Published
The National Trauma Data Bank and its successor, the American College of Surgeons Trauma Quality Programs participant use file, now draw on more than 900 participating trauma centres across the United States. Scale gets treated as a substitute for rigour more often than it should. A 2018 systematic review of 539 trauma registry manuscripts published between January 2015 and March 2017 found that 38.6% never mentioned whether their data were complete, and a further 16.3% noted missing data without describing how they handled it. What follows sets out what NTDB and TQP data can support, where the design breaks down, and the analytic decisions, on injury severity coding, physiologic missingness, and facility clustering, that separate a publishable trauma registry study from a desk rejection.
What NTDB and TQP Actually Are
The American College of Surgeons established the National Trauma Data Bank in 1989, built from the trauma registries that participating centres already kept for their own quality programmes. Growth was steady rather than sudden. One million records had accumulated by 2004, and the total reached five million by 2014. In 2017 the ACS folded NTDB into the broader Trauma Quality Programs framework, and the participant use file now released each year under TQP is the direct descendant of the original NTDB extract. Researchers who cite “NTDB” today are usually working with a TQP-branded file built on the same submission pipeline, and the two names appear interchangeably across the published literature for exactly that reason.
Participation is voluntary but concentrated. Level I and Level II centres, verified by the ACS Committee on Trauma to a defined standard of resources and case volume, contribute the bulk of records, and TQIP benchmarking reports are built around that verified subset. Level III and IV centres participate at lower and more variable rates, which matters directly for anyone trying to generalise a TQP finding to trauma care outside the largest referral centres.
When TQP Data Are the Right Instrument
TQP data answer questions about in-hospital outcomes across many centres at a scale no single institution reaches on its own, mortality and complication rates by injury pattern, variation in resuscitation practice, disparities in trauma care by insurance status or geography, and the epidemiology of injury mechanisms too uncommon for a single centre to study with any precision. Benchmarking a centre’s risk-adjusted mortality against the national participant pool, using observed-to-expected ratios with appropriate shrinkage, is one of the uses the design was built for.
Anything that needs a population denominator is the wrong use for TQP. TQP is a convenience sample of centres that chose to participate, not a population-based registry, so it cannot support an incidence rate, and reviewers who work in trauma will say so on sight. Patients who died before reaching a participating centre, or who were treated entirely outside the network, are simply absent from the file, and the set of contributing hospitals changes from year to year, which turns naive multi-year trends into composition effects unless that shift is tested for directly. Follow-up after discharge is also absent. Everything stops at the hospital door, so anything about long-term function, readmission, or late mortality needs a different data source entirely. A similar convenience-sample limitation applies to public injury surveillance more broadly, and we cover the general version of the problem in our analysis of whether NEISS injury statistics are reliable.
Injury Severity Is Derived, Not Measured
Nothing in a trauma registry hands the analyst an injury severity score directly. ISS and NISS are computed from ICD-based diagnosis codes mapped to Abbreviated Injury Scale severities, and that mapping step is where a surprising share of published errors originate. Different AIS dictionary versions assign different severities to the same diagnosis code, so two studies drawing on the same underlying admissions can report materially different ISS distributions purely from a mapping choice neither paper states. A defensible trauma registry study names the mapping version it used and, where the study period crosses a version change, tests whether the choice affects the conclusion rather than assuming it does not.
Whether early deaths are included compounds the problem. Patients who die within minutes of arrival, before a full physiologic assessment is possible, get excluded from some analyses and retained in others, and each choice changes the denominator for exactly the sickest patients in the cohort. Neither choice is wrong on its own. Reporting the choice and testing its effect on the headline result is what a reviewer is checking for.
Physiologic Missingness Is the Central Problem
Emergency department GCS, systolic blood pressure, and respiratory rate carry the highest missingness of any variables routinely used for trauma risk adjustment, and the mechanism is not random. A registry analysis of 2,520 cases from the Victorian State Trauma Registry found that respiratory rate, GCS, and systolic blood pressure, together with an intubation qualifier flag, carried the greatest proportion of missing observations of any variables needed for TRISS-based analysis, and that logistic regression on predictors of missingness identified a clear clinical pattern behind the gaps rather than haphazard non-response. Patients who arrive intubated, sedated, or in extremis are the ones a paramedic or triage nurse cannot fully assess, and they are disproportionately the patients who go on to die. Analysing only the complete records therefore does not just shrink the sample. Both the numerator and the denominator lose their sickest patients as a result, and any mortality rate computed on what remains is biased toward a healthier cohort than the one that was actually treated.
Neither the mechanism nor the direction of that bias is specific to one registry. A Monte Carlo simulation built on 63,020 NTDB patients across 68 hospitals showed that the statistical method chosen for handling missing GCS motor-component data changes which hospitals get flagged as quality outliers, comparing multiple imputation against complete-case analysis and against dropping the affected predictor entirely, benchmarked against the true, fully observed data. A separate multiple-imputation model built specifically for NTDB physiologic variables, and an out-of-hospital validation study that compared imputed values directly against known paramedic-recorded values, both reached the same conclusion from different registries. Complete-case analysis of physiologic trauma variables is not a neutral default. It runs in a predictable, testable direction, and that direction is toward understating how sick the cohort actually was.
We built an independent simulation, validated against two statistical engines, to put a number on that bias rather than assert it. See how much complete-case analysis actually biases trauma mortality and risk adjustment for the full result.
Facility Clustering Cannot Be Ignored
Patients treated at the same trauma centre are not statistically independent of one another. They share a surgical team, a resuscitation protocol, and an institutional case mix, and pooled logistic regression that treats every patient as an independent observation understates the uncertainty around any facility-level comparison. Mixed-effects models with a facility random intercept, or generalised estimating equations with robust standard errors, are the accepted fix, and the intraclass correlation from either approach is worth reporting on its own, since a low ICC is itself evidence that centre-level variation is not driving the result. Verification level and centre volume are confounders here as much as clustering terms, since Level I, II, and III centres treat structurally different case mixes, and a comparison that does not account for verification status is comparing populations that were never alike to begin with.
Transfers and Double-Counting
A meaningful share of severely injured patients arrive at a participating centre after transfer from another hospital, and their TQP record can reflect either institution depending on submission practice. Left unaddressed, this both double-counts some patients across the pooled file and mixes outcomes that belong to two different episodes of care into one record. A stated, tested transfer rule, applied consistently and checked against the alternative, is a required step rather than an optional refinement, and it is one of the more common gaps reviewers flag in submitted trauma registry manuscripts.
In Simple Words
NTDB and TQP are large, multi-centre trauma databases built from hospitals that volunteer their data, not a national census of every injured patient. They are strong for comparing outcomes across many centres and weak for anything needing a true population rate. Treating patients with missing vital signs as if they were simply absent from the study is the single biggest source of avoidable error, since in trauma registries the patients most likely to be missing a GCS or blood pressure reading are the ones who were too unstable to fully assess, and often the ones most likely to die.
NTDB/TQP Compared With the Other Major Registries
| Feature | NTDB / TQP | SEER | NHANES |
|---|---|---|---|
| Sample type | Convenience sample of participating centres | Population-based registry, ~half of US | Complex probability survey sample |
| Follow-up after the index event | None, ends at discharge | Long-term survival and cause of death | Linked mortality only, via NDI |
| Signature missing-data problem | Physiologic variables missing informatively by severity | Unknown stage and cause of death miscoding | Item non-response in interview and MEC components |
| Can it estimate a national rate | No, denominator is unknown | Yes, within covered registries | Yes, with survey weights applied |
Getting Access to a TQP Participant Use File
Access runs through the applicant’s institution rather than directly through an individual researcher. The American College of Surgeons requires a signed data use agreement that names the receiving institution, restricts who at that institution may handle the file, and sets terms on how findings can be shared and published. External analytic support, a statistician, a data-management contractor, or a firm like Editverse working from a client’s own extract, is generally permitted, but only where the institution’s agreement explicitly allows it, and confirming that before any analysis begins is the applicant’s responsibility, not an assumption to make. Turnaround for the application itself varies by year and by how completely the initial request is specified, so building in lead time before a target submission date is worth doing early rather than discovering the gap after a manuscript deadline is already fixed.
What a Reviewer Checks First in a TQP Manuscript
Submitted trauma registry manuscripts tend to fail review for a small, repeatable set of reasons, and most of them surface before a reviewer reads past the methods section. Four questions come up often enough to treat as a pre-submission checklist rather than an afterthought.
Does the methods section name the exact participant use file version and year? “NTDB data” without a version number tells a reviewer nothing about which submission cycle, coding standard, or set of participating centres actually produced the results, and it is one of the fastest ways to draw a methods-revision request.
Is the missingness pattern in the physiologic variables addressed at all? A manuscript that reports GCS, systolic blood pressure, or respiratory rate as covariates without stating how missing values were handled, or with a silent complete-case restriction buried in a flow diagram, reads to an experienced reviewer as either inexperience or an unexamined bias.
Is the case mix comparable across the centres or years being compared? Level I, II, and III centres treat different populations by design, and a comparison that pools across verification levels without adjusting for it, or without at least discussing why pooling is defensible, invites the obvious rebuttal.
Does the transfer-patient rule get stated anywhere? Silence here is common enough that reviewers now ask for it directly when a manuscript omits it, since an undefined transfer rule leaves open whether patients or outcomes were counted twice.
None of these four questions require sophisticated statistics to answer. Answering them clearly, in the methods section, before a reviewer has to ask, is what separates a first-round accept from a major revision.
Common Questions
Can NTDB or TQP data support a national trauma incidence estimate? No. A convenience sample with an unknown and shifting denominator cannot answer that question, and any incidence claim built on it will not survive review from anyone who works in trauma epidemiology.
Is a single year of TQP data enough for a rare mechanism? Sometimes, but pooling years raises the composition-effect problem above, since the hospital set contributing to the file changes year to year. Test for it before pooling, do not assume it away.
Does multiple imputation fix the missing-physiology problem completely? Most of the bias goes away when the outcome is included in the imputation model, but the fix is a mitigation rather than a cure, and a complete-case sensitivity analysis alongside it is still expected by reviewers.
What is the difference between NTDB and TQP in a methods section? For most purposes, none that matters. State the actual participant use file version and year you used, since that is what a reviewer or replicator needs, rather than debating which of the two names is technically correct.
Free Tools for Trauma Registry Work
- STROBE reporting checklist tutorial, for the reporting standard every TQP manuscript is checked against
- Six techniques for cleaning clinical registry data, for the data preparation steps that come before any model
- Levels of evidence in clinical research, for placing a registry finding correctly against a trial or a meta-analysis
Professional Services
Editverse builds risk-adjusted NTDB and TQP studies to the standard set out above, informative missingness handled with multiple imputation and a complete-case sensitivity analysis, facility clustering modelled rather than ignored, and injury severity remapped rather than trusted from the registry field.
Router
$450
Feasibility and Protocol Assessment. Checks whether your question, your ACS data-use agreement, and your target journal actually fit before you commit to a full study. Credited in full against any study you commission.
Fixed Price
$2,400
NTDB / TQP Trauma Registry Study. ICD-to-AIS remapping, informative-missingness handling with multiple imputation, facility-clustered risk adjustment, and a completed STROBE checklist. 15 business days from data receipt.
Quoted
From $3,500
Extended and Multi-Database Registry Study. For a TQP question linked to another database, a prediction model needing external validation, or a programme of related trauma papers. Fixed quote before any work begins.
Conclusion
NTDB and TQP data reward analysts who treat the file’s own construction as part of the methods section, not an afterthought before the results. Participating centres are volunteers, injury severity fields are derived rather than measured, and the physiologic variables most central to risk adjustment go missing in a pattern that tracks how sick a patient was on arrival. None of that makes the data unusable. Mapping version, missingness mechanism, and facility clustering, checked and stated rather than assumed, are what separate a manuscript that clears review from one that does not.
References
- Shivasabesan G, Mitra B, O’Reilly GM. Missing data in trauma registries: A systematic review. Injury. 2018;49(9):1641-1647. doi: 10.1016/j.injury.2018.03.035
- Glance LG, Osler TM, Mukamel DB, Meredith W, Dick AW. Impact of statistical approaches for handling missing data on trauma center quality. Annals of Surgery. 2009;249(1):143-148. doi: 10.1097/SLA.0b013e31818e544b
- O’Reilly GM, Cameron PA, Jolley DJ. Which patients have missing data? An analysis of missingness in a trauma registry. Injury. 2013;43(11):1917-1923. doi: 10.1016/j.injury.2012.07.185
- Moore L, Hanley JA, Turgeon AF, Lavoie A, Emond M. A multiple imputation model for imputing missing physiologic data in the national trauma data bank. Journal of the American College of Surgeons. 2009;209(5):572-579. doi: 10.1016/j.jamcollsurg.2009.07.004
- Newgard CD. The validity of using multiple imputation for missing out-of-hospital data in a state trauma registry. Academic Emergency Medicine. 2006;13(3):314-324. doi: 10.1197/j.aem.2005.09.011
- O’Reilly GM, Jolley DJ, Cameron PA, Gabbe B. Missing in action: a case study of the application of methods for dealing with missing data to trauma system benchmarking. Academic Emergency Medicine. 2011;17(10):1122-1129. doi: 10.1111/j.1553-2712.2010.00887.x
