https://orcid.org/0000-0001-8532-3500
Figures
Abstract
Background
Major trauma is a leading cause of death and disability. Specialised care in major trauma centres has been associated with improved outcomes and prehospital triage tools are used to ensure injured patients are treated in the right place and the right time. However, there is a trade-off between under- and over-triage, and this study aimed to externally validate current and newly developed major trauma triage tools.
Methods
A diagnostic case-cohort study was performed between November 2019 and February 2020 in 4 English regional trauma networks as part of the Major Trauma Triage Study (MATTS). The accuracy of 22 adult major trauma triage tools, including 3 newly developed MATTS tools was evaluated. Consecutive patients with acute non-trivial injury presenting to participating ambulance services were included and matched to data from the English national major trauma database. Theoretical accuracy was examined, with index tests assessed according to objective ambulance service data, regardless of the final triage decision or hospital destination. The primary reference standard was a consensus definition of serious injury that would benefit from expedited major trauma centre care.
Results
The case-cohort sample consisted of 2,607 patients, including 928 primary reference standard positive patients. The population weighted prevalence of major trauma meeting the primary reference standard definition was 3.1% (95% CI 2.3–4.0). Four optimally performing triage tools were identified with Pareto decision analysis: the Trauma score (sensitivity 0.1, specificity 0.99), MATTS specific tool (sensitivity 0.37, specificity 0.95), MATTS balanced tool (sensitivity 0.58, specificity 0.87), and the MATTS sensitive tool (sensitivity 0.72, specificity 0.76). This finding was unchanged in subgroup analyses of different age-groups and injury mechanisms; secondary analyses examining alternative reference standards (ISS ≥ 16, US consensus definition); and sensitivity analyses exploring missing data.
Conclusions
Four optimal triage tools, demonstrating a trade-off between sensitivity and specificity, were identified by this validation study. The choice of ideal tool will depend on prevalence of major trauma, and valuation of false positive and false negative cases. Further prospective investigation of real-life triage tool performance, including compliance and clinical judgment, is necessary.
Citation: Fuller GW, Baird J, Keating S, Miller J, Pilbery R, Scotney N, et al. (2026) Major Trauma Triage Study (MATTS): Diagnostic accuracy of major trauma triage tools in English regional trauma networks – A case-cohort study. PLoS One 21(3): e0344996. https://doi.org/10.1371/journal.pone.0344996
Editor: Rayan Jafnan Alharbi, Jazan University College of Applied Medical Science, SAUDI ARABIA
Received: April 18, 2024; Accepted: February 28, 2026; Published: March 27, 2026
Copyright: © 2026 Fuller et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: The data contains potentially sensitive patient information and are subject to data sharing agreements with West Midlands Ambulance Service NHS Foundation Trust (WMAS), Yorkshire Ambulance Service NHS Trust (YAS) and the Trauma Audit and Research Network (TARN). The anonymised data and full reproducible analysis code is available on reasonable request from the University Sheffield data access committee (contact: a.loban@sheffield.ac.uk [Head of data management]).
Funding: The study was funded by the National Institute of Health Research Health Technology Agency Assessment Programme (NIHR HTA ref: 17/16/04).
Competing interests: The authors have declared that no competing interests exist.
Introduction
Major trauma is an important public health issue. There are 20,000 cases annually in England, accounting for 5,400 deaths, 8,000 disabilities, and an estimated £400 million of health care costs [1]. The epidemiology of major trauma has evolved over recent decades, and it is now recognised to comprise two distinct disease entities: higher energy trauma in younger patients, and an increasing burden of low energy trauma in older patients (the ‘silver tsunami’) [2].
Major trauma care in England, and other developed countries, is organised into trauma networks, aiming to transport patients injured within a defined geographical area to a hospital matched to their clinical need [3]. Trauma networks consist of a specialist major trauma centre (MTC) offering advanced specialist care for the most seriously injured patients, and several lower-level hospitals capable of caring for lower-acuity injuries. Specialist care in MTCs has been associated with improved outcomes, although the benefit in older patients is less certain [4,5]. Major trauma triage tools are used to identify appropriate patients for direct transfer to MTCs, potentially bypassing other hospitals, and to inform pre-alert calls to allow activation of MTC trauma teams for expedited resuscitation and assessment [6]. Further details on English major trauma services and triage are provided in the supplementary materials (S1 File).
A wide range of triage tools are available, varying in structure and content [7]. As variables and cut-off points are altered, there is a trade-off in accuracy, with specificity falling as sensitivity increases. The optimal compromise depends on the prevalence of major trauma in injured patients presenting to ambulance services, and the relative valuation of false positives and negatives. The American College of Surgeons Committee on Trauma have recommended a target sensitivity of 95% and specificity of 65–75% [8]. However, multiple economic evaluations have consistently demonstrated that high specificity should be prioritised for cost-effective triage [9–12].
A recent systematic review of triage tool performance found poor quality studies and varying accuracy, with sensitivity ranging from 10% to 100%, and specificity from 9% to 100% [7]. The Major Trauma Triage Tool Study (MATTS) was funded by the National institute of Health Research to develop a new triage tool for use in the English National Health Service (NHS), validate its performance against existing triage tools, and evaluate its performance after implementation [13]. The current study reports validation of existing triage tools and three new MATTS triage tools. Specific objectives were to describe the characteristics of injured patients presenting to English ambulance services, determine the diagnostic accuracy of selected adult triage tools (including newly developed MATTS triage tools) and to evaluate their net benefit compared to default strategies of bypassing/pre-alerting all or no patients.
Methods
Study design
A diagnostic case-cohort study was undertaken to evaluate the accuracy of representative adult major trauma triage tools, and externally validate newly developed MATTS expert consensus triage tools. Case-cohort studies are recommended for evaluation of diagnostic accuracy in low prevalence scenarios [14]. In this study design, a random sample of individuals from the study population are included irrespective of reference standard (‘sub-cohort’), together with all reference standard positive patients (‘cases’) [14,15].
Setting
The study was undertaken in four individual English inclusive regional trauma networks: Birmingham, West Yorkshire, North West London, and Severn. These are predominantly served by four separate NHS ambulance services: West Midlands Ambulance Service University NHS Foundation Trust (WMAS); Yorkshire Ambulance Service NHS Trust (YAS); London Ambulance Service NHS Trust (LAS); and the South-Western Ambulance Service NHS Foundation Trust (SWAS) respectively. The study trauma networks were chosen as they are principally served by a single ambulance service, have a favourable record for data collection, and represent a diverse range of localities, demographic, socioeconomic and injury profiles.
Reference standards
The primary reference standard against which triage tool outcomes were assessed was injured patients who would benefit from expedited MTC care, as characterised by a previously published MATTS consensus-based definition [16]. Several alternative reference standards were considered in secondary analyses, including injury severity score (ISS) ≥16 [17], urgent trauma interventions, and the MATTS reference standard excluding open fractures. Urgent interventions comprised advanced airway interventions; transfusion of >4 units blood products; interventional radiology and/or emergency surgery occurring within 8 hours of emergency department (ED) arrival; ‘hot’ secondary transfer to the MTC within 6 hours, or critical care admission occurring within 12 hours of initial ED presentation.
Index tests
The index tests under consideration were three newly developed MATTS expert consensus triage tools and 19 representative international and UK adult major trauma triage tools published prior to 2019, as detailed in Table 1. Development of the MATTS triage tools has been reported in detail elsewhere and constituent variables are presented in the supplementary materials (S2 Table) [18]. Theoretical diagnostic accuracy was examined, with index tests assessed according to objective data present in ambulance service records, regardless of the final triage decision or hospital destination. Discretionary variables were excluded. Prediction models and clinical scores were evaluated at their recommended published threshold. For assessment of the secondary MATTS reference standard excluding open fractures, triage tool variables for open fractures were omitted.
Study population
The study sample was defined from the viewpoint of the ambulance service clinician assessing undifferentiated patients following injury where a major trauma triage tool would plausibly be used, thus providing the true denominator for assessment of diagnostic accuracy. The source population was all patients presenting with acute non-trivial injury to the four participating ambulance services and included trauma networks. The subsequent study population consisted of consecutive adult patients, aged over 16 years, conveyed to a participating trauma network hospital, between 1st November 2019 and 28th February 2020 and meeting study inclusion and exclusion criteria as detailed in Table 2. The final study sample included patients with complete data available allowing calculation of triage tool diagnostic accuracy.
Case-cohort derivation
Sub-cohort identification.
Consecutive patients presenting with injury were identified from YAS, WMAS, LAS and SWAST patient records. Working impression diagnostic codes corresponding to non-trivial injury, use of a trauma-specific intervention, or major trauma related pre-alerts were selected for inclusion. Ambulance service databases were searched for any cases meeting these criteria who were injured in the defined study area and transported directly to a participating hospital to give the parent cohort. A random sample of patients were then included in a sub-cohort.
Reference standard positive case identification.
The presence of major trauma in patients conveyed to participating hospital by ambulance following injury was determined using Trauma Audit and Research Network (TARN), the English national major trauma registry, data [19,20]. Reference standard criteria were applied to all cases submitted to TARN and positive cases then linked to ambulance service data. A census sample of all cases conveyed to a selected trauma network hospital during the study period by a participating ambulance service, meeting ambulance service inclusion criteria and reference standard definitions, and matching to an ambulance service case were included. Any included ambulance service cases without a corresponding TARN submission were assumed to be reference standard negative.
Data linkage between a) cases meeting reference standard criteria collected by TARN, and b) ambulance identified patients with non-trivial injuries was conducted deterministically based on a unique ambulance service patient report form number shared across both datasets. In cases where exact deterministic matching was not possible due to missing or inaccurate patient report form number, research paramedics performed probabilistic matching by manually reviewing each reference standard positive case. Demographic (age, sex), non-unique ambulance identifiers (computer aided dispatch numbers, ambulance call sign) and incident information (incident date/time, hospital destination, hospital arrival date/time, incident description, incident location (outward and partial inward postcode)) was used from TARN data to search for a corresponding record in ambulance service databases. All matches were independently confirmed by a second researcher and a match was not confirmed in the presence of any uncertainty or disagreement.
Data collection
Index test data.
Ambulance service records for patients sampled in the sub-cohort, or matched to non-sampled reference standard positive cases, were imported into a bespoke research database. Collected information comprised demographic, patient characteristics, physiology, incident, mechanism of injury, interventions, treatments, and clinical assessment information. Relevant electronic closed field data were imported directly where possible, with free text data coded by hand after review by research paramedics. Eligibility against inclusion and exclusion criteria was confirmed manually. Data abstraction was blinded, with all ambulance service data anonymised and reference standard status not available. A data dictionary was developed a priori to standardise all data elements, provide variable definitions, and define categorisations. Data collection was piloted, and weekly meetings were convened to review data collection, with any uncertainties resolved through consensus. Data was recorded as missing if not present in closed fields, or if not possible to infer from free text fields. Triage tools and prediction models selected as index tests were then coded algorithmically according to their stated variable thresholds against the observed data. The first recorded vitals sign was used for physiology variables. Where triage tool variables were defined as sustained physiology values, two or more consecutive values meeting the threshold were required. Two independent statisticians undertook coding to ensure accuracy.
Reference standard data.
TARN is the English national trauma registry database and collects information on patients with major injuries presenting to all trauma receiving hospitals in England. TARN data collection has been described in detail previously [21]. Briefly, patients are included in the TARN database if they sustain injury resulting in any of: hospital admission for >72 hours; critical care or high dependency unit admission; transfer for specialist care; or death. Patients with simple isolated closed injuries (e.g., radius fracture), pubic rami or femoral neck fractures and aged over 65, pre-hospital deaths, or diagnosed as dead on emergency department arrival with no further management instigated, are excluded. TARN data collectors in member hospitals collate demographic, injury, treatment, investigation, and outcome data for eligible patients from all relevant patient records. Independent, trained, TARN injury coders centrally grade individual injuries for each case according to the Abbreviated Injury Scale (AIS 2008 revision) criteria based on clinical, radiological and post mortem information. From the AIS severity scores a subsequent ISS is assigned [22].
Each submitted TARN case from participating hospitals during the study period was coded centrally by TARN data analysts according to primary and secondary reference standard criteria. Anonymised data for reference standard positive cases were then imported into a study database for review by research paramedics. Eligibility against inclusion and exclusion criteria was confirmed manually.
Statistical analyses
The analysis proceeded in 5 stages. Firstly, the derivation of study population, parent cohort, sub-cohort, and reference standard positive cases were enumerated and delineated graphically using flow charts. Secondly, the study sample was characterised, with patient demographics, injury features and missing data examined using descriptive statistics. Thirdly, in the main analysis, the diagnostic accuracy of individual adult triage tools was investigated against primary and secondary reference standards. Sensitivity, specificity, positive and negative predictive values, and likelihood ratios were calculated with their 95% confidence intervals. Results were displayed graphically using diagnostic forest plots and plots of individual triage tools paired sensitivity/specificity. Separate sensitivity/specificity plots were constructed to present the Pareto front (or receiver operating characteristic curve convex hull), including only non-dominated triage tools with higher specificity for any given sensitivity [23–25]. The Pareto front was formed by linear interpolation between adjacent points with no point lying above the final curve. A decision curve analysis was then performed to illustrate the clinical consequence of implementing Pareto optimal triage tools [26]. Clinical net benefit of the identified tools was compared to default strategies of bypassing/pre-alerting all or no patients. Net benefit was calculated across a range of threshold probabilities, defined as the minimum probability of major trauma at which bypass/pre-alerting would be warranted, calculated as net benefit = sensitivity × prevalence – (1 – specificity) × (1 – prevalence) × the odds at the threshold probability [27].
Fourthly, subgroup analyses were conducted examining triage tool performance against the primary reference standard for adults (aged 16–64 years) and elderly patients (aged 65 years or older), and for cases with penetrating versus blunt injury, and higher energy injury presentations versus ground level falls. These were chosen as clinically relevant subgroups where previous studies have reported heterogenous triage tool performance.
Fifthly, sensitivity analyses were performed to explore the influence of study selection criteria and missing data. The main analysis was repeated, including primary reference standard positive cases involved in acute non-trivial injury events, but where a non-selected working impression code had been used. The main analysis was also repeated including eligible patients with missing data under a range of missing data mechanism assumptions. To indicate the range of possible results, best- and worst-case scenario analyses were conducted under a missing not at random assumption, where eligible patients with missing data were classified either as triage tool positive, or triage tool negative respectively. Multiple imputation using chained equations was also performed to explore the impact of data being missing at random [28].
Analyses were conducted in R Statistical Software (v4.3.0; R Core Team 2023) and STATA version 17.0 (StataCorp. 2016. Stata Statistical Software: Release 17. College Station, TX: StataCorp LP). Multiple imputation was performed using the ice procedure and the mim command was subsequently used to calculate triage rule sensitivity and specificity, combining results from imputed data sets according to Rubin’s rules. Unweighted summary statistics were reported separately for the sub-cohort and case-cohort sample characteristics and reference standard prevalence. Diagnostic accuracy metrics independent of underlying prevalence (sensitivity, specificity, likelihood ratios) were also calculated without sample weights. Conversely, Borgan II inverse probability sampling weights were used for positive and negative predictive values to represent non-sampled non-trivially injured patients, and account for the case-cohort study design [29]. The unit of analysis was the individual incident. Direct patient identifiers were unavailable, and it was therefore not possible to account for clustering from in the event recurrent incidents in the same patient.
Sample size calculation
An estimated 600 reference standard positive patients were expected across the four participating ambulance service during the four-month study period. Assuming, based on ACS-COT (American College of Surgeons Committee on Trauma) criteria,[19] a triage tool required sensitivity of 95%, and anticipating that up to 5% of cases would have insufficient data, a sample of 570 cases would result in a 95% confidence interval (CI) width of ±2.0%. According to ACS-COT criteria a triage tool requires a specificity of 70%. [19] Therefore, to achieve a 95% CI width of ±2.0% a sample 2,014 reference standard negative patients would be required. This sample size would provide a 95% CI of ±5.0% and ±2% for under-triage rate in paediatric and elderly subgroups respectively.
Ethical approval and data governance
This study was undertaken as part of the Major Trauma Triage Study (MATTS) project, aiming to develop a new national triage tool [13]. Ethical approval was provided by Yorkshire and The Humber – Bradford Leeds Research Ethics Committee (Reference: 19/YH/0197). Approval to use anonymised patient data without direct patient consent was confirmed by the Health Research Authority Confidentiality Advisory Group in accordance with Regulation 5 of the English Health Service (Control of Patient Information) Regulations 2002 (Reference: 19/CAG/0119). A study protocol and statistical analysis plan were pre-specified.
Results
Sample derivation and characteristics
Between 1st November 2019 and 28th February 2020, 47,513 patients with non-trivial injury working impression codes were conveyed to included trauma network hospitals by participating ambulance services, forming the source population. In total, 1,733 eligible adults with complete data were included in the sub-cohort, comprising 54 adults patients meeting primary reference standard criteria, and 1,679 adults who were primary reference standard negative. During the same study period 928 adult TARN cases met inclusion criteria, were successfully matched to a corresponding ambulance service record, and had complete data. Of these, 54 were included in the sub-cohort as described above, and 874 were outside the sub-cohort, but matched to ambulance cases in the parent cohort. The final case-cohort study sample for evaluation of adult triage tools therefore consisted of 2,607 patients (928 primary reference standard positive patients, 1,679 primary reference standard negative patients). Derivation of the parent cohort, sub-cohort, and study samples for primary reference standard positive and negative patients are detailed in Figs 1 and 2.
Area of circles and squares is proportional to the number of patients..
The overall prevalence of major trauma meeting the primary reference standard definition in eligible sub-cohort patients with complete data was 3.1% (n = 54/1,733 95% CI 2.3–4.0). Included patients presenting to ambulance services with non-trivial injury were predominantly elderly (median age 75 years), female (53.5%), and sustained accidental (88.1%) blunt trauma (97.7%) from ground level falls (71.2% of injury mechanisms). Characteristics of the included complete case study sample are detailed in Table 3.
Main analysis
In general, an inverse relationship between specificity and sensitivity for the primary reference standard was observed across triage tools, with an increasing false positive fraction as the proportion of false negatives decreased (Table 4, Fig 3). A Pareto front, with higher specificity for a given sensitivity, of four optimal triage tools was evident, comprising: Trauma score, MATTS specific tool, MATTS balanced tool, and the MATTS sensitive tool (Fig 3). Within this set of triage tools, the sensitivity/specificity trade-off varied between 0.10/0.99, 0.37/0.95, 0.58/0.87, to 0.72/0.76 respectively. The decision curve analysis indicated that the MATTS specific triage tool was the best choice if between 6 and 16 false positives would be accepted for each true positive case meeting the primary reference standard (Fig 4). Relative tool performance was similar when evaluated against secondary reference standards, with the same four triage tools (Trauma score, MATTS specific tool, MATTS balanced tool, MATTS sensitive tool) maximising accuracy and demonstrating Pareto dominance (see supplementary materials S3 Table).
Left panel shows all 22 triage tools. Right panel shows Pareto front of optimal tools, excluding tools with worse accuracy..
Top panel shows full decision curve. Bottom panel shows the frontier of optimal triage tools at different threshold probabilities.
Main analysis findings were not materially changed in sensitivity analyses exploring the influence of study selection criteria and missing data. Including additional primary reference standard positive adult cases involved in acute non-trivial injury events, but where a non-selected working impression code was used, resulted in negligible changes in sensitivity estimates, with specificity remaining unchanged best- and worst-case missing data scenarios, and multiple imputation, resulted in only minor changes in sensitivity/specificity, with the same four triage tools remaining dominant. Further details of sensitivity analysis results are presented in the supplementary materials (S4 Table).
Clear spectrum effects were apparent in triage tool performance across subgroups of different ages or with differing injury characteristics. In elderly patients, adult triage tools demonstrated lower sensitivity and higher specificity for all reference standards. Triage tool accuracy in adults aged between 16 and 64 years demonstrated slightly higher sensitivity and lower specificity across all tools and reference standards compared to the main analysis. Triage tools demonstrated markedly higher sensitivity and lower specificity for the primary MATTS reference standard in penetrating (range: sensitivity = 0.93/specificity = 0.39 to sensitivity = 0.16/specificity = 1.00) compared to blunt injury trauma (range: sensitivity = 0.71/specificity = 0.77 to sensitivity = 0.17/specificity = 0.94). A similar pattern of difference in diagnostic accuracy across triage tools was also apparent in ground level falls (relatively lower sensitivity/higher specificity) compared to higher energy (relatively higher sensitivity/lower specificity) mechanisms of injury. Full details of subgroup analyses are reported in the supplementary materials (S5 Table).
Discussion
Summary
Patients presenting to four English trauma networks by ambulance with non-trivial injury were most commonly elderly females with blunt trauma from ground level falls. The overall prevalence of major trauma was low (3.1%). An inverse relationship between specificity and sensitivity for major trauma was observed across studied triage tools, with four optimally performing triage tools identified: Trauma score (sensitivity 0.1, specificity 0.99), MATTS specific tool (sensitivity 0.37, specificity 0.95), MATTS balanced tool (sensitivity 0.58, specificity 0.87), and the MATTS sensitive tool (sensitivity 0.72, specificity 0.76). Spectrum effects were apparent with higher sensitivity and lower specificity in patients aged 16–65, in penetrating trauma, and higher energy mechanisms of injury.
Interpretation
The theoretical performance of triage tools fell short of recommended ACS-COT targets (sensitivity 95%, specificity 65–75%) [8]. This likely reflects the challenges of assessment of injured patients in the field which suggests that an expectation of high accuracy is unrealistic. Contributory factors could include the heterogeneity of major trauma, lack of point of care tests, limited available information, assessment early in the natural history of the injury, and adverse prehospital environment. Moreover, given the low prevalence of major trauma in injured patients presenting to ambulance services, high specificity is necessary to maintain an acceptable positive predictive value [30]. In fact, economic evaluations of field triage have consistently demonstrated that the ACS-COT targets are not cost-effective, and specificity should be prioritised [10,12].
Together with the Trauma Score, expert consensus developed MATTS triage tools performed the best; with the MATTS specific triage tool (sensitivity 37%, specificity 95%) coming closest to the optimal sensitivity/specificity trade-offs indicated by previous economic evaluations [12]. Overall, the MATTS specific triage tool demonstrated a positive likelihood ratio (LR+) of 7.7 and negative likelihood ratio (LR-) of 0.66. On average, given the observed prevalence of 3.1%, the post-test probability of major trauma would be 20% if the triage tool was positive and 2% if negative. Performance across subgroups varied, with this tool acting as a ‘rule out’ test in penetrating trauma (prevalence 5%, LR+=1.83, LR- = 0.26, post-test probability 9% if positive, 1% if negative); a ‘rule in’ test in elderly patients (pre-test probability 1.9%, LR+=8.04, LR- = 0.77, post-test probability 13% if positive, 1% if negative); and a ‘rule in and rule out’ for high acuity major trauma requiring urgent interventions (pre-test probability 0.9%, LR+=12.33, LR- = 0.38, post-test probability 10% if positive, 0.4% if negative).
It should be noted that the theoretical performance of triage tools was evaluated. This assumed that triage tools were applied to all non-trivially injured patients, were scored according to observed data recorded in ambulance service records, discretionary triage variables (e.g., consider bypass to a MTC in patients taking anticoagulants) were not applied, there was full compliance with the result, and no additional clinical acumen was applied. In reality, there is evidence that triage tools are inconsistently applied, final triage decisions often differ from that indicted by the prehospital findings, and unstructured clinical judgement is commonly used [31]. Such practice may improve triage accuracy, generally improving sensitivity at a slight cost to specificity [32]. The MATTS specific triage tool therefore requires prospective evaluation after implementation into practice to gain a clear picture of real-life performance. It could be possible to develop separate triage tools for different subgroups (e.g., high versus low energy injury mechanisms), but multiple options are likely to represent a barrier to use, supporting a more parsimonious single tool approach.
The four included English trauma networks comprised a diverse mix of urbanisation, socioeconomic status, geographies, and injury profiles. Consecutive patients presenting to ambulance services, including helicopter emergency medical services (HEMS), were included. The results of this study should therefore be generalisable throughout the UK National Health Service. The exploration of spectrum effects across different adult ages and injury acuities, should also provide an indication of performance in alternative subgroups. However, external validity to other settings is less certain. Different injury patterns (e.g., higher numbers of gun-shot wounds), alternative health system models (state Vs insurance Vs private funding), and patient demographics (e.g., obesity rates) in other settings could all influence theoretical triage tool performance. Difference in physiology, injury patterns, prevalence, costs, and consequences of paediatric major trauma also prevents extrapolation of results to children. As noted above, real-life triage may differ, and could be affected by differences in prehospital provider training, medicolegal risk, and trauma system organisation (e.g., inclusive versus exclusive design).
Comparison to literature
Several systematic reviews have examined the theoretical diagnostic accuracy of major trauma triage tools [31,33–35]. These have reported very heterogenous results at high risk of bias, often due to skewed trauma registry study populations, making comparison to the current study difficult. The largest, and most robust, studies to date were performed by Newgard (2016) and Voskens (2017). Newgard et al. evaluated the ACS-COT field triage decision scheme in a sample of 53,487 patients against an ISS ≥ 16 reference standard. Their reported sensitivity of 66.2% and specificity of 87.8% is very similar to those observed herein (ISS ≥ 16 reference standard, sensitivity 70%, specificity 87%) [10]. Voskens and colleagues studied 4,950 injured adults judged by ambulance providers to be high priority (conveyed with flashing lights and siren) [7]. They evaluated the Dutch field triage protocol, demonstrating a sensitivity of 36.2% and specificity of 92.6%. After consideration of spectrum effects from a higher prevalence of major trauma (8.8% vs 3.1%) and younger population (median age 45 Vs 75), the results are consistent with those in from the current study (ISS ≥ 16 reference standard, sensitivity 26%, specificity 93%). Van Rein and colleagues (2019) developed a diagnostic prediction model for major trauma using the same dataset, however external validation is beyond the scope of this paper and requires further study [36].
Limitations
This study has several strengths. The ‘single-gate’ (‘fishing for all cases in the same pond’) case-cohort design avoids the selection bias inherent in ‘two-gate’ case-control study designs where there is separate selection into the study sample based on reference standard status [37]. Recommendations for data collection in retrospective studies using routine data were followed to minimise information bias [38,39]. Other common sources of bias in diagnostic accuracy studies, including test, diagnostic review, partial verification, incomplete verification, incorporation, and disease progression biases, were avoided [40].
However, there are some potential limitations. The use of routine data may have resulted in index test or reference standard misclassification. Furthermore, reference standard classification is dependent on encompassing TARN inclusion criteria, complete case ascertainment by TARN, and accurate matching of TARN and prehospital data. A detailed critical appraisal of potential biases, classified according to QUADAS-2 criteria, is provided in the supplementary materials (S6 Table) [40].
Conclusions
Four optimal triage tools, demonstrating a trade-off between sensitivity and specificity, were identified. The choice of ideal tool will depend on the prevalence of major trauma, and relative valuation of false positive and false negative cases; however, previous health economic evaluations would suggest that the MATTS specific tool would be most cost-effective. Further prospective investigation of real-life triage tool performance, including compliance and clinical judgment, is necessary.
Supporting information
S1 File. English NHS Organisation of Major Trauma Services.
https://doi.org/10.1371/journal.pone.0344996.s001
(DOCX)
S2 Table. Expert consensus MATTS specific, sensitive and balanced triage tools.
https://doi.org/10.1371/journal.pone.0344996.s002
(DOCX)
S3 Table. Diagnostic accuracy metrics for selected adult triage tools evaluated against secondary reference standards in patients aged over 16 years.
https://doi.org/10.1371/journal.pone.0344996.s003
(DOCX)
S4 Table. Sensitivity analyses for diagnostic accuracy metrics of selected triage tools evaluated against the primary MATTS reference standard in patients aged over 16 years.
https://doi.org/10.1371/journal.pone.0344996.s004
(DOCX)
S5 Table. Diagnostic accuracy metrics for selected triage tools evaluated against the primary MATTS reference standard in different subgroups.
https://doi.org/10.1371/journal.pone.0344996.s005
(DOCX)
S6 Table. QUADAS-2 Internal and external validity assessment.
https://doi.org/10.1371/journal.pone.0344996.s006
(DOCX)
Acknowledgments
We are grateful to the participating trauma networks, hospital, and Trauma Audit and Research Network staff for their support, help and work. This research was supported by the Major Trauma Triage Study (MATTS) study management group, including: Ian Maconochie, Imperial College Healthcare NHS Trust; Mathew Ward, West Midlands Ambulance Service NHS Foundation Trust; Mark Millins, Yorkshire Ambulance Service NHS Trust; Emily Turton, University of Sheffield; Simon Waterhouse, University of Sheffield; Matt Stevenson, University of Sheffield; Daniel Pollard, University of Sheffield; Abdullah Pandor, University of Sheffiel; Maria Robinson, South West Ambulance NHS Foundation Trsut; Di Charles. Sheffield Emergency Care Forum; The lead author for the MATTS study management group is Dr Gordon Fuller, University of Sheffield.
References
- 1.
Office NA. NHS Ambulance Services. 2017.
- 2. Coats TJ, Lecky F. ‘Major trauma’: now two separate diseases? Emerg Med J. 2017;34(8):494.
- 3. McCullough AL, Haycock JC, Forward DP, Moran CG. II. Major trauma networks in England. Br J Anaesth. 2014;113(2):202–6. pmid:25038152
- 4. Moran CG, Lecky F, Bouamra O, Lawrence T, Edwards A, Woodford M, et al. Changing the system - major trauma patients and their outcomes in the NHS (England) 2008-17. EClinicalMedicine. 2018;2–3:13–21.
- 5. Mackenzie EJ, Rivara FP, Jurkovich GJ, Nathens AB, Frey KP, Egleston BL, et al. The National Study on Costs and Outcomes of Trauma. J Trauma. 2007;63(6 Suppl):S54–67.
- 6. Bhaumik S, Hannun M, Dymond C, DeSanto K, Barrett W, Wallis LA, et al. Prehospital triage tools across the world: a scoping review of the published literature. Scand J Trauma Resusc Emerg Med. 2022;30(1):32. pmid:35477474
- 7. Voskens FJ, van Rein EAJ, van der Sluijs R, Houwert RM, Lichtveld RA, Verleisdonk EJ, et al. Accuracy of Prehospital Triage in Selecting Severely Injured Trauma Patients. JAMA Surg. 2018;153(4):322–7. pmid:29094144
- 8. Newgard CD, Fischer PE, Gestring M, Michaels HN, Jurkovich GJ, Lerner EB, et al. National guideline for the field triage of injured patients: Recommendations of the National Expert Panel on Field Triage, 2021. Journal of Trauma and Acute Care Surgery. 2022;93(2):e49–60.
- 9. Maughan BC, Lin A, Caughey AB, Bulger EM, McConnell KJ, Malveau S, et al. Field Trauma Triage among Older Adults: A Cost-Effectiveness Analysis. J Am Coll Surg. 2022;234(2):139–54. pmid:35213435
- 10. Newgard CD, Yang Z, Nishijima D, McConnell KJ, Trent SA, Holmes JF, et al. Cost-Effectiveness of Field Trauma Triage among Injured Adults Served by Emergency Medical Services. J Am Coll Surg. 2016;222(6):1125–37. pmid:27178369
- 11. Nishijima DK, Yang Z, Newgard CD. Cost-effectiveness of field trauma triage among injured children transported by emergency medical services. Am J Emerg Med. 2021;50:492–500. pmid:34536721
- 12. Pollard D, Fuller G, Goodacre S, van Rein EAJ, Waalwijk JF, van Heijl M. An economic evaluation of triage tools for patients with suspected severe injuries in England. BMC Emerg Med. 2022;22(1):4. pmid:35016621
- 13.
GW F. ISRCTN17968752 - MATTS: Major Trauma Triage Tool Study. ISRCTN Registry. 2018.
- 14. O’Brien KM, Lawrence KG, Keil AP. The Case for Case-Cohort: An Applied Epidemiologist’s Guide to Reframing Case-Cohort Studies to Improve Usability and Flexibility. Epidemiology. 2022;33(3):354–61. pmid:35383643
- 15. Holtman GA, Berger MY, Burger H, Deeks JJ, Donner-Banzhoff N, Fanshawe TR, et al. Development of practical recommendations for diagnostic accuracy studies in low-prevalence situations. J Clin Epidemiol. 2019;114:38–48. pmid:31150837
- 16. Fuller G, Keating S, Turner J, Miller J, Holt C, Smith JE, et al. Injured patients who would benefit from expedited major trauma centre care: a consensus-based definition for the United Kingdom. Br Paramed J. 2021;6(3):7–14. pmid:34970078
- 17. Baker SP, O’Neill B, Haddon W Jr, Long WB. The injury severity score: a method for describing patients with multiple injuries and evaluating emergency care. J Trauma. 1974;14(3):187–96. pmid:4814394
- 18.
GW F. Major trauma triage tool study (MATTS) expert consensus derived injury assessment tool.
- 19. Edwards A. Top 10 TARN research publications. Emerg Med J. 2015;32(12):966–8. pmid:26598632
- 20. Network TA. Trauma Audit and Research Network. http://www.tarn.ac.uk. 2023.
- 21. Fuller G, Hasler RM, Mealing N, Lawrence T, Woodford M, Juni P, et al. The association between admission systolic blood pressure and mortality in significant traumatic brain injury: a multi-centre cohort study. Injury. 2014;45(3):612–7. pmid:24206920
- 22.
Gennarelli TAWE. Abbreviated Injury Scale 2005. Barrington, IL: Association for the Advancement of Automotive Medicine. 2005.
- 23.
Miettinen K. Nonlinear multiobjective optimization. Boston: Kluwer Academic Publishers. 1999.
- 24. Obuchowski NA, Bullen JA. Receiver operating characteristic (ROC) curves: review of methods with applications in diagnostic medicine. Phys Med Biol. 2018;63(7):07TR01. pmid:29512515
- 25. Lasko TA, Bhagwat JG, Zou KH, Ohno-Machado L. The use of receiver operating characteristic curves in biomedical informatics. J Biomed Inform. 2005;38(5):404–15. pmid:16198999
- 26. Vickers AJ, van Calster B, Steyerberg EW. A simple, step-by-step guide to interpreting decision curve analysis. Diagn Progn Res. 2019;3:18. pmid:31592444
- 27. Mallett S, Halligan S, Thompson M, Collins GS, Altman DG. Interpreting diagnostic accuracy studies for patient care. BMJ. 2012;345:e3999. pmid:22750423
- 28. Rubin D. Multiple imputation for nonresponse in surveys. Computers, Environment and Urban Systems. 1990;14(1):75.
- 29. Borgan Ø, Samuelsen SO. A review of cohort sampling designs for Cox’s regression model: Potentials in epidemiology. Nor J Epidemiol. 2009;13(2).
- 30. Akobeng AK. Understanding diagnostic tests 1: sensitivity, specificity and predictive values. Acta Paediatr. 2007;96(3):338–41. pmid:17407452
- 31. van Rein EAJ, van der Sluijs R, Raaijmaakers AMR, Leenen LPH, van Heijl M. Compliance to prehospital trauma triage protocols worldwide: A systematic review. Injury. 2018;49(8):1373–80. pmid:30135040
- 32. Newgard CD, Kampp M, Nelson M, Holmes JF, Zive D, Rea T, et al. Deciphering the use and predictive value of “emergency medical services provider judgment” in out-of-hospital trauma triage: a multisite, mixed methods assessment. J Trauma Acute Care Surg. 2012;72(5):1239–48. pmid:22673250
- 33. Gianola S, Castellini G, Biffi A, Porcu G, Fabbri A, Ruggieri MP, et al. Accuracy of pre-hospital triage tools for major trauma: a systematic review with meta-analysis and net clinical benefit. World J Emerg Surg. 2021;16(1):31. pmid:34112209
- 34. Tohira H, Jacobs I, Mountain D, Gibson N, Yeo A. Systematic review of predictive performance of injury severity scoring tools. Scand J Trauma Resusc Emerg Med. 2012;20:63. pmid:22964071
- 35. van Rein EAJ, Houwert RM, Gunning AC, Lichtveld RA, Leenen LPH, van Heijl M. Accuracy of prehospital triage protocols in selecting severely injured patients: A systematic review. J Trauma Acute Care Surg. 2017;83(2):328–39. pmid:28452898
- 36. van Rein EAJ, van der Sluijs R, Voskens FJ, Lansink KWW, Houwert RM, Lichtveld RA, et al. Development and Validation of a Prediction Model for Prehospital Triage of Trauma Patients. JAMA Surg. 2019;154(5):421–9. pmid:30725101
- 37. Rutjes AWS, Reitsma JB, Vandenbroucke JP, Glas AS, Bossuyt PMM. Case-control and two-gate designs in diagnostic accuracy studies. Clin Chem. 2005;51(8):1335–41. pmid:15961549
- 38. Kaji AH, Schriger D, Green S. Looking through the retrospectoscope: reducing bias in emergency medicine chart review studies. Ann Emerg Med. 2014;64(3):292–8. pmid:24746846
- 39. Vassar M, Holzmann M. The retrospective chart review: important methodological considerations. J Educ Eval Health Prof. 2013;10:12. pmid:24324853
- 40. Whiting PF, Rutjes AWS, Westwood ME, Mallett S, Deeks JJ, Reitsma JB, et al. QUADAS-2: a revised tool for the quality assessment of diagnostic accuracy studies. Ann Intern Med. 2011;155(8):529–36. pmid:22007046