rose from 1.7k in 2016 to 8.8k in 2025, a fivefold increase that includes 11.5% growth over 2024. This suggests some combination of more accessible magnets in consumer products and greater awareness and reporting of these incidents.

Spherical rare-earth magnet (SREM) ingestions, identified with CPSC’s 2014 method for finding NEISS in-scope magnet incidents (narratives that indicate a powerful or round magnet), are volatile. They peaked in 2008 and 2012, stayed low through 2013–2017, then surged to 2.2k in 2020. Since 2022 they have fallen three years in a row, to about 702 in 2025, the lowest since 2017. The decline may reflect reduced supply after regulatory action and enforcement, or reduced demand from greater public awareness of the injuries high-powered magnets can cause [Related: SREM supply-side tracking]. The 2025 estimate rests on 30 sample cases and is below CPSC’s 1,200 reliability threshold, so the size of any single-year change is uncertain.

Battery ingestions have risen steadily, from about 2.7k in 2003 to 15.9k in 2025, with a pronounced increase since 2016 and a 29% jump in 2025 alone. Much of this follows the spread of small button batteries across consumer electronics. Clinicians often discuss battery and magnet ingestions together because both cause more severe internal injuries than most other swallowed objects.

The magnet share of all foreign-body ingestions has climbed overall since 2003. It reached an early peak of 3.8% in 2007, before SREMs entered the market, and an all-time high of 6.4% in 2020. It has held between 5.5% and 5.8% in each of the last four years (2022–2025), so magnets remain a significant and persistent part of foreign-body injuries.

SREM ingestions show marked volatility, with an initial surge to 1.2% of all ingestions in 2008, before Buckyballs appeared in 2009. In the 2014 magnet-set ban, CPSC believed Buckyballs (by Maxfield and Oberton) controlled over 90% of the US SREM market. CPSC compliance work began in 2011, intensified in 2012 and continued through the 2015 magnet-set ban [CPSC Magnets Enforcement Page, archived Jul 2021]. Leaving out 2003–2005, before SREM sets were widely sold, four of the five lowest SREM ratios occurred during 2013–2017, mostly after the Buckyballs recall; the fifth is 2025. After spiking to 2.4% in 2020, the ratio has fallen in each of the last three years, from 1.8% in 2022 to 0.46% in 2025, the lowest since 2017. The decline coincides with renewed notices of violation starting in 2021 [CPSC Magnets Page archived Aug 2021]. Some of this drop is likely due to CPSC compliance efforts affecting supply. Supply-chain monitoring and buyer-sentiment studies are needed to separate reduced availability from waning consumer interest.

The In Scope 2022 estimates come from Table 5 of CPSC’s 2022 magnet-set rule and represent the incidents the agency believes the rule would prevent. CPSC has not released the NEISS records behind those in-scope injuries, so the series still ends in 2021 and cannot be extended or checked. The 2022 and 2014 rules cover largely overlapping products, yet the injuries attributed to them differ widely, which limits any assessment of the newer rule’s efficacy.

Rulemaking only claims to target “in scope” incidents, so in-scope estimates as a share of all magnet ingestions are a useful gauge of regulatory effectiveness. If the in-scope share drops sharply after a ban, the rule is affecting the problem it set out to solve relative to other magnet ingestions. The share also shows whether the problem is shifting to out-of-scope magnet products, or whether the products of concern still dominate.

The SREM share of all magnet ingestions has fluctuated widely. The first magnet ingestion articles emerged in 2004, and 2005 coincides with the expiration of the main NdFeB magnet patent (US No. 4,588,439). The share reached its all-time high in 2008 (37.0%), before high-powered magnet sets were widely available, and nearly matched it in 2020 (36.9%). Over the last four years it has fallen sharply, from 31.98% in 2022 to 20.48%, 10.69% and 7.97% in 2025, the lowest of any year except 2003. Through the in-scope lens used for the 2014 magnet ban, injuries from these magnets kept falling after the 2022 ban while magnet ingestions overall grew.

The share of magnet ingestions counted as in scope for the 2022 ban was remarkably high, often above 80% between 2010 and 2013. That suggests CPSC attributed the vast majority of magnet incidents in that period to the products the ban eventually targeted. The share declined after 2014, from 82.19% in 2013 to 36.82% in 2021. The two in-scope series nearly coincide only in 2020.

3. Dataset Definitions

Four primary NEISS datasets are evaluated. Section 4 gives the methodology and reproduction steps.

• ALL Ingestions (ALL_Ingest). Every NEISS incident coded ingestion (41) or aspiration (42). Aspiration is included because of mislabeling, CPSC’s own criteria, and related magnet hazards.
• Magnet Ingestions (MAGN). Incidents in ALL_Ingest whose narrative contains “MAGN”. This deterministic text match is applied identically to every year from 2003 to 2025 (see section 6).
• Battery Ingestions (BATT). Incidents in ALL_Ingest whose narrative contains “BATT”.
• Spherical Rare Earth Magnet Ingestions (SREM). An approximation of the “in-scope” injuries for the 2014 magnet ban, which mainly targeted spherical rare-earth magnets. It follows CPSC’s 2012 draft NPR criteria for “High-Powered and/or Ball-Shaped Magnet Ingestions”, based on 72 specific NEISS incidents.

Annual national estimates, derived from the sum of weights, are in the spreadsheet’s Analysis tab. The ALL_Ingest sheet carries the MAGN and BATT flags; SREM results are in the MAGN_Only tab of the MagnetSafety.org 2003-2025 NEISS Analysis spreadsheet.

4. Methodology and Data Reproduction

ALL_Ingest, MAGN and BATT can be regenerated in minutes with Excel, Google Sheets or OnlyOffice and contain no subjective filtering. Only the SREM dataset involves judgment.

Collecting NEISS ingestion data

In the NEISS Query Builder, select Diagnosis 41 and 42 and the treatment year, group by year, include CV, 95% confidence interval and number of cases, then download in Excel format. Query one year at a time: the Query Builder widens the CV and interval when years are pooled (section 6). Alternatively, take the year’s archived NEISS file from the same page and keep rows with Diagnosis or Diagnosis_2 equal to 41 or 42, upper-casing the narratives to match the Query Builder export. To skip this step, use the ALL_Ingest tab of the 2025 spreadsheet, which holds every diagnosis 41 and 42 case from 2003 to 2025.

Generating national estimates for all ingestions

Add a Treatment Year column with =YEAR(B2). List the years in a separate column (e.g., AH), then sum weights with =SUMIF(C:C,AH2,Z:Z). The result should match the Query Builder estimates.

Magnet ingestions

Flag magnet mentions with =IF(ISNUMBER(FIND("MAGN",W2)),1,0) in a new column (e.g., AA), then compute yearly estimates with =SUMIFS(Z:Z,C:C,AH2,AA:AA,1).

Battery ingestions

Flag battery mentions with =IF(ISNUMBER(FIND("BATT",W2)),1,0) (e.g., column AB), then compute yearly estimates with =SUMIFS(Z:Z,C:C,AH2,AB:AB,1).

SREM ingestions: LLM quorum

SREM identification emulates CPSC’s 2012 selection of 72 in-scope incidents, the only in-scope list CPSC has published. MAGN narratives are first pre-matched on POWER, RARE, MARB, BALL, "BB ", BEARING, BEAD, SPHER, " ROUND", PIERCE, TONGUE, TOUNGE, " LIP". Each pre-match is then judged by three independent LLM instances using the SREM Identification Blueprint prompts, and a 2-of-3 vote decides. Through 2024 the quorum ran on GPT-4. The 2025 rows used Claude Opus 5.5 with the same three prompts, calibrated against the GPT-4 verdicts (section 6).

5. Dataset Limitations

The limitations of the ALL Ingestions dataset are those inherited from the design of the NEISS database, and apply to all NEISS data:

• Sampling. NEISS samples hospitals and weights the results, so estimates carry sampling error. CPSC notes that an estimate is unreliable if it is below 1,200, rests on fewer than 20 cases, or has a CV above 33%. The CV method is described in the NEISS sample design documentation.
• Detail. Narratives are brief and often lack context.
• Public data. Public narratives may be redacted for patient privacy, brand confidentiality or ongoing investigations.
• Accuracy and consistency. Reporting practices differ between hospitals, and data-entry errors occur.
• Inpatient data. NEISS captures emergency departments only, which is why CPSC also uses ICM estimates for rulemaking.

Magnet ingestions. The “MAGN” match catches misspellings such as “MAGNATIC”, “MAGNECTIC” and “MAGNTIC”. It misses others such as “MAGENT”, and has no false-positive filter for phrases like “NON-MAGNETIC” or words like “MAGNESIUM”. A manual reading suggests it is at least 97% accurate.

Battery ingestions. “BATT” catches “BATTRY” but also negations (“NOT A BATTERY”) and unrelated words (“BATTLESHIP”). It is also over 97% accurate.

SREM ingestions. The blueprint follows CPSC’s 2012 analysis, and there is evidence its criteria do not identify SREMs consistently. For example, it finds more in-scope injuries in 2006–2008, before SREMs were prevalent, than in 2009–2011. The table below applies NEISS’s stratified PSU variance method (single-year design) to each 2025 series. For all ingestions, this method reproduces the Query Builder’s intervals to within one case.

6. 2025 Update Notes

Data source and validation

CPSC released the 2025 data on May 7, 2026. The 2025 case rows come from the archived neiss2025.tsv file, filtered to Diagnosis or Diagnosis_2 equal to 41 or 42. That is the same selection the Query Builder returns. Applied to the 2023 and 2024 files, the filter returns exactly the 4,213 and 4,650 cases already in the workbook, and every 2024 field matches. The 2025 national estimate, CV and confidence interval are pasted from a 2025-only Query Builder query (152,573; 5,516 cases; CV 0.10; 120,687–184,459).

The Query Builder’s CV depends on the years queried

MAGN restated for 2023–2024

The 2024 edition coded 2023 and 2024 with a variant formula that also required diagnosis 41, which dropped magnet cases coded as aspiration. This edition applies the documented definition to every year. It counts any “MAGN” narrative among diagnosis 41 and 42 cases, as in 2003–2022.

The dropped cases included clear swallowings coded as aspiration, such as a toddler “witnessed to ingest a stack of about 8 magnets” with a diagnosis of foreign body in airway. Restated MAGN is 6,663 for 2023 (was 6,593) and 7,898 for 2024 (was 7,777, reported as 7.8k in the 2024 article). The magnet share becomes 5.51% and 5.74%. SREM is unchanged, because none of the added cases matches an SREM keyword.

SREM quorum model

The keyword pre-match flagged 37 of 390 magnet cases in 2025. Three independent Claude Opus 5.5 instances judged each one with the original Run1, Run2 and Run3 prompts. Thirty were included, 28 of them unanimously.

All 108 pre-matched 2023–2024 narratives were mixed blind into the same batch. The new quorum agreed with the earlier GPT-4 verdicts on 91.7% of them (98.1% for 2023, 85.7% for 2024). Disagreements went both ways: Claude was stricter on hedged narratives (“possibly swallowed”, “suspected … not found”) and more inclusive for “magnetic marble” or “magnetic ball” descriptions. Under the new quorum the 2024 SREM estimate would be 888 rather than 844, so the 2025 decline is not caused by the model change.

Interactive charts

The four charts keep the 2024 titles and order in a refreshed format. Each line is labelled at its latest value, and the All values switch shows every label. One hover box lists every series for a year. The 95% confidence interval for all ingestions is now a shaded band instead of two grey lines. On phones the legend sits above the chart, and dragging scrolls the page.

US Magnet & Battery Ingestions now shows all ingestions in a top panel and magnet and battery ingestions on their own scale below it, so their trends are readable. The Split scale switch restores the single axis. The three magnet charts also have a Policy timeline switch, off by default. It shades the periods of CPSC action: compliance work from 2011, the 2014 magnet-set rule (in effect from April 1, 2015), its vacatur by the Tenth Circuit on November 22, 2016, and the 2022 magnet rule (effective October 21, 2022). It also reports each period’s yearly averages. Those averages describe the estimates; they don’t show by themselves what caused a change.

Workbook fixes

• Quorum consensus formula. MAGN_Only!AZ used the Google-Sheets-only syntax COUNTIF({AU2, AW2, AY2},"1"), which OnlyOffice evaluates to #NAME?. All 2,892 instances were rewritten in an equivalent portable form. Recalculated in OnlyOffice, the workbook reproduces every 2003–2022 value of the 2024 edition and the restated 2023–2024 values.
• POWER keyword column. In the 2024 rows this column held a copy of the MAGN formula and always returned 0. It was restored; no 2024 pre-match changes.

Sources: CPSC NEISS Query Builder & archived annual data · 2024 NEISS observations · SREM identification blueprint · LLM quorum method.