NEISS Research Methods: How to Run an Analysis That Survives Peer Review

Published June 20, 2026  |  Last updated September 10, 2026

A NEISS project usually fails in one of three places. The case definition does not survive contact with the code book, the national estimate is produced without the survey design attached, or a trend turns out to be an artefact of a product code retired in 1995 and quietly restored five years later. NEISS research methods are specific rather than complicated, and almost none of the specifics are visible from the data file alone. This guide walks the analysis end to end, from case definition through weighted estimation, variance, narrative coding and trend testing, using the rules the Consumer Product Safety Commission publishes.

Two Numbers Every NEISS Analysis Has to Keep Apart

The most common error in NEISS work is treating the number of collected cases and the national weighted estimate as one quantity at two levels of precision. They are different quantities with different shapes.

The collected number is small and concrete. CPSC’s FY2023 Annual Report to the President and Congress states that “annually, NEISS supplies information on almost 400,000 product-related cases,” and the agency’s supporting statement to the Office of Management and Budget plans on about 470,000 product-related cases per year from the sampled hospitals, of which roughly 2,375 are selected for follow-back investigation. Those are records sitting in a file you can download.

The national estimate is large and derived. CPSC’s Directorate for Epidemiology put all consumer product-related emergency-department-treated injuries at 10,994,000 for calendar year 2020, a figure needing two caveats wherever it is cited, since 2020 was a pandemic year with depressed emergency department volumes and the number aggregates across product groups. Single-category figures are safer. From CPSC’s 2024 Data Highlights, stairs, ramps, landings and floors carry an estimated 3,101,789 injuries, and beds, mattresses and pillows 1,114,811.

Reviewers of complex survey data look for exactly this distinction. If NEISS is still a candidate rather than a decision, the comparison of which health database your study should use settles that first, and how an ER visit becomes a NEISS record explains where the raw material comes from.

The Three Design Variables NEISS Research Methods Depend On

NEISS is a stratified probability sample of 100 hospitals drawn from a 2020 frame of 4,723 eligible United States hospital emergency departments representing 155,707,175 annual visits. Frame eligibility requires at least six inpatient beds and a 24-hour emergency department. The strata are four emergency-department size bands plus a children’s hospital stratum.

Stratum	Annual ED visits	Hospitals sampled	Hospitals on frame
Small	1 to 33,280	43	2,984
Medium	33,281 to 56,995	26	860
Large	56,996 to 92,195	12	528
Very Large	92,196 and above	11	286
Children’s hospitals	Children’s designation	8	65
Total		100	4,723

That design reaches your analysis through three columns in the public file, named exactly as follows in the CPSC Data Dictionary. Stratum identifies the size band. PSU is the primary sampling unit, meaning the sampled hospital. Weight, abbreviated Wt in the SAS-format file, is the statistical weight for national estimates.

The weight is not a constant and it is not approximately sixty. Its base value is the inverse of a hospital’s selection probability within its stratum, adjusted for non-response and hospital mergers, following the CPSC formulation NEISSwt = (N_h × n'_h) / (n_h × r_h), where within stratum h the terms are frame hospitals, hospitals selected, hospitals in scope and hospitals participating. A record from a small-stratum hospital can carry a far larger weight than one from a very large emergency department. An unweighted row count is therefore not a scaled-down national number but a different statistic, whose bias depends on which strata your cases fall in.

The consequence has been quantified. Durand and colleagues used NEISS itself as the test case, analysing rug-related injuries in adults aged 65 and over by sex, and set design-corrected logistic regression and jackknife re-weighted linear regression against weighted least squares and naive approaches. The design-corrected techniques performed comparably, and the authors reported that “in almost all cases, design-corrected approaches contrasted profoundly with standard statistical techniques.” Cite it when a co-author asks why the design variables cannot be dropped. Our walkthrough of what happened when we ran the same NHANES analysis three ways shows the same effect on a different survey, and how NEISS weights scale a sample to a national estimate covers the sampling logic.

Turning Weights Into a Defensible National Estimate

The estimate is a Horvitz-Thompson sum of weights over the selected cases, so the arithmetic is trivial. The work is declaring the design once, correctly, and letting survey software propagate it through every downstream estimate.

library(survey)

neiss_design <- svydesign(
  id      = ~PSU,        # primary sampling unit, the sampled hospital
  strata  = ~Stratum,    # four ED-size strata plus children's hospitals
  weights = ~Weight,     # the CPSC statistical weight
  data    = neiss,
  nest    = TRUE         # PSU identifiers are nested within strata
)

# National estimate with a design-based standard error.
# Diagnosis code 57 is Fracture in the NEISS diagnosis code set.
svytotal(~I(Diagnosis == 57), neiss_design)

# Convert the standard error to a coefficient of variation, then apply
# CPSC's reportability rules before writing the number down.

The Stata equivalent is svyset PSU [pweight=Weight], strata(Stratum), and in SAS the same declaration sits inside PROC SURVEYMEANS or PROC SURVEYFREQ as STRATA Stratum; CLUSTER PSU; WEIGHT Wt;. Watch the names. Snippets borrowed from a NEDS or HCUP tutorial reference discwt and hosp_ed, which do not exist in NEISS.

Summers and colleagues show what finished output looks like. From 1,502 unique NEISS cases, weighted analysis in Stata gave an estimated 52,118 pediatric nonballistic firearm injuries to the upper extremity over 2000 to 2017, an average of 2,895 per year, more than 91% in males. Both numbers appear, the records and the estimate.

CPSC’s Own Reportability Rules, and Why NEISS Research Methods Should Mirror Them

CPSC does not publish every estimate its own system can produce. The agency’s explanation of NEISS estimates sets out the rules. Variance is computed from the sample design and expressed as a coefficient of variation, the square root of the variance divided by the estimate. The 95% confidence interval is the estimate plus or minus 1.96 times the estimate times the CV. An estimate is unstable, and goes unreported, when it falls below 1,200, when fewer than 20 cases contribute, or when the CV exceeds 33%.

Adopting those thresholds as your own suppression rule is the cheapest credibility available in a NEISS paper. It answers the reviewer who asks why a subgroup was collapsed, and it stops you publishing a cell that will not replicate.

It also puts the thinness of some published work in perspective. Blakney and colleagues at the Food and Drug Administration estimated 2,043 emergency department visits (95% CI 1,326 to 2,760) for electronic nicotine delivery system battery explosions and burns over 2020 to 2024, resting on 52 unweighted cases, 26.3% of which required admission. Lacher and colleagues derived a 20-year estimate of 3,112 pickleball-related ocular injuries from 73 unweighted cases in JAMA Ophthalmology. Both are legitimate analyses in good venues, and both show how little raw data can sit under a large-looking number.

Report the unweighted count beside the estimate, with an interval. Neuroth and colleagues reported an estimated 60,554 electric scooter injuries (95% CI 37,525 to 84,594) over 2015 to 2019, roughly 19% involving a motor vehicle. An interval that wide is informative, and hiding it would not have strengthened the paper. Where a subgroup looks too small, our guide to whether you have the power to detect it covers the calculation.

Defining the Case Without ICD-10

NEISS does not use ICD-10-CM or ICD-10-PCS. A research plan built on ICD-10 case definitions collapses the moment the file is opened. NEISS uses closed internal code sets maintained by CPSC, every permitted value enumerated in the Data Dictionary and the January 2025 NEISS Coding Manual. NEDS, distributed through AHRQ’s HCUP programme, is the ICD-10 database. NEISS is not.

What you get instead is a two-digit body part code with 26 current values plus four legacy codes marked old, a two-digit diagnosis code with 30 values running from 41 Ingestion through 57 Fracture and 59 Laceration to 74 Dermatitis, a four-digit product code, a seven-value disposition code and a nine-value location code.

Two traps sit here. Age is not a plain integer. Zero means age not recorded, 2 to 120 are years, and 201 to 223 encode ages of 1 to 23 months, so treating the column as numeric without recoding turns every infant into a two-hundred-year-old, silently destroying any age-stratified result. And the case definition needs validating. Thompson and colleagues linked 2,890 NEISS cases at a children’s hospital back to their medical records and found 13.32%, or 385 cases, were not injuries at all by the ICD-9-CM diagnosis in the same chart. Among genuine injuries agreement was good, kappa 0.87 (95% CI 0.85 to 0.88), but the authors concluded that best practice “should include manual review of case narratives.”

The same group tested one specific definition. Measured against the CDC ICD-9-CM definition of traumatic brain injury as the reference standard, the recommended NEISS TBI definition achieved sensitivity of 91.0% (95% CI 89.6 to 92.4) and positive predictive value of 76.5% (95% CI 74.6 to 78.4). Roughly one in four cases it captured was not a TBI by the ICD standard. Carry that measurement-error term into the discussion rather than leaving it for a reviewer. The cluster piece on how NEISS product and diagnosis codes are assigned covers where these classifications come from at the bedside.

Handling Up to Three Product Codes Per Case

NEISS permits up to three products per injury, held in Product_1, Product_2 and Product_3, with unused slots filled 0000. Product_3 exists only from 2019 onward. This one design feature generates more arithmetic errors than anything else in the file.

CPSC states the consequence in every Data Highlights release. Because up to three products can be coded for one person’s injury, that injury may be counted in three product groups, so the published category estimates cannot be summed. Anyone adding them together has double and triple counted people. The same rule binds your own work. If the definition is “any case where the product of interest appears,” a case mentioning two other products contributes its full weight to your estimate and to two others, which is correct for a product-specific question and wrong for anything framed as a share of all injuries.

Decide and state which position you take. Product_1 alone gives a narrow definition anchored to the coder’s judgement about the principal product. All three fields give the broad definition most published product studies use. Excluding cases where a competing product also appears gives a clean but smaller sample. None is wrong. Failing to say which you used is.

The 400-Character Narrative Field Is Where NEISS Research Methods Earn Their Keep

The Narrative field is why many researchers choose NEISS over larger administrative databases. NEDS has no free text. NEISS carries a description of the injury event capped at 400 characters, written to a defined convention. The coding manual requires it to open with an age and sex abbreviation such as 25YOM or 19MOF, to name the specific product, to place the clinical diagnosis at the end preceded by DX, to record blood alcohol in three-digit form such as BAC 080, and to carry no personally identifiable information.

That structure is exploitable, and the literature shows three approaches at different cost points. Manual review is the baseline. Shields and colleagues read pressure cooker narratives to classify precipitating factors into four categories, burning agent, struck by product, injured while opening and other, none of which exists as a coded field in NEISS. Supervised learning with a human in the loop sits in the middle. Chen and colleagues used the interactive DUALIST tool to sort NEISS traumatic brain injury narratives into sport and non-sport categories, and after roughly 60 minutes of training the classification time for tens of thousands of narratives fell from days to minutes. Full NLP pipelines sit at the top. Omaki and colleagues coded 4,546 narratives by hand to train a model, then yielded 91,325 coded pediatric fall cases from 2012 to 2016, finding 33% of infant fall injuries involved falling from a bed and odds of hospitalisation 2.1 times higher (95% CI 1.6 to 2.7) after a fall from another person.

The narrative can change a headline result, not merely enrich it. Parker applied NLP to 59,025 unweighted firearm injury narratives from 1993 to 2015 to recover injury location where the coded field was uninformative. Existing estimates placed the home first at 23.4% and street or highway second at 22.9%; after classification street or highway was most frequent at 46.1% and the home second at 37.7%. Dell and colleagues used the narrative to identify a population NEISS does not code at all, flagging probable homelessness in 0.10% of firearm injury cases (n=3,225), among whom assault accounted for 82.64% of injuries.

Two cautions belong here. Ehlman and colleagues, evaluating NEISS-AIP self-directed violence data for 2018 for CDC, found more than 99% of the cause and intent variables coded consistently with the open-text field capturing the chart narrative, while flagging that many of those variables rely on open-ended responses difficult to analyse efficiently. That is internal coding consistency for two variables in one subsystem in one year, not a NEISS-wide accuracy rate. Whatever method you choose, the classification rules need documenting and ideally double-coding, because a narrative-derived variable is one you created. The sibling post on the free-text narrative field goes deeper.

Trend Analysis Traps in NEISS Research Methods

Multi-year work is where quietly wrong papers get published, because the file will happily return a smooth series reflecting administrative change rather than injury change. Four traps recur.

Product codes are not stable. CPSC’s Product Code Comparability Table documents every code merged, split, deleted or reactivated since 1972. Codes 0104 and 0105 merged into 0135 in 1978 and later into 0140 by 2003. Code 0416 for corkscrews was deleted in 1983, restored in 1994, deleted again in 1995 and restored in May 2000. Code 0825 for stationary power saws was split into 0841 through 0845 from 1974. Query one code across twenty years without that table and the trend partly records CPSC’s filing decisions.

Variables appear mid-series. Body_Part_2, Diagnosis_2, Other_Diagnosis_2, Product_3, Hispanic, Alcohol and Drug are available only from 2019. A Sex value of 3 for non-binary or other is not available before 2021. Any series crossing 2018 to 2019 or 2020 to 2021 has a structural break in the variables themselves.

The reporting base moves. Ninety-six hospitals reported in 2019 and again in 2023, while 77 of the sampled 100 reported in 2024. Weights are adjusted for non-response, so estimates remain defensible, but a reader comparing 2024 with 2019 deserves to know the base shifted.

And the trend model has to respect the design. Lacher and colleagues used survey-weighted linear regression and reported pickleball ocular injuries rising by 405 cases per year (95% CI 288 to 522, P=0.004) from 2021 to 2024. Shields and colleagues used negative binomial regression across 2003 to 2019. An ordinary least squares line fitted to seventeen annual point estimates, each treated as one observation with no variance, is the commonest shortcut and the easiest to reject. Our free Poisson regression calculator sanity-checks a count trend, and the biases baked into NEISS sampling covers what else can masquerade as one.

One more item belongs in any methods section written now. CPSC announced in 2026 a modernization, NEISS-Remodel, moving from manual medical-record abstraction at roughly 70 hospitals to standardized electronic data exchange covering all 50 states, expected to become fully effective by the beginning of 2027. Under the legacy design 14 states had no participating hospital at all. State which NEISS vintage you used and the paper ages gracefully.

A Worked NEISS Study From Case Definition to Estimate

The pressure cooker study is the cleanest published example to narrate end to end, because every step is documented and the numbers reconcile.

Step	What the authors did	Result
1. Define the case	Query NEISS for the relevant product codes, 1 January 2003 to 31 December 2019	Definition fixed before extraction
2. Read the narratives	Manual review to classify precipitating factors	Four categories in no coded field
3. Count unweighted cases	Count the records contributing	759 records
4. Apply the weights	Sum Weight, with Stratum and PSU declared	28,337 injuries (95% CI 24,588 to 32,086)
5. Model the trend	Negative binomial regression over the period	Change tested, not eyeballed
6. Report both numbers	Records and estimate side by side	759 records, 1,667 estimated injuries per year

Budnitz and colleagues are worth reading as a reporting exemplar. Publishing NEISS-CADES data in JAMA, they named the participating emergency departments (60) and the cases (96,925), reporting 6.1 visits for medication harms per 1,000 population annually (95% CI 4.8 to 7.5), of which 38.6% (95% CI 35.2 to 41.9) led to hospitalisation. Once you know what you are extracting, the guide to how to get the NEISS data files covers the download. Files are free and public, covering data years 2006 through 2025 in Excel, tab-delimited and SAS formats, reachable through the NEISS Estimates Query Builder with no registration.

What NEISS Cannot Answer

A limitations paragraph naming real constraints reads as competence. These are the ones that bind.

NEISS covers emergency-department-treated injuries at sampled hospitals only. Injuries handled in urgent care, in physician offices, at home or not at all are invisible, as are deaths that never reach an emergency department. There is no patient identifier, so no linkage across visits, no readmission, no post-discharge outcome and no cost field. Questions about recovery, recurrence or expenditure belong elsewhere, and the limitations of observational real-world data is a useful frame for saying so.

Geography is the constraint most often raised in print. Navon and colleagues screened 6,944 citations and identified 167 journal articles using NEISS-AIP as the primary data source between 2001 and 2021, an average of 8.0 per year across 72 journals. The most commonly identified limitation, named by 38 of them, was that the programme produces national estimates only and cannot support state or county-level surveillance.

CPSC’s phrasing on causation is worth adopting verbatim. The system estimates injuries “associated with (but not necessarily caused by) specific consumer products.” Any results sentence reading as a causal claim will attract a comment.

Do not treat the subsystems as interchangeable either. NEISS covers product-related injuries. NEISS-AIP, launched with CDC in 2000, captures all nonfatal injuries at a subset of hospitals. NEISS-Work, run with NIOSH, draws on a national probability-based sample of approximately 67 emergency departments, a figure Kennedy and colleagues state directly in a sharps injury analysis that found health care workers experiencing rates up to 16 times those of all United States workers. NEISS-CADES used 60 emergency departments over 2017 to 2019. Different subsets, definitions and weights, so citing a NEISS-Work rate as a NEISS figure is a factual error even when the paper is sound. Hendricks and colleagues put occupational inhalation injuries at 2.2 per 10,000 full-time equivalents (95% CI 1.6 to 2.8) from NEISS-Work over 2014 to 2017, nearly four times the Bureau of Labor Statistics rate.

Carpenito and colleagues state the NEISS and NEDS trade-off cleanly. HCUP-NEDS “provides accurate data for the number of nonfatal firearm injuries, [but] injury intent is not coded reliably,” while the NEISS firearm supplement codes intent reliably yet “is too small to produce stable estimates of the number of these events.” Their multiple imputation by super learning on undetermined-intent cases recovered trends differing from complete-case analysis.

Reporting a NEISS Study So Reviewers Accept It

A NEISS analysis is a cross-sectional study of a complex survey sample, so the reporting checklist is STROBE, covered in full in our tutorial on the STROBE checklist for observational studies. Six items carry disproportionate weight here.

Name the data years and the NEISS vintage, since the system is mid-transition. Give the case definition as executable logic, meaning the product codes searched, which product fields were used, the diagnosis and body part codes included, and any narrative-based inclusion rule. State that Stratum, PSU and Weight were used, and name the software procedure. Report unweighted counts alongside every weighted estimate with confidence intervals. State the suppression rule and attribute it to CPSC. Describe how narrative-derived variables were created, who coded them and what agreement was achieved.

Volume of published NEISS work is high enough that novelty needs checking before analysis. A PubMed query on "National Electronic Injury Surveillance System"[Title/Abstract] OR NEISS[Title/Abstract], run on 9 September 2026, returns 1,268 records, with 114 published in 2023, 140 in 2024 and 163 in 2025. Re-run it for your own product and outcome before writing a protocol, because a descriptive paper on a well-worked product category is a difficult sell.

Tools and Resources for NEISS Research Methods

The primary documentation is short and worth reading before extraction. The CPSC NEISS FAQ covers release cadence, with the most recent treatment year generally available in April of the following year, although the 2025 file was posted on 7 May 2026. The NEISS-R modernization announcement sets out what changes in 2027.

Professional Support for a NEISS Study

Most NEISS papers that fail do so for reasons visible before extraction. The question had already been published, the variables did not exist across the years required, or the analytic sample could not support the estimate once subgroups were cut. Editverse works on registry and real-world data studies of exactly this shape.

If you already hold your extract and need only the survey-design analysis run and documented properly, the Just Data Service covers analysis without manuscript work. Everything in this category is listed under Registry and Real-World Data Studies.

Frequently Asked Questions About NEISS Research Methods

Does NEISS use ICD-10 codes?

No. NEISS uses closed internal code sets maintained by CPSC, comprising a 26-value body part code, a 30-value diagnosis code, four-digit product codes with up to three per case, plus disposition and location codes and a 400-character narrative. Every permitted value is listed in the CPSC Data Dictionary. NEDS, distributed through AHRQ’s HCUP programme, is the ICD-10 database. A design built on ICD-10 case definitions cannot be executed in NEISS.

What happens if I analyse NEISS without the survey weights?

You produce a count of records at roughly 100 hospitals, not a national number, and because each hospital’s weight is the inverse of its selection probability within its stratum, that count is not a proportional scaling of the national estimate. Applying the weight but omitting Stratum and PSU gives standard errors that are usually too small. Durand and colleagues, using NEISS as their test case, found design-corrected approaches contrasted profoundly with standard techniques in almost all cases.

When is a NEISS estimate too unstable to publish?

CPSC applies three thresholds and does not report an estimate that fails any of them. The estimate is below 1,200, fewer than 20 cases contribute to it, or the coefficient of variation exceeds 33%. Adopting the same rule gives you an externally sourced justification for collapsing or suppressing a subgroup, which is far stronger than an arbitrary internal cut-off.

Can NEISS give me state-level or county-level estimates?

No, and this is the most frequently reported limitation in the literature. In a scoping review of 167 articles using NEISS-AIP over 2001 to 2021, 38 named the national-estimates-only constraint as their principal limitation. Under the legacy design, 14 states had no participating hospital. NEISS-R extends coverage to all 50 states from 2027, but that does not make historical files capable of subnational estimates.

How do I handle the three product code fields?

Decide whether your case definition searches Product_1 only, all three fields, or all three with competing products excluded, and state the choice in the methods section. Product_3 exists only from 2019 onward, and because one injury may be coded to three product groups, CPSC’s published category estimates cannot be summed. The same applies to any denominator you construct yourself.

Is NEISS changing?

Yes. CPSC announced in 2026 a modernization programme, NEISS-Remodel, replacing manual medical-record abstraction at roughly 70 hospitals with standardized electronic data exchange and nationwide coverage, fully effective by the beginning of 2027. State which NEISS vintage you analysed, and treat any trend spanning the transition with a sensitivity analysis.

Conclusion

A NEISS analysis is defensible when four things are true. The case definition is written as executable logic and validated against narratives rather than assumed from code labels. The estimate is produced with Weight applied and Stratum and PSU declared, reported with its unweighted case count and a confidence interval. Estimates failing CPSC’s stability thresholds are suppressed. And any multi-year comparison has been checked against the Product Code Comparability Table, the 2019 and 2021 variable additions, and movement in the reporting base.

None of that is hard, all of it is invisible if you skip it, and all of it is the first thing a methods reviewer looks for. The starting point for anyone still deciding is the comparison of what each public health database can and cannot answer.

References

1. Durand WM, Stey PC, Chen ES, Sarkar IN. Trend analysis of aggregate outcomes in complex health survey data. AMIA Jt Summits Transl Sci Proc. 2018;2017:349-58. PMID: 29888093.
2. Thompson MC, Wheeler KK, Shi J, Smith GA, Xiang H. An evaluation of comparability between NEISS and ICD-9-CM injury coding. PLoS One. 2014;9(3):e92052. doi:10.1371/journal.pone.0092052. PMID: 24658100.
3. Thompson MC, Wheeler KK, Shi J, Smith GA, Groner JI, Haley KJ, Xiang H. Surveillance of paediatric traumatic brain injuries using the NEISS: choosing an appropriate case definition. Brain Inj. 2014;28(4):431-7. doi:10.3109/02699052.2014.887146. PMID: 24564802.
4. Navon L, Chen LH, Cowhig M, Wolkin AF. Two decades of nonfatal injury data: a scoping review of the National Electronic Injury Surveillance System-All Injury Program, 2001-2021. Inj Epidemiol. 2023;10(1):44. doi:10.1186/s40621-023-00455-4. PMID: 37679835.
5. Carpenito T, Miller M, Manjourides J, Azrael D. Using multiple imputation by super learning to assign intent to nonfatal firearm injuries. Prev Med. 2022;163:107183. doi:10.1016/j.ypmed.2022.107183. PMID: 35964778.
6. Parker ST. Estimating nonfatal gunshot injury locations with natural language processing and machine learning models. JAMA Netw Open. 2020;3(10):e2020664. doi:10.1001/jamanetworkopen.2020.20664. PMID: 33052403.
7. Omaki E, Shields W, Rouhizadeh M, Delgado-Barroso P, Stefanos R, Gielen A. Understanding the circumstances of paediatric fall injuries: a machine learning analysis of NEISS narratives. Inj Prev. 2023;29(5):384-8. doi:10.1136/ip-2023-044858. PMID: 37399309.
8. Chen W, Wheeler KK, Lin S, Huang Y, Xiang H. Computerized “Learn-As-You-Go” classification of traumatic brain injuries using NEISS narrative data. Accid Anal Prev. 2016;89:111-7. doi:10.1016/j.aap.2016.01.012. PMID: 26851618.
9. Shields W, Dong Y, Jager L, Shiang E, Frattaroli S, Omaki E. Using the NEISS database to understand pressure cooker related injuries in the USA. Inj Prev. 2023;29(6):506-10. doi:10.1136/ip-2023-044910. PMID: 37666516.
10. Lacher CR, Koc I, Tsui JC. Pickleball-related ocular injuries among patients presenting to emergency departments. JAMA Ophthalmol. 2025;143(11):938-44. doi:10.1001/jamaophthalmol.2025.3577. PMID: 41100112.
11. Blakney AM, Griggs E, Farrell S, Lindsey K, Chang JT. Electronic nicotine delivery system (ENDS) battery-related explosions and burns presenting to United States emergency departments, 2020-2024. Prev Med. 2026;207:108563. doi:10.1016/j.ypmed.2026.108563. PMID: 41875938.
12. Budnitz DS, Shehab N, Lovegrove MC, Geller AI, Lind JN, Pollock DA. US emergency department visits attributed to medication harms, 2017-2019. JAMA. 2021;326(13):1299-1309. doi:10.1001/jama.2021.13844. PMID: 34609453.
13. Summers AR, Cheema AN, Pirruccio K, Kazmers NH, Gray BL. Epidemiology of pediatric nonballistic firearm injuries to the upper extremity in the United States from 2000 to 2017. Hand (N Y). 2022;17(2):293-7. doi:10.1177/1558944720926626. PMID: 32452229.
14. Ehlman DC, Haileyesus T, Lee R, Ballesteros MF, Yard E. Evaluation of the National Electronic Injury Surveillance System – All Injury Program’s self-directed violence data, United States, 2018. J Safety Res. 2021;76:327-31. doi:10.1016/j.jsr.2020.12.002. PMID: 33653565.
15. Kennedy EJ, Hendricks KJ, Casey M. Sharps injury rates reported among US workers: National Electronic Injury Surveillance System-Occupational Supplement 2006 to 2020. J Occup Environ Med. 2023;65(6):495-501. doi:10.1097/JOM.0000000000002816. PMID: 36806612.
16. Hendricks KJ, Layne LA, Schleiff PL, Javurek ABR. Surveillance of acute nonfatal occupational inhalation injuries treated in US hospital emergency departments, 2014-2017. Am J Ind Med. 2022;65(8):690-6. doi:10.1002/ajim.23400. PMID: 35633303.
17. Neuroth LM, Humphries KD, Wing JJ, Smith GA, Zhu M. Motor vehicle-related electric scooter injuries in the US: a descriptive analysis of NEISS data. Am J Emerg Med. 2022;55:1-5. doi:10.1016/j.ajem.2022.02.004. PMID: 35228017.
18. Dell NA, Vaughn MG, Salas-Wright CP. Firearm injury among people experiencing homelessness: cross-sectional evidence from a national survey of United States emergency departments. Public Health Pract (Oxf). 2023;6:100446. doi:10.1016/j.puhip.2023.100446. PMID: 37954558.
19. Schroeder T, Ault K. The NEISS Sample (Design and Implementation) From 1979 to 1996. Bethesda (MD): US Consumer Product Safety Commission; 2001.
20. Schroeder T. Consumer Product-Related Injuries and Deaths in the United States: Estimated Injuries Occurring in 2020 and Estimated Deaths Occurring in 2019. Bethesda (MD): US Consumer Product Safety Commission, Directorate for Epidemiology; 2021.